Application of multi-algorithm mixed feature extraction model in underwater acoustic signal

Feature extraction of underwater acoustic signal (UAS) plays an important role in the analysis, classification and detection of UAS. It can extract key information in the signal, help us to understand and identify the specific feature of the signal, and thus realize the effective utilization of UAS. To efficiently and accurately extract the feature of UAS, wavelet soft threshold denoising (WSTD), variational mode decomposition (VMD) and Mel frequency cepstral coefficient (MFCC) are improved, respectively, and a multi-algorithm mixed feature extraction model in UAS based on Bayesian WSTD (BWSTD), regularized VMD (RVMD), fractional-order Mel frequency cepstral coefficient (FMFCC) and support vector machine (SVM) is proposed. Firstly, the collected UAS is normalized, and the signal is denoised by BWSTD. Secondly, RVMD is used to decompose the denoised signal into a series of intrinsic mode functions (IMFs), and the IMF with the largest mutual information with the denoised signal is selected as the eigenvector. Then, the eigenvector is input to FMFCC for feature extraction, and the corresponding feature coefficients are output. Finally, the feature coefficients are input into SVM for classification and recognition. The classification accuracy of the proposed model is 97.0 %, which is 16.5 %, 11.6 % and 4.3 % higher than that of other models such as linear predictive cepstral coefficient (LPCC)+ relevance vector machine (RVM), MFCC+RVM, FMFCC+RVM, respectively. Therefore, this research is of great significance to feature extraction and classification of UAS, and provides a potential solution to the challenges in related application fields.