Segmentation and Classification of Side-Scan Sonar Data

Abstract

Side scan sonar is an acoustic sensor which uses sound waves to generate side scan sonar images. Most adaptive behavior of AUVs would require that the vehicle be able sense the environment, detect objects of interest, localize and then change its current behavior. The first step toward this process would be the real time processing of its sensor data for object identification. In this paper we present an approach to real time processing of side scan sonar data using texture segmentation and classification. Given a side scan sonar image, texture is used to classify the image into four major categories - rocks, wreckage, sediments and sea floor. The image is first broken into relevant areas based on edge density and edge orientation statistics. Laws texture energy measures are then computed on these areas. The texture energy feature vector for each sub region is then classified using clustering algorithms.