Abstract: The crop growth status of mining subsidence farmland is the key to determine the extent of land damage and then affect the difficulty of land reclamation and the choice of reclamation plan，and also the basis for compensation of mining enterprises.Using image recognition method to determine the crop yield boundary of mining subsidence farmland can reduce the workload of field measurement and the influence of human factors.Based on the principle that the height difference of crops in different growth conditions can be expressed as the height difference of digital surface model，the height difference of crops was taken as the indicator of yield boundary.Using UAV to obtain high-resolution image and digital surface model (DSM).Using crop elevation as the threshold of image segmentation improved OTSU algorithm，it included two aspects：the discretization of elevation data and the change from gray value operation to elevation value operation.Combined with Canny operator edge detection，proposed a method of recognition out-of-production boundary in mining subsidence farmland.And taken a mining subsidence area in Jining City of Shandong Province as the research area to verified the method.The study results showed that the improved OTSU algorithm could calculate the elevation value of image，and iterated the optimal threshold value of elevation segmentation.The digital surface model (DSM) was segmented by the threshold value of elevation segmentation，and the out-ofproduction boundary could be recognition by edge detection algorithm.Through the verification of DOM sampling in the study area，the crop yield boundary recognized by this method was accurate，the effectiveness of method was proved.