Research on Defect Detection Method of Drainage Pipe Network Based on Deep Learning

Abstract

In the daily life of the city, the normal operation of underground drainage pipes is a necessary condition to ensure the normal life of residents. However, with the increase of the service life of the drainpipe and the improvement of the function of water transmission and sewage, it is particularly important to evaluate the state of the drainpipe. However, the traditional pipe network detection methods such as CCTV and periscope detection are not only inefficient but also cost high. Nowadays, the Object Detection technology is becoming more and more mature, and the application of image detection technology to the defect detection of drainage pipe network is also a hot research direction. Therefore, an improved YOLOv5 Object Detection method was selected in this paper to realize the defect detection of drainage pipe network. In addition, in order to better complete the detection task in the complex image background of the waterway, the multi-head attention mechanism was incorporated into the backbone network of YOLOv5, and the FPN+PAN structure of YOLOv5 was replaced by BiFPN structure. Finally, through simulation experiments, the Precision(P) of the YOLOv5-TB model used in this paper reached 93.1%, the Recall(R) reached 85.5%, and the Mean Average Precision(mAP) reached 88.4%. Moreover, the mAP increased by 1.1% on the basis of YOLOv5. The simulation results show that the model used in this paper can well complete the task of drainage network defect detection.