ANOMALY DETECTION OF REVETMENT BY UNSUPERVISED DEEP LEARNING USING VARIATIONAL AUTO-ENCODER

DOI Web Site 7 References

Bibliographic Information

Other Title
  • VAE を用いた教師なし深層学習による変状種別に依存しない河川護岸の異常検知

Description

<p>Anomaly detection for infrastructure by deep learning using supervised learning has problems that it is difficult to prepare data set on abnormal data, consequently unlearned deterioration cannot be detected. On the other hand, several prior works have shown the effective for the anomaly detection based on generative models training using only normal data in image recognition tasks with little training data and diversity of the deteriorations to be detected. In particular, VAE-based methods have been used for anomaly detection on uniform parts. For the purpose of building the anomaly detection method of revetment that is independent of the type of deterioration, we examined and considered for the anomaly detection of revetment in block units using VAE. This paper empirically clarifies the guideline for preparation of data set and proposes an anomaly calculation method suitable for revetment. We also discussed t detecting various types of deteriorations.</p>

Journal

References(7)*help

See more

Keywords

Details 詳細情報について

Report a problem

Back to top