Building detection from SAR (synthetic aperture radar) images using deep learning methods

SAR images are different from the optical images in terms of image properties with the values of scattering instead of reflectance. This makes SAR images difficult to apply the traditional object detection methodologies. In recent years, deep learning models are frequently used in segmentation and object detection purposes. In this study, we have investigated the potential of U-Net models for building detection from SAR and optical image fusion. The datasets used are Sentinel 1 SAR and Sentinel-2 multispectral images, provided from 'SpaceNet 6 Multi Sensor All- Weather Mapping' challenge. These images cover an area of 120 km² in Rotterdam, the Netherlands. As training datasets 20 pieces of 900 by 900 pixel sized HV polarized and optical image patches have been used together. The calculated loss value is 0.4 and the accuracy is 81%. The second dataset covers a part of İzmir province and consists of 2 SAR images, before and after the 2020 Aegean Sea earthquake. The image to be used as a mask in the U-Net algorithm was obtained from vector format. The model produced on the first dataset was applied on the second dataset and the results were discussed. Model has been applied on 3 different test dataset and classification results calculated 79%, 65% and 54% respectively.