Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector - Intelligent and Sustainable Manufacturing - Full-Text HTML

<!DOCTYPE html> <html class="no-js" lang="en"> <head> <meta charset="utf-8" /> <meta http-equiv="x-ua-compatible" content="ie=edge" /> <title>Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector - Intelligent and Sustainable Manufacturing - Full-Text HTML - SCIEPublish</title> <meta name="keywords" content="SCIE Publishing, SCIEPublish, open access, fast publication" /> <meta name="description" content="SCIEPublish is an international open-access journal publishing service provider run by SCIE Publishing Limited, dedicated to supporting and inspiring scientists or institutions/societies who aspire to publish high-quality journals." /> <meta name="viewport" content="width=device-width,user-scalable=no,initial-scale=1.0,maximum-scale=1.0,minimum-scale=1.0"> <meta name="title" content="Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector"> <link rel="image_src" href="https://www.sciepublish.com/uploads/2024/05/28/1716875145yduw.jpg"> <meta name="dc.title" content="Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector"> <meta name="dc.creator" content="Zhenjing Duan"> <meta name="dc.creator" content=" Shushu Xi"> <meta name="dc.creator" content="Shuaishuai Wang"> <meta name="dc.creator" content="Ziheng Wang"> <meta name="dc.creator" content="Peng Bian"> <meta name="dc.creator" content="Changhe Li"> <meta name="dc.creator" content="Jinlong Song"> <meta name="dc.creator" content="Xin Liu"> <meta name="dc.type" content="Article"> <meta name="dc.source" content="Intelligent and Sustainable Manufacturing 2024, Vol. 1, Page 10020"> <meta name="dc.date" content="2024-11-26"> <meta name="dc.identifier" content="10.70322/ism.2024.10020"> <meta name="dc.publisher" content="SCIEPublish"> <meta name="dc.rights" content="http://creativecommons.org/licenses/by/4.0/"> <meta name="dc.format" content="application/pdf"> <meta name="dc.language" content="en"> <meta name="dc.description" content="In the mechanical cutting process, the surface defects of the workpiece are an important indicator of cutting quality and also reflect the condition of both the machine tool and the cutting tool. Effective detection of defects on the surface of the workpiece plays an important role in adjusting the processing conditions promptly, reducing losses, improving the utilization rate of the workpiece, and maintaining the normal operation of the equipment. To address the challenge of detecting surface defects on workpieces, an inspection method based on an improved Single Shot Multibox Detector (SSD) model is proposed. The method simplifies the detection model and reduces the computation by proposing a DH-MobileNet network instead of a VGG16 network in the SSD structure. The inverse residual structure is also used for position prediction, and null convolution is used instead of a down-sampling operation to avoid information loss. A scanning electron microscope was used to obtain the surface image of the workpiece. A dataset of workpiece surface defects was constructed and expanded, then used to train and test the model for detecting three common types of high-frequency defects: peel-off, chip adhesion, and scratches. The effect was compared with YOLO, Faster R-CNN, and the original SSD model. The detection results show that the method can detect the defects on the surface of the workpiece more accurately and quickly, which provides a new idea for defect detection in real industrial scenarios."> <meta name="dc.subject" content="Cutting machining"> <meta name="dc.subject" content="Workpiece surface defects"> <meta name="dc.subject" content="SSD"> <meta name="dc.subject" content="DH-MobileNet"> <meta name="prism.issn" content="3005-8066"> <meta name="prism.publicationName" content="Intelligent and Sustainable Manufacturing"> <meta name="prism.publicationDate" content="2024-11-26"> <meta name="prism.volume" content="1"> <meta name="prism.number" content="2"> <meta name="prism.section" content="Article"> <meta name="prism.startingPage" content="10020"> <meta name="citation_issn" content="3005-8066"> <meta name="citation_journal_title" content="Intelligent and Sustainable Manufacturing"> <meta name="citation_publisher" content="SCIEPublish"> <meta name="citation_title" content="Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector"> <meta name="citation_publication_date" content="2024/11"> <meta name="citation_online_date" content="2024/11/26"> <meta name="citation_volume" content="1"> <meta name="citation_issue" content="2"> <meta name="citation_firstpage" content="10020"> <meta name="citation_author" content="Duan, Zhenjing"> <meta name="citation_author" content="Xi, Shushu"> <meta name="citation_author" content="Wang, Shuaishuai"> <meta name="citation_author" content="Wang, Ziheng"> <meta name="citation_author" content="Bian, Peng"> <meta name="citation_author" content="Li, Changhe"> <meta name="citation_author" content="Song, Jinlong"> <meta name="citation_author" content="Liu, Xin"> <meta name="citation_doi" content="10.70322/ism.2024.10020"> <meta name="citation_abstract_html_url" content="https://www.sciepublish.com/article/pii/349"> <meta name="citation_pdf_url" content="https://www.sciepublish.com/index/article/view_full_text/id/349"> <link rel="alternate" type="application/pdf" title="PDF Full-Text" href="https://www.sciepublish.com/index/article/view_full_text/id/349"> <meta name="fulltext_pdf" content="https://www.sciepublish.com/index/article/view_full_text/id/349"> <link rel="alternate" type="text/html" title="HTML Full-Text" href="https://www.sciepublish.com/article/pii/349"> <meta name="citation_fulltext_html_url" content="https://www.sciepublish.com/article/pii/349"> <meta name="fulltext_html" content="https://www.sciepublish.com/article/pii/349"> <meta property="og:site_name" content="SCIEPUBLISH" /> <meta property="og:type" content="Article" /> <meta property="og:url" content="https://www.sciepublish.com/article/pii/349" /> <meta property="og:title" content="Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector" /> <meta property="og:description" content="In the mechanical cutting process, the surface defects of the workpiece are an important indicator of cutting quality and also reflect the condition of both the machine tool and the cutting tool. Effective detection of defects on the surface of the workpiece plays an important role in adjusting the processing conditions promptly, reducing losses, improving the utilization rate of the workpiece, and maintaining the normal operation of the equipment. To address the challenge of detecting surface defects on workpieces, an inspection method based on an improved Single Shot Multibox Detector (SSD) model is proposed. The method simplifies the detection model and reduces the computation by proposing a DH-MobileNet network instead of a VGG16 network in the SSD structure. The inverse residual structure is also used for position prediction, and null convolution is used instead of a down-sampling operation to avoid information loss. A scanning electron microscope was used to obtain the surface image of the workpiece. A dataset of workpiece surface defects was constructed and expanded, then used to train and test the model for detecting three common types of high-frequency defects: peel-off, chip adhesion, and scratches. The effect was compared with YOLO, Faster R-CNN, and the original SSD model. The detection results show that the method can detect the defects on the surface of the workpiece more accurately and quickly, which provides a new idea for defect detection in real industrial scenarios." /> <meta property="og:image" content="https://www.sciepublish.com/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png" /> <meta property="og:image:secure_url" content="https://www.sciepublish.com/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png" /> <meta name="twitter:site" content="@sciepublish"> <meta name="twitter:image" content="https://www.sciepublish.com/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png" /> <meta name="twitter:card" content="summary_large_image"> <link rel="shortcut icon" type="image/x-icon" href="/favicon.ico" /> <link rel="stylesheet" href="/static/css/bootstrap.css" /> <link rel="stylesheet" href="/static/font/bootstrap-icons.css"> <link rel="stylesheet" href="/static/css/lightbox.css" /> <link rel="stylesheet" href="/static/css/theme.css?74" /> <link rel="stylesheet" href="/static/css/common.css?90" /> <link rel="stylesheet" href="/static/css/ckeditorStyle.css?2" /> <link rel="stylesheet" href="/static/css/select-mania.min.css" /> <script src="/static/js/jquery-3.7.1.min.js"></script> <script src="/static/js/lightbox.js"></script> <script src="/static/js/bootstrap.bundle.min.js"></script> <script src="/static/js/select-mania.js"></script> <link rel="stylesheet" href="/static/katex/katex.min.css"> <script src="/static/katex/katex.min.js"></script> <script src="/static/katex/contrib/auto-render.min.js"></script> <style> .katex{ font-size: 1.5rem!important; } .tooltip-inner{ /* min-width: 40rem; */ width: auto; max-width: 40rem; padding: 1rem; text-align: left; } .tooltip-inner .html-fig_img, .tooltip-inner .html-figpopup-table { padding: 1.5rem 1.5rem 1rem; background-color: #FFFeFa; border: 1px solid #e9ded8; border-radius: 5px; margin: 1.5rem 0 2rem; } .tooltip-inner .html-figpopup-table{margin-top:1rem;} .tooltip-inner .html-fig_img img, .tooltip-inner .html-figpopup-table img{ width: 100%; } .centered-div { max-width: 60%; position: fixed; top: 50%; left: 50%; transform: translate(-50%, -50%); z-index: 1000; background-color: rgba(255, 255, 255, 1); padding: 20px; border: 1px solid #ccc; box-shadow: 0 2px 5px rgba(0,0,0,0.2); } .topic-collection a{ display: block; font-size: .85rem; line-height: 1.5; } topic-collection h6.article-menu-collapse{ font-weight: 700; } .popover-content { width: 500px; } .popover { max-width:95%; } </style> </head> <body class="header-no-fixed"> <header> <nav class="navbar navbar-expand-lg navbar-light my-body-container"> <div class="navbar-left"> <a href="/" class="uk-navbar-item" alt="Back to the homepage"> <svg xmlns="http://www.w3.org/2000/svg" class="navbar-logo" xml:space="preserve" version="1.0" viewBox="0 0 5.08 1.933"> <path d="M1.021 1.245a.29.29 0 0 1-.211-.054l-.027-.023-.003-.003.056-.066.003.004a.3.3 0 0 0 .043.033.2.2 0 0 0 .128.027l.024-.007.019-.01a.07.07 0 0 0 .022-.032.1.1 0 0 0 0-.036l-.004-.014a.1.1 0 0 0-.016-.02.1.1 0 0 0-.027-.017L.994 1.01.919.98a.3.3 0 0 1-.076-.05.14.14 0 0 1-.034-.067.2.2 0 0 1 0-.06.13.13 0 0 1 .027-.056.2.2 0 0 1 .049-.041A.2.2 0 0 1 .95.683a.3.3 0 0 1 .07-.001.2.2 0 0 1 .06.017.3.3 0 0 1 .075.05l.003.003-.05.061L1.103.81a.2.2 0 0 0-.053-.034.2.2 0 0 0-.063-.013.1.1 0 0 0-.046.008L.925.78a.06.06 0 0 0-.023.047l.001.015.006.012a.1.1 0 0 0 .018.02.1.1 0 0 0 .027.017L.97.899l.016.006.074.032a.3.3 0 0 1 .058.033.14.14 0 0 1 .051.08.2.2 0 0 1 0 .07.15.15 0 0 1-.048.081.2.2 0 0 1-.062.035zm.527-.002a.24.24 0 0 1-.1 0 .23.23 0 0 1-.125-.069.2.2 0 0 1-.051-.089.3.3 0 0 1-.019-.119.4.4 0 0 1 .02-.12.3.3 0 0 1 .052-.09.23.23 0 0 1 .176-.076.3.3 0 0 1 .064.01.3.3 0 0 1 .053.025.2.2 0 0 1 .04.034l.002.003-.052.061L1.605.81A.2.2 0 0 0 1.56.776a.2.2 0 0 0-.037-.012.15.15 0 0 0-.082.013.13.13 0 0 0-.049.039l-.018.029a.2.2 0 0 0-.022.073.4.4 0 0 0 .002.104.2.2 0 0 0 .014.05.2.2 0 0 0 .023.04.14.14 0 0 0 .066.047.15.15 0 0 0 .107-.009l.028-.017.025-.023.003-.004.051.06-.002.003a.3.3 0 0 1-.076.059.2.2 0 0 1-.045.015m.314-.004h-.09V.69h.095v.549zm.485 0h-.331V.69h.328v.08H2.11v.141h.198v.082H2.11v.165h.242v.08zm.215 0h-.09V.69h.169a.4.4 0 0 1 .083.008L2.76.71l.031.016a.13.13 0 0 1 .044.053q.009.015.012.036a.22.22 0 0 1-.012.121l-.018.03a.176.176 0 0 1-.09.057.3.3 0 0 1-.084.011h-.076v.205zm.005-.472v.19h.068a.2.2 0 0 0 .056-.006.1.1 0 0 0 .038-.018.1.1 0 0 0 .022-.031.1.1 0 0 0 .007-.045l-.003-.03a.07.07 0 0 0-.028-.04L2.703.776 2.671.769 2.633.767zm.539.48a.2.2 0 0 1-.067-.003.1.1 0 0 1-.075-.07.3.3 0 0 1-.013-.09V.827h.093v.248a.3.3 0 0 0 .007.055l.008.017.012.012.016.007.02.002a.1.1 0 0 0 .033-.006.1.1 0 0 0 .03-.019.2.2 0 0 0 .03-.032V.826h.093v.413h-.078l-.006-.057a.2.2 0 0 1-.078.057zm.548-.002-.034.003-.03-.003-.029-.01A.2.2 0 0 1 3.51 1.2l-.007.039h-.075V.645h.093q0 .11-.002.219l.01-.008A.2.2 0 0 1 3.59.823a.2.2 0 0 1 .043-.007.2.2 0 0 1 .07.015.15.15 0 0 1 .07.074.2.2 0 0 1 .022.076.4.4 0 0 1 0 .095.3.3 0 0 1-.029.082.2.2 0 0 1-.079.075zm-.07-.077a.1.1 0 0 0 .046-.002.1.1 0 0 0 .043-.032l.015-.028a.2.2 0 0 0 .01-.036.3.3 0 0 0-.006-.114A.1.1 0 0 0 3.68.93.07.07 0 0 0 3.64.9.1.1 0 0 0 3.59.899l-.023.008-.023.015-.023.02v.193a.2.2 0 0 0 .043.027zm.424.08h-.015a.1.1 0 0 1-.051-.013l-.017-.016-.011-.022-.007-.026-.002-.031V.645h.093v.5L4 1.16l.003.004.003.003.008.002h.008l.005-.001h.004l.013.071-.004.002-.009.002zm.22-.01H4.14V.827h.093v.413zm-.026-.48L4.187.76q-.009 0-.016-.002L4.157.753a.05.05 0 0 1-.02-.02.1.1 0 0 1-.008-.027L4.13.69a.05.05 0 0 1 .027-.032L4.17.653a.07.07 0 0 1 .045.005.05.05 0 0 1 .03.048.1.1 0 0 1-.009.028.1.1 0 0 1-.02.019zm.315.488a.25.25 0 0 1-.19-.054l-.004-.003.045-.062.004.003a.3.3 0 0 0 .053.034.13.13 0 0 0 .058.012l.022-.002.016-.005.013-.008.009-.01a.05.05 0 0 0 .007-.025q0-.006-.002-.012l-.005-.01-.008-.009-.011-.008-.028-.014-.016-.006-.017-.007a.4.4 0 0 1-.079-.041.1.1 0 0 1-.028-.034L4.348.964 4.345.939a.12.12 0 0 1 .023-.07.1.1 0 0 1 .038-.033A.2.2 0 0 1 4.46.82a.2.2 0 0 1 .13.022l.037.024.003.003-.045.059-.003-.003A.2.2 0 0 0 4.54.9a.1.1 0 0 0-.065-.008.1.1 0 0 0-.027.012l-.007.01a.04.04 0 0 0-.007.022q0 .006.002.01l.005.01a.1.1 0 0 0 .017.015l.027.013.016.006.016.006.063.028.019.013a.1.1 0 0 1 .029.035l.008.023a.13.13 0 0 1-.008.077.1.1 0 0 1-.03.04.14.14 0 0 1-.05.027zm.307-.007h-.088V.645h.092q0 .115-.002.229a.3.3 0 0 1 .05-.038.2.2 0 0 1 .048-.017.2.2 0 0 1 .068.002.1.1 0 0 1 .057.038q.01.014.018.033A.314.314 0 0 1 5.08.98v.258h-.093V.99L4.985.96 4.98.936 4.97.92 4.96.907 4.943.9a.1.1 0 0 0-.037 0 .1.1 0 0 0-.03.011l-.016.01-.032.03v.288z" /> <path d="M1.844 1.377a.97.97 0 0 1-.878.563A.963.963 0 0 1 0 .974.964.964 0 0 1 .966.007a.96.96 0 0 1 .865.536l-.048.024A.92.92 0 0 0 .966.062a.91.91 0 0 0-.912.912.91.91 0 0 0 .912.912.91.91 0 0 0 .83-.532z" class="logo-circle" /> </svg> </a> </div> <button class="navbar-toggler" type="button" data-toggle="collapse" data-target="#navbarSupportedContent" aria-controls="navbarSupportedContent" aria-expanded="false" aria-label="Toggle navigation"> <div class="btn-menu"> <span></span> <span></span> <span></span> </div> </button> <div class="collapse navbar-collapse" id="navbarSupportedContent"> <div class="nav-form-search"> <form class="form-search" method="get" action="/index/search/index.html"> <input class="search-input" type="text" name="search" placeholder="What are you looking for?" autofocus /> <button type="submit">Search</button> </form> </div> <ul class="navbar-nav uk-navbar-nav"> <li class="nav-item "> <a class="nav-link" href="/">Home</a> </li> <li class="nav-item "> <a class="nav-link" href="/About_SCIEPublish">About</a> </li> <li class="nav-item uk-active"> <a class="nav-link" href="/index/journals/index">Journals</a> </li> <li class="nav-item "> <a class="nav-link" href="/news/list">News</a> </li> </ul> <a class="js-navbar-toggle sc-icon-button sc-transition" title="Search" href="#"> <svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="currentColor" class="bi bi-search" viewBox="0 0 16 16"> <path d="M11.742 10.344a6.5 6.5 0 1 0-1.397 1.398h-.001c.03.04.062.078.098.115l3.85 3.85a1 1 0 0 0 1.415-1.414l-3.85-3.85a1.007 1.007 0 0 0-.115-.1zM12 6.5a5.5 5.5 0 1 1-11 0 5.5 5.5 0 0 1 11 0z"/> </svg> <svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="currentColor" class="bi bi-x-lg" viewBox="0 0 16 16"> <path d="M2.146 2.854a.5.5 0 1 1 .708-.708L8 7.293l5.146-5.147a.5.5 0 0 1 .708.708L8.707 8l5.147 5.146a.5.5 0 0 1-.708.708L8 8.707l-5.146 5.147a.5.5 0 0 1-.708-.708L7.293 8 2.146 2.854Z"/> </svg> </a> <a class="uk-button sc-transition" href="/my/submitting">Publish with us</a> <a class="text-login" title="Sign in" href="/index/user/login.html"> Sign in </a> </div> </nav> </header> <section class="breadcrumbs pt-3"> <div class="my-body-container"> <nav aria-label="breadcrumb"> <ol class="breadcrumb p-0 mb-0"> <li class="breadcrumb-item" title="Home"> <a href="/">Home</a> <svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="currentColor" class="bi bi-chevron-right ml-1 mr-1" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z"/> </svg> </li> <li class="breadcrumb-item" title="Journals"> <a href="/index/journals/index">Journals</a> <svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="currentColor" class="bi bi-chevron-right ml-1 mr-1" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z"/> </svg> </li> <li class="breadcrumb-item" title="ism"> <a href="/journals/ism">ism</a> <svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="currentColor" class="bi bi-chevron-right ml-1 mr-1" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z"/> </svg> </li> <li class="breadcrumb-item" title="Volume 1"> <a href="/journals/ism/roll/1">Volume 1</a> <svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="currentColor" class="bi bi-chevron-right ml-1 mr-1" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z"/> </svg> </li> <li class="breadcrumb-item" title="Issue 2"> <a href="/journals/ism/roll/1/2">Issue 2</a> <svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="currentColor" class="bi bi-chevron-right ml-1 mr-1" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M4.