Original paper: Early detection and classification of plant diseases with Support Vector Machines based on hyperspectral reflectance

摘要: Automatic methods for an early detection of plant diseases are vital precision crop protection. The main contribution this paper is a procedure the and differentiation sugar beet based on Support Vector Machines spectral vegetation indices. aim was (I) to discriminate diseased from non-diseased leaves, (II) differentiate between Cercospora leaf spot, rust powdery mildew, (III) identify even before specific symptoms became visible. Hyperspectral data were recorded healthy leaves inoculated with pathogens beticola, Uromyces betae or Erysiphe causing respectively period 21 days after inoculation. Nine indices, related physiological parameters used as features automatic classification. Early plants well among can be achieved by Machine radial basis function kernel. discrimination resulted in classification accuracies up 97%. multiple three still accuracy higher than 86%. Furthermore potential presymptomatic demonstrated. Depending type stage disease 65% 90%.