Towards Automated Regulation of Jacobaea Vulgaris in Grassland Using Deep Neural Networks
The highly poisonous ragwort (Jacobaea Vulgaris) is increasingly spreading, posing significant risks to agriculture, livestock, and nature conservation due to the production of toxic pyrrolizidine alkaloids (PAs). The current manual control methods, such as plucking weed, are labor-intensive and time-consuming. This paper introduces a workflow towards automated regulation of J. Vulgaris, which consists of the two independent tasks of deep learning-based monitoring and controlling. We aim to detect and control J. Vulgaris in an early growth stage before the plant can reseed, which challenges the data collection and the training of deep neural networks. Primarily we need to detect the green leaf rosettes on a green meadow. The main focus lies on the monitoring part with synthetic training data generation and a deep neural network-based labeling assistant.