Research on Small Target Defect Detection of New Energy Oil and Gas Pipelines Based on YOLOv5

Aiming at the problem of small target size and large difference of oil and gas pipeline defects in UAV aerial images, this paper proposes a small target detection method suitable for oil and gas pipeline UAV inspection based on YOLOv5 deep learning network framework. The coordinate Attention (CA) module is fused in the C3 structure of YOLOv5 network to effectively capture global position information and channel relationships. The SPPF module is reconstructed into the CSP network structure to form the SPPFCSPC module. The network feature extraction ability is strengthened by increasing the receptive field, and the original NMS is changed to DIoU-NMS to improve the recognition ability of overlapping occlusion targets. The annotation is carried out in the daily inspection video, the data set is constructed and the training is completed. The experimental results show that the mAP_0.5, Precision, Recall (R) and Speed indexes of the proposed method are $0.852,0.812,0.819$ and 2.9, respectively, and the comprehensive performance is the best, which effectively improves the efficiency of oil and gas pipeline inspection.