Marvel: Real-time pollen information
Together with our project partners, we develop zero-shot learning and other machine learning tools for recognising pollen particles anywhere in the world. As a result, it will be easier to create reliable pollen weather forecasts.
An estimated 10-40% of the global population has pollen allergy, and the number is expected to grow as the climate heats up. In Switzerland, MeteoSwiss helps allergy sufferers with its real-time pollen weather forecast. Our research partner, Swisens, has made this possible by providing MeteoSwiss with technologies for capturing and recognising pollen particles. The technologies are great at identifying common pollen types in Central Europe, but expanding to new geographical regions with different plant species is a hassle with the current methods.
We collaborate with Swisens to develop a pollen classification system that can be deployed anywhere in the world. Building the current system in Switzerland required a lot of manual work to prepare a labelled training dataset. This is not a feasible way to scale up globally, which is why our project will build a machine learning model that can identify a new pollen particle even if it has never seen it before. Swisens will then be able to turn this into a helpful pollen weather forecasting tool and a viable product for people suffering from pollen allergies around the world.
Our institute’s contribution to the project focuses on machine learning and artificial intelligence. We deploy deep learning with neural networks, and we expand into few-shot learning with active learning, as we collect and validate more and more pollen data.
The Marvel project is a good example of how an applied university operates: we work with state-of-the-art technologies and find real-life applications for them. As a result, our industry partner has a solution that can be put in immediate commercial use. And pollen allergy sufferers around the world will soon have access to even better forecasts and alerts.
Information
Partner | FHNW Institute for Data Science |
Duration | January 2024 – December 2025 |
Funding | |
Project Team | Martin Melchior (FHNW Institute for Data Science) |