Anomaly Detection for Partial Collapse of Reservoir Embankments Using Aerial Images in Real Environments

Abstract

Water facilities such as reservoirs are essential to our daily lives. The entire reservoir may be compromised if a reservoir’s embankment partially collapses or leaks. However, maintaining, managing, and inspecting reservoirs is a major social issue.This study focuses on the partial collapse of embankments to improve the efficiency of reservoir inspections. The purpose of this study is to develop an automatic detection method for anomalies, specifically targeting the partial collapse of reservoir embankments using aerial images. We propose a method to automatically detect partial embankment collapses employing an autoencoder for unsupervised learning.Additionally, we propose a method to quantitatively indicate the location and size of abnormalities. The effectiveness of the proposed method has been demonstrated through actual experiments using a UAV in a real reservoir environment.