Hyperspectral Imaging Study of Wheat Grains Infected with Several Fusarium Fungal Species and Their Identification with PCA-Based Approach

Abstract

Wheat is an important agricultural crop grown under various conditions on five continents. The ability to promptly detect and defeat fungal diseases has a significant impact on the volume of the obtained harvest. One of the most significant threats to human and domestic animal health is metabolites produced by Fusarium genus fungi. In this regard, this work is devoted to the possibility of the rapid differentiation between healthy grains and grains simultaneously infected with several species of Fusarium genus fungi (Fusarium graminearum Schwabe FG-30, Fusarium poae Kr-20-14, Fusarium roseum (sambucinum) St-20-3) for practical reasons. An approach based on obtaining hyperspectral data with their subsequent processing using the principal component analysis (PCA) method and determining statistically important spectral regions sensitive for grain infection at different stages (5 and 40 days) was proposed. The effects of the grain orientation and data dimensionality on the classification result were studied. For further practical application in devices for the rapid identification of wheat grains infected with Fusarium, a method based on the use of reflection at wavelengths of 400, 451, 708, 783, 801, and 863 nm, together with normalization [0, 1] and the subsequent projection of spectral data onto the first three principal components (PCs), was proposed, regardless of the grain orientation.

Keywords: Fusarium; feature wavelength; hyperspectral imaging; reflectance; wheat.

Figure 1

Examples of Raman spectra from the grains infected for 40 days (red) and healthy grains (green).

Figure 2

FTIR spectra of the grains infected for 40 days (red) and healthy grains (green).

Figure 3

Normalized reflectance spectra of healthy grains from the top (a) and bottom (b) sides (lines with symbols); normalized reflectance spectra of grains infected for 5 days from the top (c) and bottom (d) sides (lines with symbols); and normalized reflectance spectra of grains infected for 40 days from the top (e) and bottom (f) sides (lines with symbols). The thick lines show the averaged spectra for different cases.

Figure 4

Percentage of variance from the principal component number (blue), as well as the total percentage of explained data (red).

Figure 5

Score graphs for the first 3 PC pair combinations, (PC1, PC2) (a), (PC1, PC3) (b), and (PC2, PC3) (c), and the corresponding 95% probability ellipsoids, as well as the loading graphs for the first 3 PCs in the case of the full-range PCA (d).

Figure 6

Score graphs for the first 3 PC pair combinations, (PC1, PC2) (a), (PC1, PC3) (b), and (PC2, PC3) (c), and the corresponding 95% probability ellipsoids, as well as loading graphs for the first 3 PCs in the case of selected bands in the PCA (d).

Figure 7

Scores graphs for the first 3 PC pair combinations, (PC1, PC2) (a), (PC1, PC3) (b), and (PC2, PC3) (c), and the corresponding 95% probability ellipsoids, as well as loading graphs for the first 3 PCs in the case of a selected set of feature wavelengths (d).

Figure 8

Comparative photographs of healthy grains (a) and grains after 5 (b) and 40 (c) days of infection.

Figure 9

Schematic representation of the setup for taking samples, indicating the main elements.