Exploring the frontiers of computer vision techniques for the automated identification of toxic weed species

Computer vision and digitalization are becoming impactful tools in precision agriculture. In these fields, object detection using deep learning technologies has many potential applications, one of them is the automated detection of toxic weeds, which threaten food safety. This PhD study develops a methodology to automatically detect toxic weed species such as different Senecio species, which contain highly toxic pyrrolizidine alkaloids (PAs). Various multispectral cameras and camera platforms, with a focus on aerial drones, are used to achieve this task. Vision transformers and other detection models optimized for edge computing will be used to determine whether it is possible to differentiate different Senecio and other weed species. Field surveys will be conducted in Lower Austrian herb and spice fields using high-resolution UAV imagery to capture phenological variability and environmental conditions. Through interviews and surveys with Lower Austrian farmers, this study will investigate the priorities, considerations, and practical implications of automated toxic weed detection. The developed pipeline aims to reduce PA contamination, minimize resource waste, and support compliance with EU guidelines. Additionally, this study's methodology could be applied to broader applications in sustainable crop management and food safety, such as detecting other weeds, or plant diseases.