A defect detection method based on tool images

Abstract The condition of cutting tools plays a critical role in determining the overall quality of manufactured products. To address the limitations of traditional wear detection methods, this study proposes a dual-stage approach integrating U2-Net-based segmentation and Hough line detection for boundary reconstruction. The enhanced LEA-U2-Net model achieves superior performance, with 92.06% mIoU, 93.57% mPA, 92.73% Dice coefficient, 95.39% recall, and 92.73% F-score on a custom dataset of milling cutter images. The results demonstrate the accuracy of the U2 Net improvement method in extracting tool minor cutting edges, and the conversion factor from pixels to reality can calculate the actual wear width. The proposed framework exhibits robustness under different lighting conditions and is adaptable to small datasets, outperforming benchmark tests such as DeeplabV3Plus and U-Net.