Classification of Seabed Type from Underwater Video

This paper describes a method for the classification of seabed type from a video captured by a camera mounted on a towed vehicle that is dragged along the sea floor. Classification of seabed type is important for the mapping of marine habitats. Unlike other methods that are based on various sonar technologies, the proposed method is based purely on video frames. The aim is to tell from a single frame, what seabed type is present. A supervised learning approach is adopted, with a total of 5 different seabed types being represented. We developed a set of 6 image features to characterise the visual appearances of these seabed types. Both k-Nearest Neighbours (kNN) and Support Vector Machine (SVM) classifiers are implemented based on this feature set. Our analysis shows that is possible to achieve a cross-validation error of 10% for the 5-class problem.