Multi-feature Classification and Recognition Method of Underwater Acoustic Signal based on Deep Decomposition

The classification and recognition of underwater acoustic signal (UAS) is an important way to achieve surface and underwater sensing, which is of great significance to modern naval warfare and marine ecological protection. Due to the complex underwater marine environment and serious background noise interference, UAS often has ‘three non-’ features-non-Gaussian, non-stationary and non-linear. Therefore, how to effectively classify and recognize them in real time has always been a problem for hydroacoustic workers. In this paper, a multi-feature classification and recognition method of UAS based on deep decomposition is proposed. Firstly, considering the shortcomings of variational mode decomposition (VMD) such as the difficulty of determining the initial parameters and the endpoint effect of empirical mode decomposition (EMD), we use EMD to decompose the UAS once and decompose the number of layers is determined as the K value of the VMD. For the initial parameter α of the VMD, a geometric mean optimizer (GMO) is used to select the optimal value. This can not only make full use of the advantages of EMD and VMD, but also avoid their shortcomings. Next, the intrinsic mode functions (IMFs) processed by EMD is divided into low complexity and high complexity, and the component with the strongest correlation with the undecomposed signal is selected from the low complexity IMF as the Feature IMFx. Then, the high complexity IMFs are deeply decomposed using the EVMD to select the IMFs with the strongest and the second strongest correlation with the original signals as Feature IMFy and Feature IMFz. Finally, based on Feature IMFx, Feature IMFy, Feature IMFz, fluctuation-based entropy and convolutional neural network are used for classification and recognition. After numerical experiments the recognition accuracy of ShipsEar dataset and NPU dataset can be obtained as 90.2% and 91.7% respectively.