Fusarium head blight monitoring in wheat ears using machine learning and multimodal data from asymptomatic to symptomatic periods

Abstract

The growth of the fusarium head blight (FHB) pathogen at the grain formation stage is a deadly threat to wheat production through disruption of the photosynthetic processes of wheat spikes. Real-time nondestructive and frequent proxy detection approaches are necessary to control pathogen propagation and targeted fungicide application. Therefore, this study examined the ch\lorophyll-related phenotypes or features from spectral and chlorophyll fluorescence for FHB monitoring. A methodology is developed using features extracted from hyperspectral reflectance (HR), chlorophyll fluorescence imaging (CFI), and high-throughput phenotyping (HTP) for asymptomatic to symptomatic disease detection from two consecutive years of experiments. The disease-sensitive features were selected using the Boruta feature-selection algorithm, and subjected to machine learning-sequential floating forward selection (ML-SFFS) for optimum feature combination. The results demonstrated that the biochemical parameters, HR, CFI, and HTP showed consistent alterations during the spike–pathogen interaction. Among the selected disease sensitive features, reciprocal reflectance (RR=1/700) demonstrated the highest coefficient of determination (R2) of 0.81, with root mean square error (RMSE) of 11.1. The multivariate k-nearest neighbor model outperformed the competing multivariate and univariate models with an overall accuracy of R2 = 0.92 and RMSE = 10.21. A combination of two to three kinds of features was found optimum for asymptomatic disease detection using ML-SFFS with an average classification accuracy of 87.04% that gradually improved to 95% for a disease severity level of 20%. The study demonstrated the fusion of chlorophyll-related phenotypes with the ML-SFFS might be a good choice for crop disease detection.

Global Trends and Future Directions in Agricultural Remote Sensing for Wheat Scab Detection: Insights from a Bibliometric Analysis

“…The studies also used SVM and RF models, but KNN outperformed the competing models. Moreover, the studies have also concluded that the comparative performance of disease-specific or newly developed indices for model fitting is highly improved against conventional indices [45,89] that support different other studies [99]; based on these results, a precise FHB monitoring program can be developed. Moreover, the better disease estimation performance of models through disease-specific indices or bands has also been proved by previous findings [42,99].…”

“…Numerous studies have used univariate and multivariate quantitative models for disease estimation in ARS for diverse plant diseases [98]. Different studies estimated the scab disease at spike [89] and canopy scale [45] using KNN predictive model. The studies also used SVM and RF models, but KNN outperformed the competing models.…”

Wheat Fusarium Head Blight Automatic Non-Destructive Detection Based on Multi-Scale Imaging: A Technical Perspective

“…[96] and Zhang et al [97] captured wheat ears from the front and back to obtain spectral information on them with possible differences in infection levels on each side. Mustafa et al [98] captured the top, middle, and bottom of each wheat spike from both sides to obtain the chlorophyll content of each partition. Multi-directional sampling not only provides more comprehensive and accurate spectral information but also reduces the influence of natural factors such as atmosphere, lighting, and shadows through averaging processing, improving the uniformity of surface reflection of objects.…”

“…Moreover, data fusion breaks the shackles of traditional feature selection methods and provides more comprehensive considerations for decision-making under fine FHB severity grading indicators. In the heterologous data fusion, Mustafa et al [98] utilized the Machine Learning Sequential Floating Forward Selection (ML-SFFS) algorithm to classify the severity of FHB infection into nine levels based on multimodal data fused with HSI, CFI, and HTP; Mahlein et al [101] refined the classification criteria to ten levels by fusing data from three imaging techniques: IRT, CFI, and HSI. Both of them achieved an average accuracy of over 85%.…”

“…Due to the varying relative importance of each input indicator depending on the severity of the disease [108,109], many research results based on symptomatic wheat are difficult to directly apply to asymptomatic detection. Mustafa et al [98] successfully amplified the internal weak infection characteristics of asymptomatic wheat by utilizing CFI and combining sensitive features, achieving an average classification accuracy of 87.04%. The appearance of asymptomatic wheat infected with FHB and the healthy one is highly similar, so the acquisition of asymptomatic samples and high errors in manual annotation have always been the difficulties of the detection task.…”