Automated Defect Inspection in Casting X-ray Images: A DeepLabv3+ Framework with Adaptive Receptive Field and Multi- scale Fusion

Automated Defect Inspection in Casting X-ray Images: A DeepLabv3+ Framework with Adaptive Receptive Field and Multi- scale Fusion | 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 Automated Defect Inspection in Casting X-ray Images: A DeepLabv3+ Framework with Adaptive Receptive Field and Multi- scale Fusion Lin Xue, Xiaofang Song, LinHao Zou, Ri Chen This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8505951/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 9 You are reading this latest preprint version Abstract In the casting DR ray defect detection, the defective region usually has similar grey scale characteristics with the background, and the defect size span is large, which makes it difficult to accurately segment the defective region by the traditional detection method, thus affecting the detection accuracy and efficiency. For the casting defect detection semantic segmentation algorithm has the problems of low segmentation accuracy and large arithmetic capacity, DeepLabv3 + is selected as the baseline model, and targeted optimisation and improvement is carried out on the basis of this algorithm. In order to solve the problem of imbalance between positive and negative samples of data, we propose a hybrid loss function WCE-Dice Loss that combines the weighted cross-entropy loss function and the Dice loss function; we replace Xception, the backbone network, with MobileNetv2 to lighten the network architecture, and we add the ECA attention mechanism to the MobileNetv2 backbone network to enhance the model's ability to capture defective features; we integrate DeepLabv3 + as the baseline model and optimise and improve it on the basis of the algorithm. defective features; integrating the CFF multi-scale feature fusion structure to better extract contextual information, as well as comprehensively considering the dense connectivity idea and the need for lightweight improvement, the D_DenseASPP module structure is proposed. Experiments show that the improved DeepLabv3 + model improves 5.76% in mIoU index and 6.17% in Dice coefficient, and the number of parameters is only 11.81% of the original DeepLabv3 + model, which is an effective balance between model segmentation accuracy and computational complexity, and is suitable to be deployed in real industrial environments. The research in this paper provides an efficient and accurate solution for casting DR ray defect detection, which has important industrial application value and provides strong technical support for subsequent automated inspection and process optimisation. Digital radiography Deep learning Defect segmentation DeepLabv3+ MobileNetv2 Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 13 Apr, 2026 Reviews received at journal 31 Mar, 2026 Reviewers agreed at journal 31 Mar, 2026 Reviews received at journal 21 Mar, 2026 Reviewers agreed at journal 18 Mar, 2026 Reviewers invited by journal 16 Mar, 2026 Editor assigned by journal 15 Mar, 2026 Submission checks completed at journal 09 Jan, 2026 First submitted to journal 03 Jan, 2026 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-8505951","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":608059499,"identity":"491dae41-7f60-4cea-b5a7-685fa3bf70c4","order_by":0,"name":"Lin Xue","email":"","orcid":"","institution":"Dalian University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Lin","middleName":"","lastName":"Xue","suffix":""},{"id":608059500,"identity":"686f7eed-350b-41df-8d8f-d48a6740065b","order_by":1,"name":"Xiaofang Song","email":"data:image/png;base64,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","orcid":"","institution":"Dalian University of Technology","correspondingAuthor":true,"prefix":"","firstName":"Xiaofang","middleName":"","lastName":"Song","suffix":""},{"id":608059501,"identity":"c6854f54-9bf2-45ac-830a-1b2338030cbb","order_by":2,"name":"LinHao Zou","email":"","orcid":"","institution":"Longcheng Laboratory of Intelligent Manufacturing","correspondingAuthor":false,"prefix":"","firstName":"LinHao","middleName":"","lastName":"Zou","suffix":""},{"id":608059502,"identity":"1dd3f58a-6561-414b-a6e4-fe222119c255","order_by":3,"name":"Ri Chen","email":"","orcid":"","institution":"Dalian University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Ri","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2026-01-03 09:53:12","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8505951/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8505951/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":105035585,"identity":"e67ee0cb-c634-4f37-b05f-be3d0cd2df8d","added_by":"auto","created_at":"2026-03-20 07:26:17","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1107405,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8505951/v1_covered_837c88ed-7c2c-4b34-a6bc-1c6ae3a92da9.