A Sparse Channel Attention Model using 3D Mel-spectrograms for Underwater Acoustic Target Recognition

Underwater acoustic target recognition has important applications in both military and civilian fields. However, the complexity of the underwater environment and the multipath nature of the acoustic channel pose great challenges to underwater acoustic target recognition. In this paper, we propose a sparse channel attention (SCA) model using 3D Mel-spectrograms for underwater acoustic target recognition. For feature extraction, we combine Mel-spectrograms features, Delta features and Delta-delta features to construct 3D Mel-spectrograms features. For model recognition, we propose a SCA model, which not only combines convolutional features to capture both local and global information in the spectrogram. Moreover, the sparse channel attention mechanism is adopted, which significantly reduces the computational complexity. The effectiveness of the method is verified by comparing the ablation experiments on the ShipsEar dataset. The experimental results show that SCA outperforms other recognition models with low computational complexity and achieving 98.66% recognition accuracy.