Mean-standard deviation representation of sonar images for echo detection: Application to sas images

Abstract—This paper addresses the detection of underwater mines echoes with application to synthetic aperture sonar (SAS) imaging. A detection method based on local first- and second-order statistical properties of the sonar images is proposed. It consists of mapping the data onto the mean–standard deviation plane high-lighting these properties. With this representation, an adaptive thresholding of the data enables the separation of the echoes from the reverberation background. The procedure is automated using an entropy criterion (setting of a threshold). Applied on various SAS data sets containing both proud and buried mines, the pro-posed method positively compares to the conventional amplitude threshold detection method. The performances are evaluated by means of receiver operating characteristic (ROC) curves. Index Terms—Mean–standard deviation representation, seg-mentation, sonar image processing, synthetic aperture sonar (SAS), Weibull law. I