Potato Late Blight Detection at the Leaf and Canopy Level Using Hyperspectral Data

Abstract

This study aims to assess which spectral variables and at which time late blight can be detected over potato crops. Two experiments were done in a walk-in chamber under controlled environments. To determine the time, the reflectance spectra were plotted as a function of the day post inoculation (DPI), then a Principal Component Analysis (PCA) was applied and the Jeffries–Matusita distance was computed between healthy and infected leaf or canopy spectra. A spectral ratio between infected and healthy cases was used to determine the best wavelengths. Additionally, the spectra were used to compute reflectances and associated vegetation indices for the five bands of the Micasense® RedEdge camera and a Partial Least Square-Discriminant Analysis (PLS-DA) was applied to the reflectance spectra to assess how well the infected leaves or plants can be separated from the healthy ones. The leaf level spectra gives a better separability. A good separability is achieved at 2 and 5 DPI at the leaf and canopy levels, respectively. The best bands are the blue, green, red, and red-edge. The best vegetation indices are SR, Clgreen, RI, TCARI, TCARI/OSAVI-2, ClRed-Edge, and Red-Edge NDVI. The maximum overall accuracy for the PLS-DA is observed at 4 DPI (91.11%) and at 5 DPI (85.93%) at the leaf and canopy level, respectively.