Land cover classification has become a key method for understanding natural and ecological resources, as well as for sustainable land-use planning and management. Advances in deep learning (DL) and automated satellite image analysis have significantly improved the accuracy, speed and scalability of land cover map production. This paper investigates the potential of instance-based segmentation within a GeoAI workflow for high-resolution land cover classification in the area of San Vito di Cadore (Veneto), a UNESCO mountain region with high ecological heterogeneity. The dataset comprises 650 manually annotated orthophotos at a spatial resolution of 0.1-0.5 m, labelled using seven main land cover classes (Forest, Shrubland, Grassland, Cropland,Water Bodies, Artificial/Urban Areas, and Rocky/Bare Areas), subsequently harmonized with the CORINE Land Cover (CLC) categories for inter-comparison purposes. Snow and cloud were treated as auxiliary classes due to their frequent occurrence in alpine orthophotos. The YOLOv11 instance-segmentation model was trained on 1000 × 1000 px tiles. During inference, a sliced-inference approach was adopted exploiting the SAHI (Slicing Aided Hyper Inference) framework, enabling the processing of large-scale orthophotos without degrading spatial quality. The results show an overall precision of 0.847 and an overall recall of 0.575, with mAP@0.5 greater than 0.65. Quantitative comparison with the Regione Veneto land cover product (2023) shows good agreement for the dominant forest class (-1.4%), whereas the largest discrepancies occur for artificial surfaces (+20.8%) and agricultural areas (-36.3%), attributable to differences in spatial scale and training-sample imbalance.