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L10.293 8 4.646 2.354a.5.5 0 0 1 0-.708z"/> </svg> </li> <li class="breadcrumb-item active" aria-current="page" title='10.70322/ism.2024.10020'>10.70322/ism.2024.10020</li> </ol> </nav> </div> </section> <section class="fixed-title"> <div class="my-body-container d-flex align-items-center border-bottom border-dark pt-2 pb-2"> <div class="item-img"> <a class="d-flex align-items-center" href="/journals/ism"> <div class="cover-img"> <img src="/uploads/2024/05/28/1716875145yduw.jpg" alt="Intelligent and Sustainable Manufacturing-logo" /> </div> <h6 class="flex-grow-1 pl-2" title="Intelligent and Sustainable Manufacturing">Intelligent and Sustainable Manufacturing</h6> </a> </div> <h2 class="journals-text" title="Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector">Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector</h2> </div> </section> <section class="page-items mt-3"> <div class="d-flex my-body-container border-dark border-top pt-4 collapse show" id="rightCollapse"> <div class="left-items-260 mr-3"> <div class="left-content d-flex"> <div class="left-content-item"> <div class="item-img"> <a class="d-flex align-items-center" href="/journals/ism"> <div class="cover-img"> <img src="/uploads/2024/05/28/1716875145yduw.jpg" alt="Intelligent and Sustainable Manufacturing-logo" /> </div> <h6 class="flex-grow-1 pl-2">Intelligent and Sustainable Manufacturing</h6> </a> </div> <div class="btn-bottom"> <a class="sc-btn-submit d-flex align-items-center justify-content-center mt-3" href="/my/submitting/journal/52"> Submit to <svg xmlns="http://www.w3.org/2000/svg" width="12" height="12" fill="currentColor" class="bi bi-chevron-double-right ml-1" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M3.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L9.293 8 3.646 2.354a.5.5 0 0 1 0-.708z"></path> <path fill-rule="evenodd" d="M7.646 1.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1 0 .708l-6 6a.5.5 0 0 1-.708-.708L13.293 8 7.646 2.354a.5.5 0 0 1 0-.708z"></path> </svg> </a> <a class="sc-cite-modal sc-btn-submit d-flex align-items-center justify-content-center mt-2" href="/index/article/download_article/id/349.html"> <span>Download PDF</span> </a>  <div class="sc-cite-modal sc-btn-submit d-flex align-items-center justify-content-center border-0 mt-2" data-toggle="modal" data-target="#citeModal"> Cite This Article </div> </div> <div class="mt-2 topic-collection"> <h6 class="article-menu-collapse mb-1 pb-2 pt-2">Part of Topic Collection</h6> <b><a class="pt-2" href="/journals/ism/special_issues/sus_mach_green">Sustainable Machining and Green Products</a></b> </div>  <div class="article-menu-collapse d-flex justify-content-between mt-2"> <h6 class="mb-0 d-flex align-items-center">Contents</h6> <div class="btn-collapse pt-2 pb-2 pl-2" data-toggle="collapse" data-target="#articleMenuCollapse" aria-expanded="true" aria-controls="articleMenuCollapse"> <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-chevron-up" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M7.646 4.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1-.708.708L8 5.707l-5.646 5.647a.5.5 0 0 1-.708-.708l6-6z"/> </svg> </div> </div> <div class="articleMenuCollapse-column p-0" id="articleMenuCollapse"> <div class="menu pt-2"> <div class="list-group" id="article-menu-float"> <a class="list-group-item list-group-item-action" href="#link-introduction">1. Introduction</a> <a class="list-group-item list-group-item-action" href="#link-materials-and-methods">2. SEM Sample Data</a> <a class="list-group-item list-group-item-action" href="#link-theory-calculation">3. SSD Target Detection Model</a> <a class="list-group-item list-group-item-action" href="#link-results">4. Design and Optimization of Experimental Models</a> <a class="list-group-item list-group-item-action" href="#link-discussion">5. Experimental Design</a> <a class="list-group-item list-group-item-action" href="#link-conclusions">6. Results and Discussion</a> <a class="list-group-item list-group-item-action" href="#link-supplementary-materials">7. Conclusions</a> <a class="list-group-item list-group-item-action" href="#link-author-contributions">Author Contributions</a> <a class="list-group-item list-group-item-action" href="#link-ethics-statement">Ethics Statement</a> <a class="list-group-item list-group-item-action" href="#link-informed-consent-statement">Informed Consent Statement</a> <a class="list-group-item list-group-item-action" href="#link-funding">Funding</a> <a class="list-group-item list-group-item-action" href="#link-declaration-competing-interest">Declaration of Competing Interest</a> <a class="list-group-item list-group-item-action" href="#html-references_list">References</a> </div> </div> </div> </div> <span class="btn-out" data-toggle="collapse" data-target="#rightCollapse" aria-expanded="true" aria-controls="rightCollapse"> <svg xmlns="http://www.w3.org/2000/svg" width="26" height="26" fill="currentColor" class="bi bi-chevron-bar-left" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M11.854 3.646a.5.5 0 0 1 0 .708L8.207 8l3.647 3.646a.5.5 0 0 1-.708.708l-4-4a.5.5 0 0 1 0-.708l4-4a.5.5 0 0 1 .708 0zM4.5 1a.5.5 0 0 0-.5.5v13a.5.5 0 0 0 1 0v-13a.5.5 0 0 0-.5-.5z"/> </svg> </span> </div> </div> <div class="blog-detailss-wrap flex-grow-1"> <div class="details__content pb-30"> <div class="articlelei access-tags d-flex justify-content-between"> <p class="mb-0 d-flex align-items-center"> <span class="btn-text classdise mr-1">Article</span> <span class="btn-text textdise">Open Access</span> </p> <div class="share-buttons social-share d-flex align-items-center mb-2" data-initialized="true"> Share This Article： <a href="" class="share-btn social-share-icon twitter-btn icon-twitter d-flex align-items-center justify-content-center" title="Share on Twitter"> <svg t="1713929395475" class="icon" viewBox="0 0 1399 1024" fill="currentColor" version="1.1" xmlns="http://www.w3.org/2000/svg" p-id="8742" width="20" height="20"> <path d="M282.021569 119.281213l323.998199 433.216256-326.044203 352.222995h73.379441l285.451142-308.376452 230.636674 308.376452h249.713149l-342.227762-457.583832 303.479459-327.855419h-73.379441l-262.886398 284.008876-212.407111-284.008876h-249.713149z m107.909957 54.051408h114.71879l506.578928 677.328049h-114.718791l-506.578927-677.328049z" p-id="8743"></path> </svg> </a> <a href="" class="share-btn social-share-icon facebook-btn icon-facebook d-flex align-items-center justify-content-center" title="Share on Facebook"> <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-facebook" viewBox="0 0 16 16"> <path d="M16 8.049c0-4.446-3.582-8.05-8-8.05C3.58 0-.002 3.603-.002 8.05c0 4.017 2.926 7.347 6.75 7.951v-5.625h-2.03V8.05H6.75V6.275c0-2.017 1.195-3.131 3.022-3.131.876 0 1.791.157 1.791.157v1.98h-1.009c-.993 0-1.303.621-1.303 1.258v1.51h2.218l-.354 2.326H9.25V16c3.824-.604 6.75-3.934 6.75-7.951z"/> </svg> </a> <a href="mailto:?subject=Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector&body=https%3A%2F%2Fwww.sciepublish.com%2Farticle%2Fpii%2F349" class="share-btn social-share-icon email-btn icon-email d-flex align-items-center justify-content-center" id="email-share-btn" title="Share on Email"> <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-envelope-fill" viewBox="0 0 16 16"> <path d="M.05 3.555A2 2 0 0 1 2 2h12a2 2 0 0 1 1.95 1.555L8 8.414.05 3.555ZM0 4.697v7.104l5.803-3.558L0 4.697ZM6.761 8.83l-6.57 4.027A2 2 0 0 0 2 14h12a2 2 0 0 0 1.808-1.144l-6.57-4.027L8 9.586l-1.239-.757Zm3.436-.586L16 11.801V4.697l-5.803 3.546Z"/> </svg> </a> <a href="" class="share-btn social-share-icon LinkedIn-btn icon-linkedin d-flex align-items-center justify-content-center" id="email-LinkedIn-btn" title="Share on LinkedIn"> <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-linkedin" viewBox="0 0 16 16"> <path d="M0 1.146C0 .513.526 0 1.175 0h13.65C15.474 0 16 .513 16 1.146v13.708c0 .633-.526 1.146-1.175 1.146H1.175C.526 16 0 15.487 0 14.854V1.146zm4.943 12.248V6.169H2.542v7.225h2.401zm-1.2-8.212c.837 0 1.358-.554 1.358-1.248-.015-.709-.52-1.248-1.342-1.248-.822 0-1.359.54-1.359 1.248 0 .694.521 1.248 1.327 1.248h.016zm4.908 8.212V9.359c0-.216.016-.432.08-.586.173-.431.568-.878 1.232-.878.869 0 1.216.662 1.216 1.634v3.865h2.401V9.25c0-2.22-1.184-3.252-2.764-3.252-1.274 0-1.845.7-2.165 1.193v.025h-.016a5.54 5.54 0 0 1 .016-.025V6.169h-2.4c.03.678 0 7.225 0 7.225h2.4z"/> </svg> </a> <a href="" class="share-btn social-share-icon wechat-btn icon-wechat d-flex align-items-center justify-content-center" id="wechat-share-btn" title="Share on WeChat"> <svg xmlns="http://www.w3.org/2000/svg" width="16" height="16" fill="currentColor" class="bi bi-wechat" viewBox="0 0 16 16"> <path d="M11.176 14.429c-2.665 0-4.826-1.8-4.826-4.018 0-2.22 2.159-4.02 4.824-4.02S16 8.191 16 10.411c0 1.21-.65 2.301-1.666 3.036a.324.324 0 0 0-.12.366l.218.81a.616.616 0 0 1 .029.117.166.166 0 0 1-.162.162.177.177 0 0 1-.092-.03l-1.057-.61a.519.519 0 0 0-.256-.074.509.509 0 0 0-.142.021 5.668 5.668 0 0 1-1.576.22ZM9.064 9.542a.647.647 0 1 0 .557-1 .645.645 0 0 0-.646.647.615.615 0 0 0 .09.353Zm3.232.001a.646.646 0 1 0 .546-1 .645.645 0 0 0-.644.644.627.627 0 0 0 .098.356Z"/> <path d="M0 6.826c0 1.455.781 2.765 2.001 3.656a.385.385 0 0 1 .143.439l-.161.6-.1.373a.499.499 0 0 0-.032.14.192.192 0 0 0 .193.193c.039 0 .077-.01.111-.029l1.268-.733a.622.622 0 0 1 .308-.088c.058 0 .116.009.171.025a6.83 6.83 0 0 0 1.625.26 4.45 4.45 0 0 1-.177-1.251c0-2.936 2.785-5.02 5.824-5.02.05 0 .1 0 .15.002C10.587 3.429 8.392 2 5.796 2 2.596 2 0 4.16 0 6.826Zm4.632-1.555a.77.77 0 1 1-1.54 0 .77.77 0 0 1 1.54 0Zm3.875 0a.77.77 0 1 1-1.54 0 .77.77 0 0 1 1.54 0Z"/> </svg> </a> </div> </div> <h1 class="mb-2 acquiesce-title">Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector</h1> <div class="affiliation mb-1"> <div class="affiliation_author"> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Zhenjing Duan" target="_blank">Zhenjing Duan</a> <sup>1</sup> <a href="https://orcid.org/0000-0002-8948-4528" target="_blank" rel="noopener noreferrer"> <img src="/static/img/orcid.png" title="ORCID" style="position: relative; width: 13px; margin-left: 0px; max-width: 13px !important; height: auto; top: -7px;"> </a> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q= Shushu Xi" target="_blank"> Shushu Xi</a> <sup>2</sup> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Shuaishuai Wang" target="_blank">Shuaishuai Wang</a> <sup>1</sup> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Ziheng Wang" target="_blank">Ziheng Wang</a> <sup>1</sup> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Peng Bian" target="_blank">Peng Bian</a> <sup>1</sup> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Changhe Li" target="_blank">Changhe Li</a> <sup>2</sup> <a href="https://orcid.org/0000-0003-4409-956X" target="_blank" rel="noopener noreferrer"> <img src="/static/img/orcid.png" title="ORCID" style="position: relative; width: 13px; margin-left: 0px; max-width: 13px !important; height: auto; top: -7px;"> </a> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Jinlong Song" target="_blank">Jinlong Song</a> <sup>1</sup> </span> <span class="mr-2"> <a href="https://scholar.google.com/scholar?q=Xin Liu" target="_blank">Xin Liu</a> <sup>1,*</sup> <a href="mailto:xinliu@dlut.edu.cn"> <sup><i class="bi bi-envelope-fill"></i></sup> </a> <a href="https://orcid.org/0000-0001-8323-7289" target="_blank" rel="noopener noreferrer"> <img src="/static/img/orcid.png" title="ORCID" style="position: relative; width: 13px; margin-left: 0px; max-width: 13px !important; height: auto; top: -7px;"> </a> </span> </div> </div> <div class="d-flex mb-1"> <div class="information-collapse mr-4"> Author Information </div>  </div> <div class="articledz"> <div class="affiliation_zt"> <div class="information-item affiliation_at collapse pl-2 pr-2 pb-2"> <div class="affiliation d-flex"> <div class="affiliation_sup mr-2"><sup>1</sup></div> <div class="affiliation_name flex-grow-1"> State Key Laboratory of High-Performance Precision Manufacturing, Dalian University of Technology, Dalian 116024, China </div> </div> <div class="affiliation d-flex"> <div class="affiliation_sup mr-2"><sup>2</sup></div> <div class="affiliation_name flex-grow-1"> School of Mechanical and Automotive Engineering, Qingdao University of Technology, Qingdao 266520, China </div> </div> <div class="affiliation d-flex"> <div class="affiliation_sup mr-2"><sup>*</sup></div> <div class="affiliation_name flex-grow-1"> Authors to whom correspondence should be addressed. </div> </div> </div> </div> <div class="d-flex align-items-center"> <div class="mr-2 d-flex align-items-center"><span>Views:</span><b class="ml-2">49</b></div> <div class="mr-2 d-flex align-items-center"><span>Downloads:</span><b class="ml-2">10</b></div> <a tabindex="0" class="mr-2 align-items-center cursor-pointer" style="display: none;" data-bind="popover1" data-container="body" data-toggle="popover" data-trigger="focus" data-placement="bottom"> <span>Citations:</span> <b class="ml-2">0</b> </a> </div> <div class="information-item pb-3"> <div class="articledoi"> <em>Intelligent and Sustainable Manufacturing</em> <b>2024</b>, <em>1</em> (2), 10020;  <a href="https://doi.org/10.70322/ism.2024.10020" target="_blank">https://doi.org/10.70322/ism.2024.10020</a> </div> <p class="d-flex fal-title mb-0"> <i class="fal fa-calendar-alt"></i> <span class="mr-3">Received: 15 September 2024</span> <span class="mr-3">Accepted: 22 November 2024</span> <span class="mr-3">Published: 26 November 2024</span> </p> </div> <div class="articleneir pull-item d-flex"> <a href="http://creativecommons.org/licenses/by/4.0/" itemprop="license" rel="license" data-license="by" class="pull-left pt-1"> <img src="/style/image/iconby.png" alt="Creative Commons"> </a> <p class="flex-grow-1 ml-2"> © 2024 The authors. This is an open access article under the Creative Commons Attribution 4.0 International License (<a href="https://creativecommons.org/licenses/by/4.0/" target="_blank">https://creativecommons.org/licenses/by/4.0/</a>). </p> </div> <div class="meta-info text-break border-top border-dark pt-3 mt-3"> <div class="articleneir abstract"> <span>ABSTRACT:</span> In the mechanical cutting process, the surface defects of the workpiece are an important indicator of cutting quality and also reflect the condition of both the machine tool and the cutting tool. Effective detection of defects on the surface of the workpiece plays an important role in adjusting the processing conditions promptly, reducing losses, improving the utilization rate of the workpiece, and maintaining the normal operation of the equipment. To address the challenge of detecting surface defects on workpieces, an inspection method based on an improved Single Shot Multibox Detector (SSD) model is proposed. The method simplifies the detection model and reduces the computation by proposing a DH-MobileNet network instead of a VGG16 network in the SSD structure. The inverse residual structure is also used for position prediction, and null convolution is used instead of a down-sampling operation to avoid information loss. A scanning electron microscope was used to obtain the surface image of the workpiece. A dataset of workpiece surface defects was constructed and expanded, then used to train and test the model for detecting three common types of high-frequency defects: peel-off, chip adhesion, and scratches. The effect was compared with YOLO, Faster R-CNN, and the original SSD model. The detection results show that the method can detect the defects on the surface of the workpiece more accurately and quickly, which provides a new idea for defect detection in real industrial scenarios. </div> <div class="articleneir mt-2 keywords"> <span>Keywords:</span> Cutting machining; Workpiece surface defects; SSD; DH-MobileNet </div> <div class="article-item pt-4"> <h2 class="border-bottom"> <a id="link-introduction" class="anchor-66"></a> 1. Introduction </h2> <div class="link-introduction article-item-text pt-2">During the machining process, due to the influence of cooling and lubrication, tool structure, machine condition, cutting parameters, and materials, a variety of different topographic features will be formed on the surface of the machined workpiece, including tool texture and surface defects on the workpiece (plastic buildup, flaking, and scratches, <i>etc</i>.) [<a href='#B1' class='html-bibr' title='1'>1</a>,<a href='#B2' class='html-bibr' title='2'>2</a>]. With the rapid development of modern industry, the requirements for workpiece quality in the machinery manufacturing industry are gradually increasing. The size and type of defects on the surface of a workpiece are directly linked to the cost, performance, and service