A Novel Detection Method for Weak Underwater Acoustic Targets Based on RUSBoost

Abstract

A large number of clutter false alarms cause the difficulty of detecting weak underwater acoustic targets in the active detection process. This paper proposed a new detection method based on RUSBoost. First, analyzing the imbalance of underwater acoustic data. Second, the classification cost ratio of underwater acoustic target echo data and clutter false alarm data was adjusted. The higher classification cost ratio makes the machine learning model more sensitive to the underwater acoustic target echo data in the training process. Finally, a targets-clutter classifier was constructed by combining RUSBoost, which is placed before the detection module. The experimental result shows that RUSBoost can maintain high detection probability and low false alarm probability on receiver performance curve, and its performance is stable under the different number of iterations and the cost ratio. Therefore, RUSBoost can effectively improve the performance of underwater acoustic target detection.