Classification of Surface Vessels Using Underwater Acoustic Data and Machine Learning

Automatic vessel classification is a highly relevant research topic, particularly for the U.S. Navy. In this study, we consider three machine learning techniques to classify maritime vessels based on their underwater noise: Gaussian mixture models, random forest, and k-nearest neighbors. The ShipsEar database, developed by Santos-Domnguez et al., was used to conduct the study. Mel-frequency cepstrum coefficients were selected for class feature characteristics to compare with previous findings presented by Santos-Domnguez et al. in their publication titled ShipsEar: An Underwater Vessel Noise database published in the Applied Acoustics journal, volume 113. Results indicate that all three methods offer a feasible solution to the classification problem. Notably, Gaussian mixture models show significant performance improvements over results achieved by Santos-Domnguez et al.