Early Detection of Aphanomyces Root Rot in Pea Plants Using Hyperspectral Imaging

Abstract

Plant leaf and root reflectance resulting from plant–pathogen interaction can be informative about disease status, making them useful for early disease detection. In controlled conditions, this research utilized a hyperspectral imaging (HSI) system to evaluate the early response of pea plants (Pisum sativum L.) inoculated with Aphanomyces euteiches Drechs, the causal agent of Aphanomyces root rot (ARR), using hyperspectral imaging. Two ARR partially resistant lines (NIL5-7.6b and NIL8-7.6b) with the quantitative trait locus (QTL) Ae - Ps7.6 and corresponding controls (NIL5-0b and NIL8-0b, without QTL) were grown in hydroponic conditions and organized in a split-plot design using two treatments, non-inoculated and inoculated (1 × 105 zoospores ml−1) with six replications. The HSI data were collected from the youngest leaflets 3 days after inoculation (DAI). At 8 DAI, HSI data from roots were collected. The HSI hypercubes of leaflets and roots were processed to remove the background and extract the mean value of each sample across wavelengths. Leaflet hyperspectral signatures were used to calculate normalized difference spectral indices. Then, a recursive feature elimination with cross-validation and a random forest classifier was used to select important features and test them with inferential analysis. For root data, a similar approach was used, however, the selected important features were used in random forest and gradient boosting classifiers. The leaflet results showed the red-edge wavelength of 745 nm was an essential feature for treatment separability at 3 DAI. Meanwhile, root analysis displayed a high classification accuracy of 83% and 92% with random forest and gradient boosting, respectively. This research offers valuable insights into the potential of HSI for ARR detection, particularly in the early pre-symptomatic stages of the plant disease.

Keywords: Aphanomyces euteiches, plant-pathogen interaction, classification, feature extraction, phenotyping