Bridge health anomaly detection using deep support vector data description

Abstract As an extremely important part of traffic arteries, bridge structure plays an essential role in national economic construction, social development and smart city. Thus the monitoring of the bridge structure health are increasingly concerned by the bridge industry scholars and engineering people at home and aboard. In this paper, we propose a deep learning framework to evaluate the safety of the bridge structural state. More specifically, the proposed system generates a learnable transformation which attempts to map most of the data network representations into a hypersphere characterized of minimum volume. During inference, mappings of normal examples fall within the learned hypersphere, whereas mappings of anomalies fall outside the hypersphere. The whole system is end-to-end trainable and outperforms other advanced methods in real-world dataset.