Seabed classification of multibeam sonar images

Seabed images, from multibeam hydrographic systems or from single or multibeam sidescans, convey a lot of information about seabed type. Statistical processing of portions of images can generate features adequate for seabed classification that agree with both large-scale interpretation and fine details. Here we describe Quester Tangent's QTC MULTIVIEW(TM) system for statistical seabed classification, and present several results. This software uses many statistical algorithms to generate over 130 statistical features for each image patch. Principal components analysis extracts the linear combinations of features that best describe the variance in a data set of images. The selection of statistical features is thereby optimized for a particular sediment discrimination problem, rather than a set of features selected for general use. Data points are then assigned to classes by an established clustering process. Examples demonstrate the limitations imposed by sonar physics and by design details of multibeam systems. For accurate classification results, these limitations require respect, either through compensation steps in the processing or excluding pings acquired under non-standard conditions.