Underwater Target Recognition Based on Multi-Feature Fusion and Attention Mechanism

Underwater acoustic target recognition is a critical technology in ocean exploration, underwater navigation, and marine environmental monitoring. However, the complex and dynamic underwater environment poses significant challenges to traditional recognition methods, which often rely on single features and are limited in their ability to capture the multifaceted characteristics of targets. This paper proposes a novel underwater acoustic target recognition model based on multi-feature fusion and an attention mechanism. Our approach integrates three types of features-LOFAR spectrogram features, Mel Frequency Cepstral Coefficients (MFCC), and unsupervised autoencoder features-using an attention mechanism to enhance feature representation and improve recognition accuracy. The model employs a Transformer encoder network to capture long-range dependencies in acoustic data, further refining the fusion of multi-dimensional features. Experimental results based on real underwater acoustic datasets show that our model demonstrates higher accuracy and robustness in complex underwater environments.