Speaker
Description
The plant community composition and species phenology are important indicators of environmental changes and subject to numerous ecological studies. In most previous studies, data on plant communities was collected by hand, making the data collection process laborious, time-intensive and subject to human error and subjectiveness. Additionally, the amount of work required for this process prohibits a collection in a high temporal frequency, as this would require infeasible amounts of work. Weekly observations are, however, important to generate high-quality data on species phenology. Hence, most studies merely cover data collected in large intervals, leading to temporally coarse subsequent ecological analyses.
We introduce a novel tool based on Convolutional Neural Networks called PlantCAPNet, which can be utilized to extract information about the species composition of herbaceous plant communities, and the phenology of all included species automatically from images. In conjunction with automated camera systems, PlantCAPNet offers a powerful way to extract vegetation data in a high quality and temporal frequency with little manual labor required. Therewith, our tool enables temporally fine-grained ecological analyses, offering novel insights into the dynamics of plant communities and their responses.
The tool will be made available to plant ecologists as an open-access application. The system is developed in interdisciplinary cooperation between biologists and computer scientists.
| Status Group | Postdoctoral Researcher |
|---|
