An efficient deep learning approach with frequency and channel optimization for underwater acoustic target recognition

Ship radiated noise (SRN) recognition is challenging due to environmental noise and the broad frequency range of underwater signals. Existing deep learning models often include irrelevant frequencies and use red, green, and blue (RGB) channel configurations in convolutional networks, which are unsuitable for SRN data and computationally intensive. To address these limitations, we propose FCResNet5, a neural network optimized for SRN classification. FCResNet5 adopts a streamlined architecture that focuses on the critical frequency band and applies frequency channelization to enhance spectral representation. Its compact design achieves greater computational efficiency while maintaining comparable accuracy. Ablation studies confirm the contribution of each component, and comparative results demonstrate that FCResNet5 offers a more efficient alternative to existing models without compromising performance.