Improved Pipeline Disease Grading Detection Method for Visual Inspection of Sewer Pipeline Robots

As the aging of the pipes, various degrees of diseases will occur, impacting the realization of pipe functionality and shortening the life cycle. Therefore, conducting disease detection and determining the diseases severity levels are of great significance to urban pipelines construction. To address this practical issue, we propose an improved pipeline disease grading detection network based on YOLOv7. Firstly, in response to the issue of uneven accuracy in disease detection at different grades, we designed a specific disease grading module (GM) to learn grading information. Secondly, we introduced a coordinate attention (CA) module into the network to capture crucial structural information. The experimental results demonstrated that our method performs best in pipeline disease grading detection tasks. After adding GM and CA modules to YOLOv7 network, the mean average precision increased from 77.13% to 79.56%. In terms of improving the precision of sewer pipeline disease grading detection, the proposed method provides more efficient and accurate technical reference for the intelligent construction of urban drainage pipeline disease detection. The dataset and source code are available at https://github.com/celicvml/PipelineDiseaseDetection.