Underwater Acoustic Target Recognition Using Dictionary Learning-Based Feature Fusion and MD-GANs Data Augmentation

Underwater acoustic target recognition is essential for ocean monitoring, resource exploration, and military applications. However, the complexity of underwater environments and the scarcity of data present significant challenges to acoustic target recognition. To this end, this paper introduces an innovative method that combines dictionary-based feature fusion with improved sample augmentation through multiple discriminative generative adversarial networks (MD-GANs). Specifically, Morphological component analysis (MCA) and K-SVD-trained dictionaries are employed to extract and fuse target features, while G-features are introduced to simplify representation, enhancing feature distinguishability. Although the novel feature fusion method significantly improves recognition accuracy, the issue of data scarcity still significantly impacts recognition performance. Therefore, the MD-GANs are utilized to generate high-quality augmented samples. Furthermore, a soft-voting strategy is applied to dynamically adjust the weights of these generated samples based on their reliability. The collaborative work of multiple discriminators combined with the soft decision-making method enhances the quality and diversity of the augmented data, effectively reducing both false alarm and omission rates. Experimental results show that our method achieves a recognition accuracy of 95.87%, demonstrating its effectiveness and reliability in underwater acoustic target recognition.