A system for quantitative determination of species and vitalities of urban trees on color-infrared photographs

Aerial color-infrared photographs of two Swiss cities (Bern and Lausanne) were taken to evaluate the potential use of automatic image analysis for the discrimination of urban tree species and their different vitalities. A TV-based hardware system under full software control measured five different functions at each of the 63 possible intensity levels at red (600 nm), green (550 nm), blue (470 nm), and white illumination. Eighty-three parameters per tree were derived from the corresponding intensity distributions describing tone and texture. Classification by supervised learning was carried out using stepwise discriminant analysis. The overall classification error for the discrimination of four deciduous tree species (horse chestnut, Norway maple, large-leaved lime, and London plane) was 4.9%. A higher error rate (6.2%) was found for the discrimination of five classes, namely healthy trees of the above mentioned four species and damaged horse chestnut trees. The practicability of the proposed system is discussed