Ship Detection in SAR Images Using Neural Network Trained on Optical Images

The paper discusses automated ship detection in synthetic aperture radar (SAR) images by a neural network (NN) trained on grayscale optical images. It is shown that good detection quality can be achieved using this approach despite significant differences between radar and optical images. Since, unlike optical images, SAR images are always corrupted by strong multiplicative speckle noise, several speckle noise reduction filters were applied to SAR images before detection. Different filters have significantly varying effect on detection quality. NN based noise reduction filters outperformed traditional types (Lee filter, Frost filter, etc,). NN noise reduction filters were trained using grayscale optical images with artificial speckle noise.