A comparison of ground-based hyperspectral imaging and red-edge multispectral imaging for façade material classification

This paper compares the feasibility of classifying building façade materials in situ via a low-cost multispectral imaging camera with 5 bands compared to a high-cost, short-wave infrared hyperspectral unit with 240 bands. The sensors were used to classify five common façade materials: brick, mortar, stone, painted window frames, and rendering. The data were subjected to a Partial Least Squares Discriminant Analysis by selecting 95 training pixels and 65 validation pixels for each material. Across all materials, the hyperspectral imaging demonstrated a moderate advantage over multispectral imaging (with a correct classification rate of 99% versus 80%) for the selected data and stronger qualitative matching to the entire façade. While the hyperspectral data were consistently superior, the multispectral data still produced valuable results, thereby demonstrating potential as a fast, easy, and much cheaper technology compared to hyperspectral units for in situ façade materials classification.