An Underwater Acoustic Target Recognition Method Based on Multi-Stage Feature Fusion Strategy

Underwater acoustic target recognition plays a crucial role in locating, monitoring, and tracking targets in marine environments, especially in military applications. However, underwater acoustic target recognition faces several challenges, including reliance on single-feature extraction methods and the consequent limitation in feature information. To address these issues, we propose a novel approach based on a multi-stage feature fusion strategy: different features are input into the network and their corresponding stage feature maps are fused. These fused feature maps then pass through specialized fusion networks for integration, culminating in a combined output. Subsequently, these multi-stage fusion feature maps are fed into the corresponding fusion network for further training and testing. Furthermore, the proposed method incorporates a voting strategy for multi-model integration. Experimental results demonstrate that this approach enhances recognition accuracy by 1.68% compared to the best existing method in the references.