An end‐to‐end steel surface defect detection approach via Swin transformer. Issue 5 (20th December 2022)

• Record Type: Journal Article
• Authors: Tang, Bo; Song, Zi‐Kai; Sun, Wei; Wang, Xing‐Dong
• Is Part Of: IET image processing. Volume 17:Issue 5(2023)
• Page Start: 1334
• Page End: 1345
• Publication Date: 2022-12-20
• Subjects: computer vision -- feature extraction -- image classification -- image recognition; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/ipr2.12715
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract: Different from most current studies using convolutional neural network (CNN), a deep learning detection method for steel plate surface defects based on Transformer is researched. This paper presents a process, network structure and detection method for steel strip surface defect detection. A Swin Transformer was used to extract hierarchical features in the detection system. Then feature pyramid networks (FPN) was used to fuse the above features to form multi‐scale feature maps, and region proposal network (RPN) was adopted to generate the generate region of interest (ROI) of defects. Finally, the ROI head was used to generate defect category information and its precise location. Here, ablation experiments were conducted to explore the impact of different backbone networks and the number of stages of Swin Transformer and FPN on target detection performance to prove the rationality and efficiency of the network. On the NEU‐DET dataset, the proposed algorithm achieved 81.1% mAP, which is higher than that of classic CNN detection methods such as Faster R‐CNN, SSD, Yolo v3 and RepPoints. Abstract : Different from most current studies using convolutional neural network (CNN), a deep learning detection method for steel plate surface defects based on Transformer is researched. This paper presents a process, network structure and detection method for steel strip surface defect detection. The detection system uses a Swin Transformer to extract hierarchical features, then uses feature pyramid networks (FPN) to fuse the above features to form multi‐scale feature maps, adopts region proposal network (RPN) to generate the generate region of interest (ROI) of defects, and finally uses the ROI head to generate defect category information and its precise location. In this study, an ablation experiment was conducted to explore the impact of different backbone networks and the number of stages of Swin Transformer and FPN on target detection performance, and to prove the rationality and efficiency of the network. On the NEU‐DET dataset, the proposed algorithm achieved 81.1% mAP, which is higher than that of classic CNN detection methods such as Faster R‐CNN, SSD, Yolo v3 and RepPoints.