Multi-Task Attention Convolutional Neural Network for Underwater Acoustic Target Recognition in Few-shot Scenarios

Underwater acoustic target recognition (UART) holds significant importance in fields such as rescue operations, national defense security, and marine biological research. However, in real marine environments, acquiring large volumes of effective data samples remains challenging, leading existing target recognition networks to frequently encounter overfitting issues in few-shot scenarios. This paper proposes a novel fewshot underwater acoustic target recognition method, the multitask attention convolutional neural network (MT-ACNN). By integrating multi-task learning with an attention mechanism, this network establishes an efficient few-shot learning framework. Experimental results demonstrate that our method effectively enhances the model’s capability to recognize complex acoustic signals and improves generalization performance on limited datasets, thereby successfully addressing current challenges of data imbalance and few-shot learning.