life of the machinery and equipment. Therefore, on the one hand, rational cutting parameters and auxiliary technologies (minimal quantity lubrication, cryogenics, ultrasonic vibration, cold plasma, <i>etc</i>.) are used to reduce surface defects on the workpieces during machining [<a href='#B3' class='html-bibr' title='3'>3</a>,<a href='#B4' class='html-bibr' title='4'>4</a>,<a href='#B5' class='html-bibr' title='5'>5</a>,<a href='#B6' class='html-bibr' title='6'>6</a>,<a href='#B7' class='html-bibr' title='7'>7</a>]; On the other hand, effective detection of defects on the surface of the workpiece plays an important role in adjusting the machining conditions promptly, reducing wear and tear, improving the utilization of the workpiece, and maintaining the normal operation of the equipment [<a href='#B8' class='html-bibr' title='8'>8</a>]. To minimize surface defects and enhance the quality of machined workpieces, it is essential to incorporate effective cooling and lubrication during the machining process [<a href='#B9' class='html-bibr' title='9'>9</a>,<a href='#B10' class='html-bibr' title='10'>10</a>,<a href='#B11' class='html-bibr' title='11'>11</a>,<a href='#B12' class='html-bibr' title='12'>12</a>,<a href='#B13' class='html-bibr' title='13'>13</a>]. Minimal quantity lubrication (MQL) has emerged as an environmentally friendly, resource-saving, and efficient green auxiliary processing technology [<a href='#B14' class='html-bibr' title='14'>14</a>,<a href='#B15' class='html-bibr' title='15'>15</a>,<a href='#B16' class='html-bibr' title='16'>16</a>,<a href='#B17' class='html-bibr' title='17'>17</a>]. Liu et al. combined cryogenics with MQL (biological lubricant) to grind titanium alloys, and the values of normal force, tangential force, specific abrasive energy, coefficient of friction, temperature, and surface roughness of cryogenics-MQL were reduced by 47.6%, 44.3%, 45.1%, 19.8%, 43.9%, and 46.3%, respectively, compared with dry grinding [<a href='#B2' class='html-bibr' title='2'>2</a>]. Li et al. conducted milling experiments on titanium alloys using MQL-assisted technology, which showed a significant reduction in tool wear, probably due to the formation of an oil film on the tool and workpiece by the atomized lubricant, which significantly reduced friction and chip adhesion [<a href='#B18' class='html-bibr' title='18'>18</a>]. In order to improve the machining accuracy of difficult-to-machine materials, improve the surface quality of workpieces, and extend the service life of cutting tools, researchers have proposed the nano-lubricants minimal quantity lubrication technology (NMQL), which add nanoparticles that can play a role in cooling and lubrication to bio-oil to achieve micro-lubrication synergistic effects [<a href='#B19' class='html-bibr' title='19'>19</a>,<a href='#B20' class='html-bibr' title='20'>20</a>,<a href='#B21' class='html-bibr' title='21'>21</a>]. Kamal Kishore et al. conducted experiments on cutting Inconel 625 employing NMQL-assisted technology, which showed a significant reduction in surface defects on the workpiece [<a href='#B22' class='html-bibr' title='22'>22</a>]. Adopting certain methods to reduce workpiece surface defects during the cutting process has been well studied. However, detecting surface defects on the cut workpiece is equally important. In recent years, defect monitoring technologies based on deep learning and machine vision have been increasingly adopted in industrial production, significantly advancing the development of the intelligent manufacturing industry [<a href='#B23' class='html-bibr' title='23'>23</a>,<a href='#B24' class='html-bibr' title='24'>24</a>]. Deep learning algorithms have demonstrated state-of-the-art properties in classification and target detection tasks, and their use in surface defect detection is also widespread [<a href='#B25' class='html-bibr' title='25'>25</a>,<a href='#B26' class='html-bibr' title='26'>26</a>]. Deep learning, as a simple and fast method of data analysis, has made breakthroughs in areas such as image processing and target detection [<a href='#B27' class='html-bibr' title='27'>27</a>]. Xie et al. proposed deep learning based on a Convolutional Neural Network (CNN) structure for surface defect detection of magnetic tiles. In their investigation, AlexNet, Visual Geometry Group 16 (VGG16), and ResNet50 were utilized as feature extraction models.The method utilizes a traditional CNN architecture with multiple convolutional layers followed by a Softmax layer to make the final decision [<a href='#B28' class='html-bibr' title='28'>28</a>]. Tabernik et al. proposed a two-stage approach for surface anomaly segmentation based on deep learning models. The researchers developed a segmentation network that consists of 11 convolutional layers trained on pixel labels. Notably, their proposed model shows efficient learning even with a limited number of defective surfaces, as it achieves satisfactory results with only about 25–30 defective training samples [<a href='#B29' class='html-bibr' title='29'>29</a>]. Neuhauser et al. proposed a CNN-based model for classifying and detecting surface defects on extruded aluminum profiles. The model uses Faster R-CNN and ResNet50 as the feature extraction backbone to capture relevant defect features. The study demonstrated the successful application of the proposed method in accurately classifying and detecting surface defects on extruded aluminum profiles [<a href='#B30' class='html-bibr' title='30'>30</a>]. Le et al. proposed an end-to-end lightweight deep learning model for surface defect detection of industrially manufactured parts anchored by YOLOv5. They employed coordinated attention in the Cross Stage Partial (CSP) module to enhance feature extraction [<a href='#B31' class='html-bibr' title='31'>31</a>]. Niu et al. proposed a novel data enhancement strategy to enhance the focus on low-confidence regions in downstream CNNs. The CNN-based segmentation deep learning model was further improved by employing the newly proposed data enhancement method utilizing confidence maps. Experiments demonstrated the model’s high performance in accurately segmenting defects with weak features [<a href='#B32' class='html-bibr' title='32'>32</a>]. Yu et al. introduced a defect detection model for metal surface defects using a small sample learning approach. Their method utilized the matrix decomposition attention mechanism to solve the problem of limited defect samples in industrial production lines. This method could effectively detect and classify metal surface defects using a small amount of labeled data [<a href='#B33' class='html-bibr' title='33'>33</a>]. To detect workpiece surface defects more accurately and efficiently, Scanning Electron Microscope (SEM) images of the milled workpiece surfaces, captured under both dry and NMQL-assisted conditions, were used as training sample data. Under dry conditions, the workpiece surface exhibited more and larger morphological defects, whereas the number of surface defects was fewer under NMQL-assisted conditions. The sample data from these two working conditions gave variability in the number and size of defects on the surface of the workpiece, which helped the model to adapt to the identification of workpiece surfaces with different qualities. In this paper, we propose a workpiece surface defect detection method based on an improved Single Shot MultiBox Detector (SSD) network model, which proposes a DH-MobileNet network based on the MobileNet network and replaces the VGG16 in the SSD model with this network. To address the problem of information loss caused by transmission between subsequent convolutional layers of the SSD model, the inverse residual structure is introduced, while to improve the accuracy of detection of small targets, the null convolution is used instead of the downsampling operation in the inverse residual structure. SEM was utilized to measure high-resolution images to obtain the characteristics of defects on the surface of the workpiece. By optimizing the dataset and training the network model to effectively detect surface defects on workpieces, this approach offers a new solution to the challenge of defect detection in industrial production. </div> <h2 class="border-bottom"> <a id="link-materials-and-methods" class="anchor-66"></a> 2. SEM Sample Data </h2> <div class="link-materials-and-methods article-item-text">Due to the small number of publicly available datasets on workpiece defects, a self-tested dataset was used for this experiment. The sample dataset was tested using SEM. Due to the large amount of sample data, the sample data used includes SEM images from previous tests (<a href='#Figure1' class='html-fig html-figpopup'>Figure 1</a>). A camera can also be used to collect data in practical applications. The test workpieces were processed under dry and NMQL conditions, respectively. The sample material used in this study is primarily aluminum alloy 7050, with its composition and physical properties detailed in <a href='#Table1' class='html-table html-tablepopup'>Table 1</a> and <a href='#Table2' class='html-table html-tablepopup'>Table 2</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure1'><img data-large='/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png' data-original='/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png' src='/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png'><a class='html-expand html-figpopup' href='#Figure1'></a></div></div><div class='html-fig_description'><b><a href='#Figure1' class='html-fig html-figpopup'>Figure 1</a>. </b>Sample data collection tools.</div></div> <div class='html-fig-wrap'><div class='html-table_wrap_discription'><b><a href='#Table1' class='html-table html-tablepopup'>Table 1</a>. </b>Chemical composition of workpiece.<div class='html-figpopup-table html-figpopup-link'><img data-large='/uploads/2024/11/26/6db98cb06dad58182ac184402a8de045.png' data-original='/uploads/2024/11/26/6db98cb06dad58182ac184402a8de045.png' src='/uploads/2024/11/26/6db98cb06dad58182ac184402a8de045.png'><a class='html-expand html-tablepopup' href='#Table1'></a></div></div></div> <div class='html-fig-wrap'><div class='html-table_wrap_discription'><b><a href='#Table2' class='html-table html-tablepopup'>Table 2</a>. </b>Mechanical properties.<div class='html-figpopup-table html-figpopup-link'><img data-large='/uploads/2024/11/26/2ad3b385f7e21852b4a53add5f415f17.png' data-original='/uploads/2024/11/26/2ad3b385f7e21852b4a53add5f415f17.png' src='/uploads/2024/11/26/2ad3b385f7e21852b4a53add5f415f17.png'><a class='html-expand html-tablepopup' href='#Table2'></a></div></div></div> </div> <h2 class="border-bottom"> <a id="link-theory-calculation" class="anchor-66"></a> 3. SSD Target Detection Model </h2> <div class="link-theory-calculation article-item-text"><i>3.1. Model Structure</i> SSD is a target detection algorithm proposed by Wei Liu in ECCV 2016, which utilizes different convolutional layers for boundary and classification regression to achieve better detection and speed [<a href='#B34' class='html-bibr' title='34'>34</a>,<a href='#B35' class='html-bibr' title='35'>35</a>,<a href='#B36' class='html-bibr' title='36'>36</a>]. The SSD network structure is divided into two parts. The first part is the VGG16 network structure with high classification accuracy and its classification layer removed. The second part is a network structure constructed by using convolutional layers instead of two fully connected layers and adding four more convolutional layers to incorporate feature pyramid detection for multi-scale target detection [<a href='#B37' class='html-bibr' title='37'>37</a>,<a href='#B38' class='html-bibr' title='38'>38</a>]. The network model structure of SSD is shown in <a href='#Figure2' class='html-fig html-figpopup'>Figure 2</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure2'><img data-large='/uploads/2024/11/26/bfa116e62a8552906936605b14a15874.png' data-original='/uploads/2024/11/26/bfa116e62a8552906936605b14a15874.png' src='/uploads/2024/11/26/bfa116e62a8552906936605b14a15874.png'><a class='html-expand html-figpopup' href='#Figure2'></a></div></div><div class='html-fig_description'><b><a href='#Figure2' class='html-fig html-figpopup'>Figure 2</a>. </b>Structure of SSD network model.</div></div> The SSD consists of a backbone network and a multi-scale feature layer, with the backbone network (VGG-16) used for feature extraction. The model detects objects at multiple scales to capture objects of varying sizes. Where Conv4_3 has a higher resolution and is suitable for detecting small objects; FC6 is one of the fully connected layers of the VGG-16 network and is located in the deeper layers of the network; FC was changed to a convolutional layer to preserve the spatial resolution Conv8_2, Conv9_2, Conv10_2, and Conv11_2 are additional convolutional layers that were added in the backbone network (e.g., VGG-16) after it, which are used to construct multi-scale feature maps. These feature layers sequentially have smaller and smaller spatial resolutions. <i>3.2. Scale Calculation</i> In the modeling framework of SSD, each selected feature map has <i>k</i> frames with different size and width ratios, which are called default frames. <i>k</i> denotes the number of different frames at each position. The default frames on the feature maps of different solution layers are shown in <a href='#Figure3' class='html-fig html-figpopup'>Figure 3</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure3'><img data-large='/uploads/2024/11/26/5bcddcb254f70f9ff4812433eb4fe21e.png' data-original='/uploads/2024/11/26/5bcddcb254f70f9ff4812433eb4fe21e.png' src='/uploads/2024/11/26/5bcddcb254f70f9ff4812433eb4fe21e.png'><a class='html-expand html-figpopup' href='#Figure3'></a></div></div><div class='html-fig_description'><b><a href='#Figure3' class='html-fig html-figpopup'>Figure 3</a>. </b>Characteristic Graphs of Outputs of Different Convolutional Layers.</div></div> Each default box predicts the number of targets as B and four positional parameters. In this experiment, B is set to 3 because there are three typical defects in the surface of the workpiece, and the calculation of the default box for each feature map is shown in Equation (1). <div class='html-disp-formula-info'><div class='html_pics'>```latexP_i=P_{\min}+\frac{P_{\max}-P_{\min}}{n-1}(i-1),i=1,2,...,n```</div><div class='formula'><label>(1)</label></div></div> where <i>n</i> is the number of feature maps, <i>P</i><sub>min</sub> and <i>P</i><sub>max</sub> are parameters that can be set, which are taken as 0.2 and 0.9 respectively in this experiment, meaning that the scale of the bottom layer is 0.2 and the scale of the top layer is 0.9, and <i>P</i><sub><i>i</i></sub> denotes the scale of the <i>i</i>th feature map. To control the fairness of the feature vectors in the training and testing experiments, the same five aspect ratios <i>b</i> = (1, 2, 3, 0.5, 0.33) were used to generate the default boxes, each of which can be described as: <div class='html-disp-formula-info'><div class='html_pics'>```latexW_i^b=P_i\sqrt{b}```</div><div class='formula'><label>(2)</label></div></div> <div class='html-disp-formula-info'><div class='html_pics'>```latexH_i^b=\frac{P_i}{\sqrt{b}}```</div><div class='formula'><label>(3)</label></div></div> Equation (2) represents the width of the default box, and Equation (3) represents the height of the default box. When the aspect ratio is 1, it is necessary to add a default box scaled to $$P_i^{'}=\sqrt{P_iP_{i+1}}$$. Each default box is centered on $$(\frac{a+0.5}{|g_x|},\frac{a+0.5}{|g_x|})$$, and $${|g_x|}$$ is the size of the <i>i</i>th feature map, $$a,b\in\{0,1,...,|g_x|\}$$. <i>3.3. Loss Function</i> The SSD model is trained for regression on both target location and species, and the training process generates a loss volume, which is calculated using a loss function. The overall objective loss function of the SSD model is the weighted sum of the location loss function (loc) and the confidence loss function (conf), As shown in Equation (4) [<a href='#B39' class='html-bibr' title='39'>39</a>]: <div class='html-disp-formula-info'><div class='html_pics'>```latexL(x,c,l,g)=\frac1N(L_{conf}(x,c)+aL_{loc}(x,l,g))```</div><div class='formula'><label>(4)</label></div></div> where the parameter <i>a</i> is used to balance the confidence loss and position loss; <i>c</i> is the classification confidence; <i>l</i> is the offset of the prediction frame, including the translation offset of the center coordinates and the scaling offset of the width; <i>g</i> is the calibration frame for the actual position of the target; and <i>N</i> is the number of default frames matching the calibration frames of the category. When <i>N</i> equals 0, it means that the loss is 0. The position loss function is the <i>smooth<sub>L</sub></i><sub>1</sub><i>Loss</i> between the parameters of the prediction box <i>l</i> and the true labeled value box <i>g</i>. The regression to the bias of the default border <i>d</i> centered at (<i>cx</i>, <i>cy</i>) and its width (<i>w</i>) height (<i>h</i>) is defined as shown in Equation (5): <div