关键词: 钢卷端面缺陷检测, YOLOv7算法, 注意力机制, 损失函数

Abstract: An improved YOLOv7 object detection algorithm was proposed to address the issues of complex texture and small defects on the end face of steel coils,as well as slow recognition speed and low detection rate of small targets using YOLOv7 algorithm. The ELAN structure in the YOLOv7 algorithm backbone network was improved by adding a PConv convolutional layer to design a new structure to reduce the model complexity and improve the model detection speed. Due to the tendency to miss detection in small object detection,a new attention mechanism CSCA was designed to improve the network′s sensitivity to small-scale objects. On this basis,WIoUv2 loss function was used to replace the CIoU loss function in the original YOLOv7 algorithm network to optimize the loss function and improve the robustness of the network.Experiments were conducted on a self-made dataset of steel coil end face defects,and the results showed that the improved YOLOv7 algorithm improved mAP@0.5 by 4.1 % and FPS by 10.84 f/s,with better detection performance than the original algorithm.

Key words: the end face of steel coil defect detection, YOLOv7 algorithm, attention mechanism, loss function