Swisens AG produces sensors based on holographic imaging and fluorescence spectroscopy to measure pollen concentrations in the air. The determination of the type of pollen detected by the sensors is achieved by a deep learning-based classifier which has been trained on a large, annotated dataset. For expanding to new markets, the current process of adapting their system to new regions should be simplified so that the expensive collection and annotation of data with local pollen can be largely avoided. In the scope of this project we explored the use of zero-shot learning techniques to achieve this goal. First results are promising. However, further work is needed to improve the classification performance.

Project Partners:
• Erny Niederberger, Yanick Zeder (Swisens AG): Provider of data, engineering of data, domain knowledge
• Martin Melchior, Roman Studer (FHNW/HT): Data Science and AI competencies