Automated pollen recognition using 3D volume images from fluorescence microscopy

Identifying and counting of pollen grains in ambient air samples is still a demanding and time-consuming task even for an experienced microscopist. This article describes a technique which may be employed to establish a fully automated system for this task. Based on a 3D volume fluorescence image of a pollen grain taken with a confocal laser scanning microscope, the described system is able to recognize the pollen taxa. The system autonomously extracts all required information for the recognition from a data base with reference objects (self-learning system) and only needs to calculate very general purpose features of the volumetric data sets (so-called gray scale invariants). This allows for easy adaptation of the system to other conditions (e.g., pollen of a special area) or even other objects than pollen (e.g., spores, bacteria etc.) just by exchanging the reference data base. When using a reference data base with the 26 most important German pollen taxa, the recognition rate is 92%. With a special database for allergic purposes recognizing only Corylus, Alnus, Betula, Poaceae, Secale, Artemisia and “allergically non-relevant” the recognition rate is 97.4%. Abbreviations: LSM – Laser Scanning Microscope; SVM – Support Vector Machine; FFT – Fast Fourier Transform; pixel – picture element; voxel – volume element