New Advances in Quantitative Brightfield Microscopy

Fecha: Miércoles 26 de Noviembre.
Hora: 12h-13h.
Ponente: Dr. Gabriel Cristóbal - Instituto de Óptica (CSIC) 
Lugar: Aula A-21 de la Facultad de Ciencias.

Organiza: El Máster oficial Erasmus Mundus "Color in Informatics and Media Technology (CIMET)”, como parte de su actividad académica y dentro de las actividades del próximo Año Internacional de la Luz 2015.

Abstract
Manual identification of pollen samples is a time consuming process, requiring the presence of highly trained palynologists. Routine palynological studies often require analysing thousands of individual pollen grains, most frequently using transmitted brightfield microscopy (e.g. 40X magnification). Therefore, there is a need for an automated classification system which can provide an accurate analysis of the samples. The two main problems that are difficulting the current progress in this field are (a) the extraction of knowledge from expert palynologists and (b) a limited access to open pollen databases with a large number of reference pollens per taxa. Pollen analysis can be accomplished through different stages: acquisition, pre-processing, segmentation, feature extraction and classification. Robust automatic segmentation methods are needed for avoiding dust
and pollution or the presence of aggregates (e.g. by clump splitting). Previous work in the area of aero-palynology (ASTHMA EU project) used multifocus stacks and reported recognition rates around 77% for 30 pollen types. In this work, a large dataset of features taken both from the contour (exine) and from the inner part of the pollen (texture) have been extracted from few representative examples of european bee pollen. In particular 15 taxa with 120 reference pollen grains per taxa were analyzed in the context of the Apifresh EU project. The accuracy reached around 99% of recognition of the pollen grains using the leave-one-out validation model.