Seafloor segmentation based on bathymetric measurements from multibeam echosounder data

Bathymetric data depicts the geomorphology of the seabottom and allows characterization of spatial distributions of apparent benthic habitats. The variability of seafloor topography can be defined as a texture. This prompts for the application of well developed image processing techniques for automatic delineation of regions with crucially different physiographic characteristics. In the present paper histograms of biologically motivated invariant image attributes are used for characterization of local geomorphological features. This technique can be naturally applied in a range of spatial scales. Local feature vectors are then submitted to a procedure which divides the set into a number of clusters each representing a distinct type of the seafloor. Prior knowledge about benthic habitat locations allows the use of supervised classification, by training a support vector machine on a chosen data set, and then applying the developed model to a full set. The classification method is shown to perform well on the multibeam echosounder (MBES) data from Piscataqua river, New Hampshire, USA.