Abstract: An effective approach of extracting land cover information from fully polarimetric SAR images has been compensating for the lack of optical remote sensing data. This approach has also promoted the application of SAR technology in resources and ecological monitoring in coal mining areas. With abundant polarimetric information, speckle and large local heterogeneity of fully polarimetric SAR images, this study proposed an object-oriented image classification approach for hierarchical extraction of land cover information. The study area was a coal mining region in the southwest of Xuzhou City and the SAR image acquired by Radarsat-2 for a fully polarimetric image. The gray features of typical surface objects of SAR images for the study area were analyzed. The optimal segmentation scale for object-oriented image classification was proposed. The computation approach of backscatter features of segmentation objects was also presented. Also ground-truth data were collected on the land cover to verify the SAR image for the study area. In the first step, the SAR image was segmented into several scales and the optimal segmentation scale of each land cover type selected. In the second step, the backscatter feature indexes of each land cover type were calculated under the optimal segmentation scale. Then in the last step, the land cover information in the study area was hierarchically extracted using the fuzzy logic classification method. The results showed that the proposed approach accurately extracted the five types of land cover information under the optimal segmentation scales of land cover type, including farmland, road, subsidence land, building and mountain woodland. In terms of gray, shape, texture and class-related features of the segmentation objects, the membership function and nearest neighbor classification approaches weakened the "pepper phenomenon". Compared with the maximum likelihood classification method, the classification accuracy of each land cover type improved while the average classification accuracy improved by 38.3%. In particular, the classification accuracy of farmland was as high as 90.2%, an improvement of 42.0% due to use of texture features in the classification.