Proceedings of the International Conference “Underwater Acoustic Measurements: Technologies &Results” Heraklion, Crete, Greece, 28 th June – 1 st July 2005 HOW USEFUL IS BATHYMETRIC INFORMATION IN THE CLASSIFICATION OF HIGH FREQUENCY SONAR SURVEYS? Louis Atallah a , Penny Probert Smith b a The British University in Dubai/The University of Edinburgh, Institute of Informatics, PO Box 502216, Dubai, UAE, email: latallah@inf.ed.ac.uk , fax: +971-4-3664698 b Robotics Research Group, Department of Engineering Science, The University of Oxford, Parks Road, OX1 3PJ, email: pjp@robots.ox.ac.uk , fax: +44-1865-283145 (0) 1865 283145 Abstract: In a number of sonar studies, bathymetric information is used to correct and visualise amplitude data. This work proposes a method which combines amplitude features (describing backscattering strength and sonar texture) with bathymetric features (indicating seafloor variability) for sonar classification. Features are selected per window (of user defined size) and areas around grab samples in a survey are used for training. The importance of different features is investigated in this study, and highlighted by feature selection algorithms as well as by scatter plots exploring the training areas. Classification rates are significantly improved when both amplitude and bathymetry features are used. The method is applied to a sidescan bathymetric sonar dataset taken in Hopvågen bay-Norway. Keywords: seabed classification, bathymetric data, machine learning. 1. INTRODUCTION Bathymetric (depth) data serves a number of purposes when integrated into a sonar survey. The bathymetric profile provided in multibeam or sidescan bathymetric sonars can be used to map pixels of the sonar image, correct distortions and plot an accurate plot of the relief [1]. For low frequency sonar (10-15 KHz), bathymetry has been used to derive properties that describe large features in the seabed (Mitchell [2], Herzfeld et al. [4] and Fox and Hayes [5]). Recent advancements in multibeam and sidescan-bathymetric techniques have led to an improvement in resolution of the bathymetry obtained for high frequency sonars. This means that bathymetry data is useful as an addition to amplitude data in exploring the characteristics of complex areas. Visualisation methods which combine sidescan sonar data with bathymetry