ACCURACY VERIFICATION FOR EXTRACTING DAMAGE ON THE INNER SURFACE OF THE SEWER PIPES USING SEMANTIC SEGMENTATION

Abstract

In this study, we developed AI model for extracting damage on the inner surface of the sewer pipes using semantic segmentation. In order to improve the performance of the AI model with limited training data, data augmentation was applied. To investigate the effectiveness of data augmentation in extracting damage, we tested the accuracy of the model in several cases. The results confirmed the superiority of data augmentation. It was also suggested that image contrast enhancement may contribute as noise to the robustness of the AI model.