Fractal dimension based sand ripple suppression for mine hunting with sidescan sonar

Sand ripples present a difficult challenge to current mine hunting approaches. We propose a robust and adaptive method that suppresses sand ripples prior to the detection stage. The method exploits a fractal model of the seabed and the connection between: dual-tree wavelets and local, directional fractal dimension; interscale energy ratios, scale invariant frequency localised fractal dimension, and a novel wavelet shrinkage approach. Tests on a reasonably large, real synthetic aperture sonar imagery dataset show that the ripple suppression method preserves detection performance of the matched filter on non-rippled data and significantly increases the detection performance on data that contain ripples.