Classification of objects in synthetic Aperture Sonar images

This paper discusses an approach for the classification of objects in Synthetic Aperture Sonar (SAS) images and its benefit over other approaches. Our approach fully utilizes raw sonar waveforms scattered from objects. To do so, we first locate objects of interest in an image obtained by SAS processing. Then we extract the portions of the raw sonar waveforms responsible for forming those imaged objects from the whole raw sonar data. We align/straighten these extracted waveforms for localized discriminant feature analysis from which we obtain local features used for classification. We demonstrate the usefulness of our approach using real experimental sonar data.