Optimized spectral sensing for reliable façade vegetation detection

Biological colonization significantly accelerates material decay on historical façades, posing serious conservation challenges. Traditional visual inspection remains labour-intensive and subjective, while existing remote sensing approaches developed for agriculture perform poorly when applied to cryptogamic organisms. This study addressed the need for precise, cost-effective spectral sensing methods suitable for UAV deployment. Using hyperspectral imagery from various façade materials at Sint-Martinus church in Massemen, Belgium, we evaluated three approaches: full-band machine learning classification with XGBoost, recursive feature elimination for band reduction, and exhaustive two-band vegetation index optimization. Our results demonstrated that full-spectrum classification achieves near-perfect accuracy (ROC-AUC > 0.999), while a minimal configuration of just four spectral bands maintains high performance (ROC-AUC of 0.994). Furthermore, an optimized two-band index (752 nm – 674 nm)/(752 nm + 674 nm) substantially outperforms conventional NDVI for façade-specific vegetation detection. These findings pave the way for transitioning from research-grade hyperspectral systems to cost-effective multispectral sensors for routine heritage inspection.