Summary

Reverberation, which is caused by the reflexions, diffractions and/or diffusions of the transmitted signal on the surface or the bottom of ocean interfaces, is a very disturbing noise in active sonar. Signal processing algorithms allow to obtain a sonar image where the target echo is described by a few energetic pixels. But the target detection on this image is difficult since the reverberation leads to the appearance of other highly energetic features. In this paper, we show that the target and the reverberation echoes have different morphological and statistical properties. Using these differences, two image processing based detection algorithms are proposed. Starting from a geometrical model, the first algorithm is composed of different filters issued from the mathematical morphology theory. The second algorithm is derived from the contrast box filter and is based on a local statistical model. Tested on real sonar data, these algorithms both give interesting results. To take advantage of both methods and further improve the results, the detection algorithms are then combined together: they can be applied either sequentially (one after the other), or parallelly with a fusion of the results. As the experiment contains multiple recurrences, we then propose to use this information redundancy. A fusion algorithm between the different recurrences is used in order to delete ground reverberation echoes. This final algorithm leads to a robust detection of the target (good detection probability) with almost no reverberation echoes remaining (good false alarm probability)