Estimation of yellow rust severity in wheat using visible and thermal imaging coupled with machine learning models

Abstract

Reliable and quick estimation of wheat yellow rust (WYR) severity in field is essential to manage the disease and minimize the losses. Field experiments were conducted during 2017–18 and 2018–19 to obtain visible and thermal images of 24 wheat cultivars having different levels of WYR resistance at critical growth stages. Machine learning (ML) models were constructed using the combinations of image indices (IN) and partial least square regression (PLS) scores of image indices with disease severity (DS) and Yeo-Johnson (YJ) transformed values of disease severity. The results revealed that 26 visible and 2 thermal indices considered in this study have significant correlations with WYR. The models performances were evaluated using four possible dataset combinations of (1) disease severity + indices, (2) disease severity + PLS scores of indices, (3) YJ transformed disease severity + indices, and (4) YJ transformed disease severity + PLS scores. Disease severity with image derived indices was found to be the best dataset for the prediction of WYR severity using machine learning models with an R2 and d-index above 0.95 during calibration, while up to 0.67 and 0.87, respectively during validation. Cubist model with disease severity + indices dataset was the best to predict WYR severity, while the Gaussian process regression with YJ transformed disease severity + PLS scores dataset was the poorest predictor. The results obtained in the present study showed the potential of ML models for non-destructive prediction of WYR in field using visible and thermal imaging.