Research on Underwater Acoustic Target Recognition Method Based on DenseNet

Under the statistical mode, underwater acoustic target recognition relies on heavy feature engineering, and the manually extracted features are sometimes not necessarily effective. At the same time, for confidentiality reasons, the lack of underwater acoustic data will also seriously affect the performance of the underwater acoustic target recognition system. In view of the above problems, the convolutional neural network, residual neural network and densely connected convolutional neural network are introduced and improved, and a Res-DenseNet-based network model is proposed and applied to the underwater acoustic target recognition task. An experimental study was carried out on the dataset. The experimental results show that, compared with the traditional method of MFCC+SVM, using the ResNet network alone and the DenseNet network alone, the correct recognition rates of the new model proposed in this paper are increased by 9.48%, 5.09% and 5.06%, respectively. The method in this paper can be effectively used for underwater acoustic target recognition.