Evaluation of fusarium head blight infection on winter wheat using near infrared hyperspectral imaging from the laboratory to the field.

Agriculture has to face the fundamental challenge of feeding a growing world population using limited resources. Doing so, it also needs to cope with the challenges of climate change and food safety. In this context, European Agriculture has to meet the increasing demands of sustainability, which implies using less inputs such as pesticides, fertilizers and water.
Varietal breeding is considered the main way to tackle these challenges by producing varieties that tolerate stressing environments while maintaining high yields. In this process, plant phenotyping plays a key role by characterizing plants in various environmental conditions and quantifying plant traits that are desirable for breeding. This approach requires evaluating many plants in different environments and involve the use of sensors and automation. To do so, imaging sensors are particularly useful because they combine spectral and spatial information with the advantages of a non-destructive analysis. Furthermore, technological developments allow producing advanced sensors such as hyperspectral imagers. Such technology enables the detection of very subtle changes in the optical properties of the plants or the crop. Up to now, most of the research in plant phenotyping with hyperspectral imaging is focusing on the visible range of the electromagnetic spectrum. Meanwhile, the near infrared range remains scarcely studied despite some conclusive results.
This research aimed at assessing the potential of near-infrared hyperspectral imaging for the phenotyping of winter wheat with a focus on the detection of the fungal disease Fusarium Head Blight (FHB). The first step of this work consisted in acquiring a better understanding of the information contained in the spectral signature of a winter wheat ear. It investigated the relationship between the spectral signature of the ear and its constituents such as the hull and the kernels. This part of the work allowed identifying spectral bands related to presence or absence of kernels in the ear. In addition, this step helped distinguishing the spectral bands related to water.
The second step of this work consisted in developing a method using near infrared hyperspectral imaging in the laboratory for the assessment of the severity of FHB infection on winter wheat ears. The results indicate that wavelengths related to water are important for the detection of FHB. In addition, the method showed good performance for the detection of FHB-infected ears with a sensitivity of 99.4 % and a specificity of 91.9 %.
The last step of this work consisted in transferring the method and knowledge acquired in the lab towards the detection of FHB in the field. In this perspective, a dedicated field trial was designed and the plants were scanned in the field using a NIR-HSI system mounted on a mobile gantry. The NIR-HSI data were used to develop a method detecting FHB infected ears and evaluate the severity of the infection (percentage of the plot area affected by FHB). In addition, the performance of the method was assessed by comparison to the visual scoring of FHB incidence in the field (percentage of FHB-infected ears in the plot). Over the course of the trial a take-all infection occurred simultaneously with the FHB infection. This co-occurrence of two biotic stresses was considered as an opportunity to test the ability of the method in classifying both stresses. The results indicate that the method allow a good discrimination between healthy and diseased ears with accuracies > 90 %. However, the models did not allow a satisfactory classification between FHB and Take-all. This suggests that the method allow assessing the general stress status of the plants but more developments are required in order to discriminate accurately each individual stress.