Enhancing Underwater Acoustic Signal Classification with CAM++ and Change Point Features

The effective classification of vessel acoustic signals is critical for marine environment monitoring. In this paper, we introduce two innovations: First, we developed a novel preprocessing method that combines Constant Q Transform (CQT) with a feature extraction technique introduced through dynamic programming, aiming to enhance the representation of signals by identifying statistical characteristics among signal change points. Second, we applied the CAMPPlus model, originally designed for speaker verification, to the classification of vessel acoustic signals, demonstrating its capability to significantly improve classification performance. Utilizing a dataset from Ocean Networks Canada, which contains signals recorded in the Strait of Georgia during the summer and fall of 2017, we classified the acoustic data into five categories: Tug, Passenger Ship, Cargo, Tanker, and Background Noise. A comparative performance analysis of the CAMPPlus model against the traditional ResNETSE model on this dataset revealed that CAMPPlus, combined with our proposed preprocessing and feature extraction techniques, significantly outperforms ResNETSE in terms of classification accuracy. This research not only enhances the accuracy of acoustic signal classification but also successfully transfers a mature deep learning model to a new application scenario. It demonstrates the potential of deep learning in model transfer and provides a new perspective for further research in the field of marine traffic monitoring, adhering to the standards of academic presentation in the process.