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Automated Defect Inspection in Casting X-ray Images: A DeepLabv3+ Framework with Adaptive Receptive Field and Multi- scale Fusion","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"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":"journal-of-nondestructive-evaluation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jone","sideBox":"Learn more about [Journal of Nondestructive Evaluation](http://link.springer.com/journal/10921)","snPcode":"10921","submissionUrl":"https://submission.nature.com/new-submission/10921/3","title":"Journal of Nondestructive Evaluation","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Digital radiography, Deep learning, Defect segmentation, DeepLabv3+, MobileNetv2","lastPublishedDoi":"10.21203/rs.3.rs-8505951/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8505951/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eIn the casting DR ray defect detection, the defective region usually has similar grey scale characteristics with the background, and the defect size span is large, which makes it difficult to accurately segment the defective region by the traditional detection method, thus affecting the detection accuracy and efficiency. For the casting defect detection semantic segmentation algorithm has the problems of low segmentation accuracy and large arithmetic capacity, DeepLabv3\u0026thinsp;+\u0026thinsp;is selected as the baseline model, and targeted optimisation and improvement is carried out on the basis of this algorithm. In order to solve the problem of imbalance between positive and negative samples of data, we propose a hybrid loss function WCE-Dice Loss that combines the weighted cross-entropy loss function and the Dice loss function; we replace Xception, the backbone network, with MobileNetv2 to lighten the network architecture, and we add the ECA attention mechanism to the MobileNetv2 backbone network to enhance the model's ability to capture defective features; we integrate DeepLabv3\u0026thinsp;+\u0026thinsp;as the baseline model and optimise and improve it on the basis of the algorithm. defective features; integrating the CFF multi-scale feature fusion structure to better extract contextual information, as well as comprehensively considering the dense connectivity idea and the need for lightweight improvement, the D_DenseASPP module structure is proposed. Experiments show that the improved DeepLabv3\u0026thinsp;+\u0026thinsp;model improves 5.76% in mIoU index and 6.17% in Dice coefficient, and the number of parameters is only 11.81% of the original DeepLabv3\u0026thinsp;+\u0026thinsp;model, which is an effective balance between model segmentation accuracy and computational complexity, and is suitable to be deployed in real industrial environments. The research in this paper provides an efficient and accurate solution for casting DR ray defect detection, which has important industrial application value and provides strong technical support for subsequent automated inspection and process optimisation.\u003c/p\u003e","manuscriptTitle":"Automated Defect Inspection in Casting X-ray Images: A DeepLabv3+ Framework with Adaptive Receptive Field and Multi- scale Fusion","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-19 15:17:59","doi":"10.21203/rs.3.rs-8505951/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-04-13T07:26:59+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-04-01T01:34:18+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"268990687003181217199344026229172427911","date":"2026-04-01T01:14:43+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-22T02:07:45+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"43684702469807213307898243315672766334","date":"2026-03-18T07:20:57+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-16T08:08:49+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-15T19:39:16+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-01-09T11:08:37+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of Nondestructive Evaluation","date":"2026-01-03T09:41:43+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-nondestructive-evaluation","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"jone","sideBox":"Learn more about [Journal of Nondestructive Evaluation](http://link.springer.com/journal/10921)","snPcode":"10921","submissionUrl":"https://submission.nature.com/new-submission/10921/3","title":"Journal of Nondestructive Evaluation","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"4ea81614-e0cb-484b-9321-1885e6b3f804","owner":[],"postedDate":"March 19th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-04-13T07:42:56+00:00","versionOfRecord":[],"versionCreatedAt":"2026-03-19 15:17:59","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8505951","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8505951","identity":"rs-8505951","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}