A Dual-Backbone Architecture for Lightweight RT-DETR Based Steel Defect Detection

A Dual-Backbone Architecture for Lightweight RT-DETR Based Steel Defect Detection | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article A Dual-Backbone Architecture for Lightweight RT-DETR Based Steel Defect Detection Chuqing Cao, Yujie Ma, Jing Zhang This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6732066/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract To address the challenges of model complexity, slow inference speed, and insufficient edge feature extraction in industrial surface defect detection, this paper proposes a lightweight improved RT-DETR-based object detection method. The proposed approach adopts a dual-backbone architecture: Backbone A inherits the deep HGBlock stacking strategy of RT-DETR-l to extract global semantic features, while backbone B is designed using the C2f structure combined with standard convolution modules to supplement shallow edge features. Furthermore, a Local-Global Attention Fusion (LGAF) module is introduced into the backbone to enhance multiscale feature fusion, and a Selective Boundary Aggregation (SBA) module is incorporated into the detection head to strengthen semantic guidance and boundary-aware contextual modeling. Compared with the original RT-DETR-l model, the proposed model achieves a reduction of approximately 47.82% in the number of parameters and a 60.27% decrease in FLOPs, while maintaining superior detection performance on both the NEU-DET and APDDD datasets. Experimental results demonstrate that the proposed method effectively reduces computational resource consumption while achieving high detection accuracy, exhibiting strong cross-dataset generalization ability and suitability for deployment in industrial scenarios with stringent speed and resource constraints. Industrial defect detection Lightweight keyword RT-DETR Dual-backbone architecture Real-time Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6732066","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":463493867,"identity":"7c2b8c0f-d79a-4e99-ac6d-65e9ae59b4e5","order_by":0,"name":"Chuqing Cao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6UlEQVRIiWNgGAWjYJACZiDmAVIHmCH8BKK1sCWQpgWky4A4Lfzthw9+LmyzljHnX/Pxc+GOwwz87DkGDD934NYicSYtWXpmWzqP5Yy3m6VnnjnMINnzxoCx9wxuLQYMOWbMvG2HeQxunN0gDWQwGNzIMWBmbMOjhf8NTMuZx79BWuwJapGA2XK+hw1iiwQBLRI3niVL85xLB9rCZmbNeyadR+LMs4KDvXi08PcnH/zMU2Ztb3D+8OPbvDus5fjbkzc++IlHCxQAY0QigYGBsQEUpwwMBwhqAGvhPwDWMgpGwSgYBaMAAwAA3tVMa2kM6NcAAAAASUVORK5CYII=","orcid":"","institution":"School of Computer and Information, Anhui Polytechnic University","correspondingAuthor":true,"prefix":"","firstName":"Chuqing","middleName":"","lastName":"Cao","suffix":""},{"id":463493868,"identity":"f8128ced-de94-4b03-9dd2-265f3044ab71","order_by":1,"name":"Yujie Ma","email":"","orcid":"","institution":"School of Computer and Information, Anhui Polytechnic University","correspondingAuthor":false,"prefix":"","firstName":"Yujie","middleName":"","lastName":"Ma","suffix":""},{"id":463493869,"identity":"5e741ace-53e0-4237-ae36-6238233c5c84","order_by":2,"name":"Jing Zhang","email":"","orcid":"","institution":"Yangtze River Delta HIT Robot Technology Research Institute","correspondingAuthor":false,"prefix":"","firstName":"Jing","middleName":"","lastName":"Zhang","suffix":""}],"badges":[],"createdAt":"2025-05-23 10:38:18","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6732066/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6732066/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":85152399,"identity":"eb7f8642-00d1-41e5-8e32-92ebcd4b190d","added_by":"auto","created_at":"2025-06-22 16:31:40","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2334659,"visible":true,"origin":"","legend":"","description":"","filename":"ADualBackboneArchitectureforLightweightRTDETRBasedSteelDefectDetection1.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6732066/v1_covered_0ff81086-af49-4c80-b8ae-548e27b17643.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Dual-Backbone Architecture for Lightweight RT-DETR Based Steel Defect Detection","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Industrial defect detection, Lightweight keyword, RT-DETR, Dual-backbone architecture, Real-time","lastPublishedDoi":"10.21203/rs.3.rs-6732066/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6732066/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"To address the challenges of model complexity, slow inference speed, and insufficient edge feature extraction in industrial surface defect detection, this paper proposes a lightweight improved RT-DETR-based object detection method. The proposed approach adopts a dual-backbone architecture: Backbone A inherits the deep HGBlock stacking strategy of RT-DETR-l to extract global semantic features, while backbone B is designed using the C2f structure combined with standard convolution modules to supplement shallow edge features. Furthermore, a Local-Global Attention Fusion (LGAF) module is introduced into the backbone to enhance multiscale feature fusion, and a Selective Boundary Aggregation (SBA) module is incorporated into the detection head to strengthen semantic guidance and boundary-aware contextual modeling. Compared with the original RT-DETR-l model, the proposed model achieves a reduction of approximately 47.82\\% in the number of parameters and a 60.27\\% decrease in FLOPs, while maintaining superior detection performance on both the NEU-DET and APDDD datasets. Experimental results demonstrate that the proposed method effectively reduces computational resource consumption while achieving high detection accuracy, exhibiting strong cross-dataset generalization ability and suitability for deployment in industrial scenarios with stringent speed and resource constraints.","manuscriptTitle":"A Dual-Backbone Architecture for Lightweight RT-DETR Based Steel Defect Detection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-30 10:49:39","doi":"10.21203/rs.3.rs-6732066/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"0af91b1a-4572-4d12-a1f5-b3210d83c7db","owner":[],"postedDate":"May 30th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-06-22T16:23:22+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-30 10:49:39","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6732066","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6732066","identity":"rs-6732066","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}