The research was conducted by the Lectorate AI & Big Data of Fontys University of Applied Sciences and Naturalis Biodiversity Center. The research involved students from Fontys ICT, HAS Green Academy and Leiden University. This collaboration shows how the forces between a lectureship, a research institute, undergraduate and graduate students can be combined. For three years, around 40 students from the various educational institutions worked on the research.
For instance, students from Fontys ICT set to work on a prototype of an app. This smartphone app allows you to photograph one square metre of a road verge. With the user's permission, the data collected can be forwarded to national databases on which biodiversity data is collected. In this way, AI helps spot trends (which plant species are doing well or not), but if done frequently, it can also measure the extent to which interventions have an effect on the biodiversity of the roadside.
To train the AI model, the ‘Eindhoven Wildflower Dataset’ (EWD) was used. The dataset takes its name from the region where photos were taken of roadsides, city parks, meadows and natural areas. The data consists of more than 2,000 images, showing over 65,000 flowers of some 160 species of plants. The EWD has been made publicly available for further research; thus, other researchers and parties can make even better AI models and compare their results.
Promising applications
The results of this research project are promising. Thanks to the applicable elaboration, various parties can immediately start working on practice-oriented projects aimed at biodiversity monitoring. It is easy to scale up by enriching the EWD with more plant species. This makes the application much broader. A few examples: water boards can thus plant and monitor their dykes more biodiverse. In addition, in organic agriculture, weeds can be detected using image recognition and removed with a robot. This prevents the use of harmful pesticides in the environment. Finally, the application can also be used, for example, for engagement, awareness and education of citizens on biodiversity. By using the low-threshold app, the results become accessible to everyone.
Gerard Schouten, one of the authors and professor of AI & Big Data at Fontys University of Applied Sciences, says: ‘What sets our work apart is the focus on the quality of the data. The method is scalable and can be refined to support different ecological use cases. It is essential to see AI not just as a replacement for human tasks, but as a complement that promotes synergy between humans and machines. In this way, we develop applications that help us enable a more sustainable world and society.’
Want to see the full research paper? Then click here.