Classification of high-resolution SAR imagery by Random Forest classifier

Land cover classification based on SAR data has been a key point of urban researches for a long time. Recently, the successful launch of low earth orbit satellites, such as the German TerraSAR-X and the Italian COSMO-SkyMed provide the high resolution SAR imagery to researchers. Though a lot of feature analytical methods using scattering intensity, statistical values and texture are proposed for SAR urban classification, how to use these features to get a reliable and consistent land cover classification result is still a very difficult problem. In our study, we introduce a Local Index Spatial Analysis (LISA) to describe the texture of the SAR image and carry some experiments to analyze the performance of the LISA. The classification strategy based on Random Forest (RF) is also presented. A COSMO-SkyMed scene in Chengdu city and a TerraSAR-X scene in Beijing city are used in our experiments. The high accuracy of the classification shows that the high resolution SAR image has a potential to improve the classification accuracy, especially urban area.