Assessment of Fusarium Head Blight on wheat using near infrared hyperspectral imaging under field conditions
- Vincke, D. , Eylenbosch, D. , Chandelier, A. , Baeten, V. , Mercatoris, B. & Vermeulen, P. (2025). Assessment of Fusarium Head Blight on wheat using near infrared hyperspectral imaging under field conditions. Smart Agricultural Technology, 12: 101426.
| Type | Journal Article |
| Year | 2025 |
| Title | Assessment of Fusarium Head Blight on wheat using near infrared hyperspectral imaging under field conditions |
| Journal | Smart Agricultural Technology |
| Label | U12-0270-Vincke-2025 |
| Volume | 12 |
| Pages | 101426 |
| Date | 2025 |
| Abstract | Recent reviews pointed out several gaps in the current research on plant stress detection using hyperspectral imaging (HSI). Among these gaps are the limited number of HSI applications in field conditions and the limited validation of models’ specificity. Most studies focus on discriminating healthy plants from those affected by a single stress, without testing the models against other biotic or abiotic stresses that could cause false positives. Our work investigates these research questions by applying near infrared hyperspectral imaging (NIR-HSI) in field conditions to detect Fusarium Head Blight (FHB) in wheat and assess the technique’s ability to differentiate FHB from a take-all infection. A field trial was designed involving six winter wheat varieties sown outdoors in a cultivation bed. The trial was inoculated with Fusarium graminearum (causing FHB), and the plants were scanned on seven different dates using a mobile NIR-HSI system. During the growing season, take-all infections caused by a soil fungus (Gaeumannomyces graminis tritici) were spotted along with FHB. Therefore, models based on Partial Least Squares Discriminant Analysis (PLS-DA) were applied in a dichotomous classification tree to assess the ear’s general health status and attempt to discriminate both diseases. The models demonstrated a good assessment of the diseased ears (R²≥0.8) however, due to the symptoms’ similarity of both diseases, it failed to reliably differentiate them. This work also suggests potential avenues for further studies such as using Deep Learning methods and exploiting the differences in spatial distribution of both diseases to improve the model specificity. |
| Fichier | |
| Lien | https://doi.org/10.1016/j.atech.2025.101426 |
| Authors | Vincke, D., Eylenbosch, D., Chandelier, A., Baeten, V., Mercatoris, B., Vermeulen, P. |
Team
