Feature extraction methods of ship-radiated noise: From single feature of multi-scale dispersion Lempel-Ziv complexity to mixed double features

Abstract

Dispersion Lempel-Ziv complexity (DLZC) has been introduced into the field of underwater acoustic with great performance, but it only reflects complexity information from single-scale. To address this problem, we proposed multi-scale dispersion Lempel-Ziv complexity (MDLZC), which reflects the complexity of ship-radiated noise signals (SNSs) more comprehensively. Based on the proposed MDLZC, we put forward a single feature extraction method for SNSs, which denoted as 1-MDLZC; to enhance the recognition validity of 1-MDLZC, we came up with a double features extraction method based on MDLZC under two scale factors (SFs), which denoted as 2-MDLZC; to associate the complementarity of MDLZC and multi-scale dispersion entropy (MDE), we advanced mixed double features extraction method based on MDLZC and MDE under two SFs, which denoted as 1-MDLZC&1-MDE. The experimental results demonstrated that, 1-MDLZC has the highest recognition rate compared to single feature extraction method based on MDE and multi-scale Lempel-Ziv complexity (MLZC), which up to 97%; 2-MDLZC improves the recognition rate of 1-MDLZC by 2.75%; 1-MDLZC&1-MDE makes the recognition rate achieve 100% in many cases under double features.