, A. S. J.
2000 Second European Symposium on Aerobiology, Vienna, Austria, 5-9 Sep
, A. S. J., (2000), "Image analysis applied for aerobiology", Second European Symposium on Aerobiology, Vienna, Austria, 5-9 Sep.
The need for an automated identification scheme for airborne particles such as pollen has been widely recognised for some time. As demand for aerobiological studies steadily increases this requirement will eventually become a barrier that must be overcome in order to undertake larger studies and routine monitoring in a financially viable way. Image analysis is one of several approaches to this problem that offers some promise. This approach involves digital measurement of features from each pollen grain, including morphological features such as size and shape, and surface characteristics such as texture and roughness. A study based on analysis of over 16,000 images of individual pollen grains from 80 plant species, where several hundred features were measured from each grain, has demonstrated that it is possible to construct classification keys based on numeric data alone that have accuracy's approaching 70-80%. The major difficulty in developing these automated identification keys is determining what combination of measurable features of the pollen grains are sufficiently characteristic and of low enough variance to provide adequate differentiation of each pollen species from all the others. This requires measurement of feature variances from a large number of pollen grains, followed by careful statistical analysis of an extremely large data set (Over 9 million individual measurements in this case). The results of this study and issues of feature measurement will be discussed.