Underwater Acoustic Target Classification Based on LOFAR Spectrum and Convolutional Neural Network

Abstract

The underwater acoustic target classification task has always been an important research direction of acoustic recognition and classification. The acoustic classification models include traditional models such as Gaussian Mixture Model (GMM), and deep learning models such as Convolutional Neural Network (CNN) and Long and Short Time Memory Network (LSTM). This paper proposes a deep sound feature extraction network based on VGGNet. An underwater acoustic target classification framework based on LOFAR spectrum and CNN is proposed. Although ordinary CNN can also extract underwater acoustic features, too few or too many network layers will cause problems such as insufficient features or increased calculations. Therefore, we draw on the excellent structure of VGGNet in feature extraction and delete several layers for feature extraction and classification of underwater acoustic targets. The accuracy are 94%, 98% and 96% respectively in three real data sets of civil ships, and the accuracy were improved com-pared with the traditional methods.