SAR image classification using a neural classifier based on Fisher criterion

A supervised neural classifier based on Fisher criterion is implemented to classify two regions in a real speckled SAR image. Regions around pre-classified pixels are presented to train the neural network that learns a sub-optimal set of masks via back-propagation algorithm. Classification performance is evaluated by using the ground truth. Results with higher than 90% of correct classification are obtained. The results are also compared with a statistical classifier based on Kullback-Liebler distance via the Kappa coefficient.