class='html-disp-formula-info'><div class='html_pics'>```latexL_{loc}(x,l,g)=\sum_{i\in Pos}^N\sum_{m\in\{cx,cy,w,h\}}x_{ij}^ksmooth_{L1}(l_i^m-g_j^m)```</div><div class='formula'><label>(5)</label></div></div> where $$smooth_{L1}(x)=\begin{cases}0.5x^2,\quad\left|x\right|<1\\\left|x\right|-0.5,otherwise\end{cases},\, l_i^m$$ is the prediction box, and $$g_j^m$$ is the true box. The confidence loss function is defined as shown in Equation (6). <div class='html-disp-formula-info'><div class='html_pics'>```latexL_{locf}(x,c)=\sum_{i\in Pos}^Nx_{ij}^P\log(\hat{c}_i^p)-\sum_{i\in Neg}\log(\hat{c}_i^0)```</div><div class='formula'><label>(6)</label></div></div> where $$\hat{c}_i^p\quad\mathrm{is}\quad\frac{\exp(c_i^p)}{\sum_p\exp(c_i^p)}$$ , which is generated through Softmax, which is the activation function. Unlike the original MultiBox matching method, the matching strategy during SSD model training is to first match the default bounding box with the real bounding box. If the Intersection over Union (IoU) of the two is higher than the threshold (the default threshold is 0.5), then the default detection box matches with the real box; on the contrary, if the overlap rate is less than 0.5, then the two are not matched match [<a href='#B40' class='html-bibr' title='40'>40</a>]. </div> <h2 class="border-bottom"> <a id="link-results" class="anchor-66"></a> 4. Design and Optimization of Experimental Models </h2> <div class="link-results article-item-text">The SSD model shows better performance in target detection, taking advantage of multiple convolutional layers for target detection applications. However, the SSD model requires more parameters in the detection process as well as the subsequent convolutional layers ignore the connection relationship between layers, thus increasing the computational volume, while the detection of small targets is poor and is prone to problems such as leakage and false detection. The MobileNet network uses fewer parameters, which reduces training time. However, this comes at the cost of lower detection accuracy. To improve the network accuracy while reducing the training parameters, firstly, the DH-MobileNet network is proposed by combining Dilated Convolution and Hierarchical Feature Fusion and used to replace VGG16 in the SSD model; then, the inverse residual structure is utilized to improve the subsequent convolutional layers of the SSD model by using the inverse residual structure to improve the subsequent convolutional layers of the SSD model. The structure of the optimized SSD model is shown in <a href='#Figure4' class='html-fig html-figpopup'>Figure 4</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure4'><img data-large='/uploads/2024/11/26/5e9727f222308b7e1f462cebe2ea3a38.png' data-original='/uploads/2024/11/26/5e9727f222308b7e1f462cebe2ea3a38.png' src='/uploads/2024/11/26/5e9727f222308b7e1f462cebe2ea3a38.png'><a class='html-expand html-figpopup' href='#Figure4'></a></div></div><div class='html-fig_description'><b><a href='#Figure4' class='html-fig html-figpopup'>Figure 4</a>. </b>Optimized SSD Model.</div></div> The optimized model could be improved by adding a dilation layer to the convolutional network to enhance the model’s receptive field. Dilated Convolution extended the range of the receptive field by inserting intervals between the elements of the convolutional kernel without increasing the number of parameters or decreasing the resolution of the feature map. <i>4.1. DH-MobileNet Network Feature Extraction Principle</i> MobileNet was a lightweight convolutional neural network proposed by Howard et al. The basic unit is a deeply separable convolution, which requires fewer parameters for training and can save and enable the deployment of mobile devices [<a href='#B41' class='html-bibr' title='41'>41</a>]. The structure of the MobileNet model is shown in <a href='#Figure5' class='html-fig html-figpopup'>Figure 5</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure5'><img data-large='/uploads/2024/11/26/2bdddfd3236042b0051a6bcbe334ac62.png' data-original='/uploads/2024/11/26/2bdddfd3236042b0051a6bcbe334ac62.png' src='/uploads/2024/11/26/2bdddfd3236042b0051a6bcbe334ac62.png'><a class='html-expand html-figpopup' href='#Figure5'></a></div></div><div class='html-fig_description'><b><a href='#Figure5' class='html-fig html-figpopup'>Figure 5</a>. </b>MobileNet structure (<b>a</b>) Network structure of MobileNet; (<b>b</b>) Depth-separable convolutional structure; (<b>c</b>) Point convolution kernel.</div></div> <a href='#Figure5' class='html-fig html-figpopup'>Figure 5</a>a shows the network structure of MobileNet, where Conv-Dw-Pw is a depth-separable convolutional structure, which consists of two parts: depth-wise (DW) and point-wise (PW) convolution. The DW is constructed using a 3 × 3 deep convolutional kernel, as shown in <a href='#Figure5' class='html-fig html-figpopup'>Figure 5</a>b; The PW is composed using a 1 × 1 point convolution kernel, as shown in <a href='#Figure5' class='html-fig html-figpopup'>Figure 5</a>c. BN stands for batch normalization, and ReLU6 for activation function. During the model training process, the receptive field is expanded through pooling operations to enhance detection accuracy. Still, the pooling operation will cause the problem of partially missing image information. Dilated Convolution can increase the receptive field of the input image without losing the image information; therefore, introducing the Dilated Convolution (DC) operation into the MobileNet network not only ensures the integrity of the image information but also increases the range of the receptive field of the convolutional layer, thus ensuring the detection accuracy [<a href='#B42' class='html-bibr' title='42'>42</a>]. A cavity convolution operation with a convolution kernel size of 3 × 3 and a dilation rate of 2 is shown in <a href='#Figure6' class='html-fig html-figpopup'>Figure 6</a>, where nine points in a 7 × 7 region are convolved with a 3 × 3 convolution kernel, the rest of the points contain a parameter set to 0, and the size of the sensory field after convolution is $$F_{dilation}=\left[\begin{array}{c}{2^{(\frac{dilation}{2})+2}-1}\\\end{array}\right]$$. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure6'><img data-large='/uploads/2024/11/26/4646e423353c6436487811dfd57f35db.png' data-original='/uploads/2024/11/26/4646e423353c6436487811dfd57f35db.png' src='/uploads/2024/11/26/4646e423353c6436487811dfd57f35db.png'><a class='html-expand html-figpopup' href='#Figure6'></a></div></div><div class='html-fig_description'><b><a href='#Figure6' class='html-fig html-figpopup'>Figure 6</a>. </b>Dilated Convolution operation.</div></div> Hierarchical Feature Fusion (HFF) is to sum the outputs of each convolutional unit of a null convolutional layer in turn and apply the concatenation operation to each summed result to obtain the final output, which can expand the learning parameters without increasing the complexity of the network and strengthen the network continuity [<a href='#B40' class='html-bibr' title='40'>40</a>]. The HFF structure is shown in <a href='#Figure7' class='html-fig html-figpopup'>Figure 7</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure7'><img data-large='/uploads/2024/11/26/8369db87b04ec4cc00cd9146aebddc93.png' data-original='/uploads/2024/11/26/8369db87b04ec4cc00cd9146aebddc93.png' src='/uploads/2024/11/26/8369db87b04ec4cc00cd9146aebddc93.png'><a class='html-expand html-figpopup' href='#Figure7'></a></div></div><div class='html-fig_description'><b><a href='#Figure7' class='html-fig html-figpopup'>Figure 7</a>. </b>Hierarchical Feature Fusion.</div></div> The depth-separable convolutional structure of DH-MobileNet requires only 1/34 of the VGG16 network parameters to achieve the same classification accuracy in the classification task, and the same output as the standard convolution can be obtained through operations such as dimensionality reduction, which can reduce the amount of computation and improve the effect of training. The network structure of DH-MobileNet is shown in <a href='#Figure8' class='html-fig html-figpopup'>Figure 8</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure8'><img data-large='/uploads/2024/11/26/4ec84717f4f3a83bb734c81acb836927.png' data-original='/uploads/2024/11/26/4ec84717f4f3a83bb734c81acb836927.png' src='/uploads/2024/11/26/4ec84717f4f3a83bb734c81acb836927.png'><a class='html-expand html-figpopup' href='#Figure8'></a></div></div><div class='html-fig_description'><b><a href='#Figure8' class='html-fig html-figpopup'>Figure 8</a>. </b>Architecture of DH-MobileNet.</div></div> In <a href='#Figure8' class='html-fig html-figpopup'>Figure 8</a>, Dilated denotes the cavity convolution, <i>R</i> denotes the dilation rate, and Conv-Dw-Pw represents the depth-separable convolution structure. <i>4.2. SSD Convolutional Network Layer Optimization</i> Since the subsequent convolutional layers in the SSD model ignore the connectivity between layers, this leads to poor detection of small targets. Therefore, the introduction of the inverse residual structure reduces the loss of information caused by nonlinear transformations at low latitudes during the learning process. To prevent the loss of feature information during the down-sampling process, a depthwise separable convolution is used to replace traditional down-sampling in the inverse residual structure. Additionally, the ReLU activation function is replaced with the higher-performance ReLU6. The ReLU6 activation function can be expressed as: $$Y=\min(\max(X,0),6)$$. Where <i>Y</i> is the output of ReLU6 and <i>X</i> is the value of each pixel in the feature map. Batch normalization (BN) algorithm is used for normalization. The BN algorithm supports automatic data distribution adjustment and prevents gradient vanishing and complex variable adjustment by setting two learning parameters [<a href='#B43' class='html-bibr' title='43'>43</a>]. The inverse residual structure after adding the null convolution is shown in <a href='#Figure9' class='html-fig html-figpopup'>Figure 9</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure9'><img data-large='/uploads/2024/11/26/0ac6cd20daef51eb9751bc517e4a8dba.png' data-original='/uploads/2024/11/26/0ac6cd20daef51eb9751bc517e4a8dba.png' src='/uploads/2024/11/26/0ac6cd20daef51eb9751bc517e4a8dba.png'><a class='html-expand html-figpopup' href='#Figure9'></a></div></div><div class='html-fig_description'><b><a href='#Figure9' class='html-fig html-figpopup'>Figure 9</a>. </b>Reverse residual structure before and after improvement.</div></div> </div> <h2 class="border-bottom"> <a id="link-discussion" class="anchor-66"></a> 5. Experimental Design </h2> <div class="link-discussion article-item-text">The configuration of the computer used for the experiments is as follows: the operating system is Windows 10, the CPU model is Intel(R) Core(TM) i7-8750H, the frequency of the CPU is 2.20 GHz, the RAM is 8.0 GB, and the model of the graphic card (GPU) is GeForce GTX 1060, and TensorFlow is set up based on the above hardware environment on the above hardware. The process of training the model starts with training the training set and eliminates the discarded below-threshold detection frames using a non-great-value suppression algorithm [<a href='#B43' class='html-bibr' title='43'>43</a>]. After iterating and adjusting the parameter settings to optimize the model’s performance, it is subsequently tested for accuracy using the test set from the experimental dataset. The model’s performance is examined by analyzing the detection accuracy, detection speed, and number of parameters of the model in the test set. </div> <h2 class="border-bottom"> <a id="link-conclusions" class="anchor-66"></a> 6. Results and Discussion </h2> <div class="link-conclusions article-item-text"><i>6.1. Structure of Data Samples</i> Three common defects (chip adhesion, flaking, and pear gouge) were selected for this experiment. In this paper, while satisfying the number of samples and fully considering the specification factor of the samples, images containing three kinds of defects: chip adhesion, flaking, and pear gouge are collected by SEM, such as typical defects on the surface of the workpiece as shown in <a href='#Figure10' class='html-fig html-figpopup'>Figure 10</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure10'><img data-large='/uploads/2024/11/26/55950d4f90482eea8e00307146167bd5.png' data-original='/uploads/2024/11/26/55950d4f90482eea8e00307146167bd5.png' src='/uploads/2024/11/26/55950d4f90482eea8e00307146167bd5.png'><a class='html-expand html-figpopup' href='#Figure10'></a></div></div><div class='html-fig_description'><b><a href='#Figure10' class='html-fig html-figpopup'>Figure 10</a>. </b>Surface defects in cutting processing (<b>a</b>) Peel off; (<b>b</b>) Chip adhesion; (<b>c)</b> Scratch.</div></div> The collected defective images are subjected to data expansion, and a data set is established. In data expansion, the main operations performed are rotation, horizontal migration, vertical migration, and scaling in four ways, and the specific implementation method of data expansion is shown in <a href='#Table3' class='html-table html-tablepopup'>Table 3</a>. <div class='html-fig-wrap'><div class='html-table_wrap_discription'><b><a href='#Table3' class='html-table html-tablepopup'>Table 3</a>. </b>Data expansion implementation methods.<div class='html-figpopup-table html-figpopup-link'><img data-large='/uploads/2024/11/26/05e4efdbfb974822699bd72ad4bf0826.png' data-original='/uploads/2024/11/26/05e4efdbfb974822699bd72ad4bf0826.png' src='/uploads/2024/11/26/05e4efdbfb974822699bd72ad4bf0826.png'><a class='html-expand html-tablepopup' href='#Table3'></a></div></div></div> The number of training samples is increased after data augmentation, which strongly improves the training performance of the dataset. 80% of the data-augmented dataset is used as a training set and 20% as a test set. <i>6.2. Model Accuracy Analysis</i> 6.2.1. Convergence Rate Analysis The proposed DH-MobileNet-SSD model was compared with three models, Faster R-CNN, SSD, and YOLO, in the same dataset for experiments. Each experimental model was iterated 36,000 times, as this number was deemed sufficient based on hardware limitations and practical application requirements. Beyond 36,000 iterations, further training showed minimal significance since the loss values had already stabilized. The loss function image during training is shown in <a href='#Figure11' class='html-fig html-figpopup'>Figure 11</a>. As can be seen from <a href='#Figure11' class='html-fig html-figpopup'>Figure 11</a>, during the training of the dataset of this experiment, the loss values of each model converged continuously with the increase of the number of iterations. After 25,000 iterations, each model stabilized. Among them, the DH-MobileNet-SSD could better accelerate the convergence speed of the network. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure11'><img data-large='/uploads/2024/11/26/79c412ab35bf020969b8ffb26a991295.png' data-original='/uploads/2024/11/26/79c412ab35bf020969b8ffb26a991295.png' src='/uploads/2024/11/26/79c412ab35bf020969b8ffb26a991295.png'><a class='html-expand html-figpopup' href='#Figure11'></a></div></div><div class='html-fig_description'><b><a href='#Figure11' class='html-fig html-figpopup'>Figure 11</a>. </b>Loss function training image.</div></div> 6.2.2. Performance Evaluation Precision, Average Precision (AP) and Mean Average Precision (mAP) were the main commonly used parameters to measure the performance of a target detection model, where precision was the proportion of defect types that were truly judged to be accurate in the images judged to be accurate; Average Precision was the average of the precision, reflecting the average detection accuracy of each defect; mean Average Precision represents the average accuracy of the three categories of defects, reflecting the effectiveness of the model in detecting the entire dataset. The formulas for the three commonly used parameters are shown in Equations (7)–(9). <div class='html-disp-formula-info'><div class='html_pics'>```latex\text{Precision}=\frac{Num(TruePosition)}{Num(TotalObjects)}```</div><div class='formula'><label>(7)</label></div></div> <div class='html-disp-formula-info'><div class='html_pics'>```latex\mathrm{AP}=\frac{\sum Precision}{Num(TotalImages)}```</div><div class='formula'><label>(8)</label></div></div> <div class='html-disp-formula-info'><div class='html_pics'>```latex\mathrm{mAP}=\frac{\sum AP}{Num(Classes)}```</div><div class='formula'><label>(9)</label></div></div> The test set was detected using the trained network. The test set images contain three different types of defect samples, each with 60 different images. The results of the precision of the trained detection network for the three defective samples are shown in <a href='#Table4' class='html-table html-tablepopup'>Table 4</a>. <div class='html-fig-wrap'><div class='html-table_wrap_discription'><b><a href='#Table4' class='html-table html-tablepopup'>Table 4</a>. </b>Precision table for detection of three types of defects.