The underwater acoustic target timbre perception and recognition based on the auditory inspired deep convolutional neural network

Abstract

Targeted at solving the problem in extracting the line spectrum features of ship radiated noise in complex marine environments, this paper proposes a new deep convolutional neural network model for underwater acoustic target radiated noise timbre perception, which is inspired by the timbre perception mechanism of the auditory system. The model first realizes the adaptive decomposition of the line spectrum of the underwater acoustic target radiated noise through one-dimensional deep convolution with multiple receptive fields, and then models the energy distribution structure of the line spectrum through the normalization of the decomposed signals and the point-by-point convolution, which expresses the frequency domain characteristics of ship radiated noise timbre. Next, a two-dimensional deep convolutional network is used to extract the deep representation of the difference in frequency domain characteristics of radiated noise timbres of different targets. And finally, the softmax layer is used to predict the target category. Validation experiments were carried out using measured data of five types of underwater acoustic targets. The results show that the proposed model is able to perceive the frequency domain characteristics of the radiated noise timbres of underwater acoustic targets, selectively enhance the signal energy of the radiated noise line spectrum, suppress the energy of interference noise, and extract stable line spectrum features of underwater acoustic targets. The recognition accuracy of the proposed model reaches up to 78.2%, which is 2.1%∼13.5% higher than the 6 control deep learning models.