Underwater acoustic target recognition using line enhancement and deep neural network

To enhance the features of underwater acoustic target signals, and improve the performance of underwater acoustic target recognition based on deep neural network, a target recognition method using line enhancement and deep neural network is proposed. This method focuses on the narrowband information enhancement and sets an adaptive line enhancement filter at the front end of the VGGish network. The signals are processed by the line enhancement filter and input into the network to extract deep features, and then these features are classified by a classifier. The effectiveness of the method is verified by the actual underwater acoustic dataset. Principal component analysis is performed on the deep feature set of the underwater acoustic signals, and the results show that the compactness of the deep feature set obtained after line enhancement is significantly improved. The proposed method can obtain a recognition accuracy of 94.83% on the test dataset, which is improved by 5.48% compared to the case without line enhancement, and it is also more robust under the condition of low signal-to-noise ratio.