CAD Fabric Model Defect Detection Based on Improved Yolov5 Based on Self-Attention Mechanism

In CAD fabric, there is severe problem of low speed and poor generalization performance in the defect detection algorithms. To solve this problem, improved YOLOv5 based on self-attention mechanism is proposed to detect CAD fabric defects. In the proposed method, concepts of Yolov5 have been used such as extraction of the key information from the feature map and improved target detection network. Aiming at the conflict caused by the unevenness of the special scale in the network feature fusion stage, an adaptive difference fusion model is formulated to propose the algorithm. In the proposed model, the transfer learning has been used to speed up the training process. The experimental results show that the proposed detection scheme can improve the network accuracy by 98.8% and the improve detection rate by 83 frames/s when compared with the existing non-adaptive Yolov5 algorithm. The results show that the proposed detection method can perform well with the required parameters. © 2024 U-turn Press LLC.