The Loess Plateau, due to its unique geomorphological structure, loose loess properties, and the combined effects of monsoonal heavy rainfall, experiences frequent landslide disasters characterized by small scale and high quantity, seriously threatening regional ecological security and infrastructure stability. In recent years, although landslide recognition methods integrating InSAR technology with deep learning have shown promising prospects, they still face challenges such as high false alarm rates and insufficient ability to detect weak deformations and small-scale targets. To address these challenges, this study proposes an intelligent recognition method for potential landslide slope deformation zones by integrating InSAR deformation information with the YOLOv8-PLDC model. Based on the Stacking-InSAR technique, Sentinel-1 SAR data were processed to obtain the regional annual average surface deformation rate, and sample datasets of landslide hazard and non-hazard areas were constructed. Using Ansai District as a demonstration area, experiments combining Google Earth 3D optical imagery and DEM data were conducted. While improving identification accuracy, the method effectively eliminated non-landslide deformation areas caused by human engineering activities and other factors. Results show that the method successfully identified 179 potential slope deformation zones, with an accuracy of 93.3% and a recall rate of 96.3%. This approach fully utilizes the advantages of Stacking-InSAR in acquiring high spatiotemporal resolution deformation information and combines it with the multi-scale target detection and feature extraction capabilities of the YOLOv8-PLDC model, achieving accurate identification of weak deformation and small-scale landslide hazards, thereby providing critical technical support for intelligent landslide hazard identification in the Loess Plateau region.