Title:

Cross-Ocean Acoustic Classification using Machine-Learning

This project presents a machine learning framework for the classification of underwater acoustic signals with an emphasis on cross-environment performance. Underwater acoustic propagation is highly dependent on environmental factors such as temperature, salinity, and depth, causing significant variability in recorded signals across different ocean regions. As a result, many existing models achieve high accuracy only within the specific datasets on which they are trained and fail to generalize to new environments. To address this challenge, a structured processing pipeline was developed that segments raw audio into fixed-duration samples and converts them into log-mel spectrograms for time-frequency analysis. A convolutional neural network (CNN) was then trained to classify vessel-related acoustic signatures. Regularization techniques, including dropout and L2 penalties, were incorporated to improve model stability and reduce overfitting. The model achieved up to 100% classification accuracy on an unseen passenger ship dataset collected from a different ocean environment, demonstrating strong feature extraction and transfer capability for this class. However, performance was found to be highly dependent on dataset diversity, highlighting that limitations in labeled data remain a primary barrier to broader generalization. Overall, this work demonstrated the feasibility of cross-environment acoustic classification and represents a step toward developing robust, globally applicable underwater acoustic analysis systems.

Name

Comment*

Comments are viewable only by submitter

Captcha*