Automatic detection method of dam pop-out by deep learning

Abstract

In recent years, the aging of infrastructures including dams has become a problem, and it is urgent to develop appropriate and efficient inspection methods. In this research, we propose a method that enables deep learning to accurately and efficiently understand the positional distribution of dam pop-out. In this method, we used a semantic segmentation which is one of the object recognition methods for the image captured by the dam body with UAV, and made a pop out judgment on a pixel unit. As a result, it was possible to detect the position of the pop out almost.