A Fabric Defect Detection Method Based on Improved YOLOv5

Abstract

Fabric defect detection plays an important role in quality control. The traditional manual detection method is inefficient and costly, therefore many deep learning algorithms have been proposed. With the continuous development of object detection technology, the YOLO series of algorithms have been used in various detection tasks. In this paper, we propose a Squeeze-and-Excitation(SE)-module-based YOLOv5(SE-YOLOv5) to establish an efficient fabric detection system. SE-YOLOv5 develops from the YOLOv5 algorithm, but adds the SE module to the YOLOv5 backbone, and replaces the conventional Leaky Rectified Linear Unit(ReLU) activation function of YOLOv5 cross stage partial(CSP) with the ActivateOrNot(ACON) activation function. Experimental results demonstrate that the proposed model SE-YOLOv5 improves the accuracy, generalization ability, and robustness compared with YOLOv5, which can meet the need of detecting fabric defects.