<div class='html-figpopup-table html-figpopup-link'><img data-large='/uploads/2024/11/26/29349d4195685163e3860c24989bd283.png' data-original='/uploads/2024/11/26/29349d4195685163e3860c24989bd283.png' src='/uploads/2024/11/26/29349d4195685163e3860c24989bd283.png'><a class='html-expand html-tablepopup' href='#Table4'></a></div></div></div> The inspection results showed that among the three types of defects, chip adhesion was detected with higher accuracy, but it also had a higher number of missed detections. Both types of defects, spalling and pear gouge, are still missed and wrongly detected. Overall, it can be concluded that the DH-MobileNet-SSD model has a higher detection accuracy of 88.9% for these three defects. The Average Precision (AP) values of the DH-MobileNet-SSD model with the three models Faster-RCNN, SSD, and YOLO at the same number of iteration steps are shown in <a href='#Figure12' class='html-fig html-figpopup'>Figure 12</a>. <a href='#Figure12' class='html-fig html-figpopup'>Figure 12</a> shows that the AP value of the DH-MobileNet-SSD model is much higher than that of the Faster R-CNN, SSD, and YOLO models, which further validates that the model used has better results in the detection of defects on the surface of workpieces. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure12'><img data-large='/uploads/2024/11/26/d452638e7f31087847d0678d5e89d1d4.png' data-original='/uploads/2024/11/26/d452638e7f31087847d0678d5e89d1d4.png' src='/uploads/2024/11/26/d452638e7f31087847d0678d5e89d1d4.png'><a class='html-expand html-figpopup' href='#Figure12'></a></div></div><div class='html-fig_description'><b><a href='#Figure12' class='html-fig html-figpopup'>Figure 12</a>. </b>Detection results of defect AP on the workpiece surface.</div></div> With the same number of training steps and the same training set, the DH-MobileNet-SSD model used in this paper is compared with the Faster R-CNN, SSD, and YOLO models in terms of mAP and average detection time, and the results are shown in <a href='#Figure13' class='html-fig html-figpopup'>Figure 13</a>. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure13'><img data-large='/uploads/2024/11/26/1e29566d48adede076b3b421f802927c.png' data-original='/uploads/2024/11/26/1e29566d48adede076b3b421f802927c.png' src='/uploads/2024/11/26/1e29566d48adede076b3b421f802927c.png'><a class='html-expand html-figpopup' href='#Figure13'></a></div></div><div class='html-fig_description'><b><a href='#Figure13' class='html-fig html-figpopup'>Figure 13</a>. </b>Comparison of defect detection methods (<b>a</b>) mAP; (<b>b</b>) Average detection time.</div></div> From <a href='#Figure13' class='html-fig html-figpopup'>Figure 13</a>, it can be seen that the average accuracy mean of DH-MobileNet-SSD is larger and more accurate than the average accuracy mean of the Faster R-CNN-based model, YOLO-based model, and SSD-based model. Analyzing the detection speed, the average detection time of this paper’s model is 0.122 s, which is lower than the other three detection algorithms, indicating that this paper’s method is more time-saving than the Faster R-CNN model, SSD model, and YOLO model. <i>6.3. Data Set Sample Detection</i> One type of defect was selected to evaluate detection performance across the dataset using four models. The detection results are presented in <a href='#Figure14' class='html-fig html-figpopup'>Figure 14</a>. Specifically, <a href='#Figure14' class='html-fig html-figpopup'>Figure 14</a>a displays the detection results of the Faster R-CNN model, <a href='#Figure14' class='html-fig html-figpopup'>Figure 14</a>b shows the results of the YOLO model, <a href='#Figure14' class='html-fig html-figpopup'>Figure 14</a>c presents the detection results of the original SSD model, and <a href='#Figure14' class='html-fig html-figpopup'>Figure 14</a>d illustrates the detection results of the experimental model, DH-MobileNet-SSD. The detection plots reveal that the improved SSD model (DH-MobileNet-SSD) achieves higher detection accuracy, with the detection frames aligning more precisely with the defective regions compared to the other models. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure14'><img data-large='/uploads/2024/11/26/d32a9ea38f44d1c567b11ef7cf53bbee.png' data-original='/uploads/2024/11/26/d32a9ea38f44d1c567b11ef7cf53bbee.png' src='/uploads/2024/11/26/d32a9ea38f44d1c567b11ef7cf53bbee.png'><a class='html-expand html-figpopup' href='#Figure14'></a></div></div><div class='html-fig_description'><b><a href='#Figure14' class='html-fig html-figpopup'>Figure 14</a>. </b>Detection Result Diagram (<b>a</b>) SSD; (<b>b</b>) YOLO; (<b>c</b>) Faster R-CNN;(<b>d</b>) DH-MobileNet-SSD.</div></div> The three defects were detected separately using the DH-MobileNet-SSD model, and the detection results are shown in <a href='#Figure15' class='html-fig html-figpopup'>Figure 15</a>, with Peel off in <a href='#Figure15' class='html-fig html-figpopup'>Figure 15</a>a, Chip adhesion in <a href='#Figure15' class='html-fig html-figpopup'>Figure 15</a>b, and Scratch in <a href='#Figure15' class='html-fig html-figpopup'>Figure 15</a>c. It could be seen from the detection plots that three defects can be detected with a high rate of correctness. Therefore, the DH-MobileNet-SSD model proposed in this study demonstrates superior detection performance and exhibits good stability when applied to this dataset. <div class='html-fig-wrap'><div class='html-fig_img'><div class='html-figpopup html-figpopup-link' href='#Figure15'><img data-large='/uploads/2024/11/26/32061512b6c9097f6099a9ab67231d8b.png' data-original='/uploads/2024/11/26/32061512b6c9097f6099a9ab67231d8b.png' src='/uploads/2024/11/26/32061512b6c9097f6099a9ab67231d8b.png'><a class='html-expand html-figpopup' href='#Figure15'></a></div></div><div class='html-fig_description'><b><a href='#Figure15' class='html-fig html-figpopup'>Figure 15</a>. </b>Three Defect Detection Results (<b>a</b>) Peel off (<b>b</b>); Chip adhesion; (<b>c</b>) Scratch.</div></div></div> <h2 class="border-bottom"> <a id="link-supplementary-materials" class="anchor-66"></a> 7. Conclusions </h2> <div class="link-supplementary-materials article-item-text">To improve the detection of defects on the surface of the cut workpiece, Scanning Electron Microscope (SEM) images of the mechanical workpiece surface were first acquired. Then DH-MobileNet-SSD was proposed as a method to detect the defects on the surface of the workpiece based on the SSD network model for three kinds of high-frequency defects (peel off, chip adhesion, and scratch). The method utilized the deep convolutional features, effectively avoiding the problem of traditional target detection models relying on manual features. The conclusion is as follows: (1) The DH-MobileNet-SSD inspection model proposed in this paper can effectively detect surface defects (peel off, chip adhesion, and scratch) on workpieces. (2) DH-MobileNet-SSD achieves 88.9% precision in workpiece defect detection, which is higher than Faster R-CNN, YOLO, and the original SSD method under the same conditions. However, the training network is not yet fully optimized; its accuracy requires further improvement, and its overall performance still needs enhancement. This remains an area for future research and development. (3) The mAP of the DH-MobileNet-SSD model (85.31%) is higher and more accurate compared to the mAP of the Faster R-CNN-based model, the YOLO-based model, and the SSD-based model. Analyzed in terms of detection speed, the average detection time of DH-MobileNet-SSD is 0.122 s, which is lower than the other three detection algorithms. </div> <div class="link-appendix article-item-text"></div> <div class="link-appendix-b article-item-text"></div> <div class="link-acknowledgments article-item-text"></div> <h2 class="border-bottom"> <a id="link-author-contributions" class="anchor-66"></a> Author Contributions </h2> <div class="link-author-contributions article-item-text">Z.D.: Preparation, Experiment, Writing. S.X.: Programming, Writing. S.W.: Visualization. Z.W.: Visualization. P.B.: Project administration. C.L.: Methodology. J.S.: Formal analysis. X.L.: Funding acquisition, Supervision, Resources.</div> <h2 class="border-bottom"> <a id="link-ethics-statement" class="anchor-66"></a> Ethics Statement </h2> <div class="link-ethics-statement article-item-text">Not applicable.</div> <h2 class="border-bottom"> <a id="link-informed-consent-statement" class="anchor-66"></a> Informed Consent Statement </h2> <div class="link-informed-consent-statement article-item-text">Not applicable.</div> <h2 class="border-bottom"> <a id="link-funding" class="anchor-66"></a> Funding </h2> <div class="link-funding article-item-text">This research was funded by [Fundamental Research Funds for the Central Universities] grant number [DUT19ZD202] and [National Natural Science Foundation of China] grant number [52475430].</div> <h2 class="border-bottom"> <a id="link-declaration-competing-interest" class="anchor-66"></a> Declaration of Competing Interest </h2> <div class="link-declaration-competing-interest article-item-text">The authors declare that they have no known competing interests or personal relationships that may affect the work reported in this paper.</div> <h2 class="border-bottom"> <a id="html-references_list" class="anchor-66"></a> References </h2> <div class="references_zt"> <div class="references_at"> <div class="references html-x" id='B1'> <div class="references_sup">1.</div> <div class="references_name"> Yang M, Hao JC, Wu WT, Li ZH, Ma YQ, Zhou ZM, et al. Critical<i> </i>cutting<i> </i>thickness<i> </i>model<i> </i>considering<i> </i>subsurface<i> </i>damage<i> </i>of<i> </i>zirconia<i> </i>grinding<i> </i>and<i> </i>friction—Wear<i> </i>performance<i> </i>evaluation<i> </i>applied<i> </i>in<i> </i>simulated<i> </i>oral<i> </i>environment.<i> </i><i> Tribol. Int.</i><b> 2024</b>,<i> 198,</i> 109881. doi:10.1016/j.triboint.2024.109881.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Critical<i> </i>cutting<i> </i>thickness<i> </i>model<i> </i>considering<i> </i>subsurface<i> </i>damage<i> </i>of<i> </i>zirconia<i> </i>grinding<i> </i>and<i> </i>friction—Wear<i> </i>performance<i> </i>evaluation<i> </i>applied<i> </i>in<i> </i>simulated<i> </i>oral<i> </i>environment.<i>&nbsp;</i>&publication_year=2024&author=Yang M&author=Hao JC&author=Wu WT&author=Li ZH&author=Ma YQ&author=Zhou ZM&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B2'> <div class="references_sup">2.</div> <div class="references_name"> Liu MZ, Li CH, Yang M, Gao T, Wang XM, Cui X, et al. Mechanism<i> </i>and<i> </i>enhanced<i> </i>grindability<i> </i>of<i> </i>cryogenic<i> </i>air<i> </i>combined<i> </i>with<i> </i>biolubricant<i> </i>grinding<i> </i>titanium<i> </i>alloy.<i> </i><i> Tribol. Int.</i><b> 2023</b>,<i> 187,</i> 108704. doi:10.1016/j.triboint.2023.108704.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Mechanism<i> </i>and<i> </i>enhanced<i> </i>grindability<i> </i>of<i> </i>cryogenic<i> </i>air<i> </i>combined<i> </i>with<i> </i>biolubricant<i> </i>grinding<i> </i>titanium<i> </i>alloy.<i>&nbsp;</i>&publication_year=2023&author=Liu MZ&author=Li CH&author=Yang M&author=Gao T&author=Wang XM&author=Cui X&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B3'> <div class="references_sup">3.</div> <div class="references_name"> Duan ZJ, Li CH, Zhang YB, Dong L, Bai XF, Yang M, et al. Milling surface roughness for<i> </i>7050<i> </i>aluminum<i> </i>alloy<i> </i>cavity<i> </i>influenced<i> </i>by<i> </i>nozzle<i> </i>position<i> </i>of<i> </i>nanofluid<i> </i>minimum<i> </i>quantity<i> </i>lubrication.<i> </i><i> Chin. J. Aeronaut. </i><b> 2021</b>,<i> 34,</i> 33–53. doi:10.1016/j.cja.2020.04.029.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Milling surface roughness for<i> </i>7050<i> </i>aluminum<i> </i>alloy<i> </i>cavity<i> </i>influenced<i> </i>by<i> </i>nozzle<i> </i>position<i> </i>of<i> </i>nanofluid<i> </i>minimum<i> </i>quantity<i> </i>lubrication.<i>&nbsp;</i>&publication_year=2021&author=Duan ZJ&author=Li CH&author=Zhang YB&author=Dong L&author=Bai XF&author=Yang M&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B4'> <div class="references_sup">4.</div> <div class="references_name"> Cao Y, Ding WF, Zhao BA, Wen XB, Li SP, Wang JZ. <i> </i>Effect of intermittent cutting behavior on the ultrasonic vibration-assisted grinding performance of Inconel718<i> </i>nickel-based<i> </i>superalloy.<i> </i><i> Precis. Eng. </i><b> 2022</b>,<i> 78,</i> 248–260. doi:10.1016/j.precisioneng.2022.08.006.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>Effect of intermittent cutting behavior on the ultrasonic vibration-assisted grinding performance of Inconel718<i> </i>nickel-based<i> </i>superalloy.<i>&nbsp;</i>&publication_year=2022&author=Cao Y&author=Ding WF&author=Zhao BA&author=Wen XB&author=Li SP&author=Wang JZ." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B5'> <div class="references_sup">5.</div> <div class="references_name"> Cao Y, Yin JF, Ding WF, Xu JH. Alumina abrasive wheel wear in ultrasonic vibration-assisted creep-feed grinding of Inconel 718<i> </i>nickel-based<i> </i>superalloy.<i> </i><i> J. Mater. Process Tech.</i><b> 2021</b>,<i> 297,</i> 117241. doi:10.1016/j.jmatprotec.2021.117241.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Alumina abrasive wheel wear in ultrasonic vibration-assisted creep-feed grinding of Inconel 718<i> </i>nickel-based<i> </i>superalloy.<i>&nbsp;</i>&publication_year=2021&author=Cao Y&author=Yin JF&author=Ding WF&author=Xu JH. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B6'> <div class="references_sup">6.</div> <div class="references_name"> Qu SS, Yao P, Gong YD, Chu DK, Yang YY, Li CW, et al. Environmentally<i> </i>friendly<i> </i>grinding<i> </i>of<i> </i>C/SiCs<i> </i>using<i> </i>carbon<i> </i>nanofluid<i> </i>minimum<i> </i>quantity<i> </i>lubrication<i> </i>technology.<i> </i><i> J. Clean. Prod.</i><b> 2022</b>,<i> 366,</i> 132898. doi:10.1016/j.jclepro.2022.132898.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Environmentally<i> </i>friendly<i> </i>grinding<i> </i>of<i> </i>C/SiCs<i> </i>using<i> </i>carbon<i> </i>nanofluid<i> </i>minimum<i> </i>quantity<i> </i>lubrication<i> </i>technology.<i>&nbsp;</i>&publication_year=2022&author=Qu SS&author=Yao P&author=Gong YD&author=Chu DK&author=Yang YY&author=Li CW&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B7'> <div class="references_sup">7.</div> <div class="references_name"> Liu MZ, Li CH, Zhang YB, Yang M, Gao T, Cui X, et al. <i> </i>Analysis<i> </i>of<i> </i>grain<i> </i>tribology<i> </i>and<i> </i>improved<i> </i>grinding<i> </i>temperature<i> </i>model<i> </i>based<i> </i>on<i> </i>discrete<i> </i>heat<i> </i>source.<i> </i><i> Tribol. Int.</i><b> 2023</b>,<i> 180,</i> 108196. doi:10.1016/j.triboint.2022.108196.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>Analysis<i> </i>of<i> </i>grain<i> </i>tribology<i> </i>and<i> </i>improved<i> </i>grinding<i> </i>temperature<i> </i>model<i> </i>based<i> </i>on<i> </i>discrete<i> </i>heat<i> </i>source.<i>&nbsp;</i>&publication_year=2023&author=Liu MZ&author=Li CH&author=Zhang YB&author=Yang M&author=Gao T&author=Cui X&author=et al." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B8'> <div class="references_sup">8.</div> <div class="references_name"> Wang HJ, Mo H, Lu SL, Zhao XM. <i> </i>Electrolytic<i> </i>capacitor<i> </i>surface<i> </i>defect<i> </i>detection<i> </i>based<i> </i>on<i> </i>deep<i> </i>convolution<i> </i>neural<i> </i>network.<i> </i><i> J. King Saud Univ.-Comput. Inf. Sci.</i><b> 2024</b>,<i> 36,</i> 101935. doi:10.1016/j.jksuci.2024.101935.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>Electrolytic<i> </i>capacitor<i> </i>surface<i> </i>defect<i> </i>detection<i> </i>based<i> </i>on<i> </i>deep<i> </i>convolution<i> </i>neural<i> </i>network.<i>&nbsp;</i>&publication_year=2024&author=Wang HJ&author=Mo H&author=Lu SL&author=Zhao XM." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B9'> <div class="references_sup">9.</div> <div class="references_name"> He T, Liu NC, Xia HZ, Wu L, Zhang Y, Li DG, et al. Progress<i> </i>and<i> </i>trend<i> </i>of<i> </i>minimum<i> </i>quantity<i> </i>lubrication<i> </i>(MQL):<i> </i>A<i> </i>comprehensive<i> </i>review.<i> </i><i> J. Clean. Prod.</i><b> 2023</b>,<i> 386,</i> 135809. doi:10.1016/j.jclepro.2022.135809.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Progress<i> </i>and<i> </i>trend<i> </i>of<i> </i>minimum<i> </i>quantity<i> </i>lubrication<i> </i>(MQL):<i> </i>A<i> </i>comprehensive<i> </i>review.<i>&nbsp;</i>&publication_year=2023&author=He T&author=Liu NC&author=Xia HZ&author=Wu L&author=Zhang Y&author=Li DG&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B10'> <div class="references_sup">10.</div> <div class="references_name"> Sarikaya M, Gupta MK, Tomaz I, Danish M, Mia M, Rubaiee S, et al. Cooling<i> </i>techniques<i> </i>to<i> </i>improve<i> </i>the<i> </i>machinability<i> </i>and<i> </i>sustainability<i> </i>of<i> </i>light-weight<i> </i>alloys:<i> </i>A<i> </i>state-of-the-art<i> </i>review.<i> </i><i> J. Manuf. Process. </i><b> 2021</b>,<i> 62,</i> 179–201. doi:10.1016/j.jmapro.2020.12.013.