Developed an automated methodology for building classification by integrating multiband imagery composed of LiDAR-derived metrics, orthophotos, and elevation models. Applied segmentation algorithms (Quickshift, Felzenszwalb, and Watershed) to identify candidate objects, followed by feature extraction and supervised machine learning using Scikit-learn. The models were trained using manually digitized building footprints obtained from high-precision imagery, enabling accurate learning of spatial and spectral patterns. Among the evaluated classifiers (Random Forest, k-NN, SVM), the best performance was achieved with the combination of Quickshift segmentation and a Support Vector Machine.