Land Cover Classification of PolSAR Images Using Semantic Segmentation Networks

Abstract

With the free access to the SAR images obtained by the Sentinel 1 satellite, it provided the opportunity to make large scale land cover mapping using these images. This increased the need for fast and high-performance classification of large-scale SAR images. In this study, semantic segmentation networks such as SegNet and U-net have been proposed to obtain a fast and high performance classification. Also to train and test these recommended methods, a new data set consist of dual polarization SAR images of Turkey and ground truths has been created.