A lightweight algorithm for woven fabric defect detection

At present, most fabric defect detection algorithms are modified based on the mainstream deep learning network architecture. However, few optimize the depth learning network model according to the fabrics' characteristics; few consider the parameter size of the model. In this paper, we propose a lightweight module for the fabric characteristics— Fabric module (FM). We use YOLOX-Nano as the basic network and replace the SPP module with FM. We tested the network model on the public data of Tianchi and compared it with the best result of the preliminary contest Tianchi. The experimental results show that our method can improve the accuracy and speed, and is more suitable for the automatic detection of fabric defects.