Quantitative Analysis of Vegetation Indices as Discriminators Between Different Crop Categories

Monitoring vegetation health is crucial for assessing ecosystem stability amidst climate change and environmental degradation. Hyperspectral imaging provides detailed data that can be systematized using various vegetation indices. Vegetation indices represent expert-determined values with relevance in analyzing plant growth and vigor dynamics or in sensing plant water status, among others. Many practical applications are built upon analysis of vegetation indices values. In this paper we investigate their discriminative power as features of a classifier and what is their relative importance in this discrimination. The proposed method involves a random forest classifier and is evaluated in context of agriculture crop segmentation on UAV-HSI database. Results indicate medium-to-strong discriminative power and relatively comparable importance, with none of the indices being dominant.