改进Yolov7的子午线轮胎X光图像缺陷检测

为了实现轮胎缺陷的自动检测,提高轮胎缺陷检测的精度，提出了一种子午线轮胎缺陷检测方法（YOLOv7-DCA）。首先，基于协调注意力和空洞卷积设计了DCA-MP(dilated coordinated attention-max pooling)模块，加强下采样的感受野同时提高缺陷所占权重，提高缺陷检测精度，其次，在Neck层中融合了CARAFE(content aware reassembly of features)模块进一步提高缺陷检测精度。实验结果表明，YOLOv7-DCA模型平均检测精度可以达到97.77%，相比于原YOLO(you only look once)算法提高了3.27%。与当前主流的Faster-RCNN,YOLOv5，YOLOv7-tiny系列模型相比，综合表现效果最好。可见，该模型对于轮胎缺陷自动检测研究提供了参考意义。

Improved defect detection in radial tire X-ray images of Yolov7

In order to realize automatic tire defect detection and improve the accuracy of tire defect detection, a radial tire defect detection method (YOLOv7-DCA) was proposed. Firstly, the DCA-MP(dilated coordinated attention-max pooling) module was designed based on coordinated attention and void convolution, which strengthens the receptive field of subsampling while increasing the weight of defects and improving the defect detection accuracy. Secondly, the CARAFE(content aware reassembly of features) module was integrated in the Neck layer to further improve the defect detection accuracy. Experimental results show that the average detection accuracy of YOLOV7-DCA model can reach 97.77%, which is 3.27% higher than that of the original YOLOv7 algorithm. Compared with the current mainstream Faster-RCNN,YOLOv5, YOLOv7-tiny series models, the comprehensive performance is the best. It can be seen that this model provides a reference for the research of tire defect automatic detection.