Surface ship-wake detection using active sonar and one-class support vector machine

Active sonar systems with high-frequency signals can detect a ship's wake based on the existence of wake bubbles behind a passing ship. However, it is hard for a fixed threshold method to reflect the various conditions of the ocean environment. Therefore, an adaptive detector is needed for the effective detection of wake bubbles under various conditions in a real ocean environment. Normally, many measured signals are required to design a detector with the desired level of performance as posited by pattern recognition studies. However, obtaining experimental data for the passing of a real ship over an upward-facing active sonar system in various situations is unrealistic. Therefore, this paper proposes a new bubble-wake detector using a pattern recognition technique such as the one-class kernel support vector machine that only uses the data obtained from an isolated situation in the absence of a ship's bubble wake. The proposed detector shows promising performance after being tested with an upward-facing sonar system in a real ocean environment and then artificially adds various noise levels to ship data to verify the robustness of the detector in a low signal-to-noise ratio. Thus, in the proposed ship-wake detector, the bubble-wake signals are detected and classified as the outlier class, while the normal signals are detected and classified as the trained class.