Abstract: In open-pit coal mining, the mining environment is complex and changeable, and the deformation of the mine slope caused by the change of the overall structure of the mine brings major hidden dangers to safety production and will also bring certain damage to the ecological environment. At present, although the deformation monitoring of mines is gradually becoming automated, the whole process still needs to rely on manual or various monitoring equipment, and the maintenance of the equipment is relatively difficult, and the operation of some equipment is more complex and the cost is higher. In order to monitor the deformation information of mine slope more easily and intuitively, a deformation monitoring method of mine slope based on laser point cloud registration was realized by combining the learning-based point cloud registration method. Firstly, a deep learning model SA-RPE (Self-Attention with Relative Position Encoding) was proposed, which realized the point cloud registration on the mine dataset and was verified by experimental data. Then, according to the registration results, the deformation analysis of the mine slope was carried out, and the deformation degree of each section was further analyzed by intercepting the sections of the mine slope point cloud in different directions, and the experimental results of the two-dimensional cross-section and the three-dimensional point cloud rendering showed that the deep learning model SA-RPE could accurately realize the registration task of the mine laser point cloud. By analyzing the errors of the rotation matrix and the translation vector in the registration results predicted by the deep learning model, the overall deformation information of the mine can be well grasped, and the cross-sectional diagram of the mine slope point cloud in different directions intuitively shows the degree of deformation in each place, and the average distance between the corresponding points of the cross-section point cloud in different periods can quantitatively describe the deformation degree of each section, and the outliers detected by the threshold reflect the area with large deformation on the cross-section. The experimental results show that the proposed method can accurately and intuitively represent the deformation information of mine slope on the basis of satisfying real-time performance.