Teacher-student Network Yarn-dyed Fabric Defect Detection Based on Attention Residual Block Guidance
Aiming at traditional yarn-dyed fabric defect detection reconstruction models,there are problems such as difficulty to ensure the reconstruction effect of defective regions,missed detection and high false detection rate,an unsupervised teacher-student network yarn-dyed fabric defect detection algorithm based on attention residual block guidance is proposed.Firstly,from the perspec-tive of knowledge distillation,a teacher-student model with encoding-decoding structure based on Wide_Resnet50_2 network is de-signed,and the student network enhances the reconstruction capability by recovering the multi-scale features of pre-trained teacher network.Secondly,a dual attention residual module(DARM)is proposed to incorporate the dual attention,remove the teacher net-work output redundant information in the dual weight assignment mode of feature information,further expand the representation differences of defective regions between the teacher-student networks,and improve the defect detection and localization ability of the model.The experimental results show that the AU PRO of the proposed algorithm reaches by 85.8％,the pixel-level AU ROC by 96.3％,and the image-level AU ROC by 98.3％on the YDFID-1 dataset,the AU PRO and AU ROC of the proposed algorithm de-crease by no more than 4.5％on the MV Tec dataset under few sample conditions,the experimental results verify that the algorithm has the effectiveness and stability of dealing with color fabric defect detection.
万方数据
维普
