Pollen Classification Based on Binary 2D Projections of Pollen Grains
Halil Akcam, Volker Lohweg
Abstract
Abstract Pollen is one of the main causes of allergic diseases in humans. Therefore, it is indispensable to develop and conduct effective treatment and prevention measures. For this purpose, detailed and differentiated information about the respective local exposure profiles for the individual patients is required. The present paper serves the purpose of testing a new approach which aims at detecting and classifying individual pollen grains by using binary 2D projection. This paper explores the question of whether and to what extent a classification of individual pollen grains is possible using this new imaging technology. To this end, using artificial pollen grains, binary 2D projections with different levels of resolution are simulated. To extract the respective features, both shape-based Fourier descriptors and topological features are used. Apart from that, Zernike moments for different orders are measured to extract the respective characteristics of the pollen grains. While the feature selection is conducted by means of a feature forward selection method, a kernel machine ( Support Vector Machine ) with a Gaussian kernel is used for the classification. First results of the simulation show that with a resolution of 0.1 μm, 100% of the allergologically relevant artificial pollen are classified correctly. Conversely, a lower resolution corresponds with a higher error rate in the classification.