The sidescan sonar records the energy of an emitted acoustical wave backscattered by the seabed for a large range of grazing angles. The statistical analysis of the recorded signals points out a dependence according grazing angles, which penalizes the segmentation of the seabed into homogeneous regions. To improve this segmentation, classical approaches consist in compensating artifacts due to the sonar image formation (geometry of acquisition, gains, etc.) considering a flat seabed and using either Lambert’s law or an empirical law estimated from the sonar data. The approach chosen in this study proposes to split the sonar image into stripes in the swath direction; the stripe width being limited so that the statistical analysis of pixel values can be considered as independent of grazing angles. Two types of texture analysis are used for each stripe of the image. The first technique is based on the Grey-Level Co-occurrence Matrix (GLCM) and various Haralick attributes derived from. The second type of analysis is the estimation of spectral attributes. The starting stripe at mid sonar slant range is segmented with an unsupervised competitive neural network based on the adaptation of Self- Organizing Feature Maps (SOFM) algorithm. Then, from the knowledge acquired on the segmentation of this first stripe, the classifier adapts its segmentation to the neighboring stripes, allowing slight changes of statistics from one stripe to the other. The operation is repeated until the beginning and the end of the slant range are reached. The study made in this work is validated on real data acquired by the sidescan sonar Klein 5000. Segmentation performances of the proposed algorithm are compared with those of conventional approaches.