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Cooling<i> </i>techniques<i> </i>to<i> </i>improve<i> </i>the<i> </i>machinability<i> </i>and<i> </i>sustainability<i> </i>of<i> </i>light-weight<i> </i>alloys:<i> </i>A<i> </i>state-of-the-art<i> </i>review.<i>&nbsp;</i>&publication_year=2021&author=Sarikaya M&author=Gupta MK&author=Tomaz I&author=Danish M&author=Mia M&author=Rubaiee S&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B11'> <div class="references_sup">11.</div> <div class="references_name"> Duan ZJ, Wang SS, Wang ZH, Li CH, Li YH, Song JL, et al. Tool<i> </i>wear<i> </i>mechanisms<i> </i>in<i> </i>cold<i> </i>plasma<i> </i>and<i> </i>nano-lubricant<i> </i>multi-energy<i> </i>field<i> </i>coupled<i> </i>micro-milling<i> </i>of<i> </i>Al-Li<i> </i>alloy.<i> </i><i> Tribol. Int. </i><b> 2024</b>,<i> 192,</i> 109337. doi:10.1016/j.triboint.2024.109337.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Tool<i> </i>wear<i> </i>mechanisms<i> </i>in<i> </i>cold<i> </i>plasma<i> </i>and<i> </i>nano-lubricant<i> </i>multi-energy<i> </i>field<i> </i>coupled<i> </i>micro-milling<i> </i>of<i> </i>Al-Li<i> </i>alloy.<i>&nbsp;</i>&publication_year=2024&author=Duan ZJ&author=Wang SS&author=Wang ZH&author=Li CH&author=Li YH&author=Song JL&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B12'> <div class="references_sup">12.</div> <div class="references_name"> Duan ZJ, Wang SS, Li CH, Wang ZH, Bian P, Sun J, et al.  Cold plasma and different nano-lubricants multi-energy field coupling-assisted micro-milling of Al-Li alloy<i> </i>2195-T8<i> </i>and<i> </i>flow<i> </i>rate<i> </i>optimization.<i> </i><i> J Manuf Process. </i><b> 2024</b>,<i> 127,</i> 218–237. doi:10.1016/j.jmapro.2024.07.146.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=&nbsp;Cold plasma and different nano-lubricants multi-energy field coupling-assisted micro-milling of Al-Li alloy<i> </i>2195-T8<i> </i>and<i> </i>flow<i> </i>rate<i> </i>optimization.<i>&nbsp;</i>&publication_year=2024&author=Duan ZJ&author=Wang SS&author=Li CH&author=Wang ZH&author=Bian P&author=Sun J&author=et al." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B13'> <div class="references_sup">13.</div> <div class="references_name"> Zhang<i> </i>Y,<i> </i>Li<i> </i>L,<i> </i>Cui<i> </i>X,<i> </i>An<i> </i>Q,<i> </i>Xu<i> </i>P,<i> </i>Wang<i> </i>W,<i> </i>et al.<i> </i>Lubricant<i> </i>activity<i> </i>enhanced<i> </i>technologies<i> </i>for<i> </i>sustainable<i> </i>machining:<i> </i>Mechanisms<i> </i>and<i> </i>processability.<i> Chin. J. Aeronaut. </i><b>2024</b>,<i> </i>doi:10.1016/j.cja.2024.08.034. </div> </div> <div class="references html-x" id='B14'> <div class="references_sup">14.</div> <div class="references_name"> Jia DZ, Zhang YB, Li CH, Yang M, Gao T, Said Z, et al. Lubrication-enhanced<i> </i>mechanisms<i> </i>of<i> </i>titanium<i> </i>alloy<i> </i>grinding<i> </i>using<i> </i>lecithin<i> </i>biolubricant.<i> </i><i> Tribol. Int. </i><b> 2022</b>,<i> 169,</i> 107461. doi:10.1016/j.triboint.2022.107461.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Lubrication-enhanced<i> </i>mechanisms<i> </i>of<i> </i>titanium<i> </i>alloy<i> </i>grinding<i> </i>using<i> </i>lecithin<i> </i>biolubricant.<i>&nbsp;</i>&publication_year=2022&author=Jia DZ&author=Zhang YB&author=Li CH&author=Yang M&author=Gao T&author=Said Z&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B15'> <div class="references_sup">15.</div> <div class="references_name"> Jia DZ, Li CH, Liu JH, Zhang YB, Yang M, Gao T, et al. Prediction<i> </i>model<i> </i>of<i> </i>volume<i> </i>average<i> </i>diameter<i> </i>and<i> </i>analysis<i> </i>of<i> </i>atomization<i> </i>characteristics<i> </i>in<i> </i>electrostatic<i> </i>atomization<i> </i>minimum<i> </i>quantity<i> </i>lubrication.<i> </i><i> Friction </i><b> 2023</b>,<i> 11,</i> 2107–2131. doi:10.1007/s40544-022-0734-2.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Prediction<i> </i>model<i> </i>of<i> </i>volume<i> </i>average<i> </i>diameter<i> </i>and<i> </i>analysis<i> </i>of<i> </i>atomization<i> </i>characteristics<i> </i>in<i> </i>electrostatic<i> </i>atomization<i> </i>minimum<i> </i>quantity<i> </i>lubrication.<i>&nbsp;</i>&publication_year=2023&author=Jia DZ&author=Li CH&author=Liu JH&author=Zhang YB&author=Yang M&author=Gao T&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B16'> <div class="references_sup">16.</div> <div class="references_name"> Cönger DB, Yapan YF, Emiroglu U, Uysal A, Altan E. Influence of singular and dual MQL nozzles on sustainable milling of Al6061-T651<i> </i>in<i> </i>different<i> </i>machining<i> </i>environments.<i> </i><i> J. Manuf. Process. </i><b> 2024</b>,<i> 109,</i> 524–536. doi:10.1016/j.jmapro.2023.12.043.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Influence of singular and dual MQL nozzles on sustainable milling of Al6061-T651<i> </i>in<i> </i>different<i> </i>machining<i> </i>environments.<i>&nbsp;</i>&publication_year=2024&author=Cönger DB&author=Yapan YF&author=Emiroglu U&author=Uysal A&author=Altan E. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B17'> <div class="references_sup">17.</div> <div class="references_name"> Krishnan GP, Raj DS. Machinability and tribological analysis of used cooking oil for MQL applications in drilling AISI 304<i> </i>using<i> </i>a<i> </i>low-cost<i> </i>pneumatic<i> </i>operated<i> </i>MQL<i> </i>system.<i> J. Manuf. Process. </i><b> 2023</b>,<i> 104,</i> 348–371. doi:10.1016/j.jmapro.2023.09.028.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Machinability and tribological analysis of used cooking oil for MQL applications in drilling AISI 304<i> </i>using<i> </i>a<i> </i>low-cost<i> </i>pneumatic<i> </i>operated<i> </i>MQL<i> </i>system.&publication_year=2023&author=Krishnan GP&author=Raj DS. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B18'> <div class="references_sup">18.</div> <div class="references_name"> Li JH, Shi WT, Lin YX, Li J, Liu S, Liu B. Comparative<i> </i>study<i> </i>on<i> </i>MQL<i> </i>milling<i> </i>and<i> </i>hole<i> </i>making<i> </i>processes<i> </i>for<i> </i>laser<i> </i>beam<i> </i>powder<i> </i>bed<i> </i>fusion<i> </i>(L-PBF)<i> </i>of<i> </i>Ti-6Al-4V<i> </i>titanium<i> </i>alloy.<i> </i><i> J. Manuf. Process.</i><b> 2023</b>,<i> 94,</i> 20–34. doi:10.1016/j.jmapro.2023.03.055.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Comparative<i> </i>study<i> </i>on<i> </i>MQL<i> </i>milling<i> </i>and<i> </i>hole<i> </i>making<i> </i>processes<i> </i>for<i> </i>laser<i> </i>beam<i> </i>powder<i> </i>bed<i> </i>fusion<i> </i>(L-PBF)<i> </i>of<i> </i>Ti-6Al-4V<i> </i>titanium<i> </i>alloy.<i>&nbsp;</i>&publication_year=2023&author=Li JH&author=Shi WT&author=Lin YX&author=Li J&author=Liu S&author=Liu B. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B19'> <div class="references_sup">19.</div> <div class="references_name"> Cui X, Li CH, Zhang YB, Said Z, Debnath S, Sharma S, et al. Grindability<i> </i>of<i> </i>titanium<i> </i>alloy<i> </i>using<i> </i>cryogenic<i> </i>nanolubricant<i> </i>minimum<i> </i>quantity<i> </i>lubrication.<i> </i><i> J. Manuf. Process. </i><b> 2022</b>,<i> 80,</i> 273–286. doi:10.1016/j.jmapro.2022.06.003.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Grindability<i> </i>of<i> </i>titanium<i> </i>alloy<i> </i>using<i> </i>cryogenic<i> </i>nanolubricant<i> </i>minimum<i> </i>quantity<i> </i>lubrication.<i>&nbsp;</i>&publication_year=2022&author=Cui X&author=Li CH&author=Zhang YB&author=Said Z&author=Debnath S&author=Sharma S&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B20'> <div class="references_sup">20.</div> <div class="references_name"> Khanna N, Agrawal C, Pimenov DY, Singla AK, Machado AR, da Silva LRR, et al. Review<i> </i>on<i> </i>design<i> </i>and<i> </i>development<i> </i>of<i> </i>cryogenic<i> </i>machining<i> </i>setups<i> </i>for<i> </i>heat<i> </i>resistant<i> </i>alloys<i> </i>and<i> </i>composites.<i> </i><i> J. Manuf. Process.</i><b> 2021</b>,<i> 68,</i> 398–422. doi:10.1016/j.jmapro.2021.05.053.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Review<i> </i>on<i> </i>design<i> </i>and<i> </i>development<i> </i>of<i> </i>cryogenic<i> </i>machining<i> </i>setups<i> </i>for<i> </i>heat<i> </i>resistant<i> </i>alloys<i> </i>and<i> </i>composites.<i>&nbsp;</i>&publication_year=2021&author=Khanna N&author=Agrawal C&author=Pimenov DY&author=Singla AK&author=Machado AR&author=da Silva LRR&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B21'> <div class="references_sup">21.</div> <div class="references_name"> Wang XM, Li CH, Zhang YB, Said Z, Debnath S, Sharma S, et al. Influence<i> </i>of<i> </i>texture<i> </i>shape<i> </i>and<i> </i>arrangement<i> </i>on<i> </i>nanofluid<i> </i>minimum<i> </i>quantity<i> </i>lubrication<i> </i>turning.<i> Int. J. Adv. Manuf. Tech. </i><b> 2022</b>,<i> 119,</i> 631–646. doi:10.1007/s00170-021-08235-4.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Influence<i> </i>of<i> </i>texture<i> </i>shape<i> </i>and<i> </i>arrangement<i> </i>on<i> </i>nanofluid<i> </i>minimum<i> </i>quantity<i> </i>lubrication<i> </i>turning.&publication_year=2022&author=Wang XM&author=Li CH&author=Zhang YB&author=Said Z&author=Debnath S&author=Sharma S&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B22'> <div class="references_sup">22.</div> <div class="references_name"> Kishore K, Chauhan SR, Sinha MK. A comprehensive investigation on eco-benign grindability improvement of Inconel 625<i> </i>using<i> </i>nano-MQL.<i> </i><i> Precis. Eng. </i><b> 2024</b>,<i> 90,</i> 81–95. doi:10.1016/j.precisioneng.2024.08.004.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=A comprehensive investigation on eco-benign grindability improvement of Inconel 625<i> </i>using<i> </i>nano-MQL.<i>&nbsp;</i>&publication_year=2024&author=Kishore K&author=Chauhan SR&author=Sinha MK. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B23'> <div class="references_sup">23.</div> <div class="references_name"> Li D, Duan Z, Hu X, Zhang D, Zhang Y. Automated<i> </i>classification<i> </i>and<i> </i>detection<i> </i>of<i> </i>multiple<i> </i>pavement<i> </i>distress<i> </i>images<i> </i>based<i> </i>on<i> </i>deep<i> </i>learning. <i> J. Traffic Transp. Eng.</i><b> 2023</b>,<i> 10,</i> 276–290. doi:10.1016/j.jtte.2021.04.008.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Automated<i> </i>classification<i> </i>and<i> </i>detection<i> </i>of<i> </i>multiple<i> </i>pavement<i> </i>distress<i> </i>images<i> </i>based<i> </i>on<i> </i>deep<i> </i>learning.&nbsp;&publication_year=2023&author=Li D&author=Duan Z&author=Hu X&author=Zhang D&author=Zhang Y. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B24'> <div class="references_sup">24.</div> <div class="references_name"> Lu HZ, Li CF, Chen WM, Jiang ZJ. <i> </i>A<i> </i>single<i> </i>shot<i> </i>multibox<i> </i>detector<i> </i>based<i> </i>on<i> </i>welding<i> </i>operation<i> </i>method<i> </i>for<i> </i>biometrics<i> </i>recognition<i> </i>in<i> </i>smart<i> </i>cities.<i> </i><i> Pattern Recogn. Lett.</i><b> 2020</b>,<i> 140,</i> 295–302. doi:10.1016/j.patrec.2020.10.016.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>A<i> </i>single<i> </i>shot<i> </i>multibox<i> </i>detector<i> </i>based<i> </i>on<i> </i>welding<i> </i>operation<i> </i>method<i> </i>for<i> </i>biometrics<i> </i>recognition<i> </i>in<i> </i>smart<i> </i>cities.<i>&nbsp;</i>&publication_year=2020&author=Lu HZ&author=Li CF&author=Chen WM&author=Jiang ZJ." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B25'> <div class="references_sup">25.</div> <div class="references_name"> Zhuang XL, Zhang TM. Detection<i> </i>of<i> </i>sick<i> </i>broilers<i> </i>by<i> </i>digital<i> </i>image<i> </i>processing<i> </i>and<i> </i>deep<i> </i>learning.<i> </i><i> Biosyst. Eng. </i><b> 2019</b>,<i> 179,</i> 106–116. doi:10.1016/j.biosystemseng.2019.01.003.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Detection<i> </i>of<i> </i>sick<i> </i>broilers<i> </i>by<i> </i>digital<i> </i>image<i> </i>processing<i> </i>and<i> </i>deep<i> </i>learning.<i>&nbsp;</i>&publication_year=2019&author=Zhuang XL&author=Zhang TM. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B26'> <div class="references_sup">26.</div> <div class="references_name"> Amemiya S, Takao H, Kato S, Yamashita H, Sakamoto N, Abe O. Automatic<i> </i>detection<i> </i>of<i> </i>brain<i> </i>metastases<i> </i>on<i> </i>contrast-enhanced<i> </i>CT<i> </i>with<i> </i>deep-learning<i> </i>feature-fused<i> </i>single-shot<i> </i>detectors.<i> </i><i> Eur. J. Radiol. </i><b> 2021</b>,<i> 136,</i> 109577. doi:10.1016/j.ejrad.2021.109577.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Automatic<i> </i>detection<i> </i>of<i> </i>brain<i> </i>metastases<i> </i>on<i> </i>contrast-enhanced<i> </i>CT<i> </i>with<i> </i>deep-learning<i> </i>feature-fused<i> </i>single-shot<i> </i>detectors.<i>&nbsp;</i>&publication_year=2021&author=Amemiya S&author=Takao H&author=Kato S&author=Yamashita H&author=Sakamoto N&author=Abe O. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B27'> <div class="references_sup">27.</div> <div class="references_name"> Liu JM, Prabuwono AS, Abulfaraj AW, Miniaoui S, Taheri N. Cognitive<i> </i>cloud<i> </i>framework<i> </i>for<i> </i>waste<i> </i>dumping<i> </i>analysis<i> </i>using<i> </i>deep<i> </i>learning<i> </i>vision<i> </i>computing<i> </i>in<i> </i>healthy<i> </i>environment.<i> </i><i> Comput. Electr. Eng.</i><b> 2023</b>,<i> 110,</i> 108814. doi:10.1016/j.compeleceng.2023.108814.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Cognitive<i> </i>cloud<i> </i>framework<i> </i>for<i> </i>waste<i> </i>dumping<i> </i>analysis<i> </i>using<i> </i>deep<i> </i>learning<i> </i>vision<i> </i>computing<i> </i>in<i> </i>healthy<i> </i>environment.<i>&nbsp;</i>&publication_year=2023&author=Liu JM&author=Prabuwono AS&author=Abulfaraj AW&author=Miniaoui S&author=Taheri N. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B28'> <div class="references_sup">28.</div> <div class="references_name"> Xie LF, Xiang X, Xu HN, Wang L, Lin LJ, Yin GF. <i> </i>FFCNN:<i> </i>A<i> </i>Deep<i> </i>Neural<i> </i>Network<i> </i>for<i> </i>Surface<i> </i>Defect<i> </i>Detection<i> </i>of<i> </i>Magnetic<i> </i>Tile.<i> </i><i> IEEE T Ind. Electron.</i><b> 2021</b>,<i> 68,</i> 3506–3516. doi:10.1109/Tie.2020.2982115.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>FFCNN:<i> </i>A<i> </i>Deep<i> </i>Neural<i> </i>Network<i> </i>for<i> </i>Surface<i> </i>Defect<i> </i>Detection<i> </i>of<i> </i>Magnetic<i> </i>Tile.<i> </i>&publication_year=2021&author=Xie LF&author=Xiang X&author=Xu HN&author=Wang L&author=Lin LJ&author=Yin GF." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B29'> <div class="references_sup">29.</div> <div class="references_name"> Tabernik D, Sela S, Skvarc J, Skocaj D. <i> </i>Segmentation-based<i> </i>deep-learning<i> </i>approach<i> </i>for<i> </i>surface-defect<i> </i>detection.<i> </i><i> J. Intell. Manuf.</i><b> 2020</b>,<i> 31,</i> 759–776. doi:10.1007/s10845-019-01476-x.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>Segmentation-based<i> </i>deep-learning<i> </i>approach<i> </i>for<i> </i>surface-defect<i> </i>detection.<i>&nbsp;</i>&publication_year=2020&author=Tabernik D&author=Sela S&author=Skvarc J&author=Skocaj D." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B30'> <div class="references_sup">30.</div> <div class="references_name"> Neuhauser FM, Bachmann G, Hora P. Surface<i> </i>defect<i> </i>classification<i> </i>and<i> </i>detection<i> </i>on<i> </i>extruded<i> </i>aluminum<i> </i>profiles<i> </i>using<i> </i>convolutional<i> </i>neural<i> </i>networks.<i> </i><i> Int. J. Mater. Form.</i><b> 2020</b>,<i> 13,</i> 591–603. doi:10.1007/s12289-019-01496-1.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Surface<i> </i>defect<i> </i>classification<i> </i>and<i> </i>detection<i> </i>on<i> </i>extruded<i> </i>aluminum<i> </i>profiles<i> </i>using<i> </i>convolutional<i> </i>neural<i> </i>networks.<i>&nbsp;</i>&publication_year=2020&author=Neuhauser FM&author=Bachmann G&author=Hora P. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B31'> <div class="references_sup">31.</div> <div class="references_name"> Le HF, Zhang LJ, Liu YX. <i> </i>Surface<i> </i>Defect<i> </i>Detection<i> </i>of<i> </i>Industrial<i> </i>Parts<i> </i>Based<i> </i>on<i> </i>YOLOv5.<i> </i><i> IEEE Access.</i><b> 2022</b>,<i> 10,</i> 130784–130794. doi:10.1109/Access.2022.3228687.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>Surface<i> </i>Defect<i> </i>Detection<i> </i>of<i> </i>Industrial<i> </i>Parts<i> </i>Based<i> </i>on<i> </i>YOLOv5.<i>&nbsp;</i>&publication_year=2022&author=Le HF&author=Zhang LJ&author=Liu YX." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B32'> <div class="references_sup">32.</div> <div class="references_name"> Niu SL, Peng YR, Li B, Qiu YH, Niu TZ, Li WF. <i> </i>A<i> </i>novel<i> </i>deep<i> </i>learning<i> </i>motivated<i> </i>data<i> </i>augmentation<i> </i>system<i> </i>based<i> </i>on<i> </i>defect<i> </i>segmentation<i> </i>requirements.<i> </i><i> J. Intell. Manuf.</i><b> 2024</b>,<i> 35,</i> 687–701. doi:10.1007/s10845-022-02068-y.