Optimized image processing for multi-beam/multi-aspect sidescan sonar application

Abstract

In the development of future sonor systems, computer-aided classification (CAC) becomes increasingly important. Adequate processing of the multidimensional image data is one important component in a multi-beam/multi-aspect sidescan sonar CAC system. Because the target strength of an object varies with aspect angle, the sonar echo and its representation in a sidescan image is rather random. To make sure that a maximum echo strength is gained, in multi-beam/multi-aspect sidescan sonars overlapping bottom areas are insonified by temporal successive pings, giving the echoes of the target under different aspect angles. Taking advantage of the multi-aspect echo structure in the images of successive pings, they are fused to one sidescan image. Existing image fusion algorithms now require an operator to set threshold values distinguishing target and shadow zones from bottom reverberation zones. An unsupervised method is proposed here for segmentation and optimal fusion of multidimensional images. Its basic steps are image segmentation by means of the EM algorithm; clustering of image segments applying the Markov random field approach; and fusion of successive pings images following a minimum/mean/maximum strategy according to the results of image segmentation and clustering.