Research on the Defect Detection Algorithm of Warp-Knitted Fabrics Based on Improved YOLOv5

To resolve the problems of low detection accuracy, slow detection speed, and high missed detection rate of traditional warp-knitted fabrics, this study researches and proposes an improved YOLOv5 algorithm for automatic detection of warp-knitted fabric defects, utilizing YOLOv5's fast detection speed and high accuracy. First, a multi-head self-attention mechanism module with an improved activation function is proposed to enhance the model's attention to the defect area of the fabric, improve the detection accuracy of warp-knitted fabric defects and reduce the missed detection rate. Second, a hybrid atrous space pyramid module is added to the backbone extraction network to enhance the receptive field, capture global feature details, and improve the model's recognition and location accuracy of warp-knitted fabric defects. Finally, the transposed convolution is used as an upsampling layer to improve the feature fusion network. The feature extraction layer can better combine fine-grained details with highly abstract information, enhance the accuracy of feature fusion, and then improve the detection accuracy of the model. Experimental results show that using the self-built warp-knitted fabric dataset, the mean average precision of the improved YOLOv5 is 91.3%, the precision rate is 89.7%, and the recall rate is 79.9%, which is 7.9%, 15.6% and 4.1% higher than the original YOLOv5 algorithm, respectively. The improved YOLOv5 defect detection algorithm has a higher accuracy, faster speed, and better robustness, which is helpful for the development and application of a warp-knitted fabric automatic inspection system.