Automatic Drainage Pipeline Defect Detection Method Using Handcrafted and Network Features

Effective maintenance of urban drainage pipeline networks is critical to the healthy development of cities. Routine CCTV surveys are costly, time-consuming and rely on technicians. In this paper, we propose a method to detect pipeline defects automatically, which can improve the efficiency of pipeline maintenance and save cost and reduce dependence on technicians. We use image restoration method to remove the interference of ropes in the pipeline image, before using Canny edge detection operator to extract the texture information of the defects. Then look for contours regionally to identify suspected defect areas. We combine the hybrid LBP handcrafted and VGG network features of suspected defect areas and use SVM classifier to predict pipeline defects. This method is more than 90% accurate on the test set. Furthermore, the high efficiency of this method enables it to compete with the performance of trained technicians. In general, this method of automatically detection pipeline defects provides a new option for the Urban Drainage Group and has a good development prospect.