Processing of AUV Sidescan Sonar Images for Enhancement and Classification

Abstract:

An arc volcano hosting a hydrothermal field was surveyed by using an autonomous underwater vehicle equipped with a sidescan sonar system and a multibeam echo sounder. The survey area is relatively small in area but has large variations in bathymetry and geology. To correct large geometric distortions in sidescan images, actual topographic cross sections cut by fan beams were taken into consideration instead of assuming flat bottoms. Beam pattern corrections were cautiously performed in combination with theoretical radiometric corrections for slant range and incident angle. These detailed geometric and radiometric corrections were efficient to patch neighboring images and build a complete picture of the whole survey area. Three textural attributes were computed from the corrected images by means of grey level co-occurrence matrices and used for the seafloor classification. As no ground truth data were available to us, we used a cluster analysis for the classification and obtained a result that seems relevant to the geological features suggested by the topography. Moreover, slopes of the caldera wall and of the central cones are clearly differentiated in the classification result, though the difference is not immediately obvious to our eyes. As one of the classes clearly delineates a known hydrothermal field, we expect by analogy that this class will highlight hydrothermal features in the survey area, helping to detect potential targets to be specifically investigated for mineral exploration. Numerical processing of sonar images effectively complements their visual inspection with human eyes and is helpful in providing a different perspective.