Optimal hyperspectral narrowbands for discriminating agricultural crops

Abstract

The main goal of this paper was to establish the best hyperspectral narrowbands for discriminating agricultural crops and to determine the accuracy with which such discrimination was possible. Six crops (wheat, barley, chickpea, lentil, vetch, and cumin) were studied. The best 12 narrowbands provided the most rapid increase in spectral discrimination. Further addition of narrowbands, only marginally increased discrimination capability reaching a plateau around 30 narrowbands. The overall accuracy (and K hat) in separating the six crops increased rapidly from 73% (K hat = 71) when 6 best bands were used to 84% (K hat = 79) when 12 best bands were used. Peak overall accuracies of 94% (K hat = 92) were achieved with about 30 narrowbands. Possibility of significant improvements by using mid infrared bands were indicated. Principal component derived hyperspectral narrowbands explained 61–92% variability in biomass and 70–87% variability in LAI. The best narrowbands to study agricultural crops have been recommended.