Acoustic feature extraction of divers based on multivariate variational mode decomposition with low signal-to-noise ratio

Aiming at improving the ability to extract feature of divers under low signal-noise-ratio, a method is proposed to extract acoustic feature of divers for vector multi-channel signals. Firstly, the multivariate variational mode decomposition is used for vector multi-channel signals to obtain intrinsic mode functions. Then, the intrinsic mode function with the highest energy is selected to calculate the cross-spectrum and its weighted permutation entropy. At last, the feature of weighted permutation entropy spectrum is extracted by the fast Fourier transform. The results of sea trial show that, the proposed method can effectively extract feature of divers with vector hydrophone without prior information, and the noise immunity is improved by 10 dB compared with the traditional algorithm with 80% detection rate.