第５部門①

[2E01-05-02] Anomaly Detection of River Revetment for Individual Blocks Using Variational Auto Encoder

*Yukino Tsuzuki¹, Ryuto Yoshida¹, Junichi Okubo¹, Junichiro Fujii¹, Takayoshi Yamashita² (1. Yachiyo Engineering Co., Ltd., 2. Chubu University)

Keywords:Anomaly detection, Variational Autoencoder, Template matching, Maintenance

The management of river revetment is mainly carried out visually by engineers. There are many types in the deformation related to the functional depression of the river revetment. If these can be detected from revetment images, the efficiency improvement of inspection can be expected. Recently, VAE has been shown to be effective as an anomaly detection technology for uniform components. Since river revetment is formed by blocks of the same pattern, it is thought that anomaly detection in individual blocks is possible. This paper proposes the method for cropping individual blocks and the effectiveness of anomaly detection by VAE.