Underwater target recognition based on CNN featute spectrum learning

Due to the complex marine environment, it is difficult to classify and identify the noise signals of underwater targets to achieve the expected results. In order to obtain higher recognition accuracy, this paper applies deep neural network to underwater target recognition, and proposes a recognition method combining auditory features and convolutional neural network. Extract the Mel spectrum of the target noise signal as the feature spectrum, and then use the DenseNet convolutional neural network to train and predict the feature spectrum to obtain a better recognition effect. Later, by adding strategies such as regularization and early stopping, the network was prevented from over-fitting, and the recognition accuracy was further improved on the test set.