Defect detection network for critical surface products based on content-position features and dynamic regression loss

View at publisher Summarise with VIRAT
Authors: Wencheng Zhang, Chen Guo, Jun Qing Guo, Baigang Du, Shunsheng Guo
Cited by: 0
References: 49

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Abstract Detecting defects on critical surface products is essential in manufacturing for ensuring high-quality standards. However, defect detection faces challenges due to the random variation in defect location and content, as well as inconsistent annotation quality. To address these issues, we propose the CCW-Net for defect detection on critical surface products. To manage positional variation, positional information is embedded within channel attention, emphasizing defect location features through coordinate attention (CA). For handling content variation, a content-aware reassembly of features (CARAFE) aggregates contextual information across a broad receptive field, enabling the capture of comprehensive defect size and shape characteristics. For the uneven annotation quality, the regression loss with dynamic focusing mechanism (Wise-IOU) module is introduced to dynamically allocate gradient gain, which enhances the model’s generalization ability. Experiments conducted on public datasets—NEU-DET (83.9% mAP) and Magnetic-Tile (94.6% mAP)—demonstrate the efficacy, real-time performance, and practical feasibility of the proposed approach.

Fields of Study & Subjects

Industrial Vision Systems and Defect Detection Advanced Neural Network Applications Infrastructure Maintenance and Monitoring

Publication details

Journal / source
Engineering Research Express
Volume
8
Issue
6
Pages
065230–065230
Publication date
1 March 2026
Language
EN