Deep Learning Applied to Underwater Mine Warfare

Abstract

In this article we are addressing the problem of automatic detection and classification of underwater mines on images generated by a Synthetic Aperture Sonar (SAS). To tackle this problem, we are investigating the use of Machine Learning techniques, in particular Deep Learning. Using this method we faced two challenges, (i) the availability of a sufficient amount of training data to learn the classification model and (ii) the design of the deep learning pipeline suited for this one-class classification problem. Our contributions in this paper are, first the synthetic generation of realistic image datasets for the training of our Machine Learning algorithm, and second the research and development of a novel Deep Learning approach for automatic underwater mines classification using sonar images. The combination of these two contributions offers a new pipeline of operation for Mine Counter Measure Automatic Target Recognition (MCM ATR) systems.