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=<i>&nbsp;</i>A<i> </i>novel<i> </i>deep<i> </i>learning<i> </i>motivated<i> </i>data<i> </i>augmentation<i> </i>system<i> </i>based<i> </i>on<i> </i>defect<i> </i>segmentation<i> </i>requirements.<i>&nbsp;</i>&publication_year=2024&author=Niu SL&author=Peng YR&author=Li B&author=Qiu YH&author=Niu TZ&author=Li WF." target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B33'> <div class="references_sup">33.</div> <div class="references_name"> Yu RY, Guo BY, Yang K. Selective<i> </i>Prototype<i> </i>Network<i> </i>for<i> </i>Few-Shot<i> </i>Metal<i> </i>Surface<i> </i>Defect<i> </i>Segmentation.<i> </i><i> IEEE T Instrum. Meas.</i><b> 2022</b>,<i> 71,</i> 3196447. doi:10.1109/Tim.2022.3196447.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Selective<i> </i>Prototype<i> </i>Network<i> </i>for<i> </i>Few-Shot<i> </i>Metal<i> </i>Surface<i> </i>Defect<i> </i>Segmentation.<i>&nbsp;</i>&publication_year=2022&author=Yu RY&author=Guo BY&author=Yang K. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B34'> <div class="references_sup">34.</div> <div class="references_name"> Qian W, Zhu Z, Zhu C, Luo W, Zhu Y. Efficient<i> </i>deployment<i> </i>of<i> </i>Single<i> </i>Shot<i> </i>Multibox<i> </i>Detector<i> </i>network<i> </i>on<i> </i>FPGAs.<i> </i><i> Integration</i><b> 2024</b>,<i> 99,</i> 102255. doi:10.1016/j.vlsi.2024.102255.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Efficient<i> </i>deployment<i> </i>of<i> </i>Single<i> </i>Shot<i> </i>Multibox<i> </i>Detector<i> </i>network<i> </i>on<i> </i>FPGAs.<i> </i>&publication_year=2024&author=Qian W&author=Zhu Z&author=Zhu C&author=Luo W&author=Zhu Y. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B35'> <div class="references_sup">35.</div> <div class="references_name"> Cai J, Makita Y, Zheng Y, Takahashi S, Hao W, Nakatoh Y. Single<i> </i>shot<i> </i>multibox<i> </i>detector<i> </i>for<i> </i>honeybee<i> </i>detection.<i> </i><i> Comput. Electr. Eng.</i><b> 2022</b>,<i> 104,</i> 108465. doi:10.1016/j.compeleceng.2022.108465.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Single<i> </i>shot<i> </i>multibox<i> </i>detector<i> </i>for<i> </i>honeybee<i> </i>detection.<i>&nbsp;</i>&publication_year=2022&author=Cai J&author=Makita Y&author=Zheng Y&author=Takahashi S&author=Hao W&author=Nakatoh Y. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B36'> <div class="references_sup">36.</div> <div class="references_name"> Qiang J, Liu W, Li X, Guan P, Du Y, Liu B, et al. Detection<i> </i>of<i> </i>citrus<i> </i>pests<i> </i>in<i> </i>double<i> </i>backbone<i> </i>network<i> </i>based<i> </i>on<i> </i>single<i> </i>shot<i> </i>multibox<i> </i>detector.<i> </i><i> Comput. Electron. Agric.</i><b> 2023</b>,<i> 212,</i> 108158. doi:10.1016/j.compag.2023.108158.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Detection<i> </i>of<i> </i>citrus<i> </i>pests<i> </i>in<i> </i>double<i> </i>backbone<i> </i>network<i> </i>based<i> </i>on<i> </i>single<i> </i>shot<i> </i>multibox<i> </i>detector.<i>&nbsp;</i>&publication_year=2023&author=Qiang J&author=Liu W&author=Li X&author=Guan P&author=Du Y&author=Liu B&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B37'> <div class="references_sup">37.</div> <div class="references_name"> Zhu W, Zhang H, Eastwood J, Qi X, Jia J, Cao Y. Concrete<i> </i>crack<i> </i>detection<i> </i>using<i> </i>lightweight<i> </i>attention<i> </i>feature<i> </i>fusion<i> </i>single<i> </i>shot<i> </i>multibox<i> </i>detector.<i> </i><i> Knowl. -Based Syst. </i><b> 2023</b>,<i> 261,</i> 110216. doi:10.1016/j.knosys.2022.110216.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Concrete<i> </i>crack<i> </i>detection<i> </i>using<i> </i>lightweight<i> </i>attention<i> </i>feature<i> </i>fusion<i> </i>single<i> </i>shot<i> </i>multibox<i> </i>detector.<i>&nbsp;</i>&publication_year=2023&author=Zhu W&author=Zhang H&author=Eastwood J&author=Qi X&author=Jia J&author=Cao Y. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B38'> <div class="references_sup">38.</div> <div class="references_name"> Sun H, Xu H, Liu B, He D, He J, Zhang H, et al. MEAN-SSD:<i> </i>A<i> </i>novel<i> </i>real-time<i> </i>detector<i> </i>for<i> </i>apple<i> </i>leaf<i> </i>diseases<i> </i>using<i> </i>improved<i> </i>light-weight<i> </i>convolutional<i> </i>neural<i> </i>networks.<i> </i><i> Comput. Electron. Agric.</i><b> 2021</b>,<i> 189,</i> 106379. doi:10.1016/j.compag.2021.106379.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=MEAN-SSD:<i> </i>A<i> </i>novel<i> </i>real-time<i> </i>detector<i> </i>for<i> </i>apple<i> </i>leaf<i> </i>diseases<i> </i>using<i> </i>improved<i> </i>light-weight<i> </i>convolutional<i> </i>neural<i> </i>networks.<i>&nbsp;</i>&publication_year=2021&author=Sun H&author=Xu H&author=Liu B&author=He D&author=He J&author=Zhang H&author=et al. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B39'> <div class="references_sup">39.</div> <div class="references_name"> Shen YF, Zhou HL, Li JT, Jian FJ, Jayas DS. Detection<i> </i>of<i> </i>stored-grain<i> </i>insects<i> </i>using<i> </i>deep<i> </i>learning.<i> Comput. Electron. Agr.</i><b> 2018</b>,<i> 145,</i> 319–325. doi:10.1016/j.compag.2017.11.039.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Detection<i> </i>of<i> </i>stored-grain<i> </i>insects<i> </i>using<i> </i>deep<i> </i>learning.&publication_year=2018&author=Shen YF&author=Zhou HL&author=Li JT&author=Jian FJ&author=Jayas DS. " target="_blank">Google Scholar</a>]</span> </div> </div> <div class="references html-x" id='B40'> <div class="references_sup">40.</div> <div class="references_name"> Erhan<i> </i>D,<i> </i>Szegedy<i> </i>C,<i> </i>Toshev<i> </i>A,<i> </i>Anguelov<i> </i>D.<i> </i>Scalable<i> </i>Object<i> </i>Detection<i> </i>Using<i> </i>Deep<i> </i>Neural<i> </i>Networks. IEEE. 2013. Available online: https://arxiv.org/pdf/1312.2249 (accessed on 21 August 2024). </div> </div> <div class="references html-x" id='B41'> <div class="references_sup">41.</div> <div class="references_name"> Howard<i> </i>AG,<i> </i>Zhu<i> </i>M,<i> </i>Chen<i> </i>B,<i> </i>Kalenichenko<i> </i>D,<i> </i>Wang<i> </i>W,<i> </i>Weyand<i> </i>T,<i> </i>et al.<i> </i>MobileNets:<i> </i>Efficient<i> </i>Convolutional<i> </i>Neural<i> </i>Networks<i> </i>for<i> </i>Mobile<i> </i>Vision<i> </i>Applications. <i>arXiv</i><i> </i><b>2017</b>, arXiv:1704.04861. </div> </div> <div class="references html-x" id='B42'> <div class="references_sup">42.</div> <div class="references_name"> Yu<i> </i>F,<i> </i>Koltun<i> </i>V,<i> </i>Funkhouser<i> </i>T.<i> </i>Dilated<i> </i>Residual<i> </i>Networks. In Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, USA, 21<b>–</b>26 July<i> </i>2017.<i> </i>doi:10.48550/arXiv.1705.09914. </div> </div> <div class="references html-x" id='B43'> <div class="references_sup">43.</div> <div class="references_name"> Ioffe S, Szegedy C. Batch<i> </i>Normalization:<i> </i>Accelerating<i> </i>Deep<i> </i>Network<i> </i>Training<i> </i>by<i> </i>Reducing<i> </i>Internal<i> </i>Covariate<i> </i>Shift.<i> </i><i> Proc. Mach. Learn. Res. </i><b> 2015</b>,<i> 37,</i> 448–456. doi:10.48550/arXiv.1502.03167.<span class="tages-text">[<a href="https://scholar.google.com/scholar_lookup?title=Batch<i> </i>Normalization:<i> </i>Accelerating<i> </i>Deep<i> </i>Network<i> </i>Training<i> </i>by<i> </i>Reducing<i> </i>Internal<i> </i>Covariate<i> </i>Shift.<i>&nbsp;</i>&publication_year=2015&author=Ioffe S&author=Szegedy C. " target="_blank">Google Scholar</a>]</span> </div> </div> </div> </div> </div> </div> </div> </div> </div> </div> </section> <div class="modal fade cite-modal" id="citeModal" tabindex="-1" aria-labelledby="exampleModalLabel" aria-hidden="true"> <div class="modal-dialog modal-dialog-centered"> <div class="modal-content border-0"> <div class="modal-body pt-5 pl-5 pr-5 pb-4"> <h3 class="pb-2 border-bottom border-dark">Cite This Article</h3> <h6 class="mt-4 d-flex justify-content-between"> SCIEPublish Style </h6> <p class="modal-text2 text-justify">Duan Z, Xi , Wang S, Wang Z, Bian P, Li C, et al. Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector. <i>Intelligent and Sustainable Manufacturing</i> <b>2024</b>, <i>1</i>, 10020. https://doi.org/10.70322/ism.2024.10020</p> <h6 class="mt-3 d-flex justify-content-between text-justify"> AMA Style </h6> <p class="modal-text2 text-justify">Duan Z, Xi , Wang S, Wang Z, Bian P, Li C, et al. Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector. <i>Intelligent and Sustainable Manufacturing</i>. 2024; 1(2):10020. https://doi.org/10.70322/ism.2024.10020</p> </div> <button type="button" class="btn btn-close" data-dismiss="modal"> <svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" fill="currentColor" class="bi bi-x-lg" viewBox="0 0 16 16"> <path d="M2.146 2.854a.5.5 0 1 1 .708-.708L8 7.293l5.146-5.147a.5.5 0 0 1 .708.708L8.707 8l5.147 5.146a.5.5 0 0 1-.708.708L8 8.707l-5.146 5.147a.5.5 0 0 1-.708-.708L7.293 8 2.146 2.854Z"></path> </svg> </button> </div> </div> </div> <div class="modal modal-amplification-img" id="imgModal" tabindex="-1" aria-labelledby="imgModal" aria-hidden="true"> <div class="modal-dialog modal-xl modal-dialog-centered modal-dialog-scrollable modal-dialog-img"> <div class="modal-content"> <button type="button" class="btn btn-close" data-dismiss="modal"> <svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="currentColor" class="bi bi-x-lg" viewBox="0 0 16 16"> <path d="M2.146 2.854a.5.5 0 1 1 .708-.708L8 7.293l5.146-5.147a.5.5 0 0 1 .708.708L8.707 8l5.147 5.146a.5.5 0 0 1-.708.708L8 8.707l-5.146 5.147a.5.5 0 0 1-.708-.708L7.293 8 2.146 2.854Z"></path> </svg> </button> <div class="modal-body"> </div> </div> </div> </div> <link rel="stylesheet" href="/static/css/share.min.css?3" /> <script src="/static/js/social-share.min.js"></script> <script> var $config = { image:"https://www.sciepublish.com/uploads/2024/11/26/da4f3c9e4878d009d85bec611226ea24.png", origin:'', wechatQrcodeTitle: "WeChat Share", // WeChat QR code prompt text wechatQrcodeHelper: '<p>Scan QR code</p> <p>This article can be shared to your WeChat.</p>', }; socialShare('.social-share', $config); </script> <script> document.addEventListener("DOMContentLoaded", function() { /**/ var stopScroll = $(document).scrollTop(), column = $('.left-content'), fixedTitle = $('.fixed-title'), acquiesceTitle = $('.acquiesce-title'), acquiesceTitleTop = acquiesceTitle.offset().top; $(window).scroll(function(){ var scrollTop = $(this).scrollTop(); if(scrollTop > acquiesceTitleTop){ column.addClass('fixed'); fixedTitle.addClass('fixed'); }else{ column.removeClass('fixed'); fixedTitle.removeClass('fixed'); } }) if (stopScroll > acquiesceTitleTop ){ column.addClass('fixed'); fixedTitle.addClass('fixed'); } //Anchor navigation $('body').scrollspy({ target: '#article-menu-float' }) //tab switch $('.information-collapse').each(function(){ $(this).click(function(){ $(this).toggleClass('active').siblings().removeClass('active'); $(".information-item").eq($(this).index()).slideToggle().siblings().slideUp(); }) }) function markdownReplace(text) { // Replace the oblique body, bold it to HTML // /article/pii/21 text = text.replace(/</g, '<'); //article/pii/166 text = text.replace(/>/g, '>'); // text = text.replace(/\*\*(.+?)\*\*/g, "<b>$1</b>"); text = text.replace(/\*(.+?)\*/g, "<i>$1</i>"); // article/pii/18 text = text.replace(/&/g, '&');//* /article/pii/181#Figure5 中的* // text = text.replace(/\\\$/g, '$'); //article/pii/19 return text; } // Treatment of non -serial list function replaceConsecutiveBulletsWithList(text) { // Regular expression matching continuously with • starting line const regex = /^(?:•\s.*\n?)+/gm; // Replace the matching line, convert it to <ul> and <li> elements return text.replace(regex, (match) => { // Remove the • and the subsequent blank characters at the beginning of each line, and convert it to <li> element const listItems = match.split('\n').map(line => { return line.replace(/^•\s/, '').trim() ? `<li>${line.replace(/^•\s/, '').trim()}</li>` : ''; }).join(''); // If there is a list item, the package is wrapped in the <ul> label return `<ul>${listItems}</ul>`; }); } // Pre -predetermined formula rendering global indexes window.formula = {}; // Treatment to find a third -level and four -level title, sort out the P label that is not available in the text document.querySelectorAll('.article-item-text').forEach(element => { element.innerHTML = replaceConsecutiveBulletsWithList(element.innerHTML); // Forwards <em> 3.1. Serine Integrase-Mediated DNA Assembly Between Two Plasmids </em> element.innerHTML = element.innerHTML.replace(/(<h[1-6].+?>)/gi, '\r\n$1'); //TODO There is a change in the middle of the H tag to cause problems element.innerHTML = element.innerHTML.replace(/(<\/h[1-6]\s*>)/gi, '$1\r\n'); element.innerHTML = element.innerHTML.replace(/<(em|i)>(\d+\.\d+\..+?)<\/(em|i)>/gi, '\r\n<h4 class="value4">$2</h4>\r\n'); // Processing replacement of level 4 title element.innerHTML = element.innerHTML.replace(/>\s*(\d+\.\d+\.\d+\.\s*.*?)(?=\s*\n|$)/g, '>\r\n<h5 class="value5">$1</h5>\r\n'); element.innerHTML = element.innerHTML.replace(/^\s*(\d+\.\d+\.\d+\.\s*.*?)(?=\s*\n|$)/gm, '\r\n<h5 class="value5">$1</h5>\r\n'); // Processing replacement three -level title element.innerHTML = element.innerHTML.replace(/\*+(\d+\.\d+\..+?)\*+\s+/g, '\r\n<h4 class="value4">$1</h4>\r\n'); // Treatment P tag element.innerHTML = element.innerHTML.replace(/<p.+?>(.+?)<\/p>/gi, '\r\n$1\r\n'); let lastP = document.createElement('p'); Array.from(element.childNodes).forEach(node=> { if (node.childElementCount == 0 && node.textContent.trim().length == 0){ // Delete the empty node element.removeChild(node); } // Handle the title if (/H[1-6]/i.test(node.nodeName)){ if (lastP.childElementCount > 0 || lastP.textContent.trim().length > 0){ // Lastp is not an empty node. element.appendChild(lastP); lastP = document.createElement('p'); } element.appendChild(node); return; } // Check the HTML-Disp-Formula-Info node if (node.classList && node.classList.contains('html-disp-formula-info')){ if (lastP.childElementCount > 0 || lastP.textContent.trim().length > 0){ element.appendChild(lastP); lastP = document.createElement('p'); } element.appendChild(node); // Treatment of the text formula in the chapter renderMathInElement(node, { delimiters: [ {left: '```latex', right: '```', display: true}, ], throwOnError : false }); const label = node.querySelector('.formula > label'); if (label === null){ return; } var numMatch = label.textContent.match(/\d+/g) if (!numMatch){ return; } const html_pics = node.querySelector('.html_pics') if (html_pics){ formula["Equation"+numMatch[0]]=html_pics.innerHTML; } return; } // Treatment textbooks if (node.nodeType === 3) { var lines = node.nodeValue.split(/\r?\n/); if (lines.length == 1){ // No change, it may be the content of the previous P tag if (node.nodeValue.trim().length > 0){ lastP.appendChild(node); } return; } lines.forEach((line, index)=> { if (index >0){ // Other changes may be the new P tag, the old P label is moved in if (lastP.childElementCount > 0 || lastP.textContent.trim().length > 0){ element.appendChild(lastP); lastP = document.createElement('p'); } } if (line.trim().length > 0){ lastP.appendChild(document.createTextNode(line)); } }); element.removeChild(node); return; } if (node.classList.contains('html-fig-wrap')){ if (lastP.childElementCount > 0 || lastP.textContent.trim().length > 0){ element.appendChild(lastP); lastP = document.createElement('p'); } } //Other situations are the content of the previous P tag lastP.appendChild(node); }); if (lastP.childElementCount > 0 || lastP.textContent.trim().length > 0){ element.appendChild(lastP); } if (element.childElementCount == 0 && element.textContent.trim().length == 0){ // Delete the empty node element.parentNode.removeChild(element); } function capitalizeFirstLetter(str) { return str.charAt(0).toUpperCase() + str.slice(1); } // Find the picture form formula without setting the A tag, and give a tag element.innerHTML = element.innerHTML.replace(/(?<!<a.+?>\s*|[a-zA-Z])(Figure|Table|Scheme|Equation)\s+($([A-Z]?\d+)$|([A-Z]?\d+))(?<!\s*<\/a>)/gi, function(match, $2,$3) { match = capitalizeFirstLetter(match); return '<a href="#'+ capitalizeFirstLetter($2) + $3.replace(/$|$/g,'') +'">'+ match +'</a>'; }); // Remove Markdown element.innerHTML = markdownReplace(element.innerHTML); // Find a super -text link without setting A tags, give a tagnk without setting A tags, give a tagnk without setting A tags, give a tag element.innerHTML = element.innerHTML.replace(/(?<!<a.+?>\s*)https?:\/\/[^\s\/]+(\/[^\s<.]*)*(?<!\s*<\/a>)/gi, function(match) { return '<a href="'+ match.replace(/[.,，。