Abstract: In this paper, Honghe Town, Pengyang County, Guyuan City, Ningxia was selected as the research area, an orthophoto with a resolution of 0.13 m and an area of 174 km² was obtained by UAV. After analyzing the image features of the landslide elements at risk, the test area is set. Two object-oriented methods, multi-scale segmentation and single-scale segmentation, are used in the experimental area, and the extraction accuracy and time efficiency of the two methods are compared. Through the optimization of the scheme, the information of landslide elements at risk in the study area is extracted. The results show that: (1)By introducing the local variance index reflecting the homogeneity of the segmentation results, the optimal segmentation scale of multi-scale segmentation was carried out for the experimental area. By constructing the characteristic rules of spectrum, range and shape, the corresponding landslide elements at risk information could be extracted at different levels successively; (2)On the basis of multi-scale segmentation, a single segmentation scale was set up in the experimental area by combining trial calculation. Although the overall accuracy and Kappa coefficient of elements at risk information extraction were lower than that of multi-scale segmentation, there was little difference, and the time required was less than a quarter of that of multi-scale segmentation; (3)Considering the spatial distribution of landslide elements at risk in the study area, the image feature information and the extraction results of landslide elements at risk in the experimental area, the single segmentation scale optimization scheme of the study area was determined. The research results are expected to provide necessary references for regional landslide elements at risk information extraction, landslide risk assessment and risk management.