Research on Pipeline Video Defect Detection Based on Improved Convolution Neural Network

Abstract

At present, the existing drainage pipeline defect detection methods cannot meet the use standards. This paper proposes an image classification method based on improved convolution neural network. By adopting multi-scale convolution kernel and splitting convolution kernel, pipeline image features can be fully extracted and accurate image classification can be realized. The data set used in this method is 6 kinds of pipeline defects collected under real scenes, including residual wall, deposition, root invasion, foreign body penetration, obstacles and hidden connection of branch pipes. Through a large number of comparative experiments, the accuracy rate of the method proposed in this paper is as high as 90.2%, which can effectively solve the complicated problem of pipeline defect classification. The method proposed in this paper has greatly improved its accuracy and efficiency, which has laid a solid foundation for efficient and accurate detection of pipeline defects.