SAR Image Change Detection Using Modified Gauss-Log Ratio Operator and Convolution Neural Network

Detection of change in synthetic aperture radar images (SAR) has been considering an area of study in remote sensing and computer vision. It is important to classify the modified feature in the different SAR images obtained during various observing periods. Imaging of synthetic aperture radar (SAR) has been recognized as effective in the recognition of change than other imagery sensors, particularly when the earth is observed with cloudy, rainy, or hazy weather. The images taken by the synthetic aperture radar are low contrast and low brightness grayscale image. So, it is difficult to find the changes in SAR images. In this article, we proposed a detection of changes system where we use a convolution neural network (CNN) as a classification model, Modified Gauss-Log Ratio used to find the difference image of the two sample images. The classification result from CNN and the preclassification outcome is used to find out the final change map. The reliability and robustness of the proposed approach were shown by investigative findings using two SAR image data sets.