Fabric Defect Detection with Fusion of Convolutional Attention Module and YOLOv5

Abstract

AbstractTo address the problems of low detection accuracy and difficulty in detecting small target defects in the current fabric defect detection, an improved YOLOv5 defect detection algorithm is proposed. Firstly, by introducing the convolutional block attention module (CBAM) into the original network to improve the network’s ability to extract small target features, the model can accurately identify small defect targets; secondly, distance-intersection over union (DIoU) loss is used to improve the accuracy of bounding box localization; Finally, the Focal Loss function is introduced to solve the problem of imbalance between foreground and background samples. The experimental results show that the improved network model improves the precision by 6.58% compared with the original YOLOv5 model, reaching 92.12%, and the mean average precision (mAP) improves by 6.95%, reaching 80.08%. The comprehensive detection capability is better than other mainstream algorithms, and meets the demand of real-time detection in textile enterprises.KeywordsDefect detectionYOLOv5Attention mechanismFocal loss