Temporal Feature Based Classification of Active Sonar Targets in a Deep Water Environment

Owing to the growing threat of quiet diesel submarines, low-frequency active sonar has emerged as an efficient surveillance sensor system for anti-submarine warfare. Most operational active sonar systems incorporate a detection strategy based on the magnitude of the echo exceeding a localized threshold. However, this conventional method leads to target-like false alarms, generally termed clutter, which are the main problem of active sonar. The clutter originates from numerous sources with geologic, biologic, and anthropogenic origins. These numerous clutter echoes increase the computational load of target trackers and delay the correct classification of each echo. The clutter may be automatically distinguished using either spatial or temporal features of echo signals. Extensive studies on temporal features to mimic sonar operator's aural classification ability are underway because sonar operation experts can discriminate targets and clutter by listening to their echo signals. Temporal features of target and clutter exhibit various characteristics depending on the pulse type of active sonar and underwater environment. The bottom-bounce path is the primary detection channel for active sonar in deep water at long ranges. A more robust approach is needed to extract temporal features from bottom-bounced echoes than conventional techniques that did not consider underwater detection channels. In this study, we modify conventional temporal feature extraction techniques by applying sub-band processing to improve the performance of bottom-bounce path detection. We design a classifier using one-class classification techniques to overcome the lack of target data for training. The proposed method is evaluated on simulated and sea trial data.