Real-time pollen monitoring across Europe with the Swisens Poleno by a deep-learning classifier: laboratory and field validation

Real-time air-flow cytometry has been rapidly expanding as a method for monitoring airborne biological particles, providing continuous measurements with high temporal resolution. This is particularly relevant for pollen monitoring, where accurate and timely information is needed for health-related applications. In this study, we present results from measurements performed with the Swisens Poleno air-flow cytometer, an automated instrument combining light-scattering, holographic imagery, fluorescence excitation, and polarization measurements to detect and classify airborne bioaerosols.

For data analysis, a deep-learning pollen-recognition classifier was used to target the most common pollen taxa in Europe. The classifier was trained on pollen samples provided to the device under laboratory conditions and achieved an average classification accuracy above 90%, with most errors occurring between morphologically similar taxa. Performance in real atmospheric measurements was expectedly lower. To evaluate and correct this, classifier-processed Poleno measurements were compared with co-located measurements from manual Hirst-type traps across Europe. A transposed confusion-matrix correction was applied to account for systematic misclassifications, improving agreement with reference data. The resulting performance was further evaluated for Poleno measurements available through the EU Horizon SYLVA project.

These results demonstrate that combining real-time cytometry with machine-learning and correction techniques provides a reliable and effective approach for automated pollen monitoring, supporting the broader advancement of bioaerosol observation and health-related applications.