Research on feature extraction method for underwater acoustic signal using secondary decomposition

Due to the high complexity of the marine environment, underwater acoustic signal (UAS) contains a lot of background noise, which makes it more difficult to extract its features. To accurately extract target features, a feature extraction method for UAS using secondary decomposition and mixed features is proposed. Firstly, an improved CEEMDAN (ICEEMDAN) is proposed to alleviate the mode aliasing problem. Secondly, the threshold for secondary decomposition is determined adaptively, and the UAS is secondarily decomposed using the enhanced beluga whale optimizer (EBWO) improved time-varying filtering empirical mode decomposition to obtain dual mode components. Thirdly, two eigenvectors are selected from the dual mode components, respectively. Fourthly, refined time-shift multiscale fuzzy dispersion entropy (RTSMFuDE) is proposed, and RTSMFuDE of the eigenvector and the original UAS are calculated to construct the three-dimensional eigenvalues. Finally, support vector machine is used to identify three-dimensional eigenvalues. The simulation results of the measured ship radiated noise show that the recognition rate of the proposed method reaches 98.43%, which proves that it can accurately classify ship signals. In addition, the proposed method has been successfully applied to feature extraction of underwater acoustic biological signals, which proves its universality.