Vegetation Detection on Heritage Facades: Limitations of NDVI and a Case-Optimized Alternative

Abstract

Remote sensing technologies have emerged as essential tools for assessing the condition of heritage buildings, with applications ranging from documentation to conservation planning. This study specifically investigates the utility of the Normalized Difference Vegetation Index (NDVI) for detecting vegetation on façades of cultural heritage buildings. Employing hyperspectral imaging data acquired via an Unmanned Aerial Vehicle (UAV), this research evaluates NDVI’s performance in the context of the Castle of Horst, a historical monument in Belgium undergoing restoration. The methodology includes photogrammetric reconstruction, radiance-to-reflectance correction using spectral reference panels, and systematic analysis of spectral data to distinguish vegetation from non-vegetated materials. The evaluation demonstrates NDVI’s capability to detect vegetation accurately, while also revealing significant limitations such as sensitivity to varying illumination conditions and misclassification due to indirect lighting effects. To address these limitations, this paper proposes and validates a novel Case Optimized Index (COI), derived through exhaustive spectral band analysis, exhibiting superior classification accuracy compared to NDVI alone. Additionally, an XGBoost classifier further confirms the effectiveness of combining hyperspectral and RGB data, emphasizing the potential of machine learning techniques in enhancing vegetation detection accuracy. This research contributes practical insights into optimizing vegetation indices specifically for cultural heritage conservation, informing future methodologies for non-invasive façade assessment.