Pipeline defect detection system based on crawling robots

摘要

For PE gas pipeline defect detection problems, a crawling robot based pipeline defect detection system is proposed. Firstly, the crawling robot is designed and modelled; And then, the pipeline defect images are preprocessed to improve the contrast between the pipeline background and the defects, so as to identify and classify the defects more efficiently;Secondly, a large number of pipeline defect pictures are used to train the YOLO v5 algorithm , but considering that in the natural environment pipeline defects can be large or small, and the obtained picture can have problems such as occlusion phenomenon, therefore, we introduce the attention mechanism to extract the characteristics of pipeline defects, and use multiple attention mech-anisms to compare the selection of the most accurate model detection framework; Finally, the robot is put into the gas pipeline to carry out real-time detection,Compared with the original algorithm,the improved target detection algorithm has better detection effect and higher robu-stness,for common defects have high recognition, the highest confidence improvement is 0.2.