Enhanced YOLOv3 Model for Automated Fabric Defect Detection

The manufacture of high-quality fabrics depends on an accurate and efficient fault detection system that can analyze data in real-time. A sophisticated defect detection system based on an enhanced YOLOv3 architecture is presented in this work to improve detection accuracy and reduce false identifications. Using a hybrid technique that combines defect size analysis and k-means clustering, the proposed approach optimizes the number and size of anchor boxes for fabric imperfections, introducing two crucial breakthroughs. Second, a multi-scale feature improvement strategy is used, combining high-level semantic insights with lower-layer spatial features. Incorporating an extra detection layer at various feature levels allows for the reliable identification of defects across multiple fabric types, including patterned and gray fabrics. Experimental verification shows that the improved model outperforms standard YOLOv3 regarding detection accuracy and dependability, achieving a defect misclassification rate below 5%. The results indicate a significant reduction in false detections, with the model performing 97.27% accuracy on gray cloth and 98.14% on lattice cloth, outperforming YOLOv3 by 2.59% and 1.91%, respectively. These findings underscore the model’s potential for industrial applications, significantly improving defect localization and fabric quality assurance.