A Comparative Study of Deep Learning-Based Models for Ship-Radiated Noise Classification from Underwater Acoustic Data

Underwater noise resulting from human activities has already caused considerable harm to the marine environment. Ship radiated noise, as the primary source of human-induced underwater noise, plays a crucial role in accurately and efficiently classifying such noise to safeguard the marine ecosystem. Although deep learning (DL)-based methods have been shown to enhance classification accuracy and efficiency, their generalization ability remains limited due to insufficient data and the absence of comparative analyses between different models and acoustic features. This study addresses these challenges by expanding the ship radiated noise signal dataset and extracting three types of acoustic features. The classification accuracy of different models with various feature combinations is evaluated, and the results show that feature fusion improves the model's classification accuracy by an average of over 8%, offering an effective method for accurately identifying ship radiated noise.