Pre-symptomatic leaf reflectance of Fusarium virguliforme infected soybean plants in greenhouse conditions

Abstract

Sudden death syndrome (SDS) of soybean is caused by a soil-borne pathogen, Fusarium virguliforme. Prior to visible foliar symptoms, a destructive technique is usually carried out to diagnose root infection. The use of hyperspectral sensors for pre-symptomatic and non-destructive plant disease diagnosis has been on the rise. This study was designed to relate leaf spectral reflectance to F. virguliforme root infection in the absence of foliar symptoms. Soybean plants were grown under controlled greenhouse conditions. The plants’ spectral reflectance was measured weekly beginning at 21 days after transplanting (DAT) up until 42 DAT using a swing hyperspectral imaging system that is fixed on a gantry. Destructive root sampling confirmed F. virguliforme root infection using Real-time PCR. The most relevant wavelengths for discrimination were selected using the ReliefF algorithm. Three machine learning models [Partial least squares discriminant analysis (PLS-DA), support vector machine, and random forest] were evaluated for classification accuracy using the selected wavelengths. Relevant wavelengths for differentiating between the healthy and F. virguliforme infected plants were found in the visible and red-edge region from 500 to 750 nm, and the shortwave infrared region from 1400 to 2350 nm. In the absence of visible foliar symptoms, classification results showed over 79% mean F1-scores for all models. PLS-DA was able to differentiate healthy and F. virguliforme infected plants with a mean F1-score of 83.1 to 85.3% and a kappa statistic of 0.43 to 0.54. This work supports the use of hyperspectral remote sensing for early pre-symptomatic disease diagnosis under controlled environment.