Applications of an image segmentation technique to multibeam echo-sounder data

A segmentation algorithm has recently been presented for seafloor characterization from multibeam echosounder sonar images, accounting for angular variations of the backscattered signals obtained from training zones. This approach allows one to suppress the image effects due to seafloor topography and angle dependance. It is applied here to some real cases both in deep and shallow water, with data from two different echo-sounders at 13 kHz and 95 kHz. For every case the raw image, the angular backscattering patterns obtained on training zones, and the segmentation results are presented. The various limitations of the method are finally discussed: difficulties around the vertical of the sonar, too low number of training data, high dependence of the results on the quality of the echo-sounder processes and corrections. It is shown that such a segmentation is really effective despite non optimal data and difficult geological configurations