An acoustic dataset augmentation generator for ship noise classification

Automated target recognition using ship noise has advanced rapidly with machine learning, yet remains constrained by a limited number of benchmark datasets due to a scarcity of labeled data. This study aims to enhance ship classifier performance by addressing dataset limitations through an innovative Acoustic Dataset Augmentation Generator. Leveraging acoustic data from Arctic hydrophones and incorporating simulated ship noise via a Passive Sonar Simulator alongside ambient noise variations, this generator simulates diverse environmental conditions to enrich training datasets. By filling gaps and introducing realistic noise sources, it aims to improve model robustness and performance. The research evaluates model effectiveness using the augmented dataset, focusing on identifying and mitigating misclassifications.