Abstract: Accurate stability evaluation of slope remains a fundamental challenge in geotechnical engineering and geological hazard mitigation, where the key lies in accurately identifying the spatial geometry and location of potential slip surface. Traditional investigation method, which rely on boreholes and subsurface geophysical exploration, are inherently limited by their “ point-to-surface” nature, making it difficult to obtain real three-dimensional slip surface in an efficient and cost-effective manner. Moreover, existing three-dimensional limit equilibrium method commonly depend on idealized slip surface assumptions, which deviate from actual geological conditions. To address this limitation, this study proposes a novel collaborative detection and non-contact identification framework for slip surface by integrating slope radar (deformation detection), threedimensional limit equilibrium analysis (mechanical inversion), and video-based technique (apparent geological recognition). First, real aperture radar is employed to acquire millimeter-level precision three-dimensional surface deformation field. Potential sliding mass boundaries are delineated using deformation gradients and spatial clustering, serving as strong constraints for subsequent three-dimensional inversion. Subsequently, deep learning and image semantic segmentation techniques are utilized to construct an apparent geological model, enabling the extraction of lithological zoning, fracture network, and weak surface distribution. These results provide geological priors and crossvalidation for determining the spatial position of slip surface. Based on these multi-source constraints, a three-dimensional critical slip surface analytical method is introduced to perform optimized searching of parameterized slip surface, ultimately identifying the most critical three-dimensional slip surface with the minimum factor of safety. A case study conducted at an open-pit mine in northwestern China demonstrates the effectiveness of our proposed method. Deformation data capturing the full evolution process of the landslide were obtained using portable real-aperture radar, and, combined with video recognition and geological profile data, a three-dimensional slip surface model consistent with field observations was successfully constructed. The results verify the feasibility and effectiveness of the proposed approach for non-contact identification of deep-seated slip surfaces. By incorporating physics-based three-dimensional analytical modeling under multi-source data constraints, the proposed method provides a robust solution for detecting complex and high-risk landslide hazard. It offers critical support for hazard early warning, reinforcement design, and emergency decision-making, and demonstrates significant potential for practical engineering applications and industrial deployment.