;\s]+$/, '') +'" target="_blank">'+ match +'</a>'; } ); // Add a box to the formula quotation element.querySelectorAll('a[href^="#Equation"]').forEach(auchor=>{ const id= auchor.getAttribute('href').slice(1); if (formula[id]){ $(auchor).tooltip({ html: true , delay: { "show": 200, "hide": 200 },title:$('<div class="formulaTip">').html(formula[id])}); } }); // Handling in -line formula renderMathInElement(element, { delimiters: [ {left: '$$', right: '$$', display: false}, ], throwOnError : false }); }); document.querySelectorAll("div.articleneir.abstract").forEach(element => { renderMathInElement(element, { delimiters: [ {left: '$$', right: '$$', display: false}, ], throwOnError : false }); }); // Processing reference document.querySelectorAll(".html-bibr").forEach(link => { link.removeAttribute('title'); const target = document.querySelector(link.getAttribute('href')+'.references > .references_name'); if (target){ $(link).tooltip({ html: true , delay: { "show": 200, "hide": 200 },title:target.innerHTML}); } }); document.querySelectorAll(".references").forEach(ref => { if (ref.id){ const a = document.createElement("a"); a.name=ref.id; a.classList.add("anchor-abs-66"); ref.removeAttribute("id"); ref.insertBefore(a, ref.firstChild); } }); // Treat the display area of the formula formula of the picture form, remove the A tag, set the anchor point document.querySelectorAll('.html-fig-wrap .html-fig_description,.html-fig-wrap .html-table_wrap_discription').forEach(element => { element.innerHTML = markdownReplace(element.innerHTML); const match = element.firstChild.textContent.match(/(Figure|Table|Scheme) [A-Z]?\d+/g); if (match){ element.firstChild.textContent = element.firstChild.textContent let id= match[0].replace(" ","") const target = document.createElement("a"); target.name=id; target.classList.add("anchor"); element.parentNode.insertBefore(target, element.parentNode.firstChild); document.querySelectorAll("a[href='#"+id+"']").forEach(auchor => { $(auchor).tooltip({ html: true , delay: { "show": 1000, "hide": 100 },title:$(element.parentNode).html()}); }); } }); // Processing keywords const keywordsContainer = document.querySelector('.keywords') if (keywordsContainer){ let title; while(child = keywordsContainer.firstChild){ // Take out the first non -empty element in Keywords if (child.textContent.trim().length>0){ title = child; title.appendChild(document.createTextNode(' ')); keywordsContainer.removeChild(child); break; } keywordsContainer.removeChild(child); } let listLinks = []; let link = document.createElement('a'); Array.from(keywordsContainer.childNodes).forEach(node=> { // Treatment text nodes if (node.nodeType === 3) { var ks = node.nodeValue.split(/(?<!&#(?:x[0-9a-fA-F]{1,5}|\d{1,6}))[;]|,/); if (ks.length == 1){ // There is no separatist, no new link is created if (link.lastChild && link.lastChild.nodeType === 3){ //Condented text nodes link.lastChild.nodeValue += ks[0]; }else{ link.appendChild(node); } return; }else{ ks.forEach((k, index)=> { if (index==0){ // In the first paragraph, do not create a new LINK, and the merger with the previous merger if (link.lastChild && link.lastChild.nodeType === 3){ //Condented text nodes link.lastChild.nodeValue += k; return; } }else{ // Means encounter; create a new link listLinks.push(link); link = document.createElement('a'); } link.appendChild(document.createTextNode(k.trim())); }); return; } }else{ link.appendChild(node); } }); if (link.innerText.trim().length>0){ listLinks.push(link); } keywordsContainer.innerHTML = ""; keywordsContainer.appendChild(title); listLinks.forEach((link, index, array)=> { link.href = "/index/search/index.html?search="+link.innerHTML.trim().replace(/ /g, '+');// TODO :Angle/lowering/small capitalization keywordsContainer.appendChild(link); if (index !== array.length - 1) { keywordsContainer.appendChild(document.createTextNode(', ')); } }); } // Treatment reference copy document.querySelectorAll('.cite-modal h6 ').forEach(h6 => { let a = document.createElement('a'); let h6Text = h6.textContent.trim(); h6.appendChild(a); a.href = "#"; a.textContent = "Copy"; a.onclick = function(e) { if (h6.nextElementSibling) { navigator.clipboard.writeText(h6.nextElementSibling.textContent.replace(/\r\n|\n|\r/g, '')).then(function() { $('#citeModal').modal('hide'); // showToast(h6Text + ' copied to clipboard!'); }, function() { showToast('Failed to copy '+h6Text+' to clipboard.'); }); } e.preventDefault(); }; }); //Picture amplification var $imgModal = $('#imgModal'), $imgModalBody = $imgModal.find('.modal-body'); $(".html-figpopup-link").each(function(){ var dataSrc = $(this).children('img').attr('data-large'); $(this).attr({ 'data-toggle': 'modal', 'data-target': '#imgModal' }) $(this).on('click', function () { $imgModalBody.append(`<img src="`+dataSrc+`" class="amplification-img">`); }) }); $imgModal.on('hidden.bs.modal', function (event) { $imgModalBody.children('img').remove(); }) //The picture is enlarged again $imgModal.on('shown.bs.modal', function (event) { $imgModalBody.children('img').on('click', function () { $(this).parent().parent().toggleClass('toggle-img'); }) }) // Process doi // Doi reference the magazine $.ajax({ type: 'GET', url: '/api/article/doi/id/349', dataType:"json", success: function(data){ var doiEL =""; $.each(data.doi_list, function(k, v) { doiEL += "<p>" + v.title + "." + " <em>" + v.source + "</em>" + " " + "<b>" + v.year + "</b>. " + "[" + "<a target='_blank' href='http://dx.doi.org/" + v.doi + "'>CrossRef</a>" + "]" +"</p>"; }) if (data.doi_count > 0) { $("[data-bind='popover1'] > b").text(data.doi_count); $("[data-bind='popover1']").addClass('d-flex'); $("[data-bind='popover1']").popover({ trigger: 'focus', html: true, delay: { "show": 100, "hide": 500 }, content: function(){ return doiEL; }, }); } }, timeout: 5000, error: function(jqXHR,textStatus){ } }) }) function showToast(message) { var toast = document.createElement('div'); toast.classList.add('toast', 'centered-div', 'show'); toast.setAttribute('role', 'alert'); toast.setAttribute('aria-live', 'assertive'); toast.setAttribute('aria-atomic', 'true'); var toastBody = document.createElement('div'); toastBody.classList.add('toast-body'); toastBody.textContent = message; toast.appendChild(toastBody); document.body.appendChild(toast); setTimeout(function () { toast.classList.remove('show'); toast.classList.add('hide'); setTimeout(function () { document.body.removeChild(toast); }, 500); }, 3000); } </script>  <footer class="mt-1"> <div class="my-body-container pt-4 pb-4"> <div class="fotter-top"> <div> <div class="item-text"> <h3>About</h3> <ul> <li> <a href="/About_SCIEPublish" title="About SClEPublish">About SClEPublish</a> </li> <li> <a href="/Management_Team" title="Management Team">Management Team</a> </li> <li> <a href="/Careers" title="Careers">Careers</a> </li> <li> <a href="/Contact" title="Contact">Contact</a> </li> </ul> </div> <div class="item-text"> <h3>Policies</h3> <ul> <li> <a href="/Peer_Review_Policy" title="Peer Review Policy">Peer Review Policy</a> </li> <li> <a href="/Open_Access_Policy" title="Open Access Policy">Open Access Policy</a> </li> <li> <a href="/Licensing_and_Copyright" title="Licensing and Copyright">Licensing and Copyright</a> </li> <li> <a href="/Editorial_Policy" title="Editorial Policy">Editorial Policy</a> </li> <li> <a href="/Advertising_Policy" title="Advertising Policy">Advertising Policy</a> </li> </ul> </div> <div class="item-text"> <h3>Information</h3> <ul> <li> <a href="/For_Authors" title="For Authors">For Authors</a> </li> <li> <a href="/For_Reviewers" title="For Reviewers">For Reviewers</a> </li> <li> <a href="/For_Editors" title="For Editors">For Editors</a> </li> </ul> </div> <div class="item-text item-text-membership"> <h3>A Member of</h3> <ul> <li class="membership-img1"> <a href="https://stm-assoc.org/" target="_blank" rel="nofollow"> <img src="/style/image/stm2024-logo.png"> </a> </li> <li class="membership-img2"> <a href="https://www.alpsp.org/" target="_blank" rel="nofollow" title=""> <img src="/style/image/alpsp-logo.png" alt=""> </a> </li>  </ul> </div> </div> <div class="item-text item-text-right"> <h3> <svg xmlns="http://www.w3.org/2000/svg" class="navbar-logo" xml:space="preserve" version="1.0" viewBox="0 0 5.08 1.933"> <path d="M1.021 1.245a.29.29 0 0 1-.211-.054l-.027-.023-.003-.003.056-.066.003.004a.3.3 0 0 0 .043.033.2.2 0 0 0 .128.027l.024-.007.019-.01a.07.07 0 0 0 .022-.032.1.1 0 0 0 0-.036l-.004-.014a.1.1 0 0 0-.016-.02.1.1 0 0 0-.027-.017L.994 1.01.919.98a.3.3 0 0 1-.076-.05.14.14 0 0 1-.034-.067.2.2 0 0 1 0-.06.13.13 0 0 1 .027-.056.2.2 0 0 1 .049-.041A.2.2 0 0 1 .95.683a.3.3 0 0 1 .07-.001.2.2 0 0 1 .06.017.3.3 0 0 1 .075.05l.003.003-.05.061L1.103.81a.2.2 0 0 0-.053-.034.2.2 0 0 0-.063-.013.1.1 0 0 0-.046.008L.925.78a.06.06 0 0 0-.023.047l.001.015.006.012a.1.1 0 0 0 .018.02.1.1 0 0 0 .027.017L.97.899l.016.006.074.032a.3.3 0 0 1 .058.033.14.14 0 0 1 .051.08.2.2 0 0 1 0 .07.15.15 0 0 1-.048.081.2.2 0 0 1-.062.035zm.527-.002a.24.24 0 0 1-.1 0 .23.23 0 0 1-.125-.069.2.2 0 0 1-.051-.089.3.3 0 0 1-.019-.119.4.4 0 0 1 .02-.12.3.3 0 0 1 .052-.09.23.23 0 0 1 .176-.076.3.3 0 0 1 .064.01.3.3 0 0 1 .053.025.2.2 0 0 1 .04.034l.002.003-.052.061L1.605.81A.2.2 0 0 0 1.56.776a.2.2 0 0 0-.037-.012.15.15 0 0 0-.082.013.13.13 0 0 0-.049.039l-.018.029a.2.2 0 0 0-.022.073.4.4 0 0 0 .002.104.2.2 0 0 0 .014.05.2.2 0 0 0 .023.04.14.14 0 0 0 .066.047.15.15 0 0 0 .107-.009l.028-.017.025-.023.003-.004.051.06-.002.003a.3.3 0 0 1-.076.059.2.2 0 0 1-.045.015m.314-.004h-.09V.69h.095v.549zm.485 0h-.331V.69h.328v.08H2.11v.141h.198v.082H2.11v.165h.242v.08zm.215 0h-.09V.69h.169a.4.4 0 0 1 .083.008L2.76.71l.031.016a.13.13 0 0 1 .044.053q.009.015.012.036a.22.22 0 0 1-.012.121l-.018.03a.176.176 0 0 1-.09.057.3.3 0 0 1-.084.011h-.076v.205zm.005-.472v.19h.068a.2.2 0 0 0 .056-.006.1.1 0 0 0 .038-.018.1.1 0 0 0 .022-.031.1.1 0 0 0 .007-.045l-.003-.03a.07.07 0 0 0-.028-.04L2.703.776 2.671.769 2.633.767zm.539.48a.2.2 0 0 1-.067-.003.1.1 0 0 1-.075-.07.3.3 0 0 1-.013-.09V.827h.093v.248a.3.3 0 0 0 .007.055l.008.017.012.012.016.007.02.002a.1.1 0 0 0 .033-.006.1.1 0 0 0 .03-.019.2.2 0 0 0 .03-.032V.826h.093v.413h-.078l-.006-.057a.2.2 0 0 1-.078.057zm.548-.002-.034.003-.03-.003-.029-.01A.2.2 0 0 1 3.51 1.2l-.007.039h-.075V.645h.093q0 .11-.002.219l.01-.008A.2.2 0 0 1 3.59.823a.2.2 0 0 1 .043-.007.2.2 0 0 1 .07.015.15.15 0 0 1 .07.074.2.2 0 0 1 .022.076.4.4 0 0 1 0 .095.3.3 0 0 1-.029.082.2.2 0 0 1-.079.075zm-.07-.077a.1.1 0 0 0 .046-.002.1.1 0 0 0 .043-.032l.015-.028a.2.2 0 0 0 .01-.036.3.3 0 0 0-.006-.114A.1.1 0 0 0 3.68.93.07.07 0 0 0 3.64.9.1.1 0 0 0 3.59.899l-.023.008-.023.015-.023.02v.193a.2.2 0 0 0 .043.027zm.424.08h-.015a.1.1 0 0 1-.051-.013l-.017-.016-.011-.022-.007-.026-.002-.031V.645h.093v.5L4 1.16l.003.004.003.003.008.002h.008l.005-.001h.004l.013.071-.004.002-.009.002zm.22-.01H4.14V.827h.093v.413zm-.026-.48L4.187.76q-.009 0-.016-.002L4.157.753a.05.05 0 0 1-.02-.02.1.1 0 0 1-.008-.027L4.13.69a.05.05 0 0 1 .027-.032L4.17.653a.07.07 0 0 1 .045.005.05.05 0 0 1 .03.048.1.1 0 0 1-.009.028.1.1 0 0 1-.02.019zm.315.488a.25.25 0 0 1-.19-.054l-.004-.003.045-.062.004.003a.3.3 0 0 0 .053.034.13.13 0 0 0 .058.012l.022-.002.016-.005.013-.008.009-.01a.05.05 0 0 0 .007-.025q0-.006-.002-.012l-.005-.01-.008-.009-.011-.008-.028-.014-.016-.006-.017-.007a.4.4 0 0 1-.079-.041.1.1 0 0 1-.028-.034L4.348.964 4.345.939a.12.12 0 0 1 .023-.07.1.1 0 0 1 .038-.033A.2.2 0 0 1 4.46.82a.2.2 0 0 1 .13.022l.037.024.003.003-.045.059-.003-.003A.2.2 0 0 0 4.54.9a.1.1 0 0 0-.065-.008.1.1 0 0 0-.027.012l-.007.01a.04.04 0 0 0-.007.022q0 .006.002.01l.005.01a.1.1 0 0 0 .017.015l.027.013.016.006.016.006.063.028.019.013a.1.1 0 0 1 .029.035l.008.023a.13.13 0 0 1-.008.077.1.1 0 0 1-.03.04.14.14 0 0 1-.05.027zm.307-.007h-.088V.645h.092q0 .115-.002.229a.3.3 0 0 1 .05-.038.2.2 0 0 1 .048-.017.2.2 0 0 1 .068.002.1.1 0 0 1 .057.038q.01.014.018.033A.314.314 0 0 1 5.08.98v.258h-.093V.99L4.985.96 4.98.936 4.97.92 4.96.907 4.943.9a.1.1 0 0 0-.037 0 .1.1 0 0 0-.03.011l-.016.01-.032.03v.288z" /> <path d="M1.844 1.377a.97.97 0 0 1-.878.563A.963.963 0 0 1 0 .974.964.964 0 0 1 .966.007a.96.96 0 0 1 .865.536l-.048.024A.92.92 0 0 0 .966.062a.91.91 0 0 0-.912.912.91.91 0 0 0 .912.912.91.91 0 0 0 .83-.532z" class="logo-circle" /> </svg> </h3> <div class="text-share"> <a href="https://x.com/SCIEPublish" class="share-btn twitter-btn" title="Share on Twitter" target="_blank"> <svg t="1713929395475" class="icon" width="32" height="32" viewBox="0 0 1399 1024" fill="#000000" version="1.1" xmlns="http://www.w3.org/2000/svg" p-id="8742"> <path d="M282.021569 119.281213l323.998199 433.216256-326.044203 352.222995h73.379441l285.451142-308.376452 230.636674 308.376452h249.713149l-342.227762-457.583832 303.479459-327.855419h-73.379441l-262.886398 284.008876-212.407111-284.008876h-249.713149z m107.909957 54.051408h114.71879l506.578928 677.328049h-114.718791l-506.578927-677.328049z" p-id="8743"></path> </svg> </a> <a href="" class="share-btn LinkedIn-btn" id="email-LinkedIn-btn" title="Share on LinkedIn" target="_blank"> <svg xmlns="http://www.w3.org/2000/svg" width="24" height="24" fill="#000000" class="bi bi-linkedin" viewBox="0 0 16 16"> <path d="M0 1.146C0 .513.526 0 1.175 0h13.65C15.474 0 16 .513 16 1.146v13.708c0 .633-.526 1.146-1.175 1.146H1.175C.526 16 0 15.487 0 14.854V1.146zm4.943 12.248V6.169H2.542v7.225h2.401zm-1.2-8.212c.837 0 1.358-.554 1.358-1.248-.015-.709-.52-1.248-1.342-1.248-.822 0-1.359.54-1.359 1.248 0 .694.521 1.248 1.327 1.248h.016zm4.908 8.212V9.359c0-.216.016-.432.08-.586.173-.431.568-.878 1.232-.878.869 0 1.216.662 1.216 1.634v3.865h2.401V9.25c0-2.22-1.184-3.252-2.764-3.252-1.274 0-1.845.7-2.165 1.193v.025h-.016a5.54 5.54 0 0 1 .016-.025V6.169h-2.4c.03.678 0 7.225 0 7.225h2.4z"/> </svg> </a> <a href="javaScript:void(0)" class="share-btn wechat-btn" id="wechat-share-btn"> <svg xmlns="http://www.w3.org/2000/svg" width="26" height="26" fill="#000000" class="bi bi-wechat" viewBox="0 0 16 16"> <path d="M11.176 14.429c-2.665 0-4.826-1.8-4.826-4.018 0-2.22 2.159-4.02 4.824-4.02S16 8.191 16 10.411c0 1.21-.65 2.301-1.666 3.036a.324.324 0 0 0-.12.366l.218.81a.616.616 0 0 1 .029.117.166.166 0 0 1-.162.162.177.177 0 0 1-.092-.03l-1.057-.61a.519.519 0 0 0-.256-.074.509.509 0 0 0-.142.021 5.668 5.668 0 0 1-1.576.22ZM9.064 9.542a.647.647 0 1 0 .557-1 .645.645 0 0 0-.646.647.615.615 0 0 0 .09.353Zm3.232.001a.646.646 0 1 0 .546-1 .645.645 0 0 0-.644.644.627.627 0 0 0 .098.356Z"/> <path d="M0 6.826c0 1.455.781 2.765 2.001 3.656a.385.385 0 0 1 .143.439l-.161.6-.1.373a.499.499 0 0 0-.032.14.192.192 0 0 0 .193.193c.039 0 .077-.01.111-.029l1.268-.733a.622.622 0 0 1 .308-.088c.058 0 .116.009.171.025a6.83 6.83 0 0 0 1.625.26 4.45 4.45 0 0 1-.177-1.251c0-2.936 2.785-5.02 5.824-5.02.05 0 .1 0 .15.002C10.587 3.429 8.392 2 5.796 2 2.596 2 0 4.16 0 6.826Zm4.632-1.555a.77.77 0 1 1-1.54 0 .77.77 0 0 1 1.54 0Zm3.875 0a.77.77 0 1 1-1.54 0 .77.77 0 0 1 1.54 0Z"/> </svg> <div class="cord"> <div class="img"> <img src="/style/image/wechat.jpg" alt="SCIEPublish wechat" /> </div> </div> </a> <a href="mailto:office@sciepublish.org" class="share-btn email-btn" id="email-share-btn" title="Share on Email" target="_blank"> <svg xmlns="http://www.w3.org/2000/svg" width="26" height="26" fill="#000000" class="bi bi-envelope-fill" viewBox="0 0 16 16"> <path d="M.05 3.555A2 2 0 0 1 2 2h12a2 2 0 0 1 1.95 1.555L8 8.414.05 3.555ZM0 4.697v7.104l5.803-3.558L0 4.697ZM6.761 8.83l-6.57 4.027A2 2 0 0 0 2 14h12a2 2 0 0 0 1.808-1.144l-6.57-4.027L8 9.586l-1.239-.757Zm3.436-.586L16 11.801V4.697l-5.803 3.546Z"/> </svg> </a> </div> </div> </div> <div class="copyright-box mt-4 mb-0 pt-3 pb-0"> <p class="text-left">Copyright © 2021-2024 SCIE Publishing Ltd. unless otherwise stated.</p> <p class="text-right"> <a href="/Privacy" title="Privacy">Privacy</a>  <a href="/Terms_of_Use" title="Terms of Use">Terms of Use</a> </p> </div> </div> </footer>  <div class="back-top"> <div class="d-flex justify-content-center align-items-center flex-wrap"> <svg xmlns="http://www.w3.org/2000/svg" width="20" height="20" fill="currentColor" class="bi bi-chevron-up" viewBox="0 0 16 16"> <path fill-rule="evenodd" d="M7.646 4.646a.5.5 0 0 1 .708 0l6 6a.5.5 0 0 1-.708.708L8 5.707l-5.646 5.647a.5.5 0 0 1-.708-.708l6-6z"/> </svg> <span>TOP</span> </div> </div>  <div class="toast-container"> <div class="toast" id="liveToast" role="alert" data-delay="3000" aria-live="assertive" aria-atomic="true"> <div class="toast-header"> <strong class="mr-auto">Message</strong> <button type="button" class="ml-2 mb-1 close" data-dismiss="toast" aria-label="Close"> <span aria-hidden="true">×</span> </button> </div> <div class="toast-body"> </div> </div> </div> <script> $(function () { $('.js-navbar-toggle').on('click', function () { $('header nav').toggleClass('nav-search-active'); }); var windowHeight = $(window).height(), header = $('header'), backTop = $('.back-top'); $(window).scroll(function(){ var scrollTop = $(this).scrollTop(); scrollTop > 0 ? header.addClass('fixed') : header.removeClass('fixed'); scrollTop > windowHeight ? backTop.addClass('fixed') : backTop.removeClass('fixed'); }) backTop.click(function(){ $("body,html").animate({scrollTop:0},300); }) $(".article-abseract.clamp").each(function () { $(this).click(function () { $(this).toggleClass("clamp-open"); }); }); //初始化select $('select.select-mania').selectMania(); }) $('.logout').click(function () { $.get('/index/user/login_out', function (res) { if (res.err == 0) { window.location.href = res.data layer.msg('exit successfully!') } else { layer.msg('Exit failed!') } }) }) </script> </body> </html>

CINXE.COM

Identification of Cutting Workpiece Surface Defects Based on an Improved Single Shot Multibox Detector - Intelligent and Sustainable Manufacturing - Full-Text HTML - SCIEPublish