Research on feature extraction of underwater acoustic signal based on hybrid entropy algorithms

• Published in: Applied acoustics Vol. 235; p. 110688
• Main Authors: Yang, Hong, Wang, Chao, Li, Guohui
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 14.05.2025
• Subjects: Decomposition; Entropy; Feature extraction; Mutual information; Ship-radiated noise; Spectral kurtosis; Underwater acoustic signal; Ship-radiated noise; Underwater acoustic signal; Feature extraction; Entropy; Decomposition; Spectral kurtosis; Mutual information
• ISSN: 0003-682X
• DOI: 10.1016/j.apacoust.2025.110688

•ZSSD is proposed. It has solved the difficult problem of VMD and SSD.•GAAPE is proposed. It has solved the problem that AAPE ignores signal phase information.•FSVE is proposed. It can capture the dynamic changes of time series more effectively.•It is proved that GAAPE and FSVE have complementary characteristics in UAS classification.•A hybrid complementary feature extraction method for UAS with ZSSD, GAAPE and FSVE is proposed. Its effective is proved. The research on feature extraction of underwater acoustic signal (UAS) is of great significance in developing and protecting marine resources. Therefore, to effectively improve the feature extraction technology, a novel hybrid entropy feature extraction method based on enhanced singular spectrum decomposition (ZSSD), generalized phase-amplitude-aware permutation entropy and fractional singular value entropy (GAAPE&FSVE) is proposed. To better capture the dynamic fluctuation components in the UAS, an enhanced SSD algorithm based on the modified Cao algorithm and ensemble fluctuation-based dispersion entropy is proposed. To make AAPE better adapted to feature extraction of UAS, phase processing and entropy parameters transformation are introduced, and GAAPE is proposed. To solve the problem that the entropy value of SVE is unstable when the quantized signal changes dynamically, a new fractional-order processing is introduced, and FSVE is proposed. Firstly, UAS is decomposed into a series of singular spectrum components (SSCs) by ZSSD. Secondly, the useful information contained in SSC is calculated from the time and frequency domains, respectively. The best two SSCs are reconstructed as feature vectors. Then, 150 samples are randomly selected from the feature vectors, and the GAAPE and FSVE of each sample are calculated, respectively. Finally, compared with at least 15 other methods, the experimental results show that the proposed method exhibits stronger synergy in UAS feature extraction and outperforms all compared methods with a 99% recognition rate.