A multiresolution cooperating multiple-swarm pso algorithm for automatic target recognition in polarimetric SAR images
V. Sahajpal, O. Overerein · 2006
The PSO framework is shown to easily incorporate and fruitfully utilize the high correlation between specific SAR channels for target localization and recognition in high resolution polarimetric SAR imagery. Target localization utilizing normalized correlation over the entire image is replaced by PSO particles sparsely spread over the image over correlated SAR channel images. The sparsity of PSO particles reduces the total number of template matchings. The global exploration ability of PSO aided by extra information over multiple channels ensures that the sought target is localized with greater accuracy in majority of PSO runs, as against PSO seeking target over only one channel. The algorithm is made more viable by operating on low-pass sub-sampled coarser images using discrete wavelet transform and utilizing de-correlating properties of DWT. The multiplicative noise called speckle which affects coherent imaging systems, in our case SAR images, tends to de-correlate over wavelet subbands whereas the target's signature stands through.