Compression of a Deep Competitive Network Based on Mutual Information for Underwater Acoustic Targets Recognition.

Sheng Shen,Meiping Sheng,Honghui Yang

doi:10.3390/e20040243

Abstract

The accuracy of underwater acoustic targets recognition via limited ship radiated noise can be improved by a deep neural network trained with a large number of unlabeled samples. However, redundant features learned by deep neural network have negative effects on recognition accuracy and efficiency. A compressed deep competitive network is proposed to learn and extract features from ship radiated noise. The core idea of the algorithm includes: (1) Competitive learning: By integrating competitive learning into the restricted Boltzmann machine learning algorithm, the hidden units could share the weights in each predefined group; (2) Network pruning: The pruning based on mutual information is deployed to remove the redundant parameters and further compress the network. Experiments based on real ship radiated noise show that the network can increase recognition accuracy with fewer informative features. The compressed deep competitive network can achieve a classification accuracy of , which is higher than deep competitive network and higher than the state-of-the-art signal processing feature extraction methods.

Highlights

Underwater acoustic targets recognition based on ship radiated noise is one of the main functions of passive sonar system
The results indicate that competitive restricted Boltzmann machine (CRBM) can learn the differences of categories
A compressed deep competitive network is presented by integrating competitive learning into the restricted Boltzmann machine learning algorithm and pruning the network based on mutual information

Summary

Introduction

Underwater acoustic targets recognition based on ship radiated noise is one of the main functions of passive sonar system. The acquired underwater acoustic signals are usually noisy due to the complexity of sound propagation in shallow sea and the frequent presence of high background noise in the sensor. Underwater acoustic targets recognition still depends on the decision of well-trained sonarmen, but it is difficult to implement continuous monitoring and recognition. An unattended underwater acoustic targets recognition system with high recognition accuracy and efficiency needs to be developed to achieve real-time targets recognition. In order to build an automatic underwater acoustic targets recognition system, various signal processing strategies were applied to extract features and design classifiers. Feature selection and compression methods were studied to improve the classification accuracy and efficiency

Methods

Findings

Discussion

Conclusion