Robust automated evaluation of drone images in plant breeding using artificial intelligence

Global food security is seriously threatened by plant diseases, with one of the most dangerous epidemic fungal diseases in cereal production being Fusarium head blight. On the one hand, the yield is reduced, on the other hand, the quality of the harvest is drastically damaged. In the worst case, there is a serious health risk from the mycotoxins in the grain resulting from the fungal infestation. The aim is therefore to develop resistant varieties in order to avert the problems caused by this disease. In plant breeding, those varieties and new breeding lines that are resistant to the disease must be identified and selected, a task that is usually carried out by trained personnel. Since classical selection is very time-consuming, expensive and prone to error due to the human factor, such a procedure is enormously time-consuming in practice for larger selection programs: an automated approach is required. To avoid these problems, the aim of the project is to use a drone to automatically capture high-resolution images of the field plots and classify the different test lines. As there is no suitable technology for the specific task (high-resolution images, extreme flight altitude, oblique images), a new system for image acquisition must be developed. In addition, existing methods for image analysis are not suitable in this setup due to the highly variable environmental conditions. A new robust and generally applicable approach is required. In addition, the effort for labeling the data must be reduced in order to ensure a practically applicable system that is accepted in plant breeding. This requires close cooperation between breeders (from a plant science perspective) and computer scientists (from a technical perspective).