The Classification of Underwater Acoustic Targets Based on Deep Learning Methods

Abstract

The underwater target classification is a challenging task due to the complexity of marine environment and the diversity of underwater target features. Most of the-state-of-theart target recognition systems depend on feature extraction schemes based on expert knowledge in order to effectively represent the target signatures. In contrast, 16 different kinds of underwater acoustic targets are categorized in this paper by using Convolution Neural Network (CNN) and Deep Brief Network (DBN), which can achieve the accuracy up to 94.75% and 96.96% respectively in both supervised and unsupervised fashions. To compare with the results of traditional machine learning methods, we also use Support Vector Machine (SVM) and Wndchrm to do the same job and the latter is originally a tool applied for the biological image analysis. The results show that deep learning methods can achieve higher recognition accuracy when classifying the underwater targets from their radiation noises.