Identifying content of low profile target in cluttered environment using the BioSonar

In the mine counter measure context image based automatic target recognition is confronted to mainly two limitations: in a cluttered and heavily cluttered environment recognition algorithms observe a drastic increase in the PFa (probability of false alarm) making the output results unpractical at the best. The second main limitation is unknown threats such as IEDs (Improvised explosive devices) or simply unknown types of mines. In this paper we present classification results from a trial done in Portland harbour, Weymouth, UK in October 2012. The test targets were 9 identical gas cylinders (65 cm height, 30 cm diameter) filled with three different contents: 3 cylinders filled with seawater, 3 with sand and 3 with gravel. The targets were placed on two different and highly reflective seabed types (mud with broken shells). The estimated SNR of those low profile targets in these difficult environments is negative making them almost undetectable in traditional sidescan images. We demonstrate here the capability of the wideband BioSonar to identify and distinguish between the 3 types of cylinders. We present in situ learning using spiral reacquisition pattern and show that the wideband classification is robust to environment variations.