Automatic Object Classification with Active Sonar using Unsupervised Anomaly Detection

This work describes an unsupervised anomaly detection method for automatic contacts classification of an active sonar system. The proposed method refers to littoral, shallow water environments where there is a significant amount of clutter contacts from the seafloor and coastal reverberation. This huge amount of undesired contacts can be exploited to learn the "finger-print" of the clutter and then to identify the object related contacts as anomalies. The paper describes the proposed classification method and shows its performance with real data collected at sea using an echo-repeater as an artificial object.