TD-YOLOA: An Efficient YOLO Network With Attention Mechanism for Tire Defect Detection

Tire quality is crucial for vehicle safety, and thus delivery inspection is necessary. However, detecting internal tire defects from X-ray images is still a challenging task due to its complex texture background, diverse types of defects, small defect areas and so on, which make it difficult for existing methods to achieve high accuracy and real-time performance simultaneously. In this article, a novel tire detection approach is proposed to address these problems by integrating the advantages of improved yolo network and attention mechanism (TD-YOLOA). In particular, i) an efficient layer aggregation network (ELAN) backbone structure is proposed to improve the ability of model detection for feature extraction, where grouped convolution is applied to enhance the information interaction and reduce computational complexity; ii) a spatial pyramid pooling with cross stage partial convolution (SPPCSPC) is adopted to improve the efficiency of feature fusion, where the SPP module is retained for enlarging receptive field and CSPC is designed to merge features from different operations; and iii) a convolutional block attention module (CBAM) is introduced to improve the detection accuracy of small tire defects, which combines channel and spatial attention. Finally, the experimental results on a tire common defects dataset have demonstrated the superiority of the proposed TD-YOLOA method over other methods, achieving a 91.3% mAP and 9.28 ms for a tire sub-image, which is 0.5% and 0.65ms better. And the actual industrial application verifies the effectiveness of proposed method.