FABRIC DEFECT DETECTION BASED ON IMPROVED CLASS ACTIVATION MAPPING

This paper proposes a fabric detect defection method based on improved class activation mapping (CAM) to realize the fabric defect detection under weak supervised condition. This paper added squeeze-excitation (SE) block to CNN and combined the deep layer and the shallow layer together to improve the classification performance of the network. In order to obtain more accurate localization results, an improved class activation map was generated by combining the class activation maps of two resolutions. Experimental results show that the recognition accuracy of the proposed algorithm is 96.88% for four categories of fabric images, including no defects, holes, stains and yarn defects. Meanwhile, it can locate fabric defects accurately when there is only image-level labeling available in the dataset.