Underwater Target Identification Method Based on Optimized 2D Variational Mode Decomposition and Transfer Learning

Intelligent identification of underwater targets using the traditional underwater acoustic target identification method has limitations owing to the complexity and variability of marine environments. The data sets that are constructed based on single-domain features cannot be used to characterize the global information of the target signal and the traditional machine learning and deep learning methods demonstrate a low generalization ability for small sample targets. To solve the problem of low accuracy and efficiency of traditional underwater acoustic target identification methods in complex marine environments, an underwater target identification method based on optimized two-dimen- sional variational mode decomposition(2D-VMD) and transfer learning is proposed herein. This method obtains the time-frequency map of the underwater target by wavelet transform and then uses the optimized 2D-VMD method to adaptively select the effective mode to achieve the separation of the target effective mode and the noise mode. The denoising of the time-frequency map is completed, and the classic texture features of the image are extracted for comparative analysis. Finally, through the transfer learning strategy being adopted, the model transfer learning verification based on InceptionV3 is realized, and the classification and identification for small sample data set of underwater targets are completed. Combined with the classification test experiment of five types of underwater targets in ShipsEar, the results show that: the underwater target identification method based on optimized 2D-VMD and transfer learning shows good feature extraction and denoising capabilities while taking into account the accuracy of underwater target identification and efficiency requirements. Moreover, it provides theoretical and technical support for the detection and identification of intelligent targets of marine equipment.