Comparative Analysis of Machine Learning Algorithms for Classification of UAV-based Photogrammetric Cultural Heritage Point Clouds

Unmanned Aerial Vehicles (UAVs) are being increasingly utilized across different fields because of their ability to deliver quick, cost-effective, and precise spatial information. Innovations in photogrammetry and computer vision techniques, especially Structure from Motion (SfM) and Multi-View Stereo (MVS), have improved the generation of orthoimages, digital surface models, and dense point clouds, rendering UAVs highly efficient for documentation and three-dimensional reconstruction. In studies focused on cultural heritage, UAV-based photogrammetry has emerged as a crucial resource for accurately preserving and representing historical sites with great detail and resolution. In this context, the current study analyzes UAV-acquired point cloud data from the Temple of Hera in Italy and performs a comparative assessment of three machine learning algorithms, Support Vector Machines (SVM), Random Forest (RF), and Extreme Gradient Boosting (XGBoost), for the purpose of semantic segmentation tasks. According to our results, the XGBoost and Random Forests (RF) methods has reached to more than 90% F1 score for all classes, and the SVM method has reached 90% F1 score only for three classes.