Detection of underwater acoustic target using beamforming and neural network in shallow water

Abstract

Due to the complexity of marine environment, underwater acoustic target detection based on experience of expert observers is not fully reliable. We proposed a two-stage approach for underwater acoustic target detection task. In the first stage, the features extracted by beamforming-based direction of arrival estimation can carry essential information of the targets, and are taken as the input of the second stage. In the second stage, a feedforward neural network (FNN) is used as the kernel network to discover the potential pattern and construct a reliable detection model. The performance of the proposed method is evaluated on ocean underwater acoustic data. Convolutional neural network (CNN) is adopted as the baseline method for comparison. In three different sea areas, the accuracy of detection with the proposed method can reach 93.81%, 97.94%, and 97.32%, respectively, while CNN presents an average accuracy of only 74.17%. The experimental results demonstrate that the proposed method is effective in complex and changing underwater environments with little precise environmental information.