DCC-CenterNet: A rapid detection method for steel surface defects. (January 2022)

• Record Type: Journal Article
• Authors: Tian, Rushuai; Jia, Minping
• Is Part Of: Measurement. Volume 187(2022)
• Publication Date: 2022-01
• Subjects: Dilated convolution -- Center-weight -- CIoU loss -- Surface defect detection; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2021.110211
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: An anchor-free defect detector based on CenterNet is proposed. A dilated feature enhancement model (DFEM) to enlarge the receptive field is proposed. A centerness function, center-weight (CW), is designed for defect center location. CIoU loss is adopted for size regression. The experiment results demonstrate the effectiveness of our method. Abstract: In recent years, surface defect detection methods based on deep learning have been widely used. A conflict between speed and accuracy, however, still exists. In this paper, a steel surface defect detector, DCC-CenterNet, is proposed to achieve the best speed-accuracy trade-off. This detector uses keypoint estimation to locate center points and regresses all other defect properties. Firstly, a dilated feature enhancement model is proposed to enlarge the receptive field of the detector. Secondly, a new centerness function center-weight is proposed to make the keypoint estimation more accurate. Then, the CIoU loss that considers the overlap area and aspect ratio of the defect is adopted in the size regression. Finally, the results of experiments show that the accuracy of DCC-CenterNet can reach 79.41 mAP, and the running speed FPS is 71.37 with input size 224 × 224 on the NEU-DET steel defect dataset. And it reaches 61.93 mAP on the GC10-DET steel sheet surface defect dataset at a running speed of 31.47 FPS with input size 512 × 512. It demonstrates that the developed detector can detect steel surface defects efficiently and effectively.