By Sabrina Matteucci
Side scan sonar image segmentation through multi-resolution texture features: a case study over the Luce Bay site during Northern Light 2003. Grasso, Raffaele ; Spina, Francesco ; Hagen, Per Espen ; Bjorn, Halvin. SR-413. August 2005.
The segmentation of a side scan sonar (SSS) data set acquired in the Luce Bay site during the Northern Light '03 exercise by a Hugin autonomous underwater vehicle (AUV) equipped with an EdgeTech SSS, is presented and discussed. The segmentation was performed through the analysis of the image texture using a set of un-decimated wavelet transform features followed by classification using a supervised classifier based on the Mahalanobis distance in the feature space. The classifier was trained on four main sea floor classes, including low and high reflectivity sediments and sand ripples having two different ripple wavelengths, which are well discriminated in the feature space. A simple post-processing with geographic information system (GIS) tools was applied to smooth the data and convert pixels into polygons for a final output in AML/S57, the format adopted for data upload in Command and Control Information Systems (CCIS). The final thematic map showing the classification of the whole data set contains two distinct areas, the first in the inner leg (IL), showing no appreciable variations in the sea bed characteristics and the second, namely the outer leg (OL), having more variability.