Abstract: Surface fractures caused by underground coal mining are the intuitive manifestation of mining subsidence failure. It is of great significance for mining su bsidence research and mining surface protection to accurately obtain fracture development characteristics in the subsidence area. At present, the main method of collect ng mine fracture information is field measurement or satellite remote sensing image interpretation,which has obvious limitations in cost efficiency and spatial and temp oral resolution. In this paper,the surface of a fully mechanized caving face of a mine in Yushan Mining area is taken as the test area,and the digital orthometric image of the mining subsidence area is obtained by means of low-altitude uav photometric measurement technology. Canny algorithm,support vector machine( SVM)and Maximu m likelihood method are respectively used to extract the characteristics of mining surface fractures. The results show that the maximum likelihood method is effective in extracting the information of crack length and width,but there is a lot of information of dry vegetation in the obtained crack image. For this purpose,the random forest( R F) algorithm was used to classify vegetation and perform corrosion operations on the images, and mask files were prepared to eliminate the influence of vegetation. The n,maximum likelihood method was used to accurately extract surface fractures. According to the gray scale characteristics of surface fracture images,the mining cracks in the test area can be divided into crack type stair steps,closed type fractures and sag type,striped cracks,cracks,dents type crack closure steps.In addition,the surface fracture information extracted from the image was compared with the measured results to verify the reliability of the low-altitude UAV image automatic fracture extractio n,which provides an effective technical means for quantitative analysis and revealing the development law of fractures in mining subsidence areas.