Towards Automated Classification of Seabed Substrates in Underwater Video

In this work, we present a system for the automated classiffication of seabed substrates in underwater video. Classiffication of seabed substrates traditionally requires manual analysis by a marine biologist, according to an established classiffication system. Accurate, consistent and robust classiffication is difficult in underwater video due to varying lighting conditions, turbidity and method of original recording. We have developed a system that uses ground truth data from marine biologists to train and test per-frame classiffiers. In this paper we present preliminary results of this using various feature representations (histograms, Gabor wavelets) and classiffiers (SVC, kNN) on both full-frame and patched-based analysis, achieving up to 93% accuracy.