ON THE UTILITY OF COHERENCE FOR UNDERWATER OBJECT DETECTION IN SAS IMAGES

Abstract: A large-scale study is undertaken to assess the capability of detecting underwater objects in sonar imagery. The data used in the analysis are synthetic aperture sonar (SAS) images containing various man-made objects. The data were collected with the MUSCLE autonomous underwater vehicle (AUV) during six large sea trials, conducted between 2008 and 2012, in different geographical locations with diverse environmental conditions. The detection algorithm for which performance is assessed is a cascaded, integral-image-based approach. The analysis examines detection performance for specific target shapes as a function of target-sensor aspect, range, image quality (based on spatio-temporal coherence), and environment (defined in terms of two seabed features). To our knowledge, this study – covering approximately 158 square kilometers of seabed imagery, and involving over 1400 target detection opportunities – represents the most extensive such systematic, quantitative assessment of object detection performance with SAS data. The analysis reveals that, for detection performance, the image quality (or spatio-temporal coherence) is a more important quantity than range, aspect, or shape. Dependence on the environment, and in particular seabed characterized by sand ripples, is also established. It is then explained how these results can be exploited, by intelligently adapting AUV survey routes, to maximize detection performance over an area of interest