Machine Classification of Sedimentary Sea Bottoms

The side-scan sonar method for study of the sea bottom enjoys widespread use at the present time. The interpretation of the records is restricted to those observers having some experience, and is offered in terms of geological and man-made features, bottom topography, and sediment types. The most subjective opinions are perhaps found in the latter category, and it is here that the attention of this paper is focused. By drawing on ideas of texture quantification, this paper explores the potential of using a number of textural features to characterize small areas of side-scan sonar record with the view of producing, via a pattern-recognition procedure, classification decisions in the context of sedimentary sea-bottom identification. In this initial investigation, high levels of correct classification probabilities were achieved using small numbers of textural features derived from relative signal level variations rather than absolute ones, together with an uncomplicated decision rule.