Leather Defect Detection Algorithm Based on an Improved YOLOv5s

In the field of intelligent manufacturing, the realization of efficient and accurate automated surface defect detection is particularly critical for enhancing product quality, especially in the leather industry, which holds a significant position in global trade. Traditional defect detection methods encounter challenges in handling complex scenes and identifying minute defects. However, the advent of deep learning technologies has provided a new pathway to address these issues. This paper is dedicated to innovative research on the precise localization and detailed classification of leather surface defects through the enhancement of deep learning models, aiming to overcome the limitations of existing technologies and improve the accuracy and efficiency of detection. By extending the YOLOv5s model, this study has developed the CMR-YOLO algorithm. This algorithm integrates the Channel Attention mechanism (CA), an improved Multi-Dimensional Precision Intersection over Union (MPDIoU), and the Cross Stage Partial Residual feature enhancement module (C3R). These enhancements not only improve the classification accuracy but also bolster the model's stability and robustness in complex scenes. Through the comprehensive application of these techniques, the CMR-YOLO algorithm demonstrates improved performance in categorizing different types of leather defects, particularly effective in handling diverse defect characteristics and accurately classifying various defects. Compared to the traditional YOLOv5s, the CMR-YOLO algorithm also exhibits improved accuracy in defect classification, with the mAP50 increasing from 62.5% to 79.5%, effectively enhancing the overall performance of leather defect detection. These research outcomes not only achieve technological breakthroughs but also provide significant technical support for quality control and intelligent manufacturing in the leather goods industry.