Underwater acoustic target recognition using RCRNN and wavelet-auditory feature

Underwater acoustic target recognition plays an essential role in sonar signal processing. Despite numerous efforts in target recognition, it remains a challenging task due to the complex nature of the underwater environment. Specifically, the overlapping acoustic feature of different target classes, coupled with the temporal-spatial variation characteristic of the marine environment, results in reduced recognition performance. This paper introduces a novel approach to address these challenges, which emphasizes the acquisition of fine-grained feature parameters and achieving high-precision classification results. Firstly, we develop a wavelet-auditory feature that comprehensively represents the underwater acoustic signal. The feature describes the time-frequency auditory information and reduces the dimensionality of the original signal. Secondly, a convolutional recurrent neural network with residual blocks module is designed, which enables the extraction of more discriminative global deep features from the wavelet-auditory feature. The network addresses the effects of time-varying and time-dependent nonuniform underwater environments. Finally, we conduct experiments on the ShipsEar database to evaluate the proposed method. Experimental results demonstrate that our method outperforms other classification methods in terms of recognition accuracy. Furthermore, the efficacity of each method component has been demonstrated via ablation studies, demonstrating that the proposed method is a significant initiative and contribution to the underwater acoustic target recognition task.