High-gradient gullies, prevalent on steep slopes in mountainous regions, drive severe soil erosion and landscape degradation. To address the limitations of conventional single-source remote sensing approaches, this study developed an automated identification framework for high-gradient gullies by integrating high-precision unmanned aerial vehicle (UAV) photogrammetry data, including digital orthophoto maps (DOM) and digital elevation models (DEM). Focusing on arid valley gullies in Liangshan Yi Autonomous Prefecture, a critical yet understudied erosion hotspot, this study employed statistically rigorous screening via Spearman's rank correlation to identify pivotal topographic indicators, and fused these with spectral features, textural features, and geometric features. Leveraging object-based image analysis (OBIA) alongside three machine learning algorithms: K-nearest neighbor (KNN), support vector machine (SVM), and random forest (RF). The results show the superior performance of the RF model, achieving highest classification accuracy with minimal overfitting risk, validated by reducing out-of-bag (OOB) error analysis at 4.87 % and 1.27 %. Integration of topographic data enhanced average accuracy by 2.11 %, increased the average Kappa coefficient by 0.092, and raised the average Area Under the Curve (AUC) value by 0.062. Feature importance on the RF model and SHAP analysis reveals that key drivers of model performance included hill shade (HS), surface cutting depth (D), and surface curvature (Curvature), which collectively resolved edge ambiguities and shadow interference. This methodology advances highprecision gully mapping in complex terrains and provides a scalable framework to integrate UAV photogrammetry with geomorphic analytics, offering practical insights for regional soil conservation and disaster mitigation strategies.