New Method using vis-NIR and MIR for Detecting Fusarium Head in Wheat

When combined with machine learning algorithms, non-destructive spectroscopy techniques such as visible near-infrared (vis-NIR) and mid-infrared (MIR) spectroscopy can improve detection of Fusarium head blight (FHB) in wheat kernels and flour, according to a recent study in Chemometrics and Intelligent Laboratory Systems (1).

FHB is one of the most devastating fungal diseases affecting cereal crops, notably reducing yield and degrading kernel quality with mycotoxins that pose serious health risks to humans and animals (1,2). Traditionally, identifying FHB at the post-harvest stage involves extensive laboratory-based analyses, which are time-consuming, costly, and labor-intensive. Lead author Abdul M. Mouazen of Ghent University and his team explored the potential of vis-NIR and MIR spectroscopy in this space and whether they can serve as faster, cost-effective alternatives.

Landscape with tractor road in wheat field | Image Credit: © Ryzhkov Oleksandr - stock.adobe.com

To assess the viability of vis-NIR and MIR spectroscopy in this field, the researchers collected 143 ear samples of winter wheat varieties, with 93 infected and 50 healthy samples, from an inoculated trial (1). Then, the team collected the spectral data using vis-NIR in the wavelength range of 400 to 1700 nm and MIR in the wavenumber range of 4000 to 650 cm⁻¹ (1). Then, these data were analyzed using two machine learning algorithms: random forest (RF) and linear discriminant analysis (LDA).

The researchers discovered good results with both the RF and LDA models. Both RF and LDA models showed high test accuracy rates, particularly for flour samples (1). The LDA model achieved a perfect 100% accuracy in classifying flour samples, while the RF model achieved 96.6% (1). For kernel samples, the MIR spectroscopy combined with LDA provided a 93.1% accuracy rate. Furthermore, the application of recursive feature elimination (RFE) improved the accuracy of the vis-NIR models for kernel samples, with the LDA model reaching 100% classification accuracy (1).

One key finding was the impact of imbalanced data sets on model performance. The researchers recommended the use of an oversampling synthesis algorithm to enhance model reliability and performance. This approach addresses the challenge of imbalanced data, which can lead to suboptimal results (1).

Additionally, the study found that disease-resistant wheat varieties exhibited distinct spectral reflectance patterns compared to lower-resistance varieties. This differentiation contributed to the high accuracy of disease classification (1). The researchers found that using vis-NIR technique was effective for kernel samples because it achieved better results than when MIR was used (1).

Feature selection, especially RFE, was also key in maintaining accuracy. This approach was instrumental in designing straightforward and customized instruments for practical application in detecting FHB-infected kernels and flour (1). Notably, the RFE method significantly improved the accuracy of vis-NIR models for kernel samples, though it had less impact on the MIR models (1).

As a result, the study demonstrated that using a limited number of wavebands within the vis-NIR spectral range can effectively classify FHB infection in wheat. This finding suggests a pathway to more efficient, streamlined detection methods that could be readily adopted in the agricultural industry for quality control and disease management (1).

Overall, the research highlights the viability of vis-NIR and MIR spectroscopy techniques, augmented by machine learning models, as efficient, non-destructive methods for detecting FHB at the post-harvest stage. These advancements not only promise to enhance the speed and accuracy of FHB detection, but it also suggests significant benefits for the wheat industry in terms of cost, labor, and health safety (1).

References

(1) Almoujahed, M. B.; Rangarajan, A. K.; Whetton, R. L.; et al. Non-Destructive Detection of Fusarium Head Blight in Wheat Kernels and Flour Using Visible Near-Infrared and Mid-Infrared Spectroscopy. Chemometrics and Intelligent Laboratory Systems 2024, 245, 105050. DOI: 10.1016/j.chemolab.2023.105050

(2) Alisaac, E.; Mahlein, A.-K. Fusarium Head Blight on Wheat: Biology, Modern Detection and Diagnosis and Integrated Disease Management. Toxins (Basel) 2023, 15 (3), 192. DOI: 10.3390/toxins15030192