SAR-image edge detection using artificial neural network

A task of edge detection in single-look synthetic aperture radar (SAR) images is considered. It is shown that edge detector performance can be improved by using an artificial neural network (NN). Selection of input local parameters is studied. As shown, it is reasonable to employ a limited number of elementary edge detectors that are the most informative (insensitive to noise) and that differ by the used principle of edge detection. ROC curves obtained for training and verification data sets are presented demonstrating efficiency of the designed NN based edge detector.