High resolution images provided by synthetic aperture sonar sensors (SAS) are of great interest, especially for the detection, location and classification of mines laying on the sea bed. For this purpose, this paper proposes a method based on local statistical characteristics of the sonar image. Its goal is to highlight specificities of the echoes of the mines by using a new representation of the data. The results can then be used for a classification of the different kinds of under-water mines.