Pollen Identification and Concentration Estimation Using High Resolution Imaging and Deep Learning

Climate change, coupled with the rise in pollen concentrations and extended pollen seasons, poses significant challenges to public health, particularly for individuals with pollen allergies and/or respiratory diseases. In response to these challenges, the work here's objective is to pioneer an automated trainable system for the recognition and counting of pollen grains. Such a system would enhance efficiency, allowing for quicker and more accurate assessments of airborne pollen concentrations, ultimately aiding in the mitigation of allergy symptoms associated with changing environmental conditions and the spread of allergenic species. This approach utilizes image processing tools to segment pollens on digitalized slides and several deep learning tools to recognize them among 17 different allergenic species in total. The system has been designed to avoid the time spent by palynologists on the microscope and to considerably increase the number of observation sites above the current European standards in allergenic pollen concentrations evaluation. In turn reducing associated errors with such analysis.