Manss, ChristophMartel, ViktorWeisgerber, Roman2024-04-082024-04-082024978-3-88579-738-82944-7682https://dl.gi.de/handle/20.500.12116/43898Process-monitoring for autonomous mowers in agriculture is crucial to establish an online quality assessment. Here, neural networks (NNs) are employed to classify ground conditions, distinguishing between dry, mowed, unplanted, and grass. The data comprises RGB images that are captured by a camera mounted on a mower. These images are then used to train various NNs, with EfficientNet_V2_s emerging as the most accurate network and with ResNet18 to be the most efficient network in terms of training duration and accuracy. The study also reveals for this use-case that employing transfer learning enhances the overall network performance. The developed NNs is intended for deployment on mowers, enabling them to adjust their mowing blades, conserve energy, and enhance the quality of mowed grass. Beyond mowing, the NN can be applied in process control and the identification of other plant species or weeds in the agricultural field, contributing to biodiversity assessments and more sustainable farming practices.enclassificationprocess controlautonomous systemsText/Conference Paper10.18420/giljt2024_371617-5468