Deep Learning for Deep Water: Robust classification of ship wakes with expert in the loop

Abstract

This work examines the applicability of the deep learning models to pattern recognition in acoustic ocean data. The features of the dataset include noise, data scarcity and the lack of labeled samples. A deep learning model is proposed for the task of automatic wake detection. It takes advantage of the availability of an expert in the marine science domain while using data generation and robustness techniques to enhance performance. The model shows encouraging results, although its performance decreases with heavily unbalanced data and the introduction of noise.