Open AccessImage analysis in automatic system of pollen recognition
Author(s) -
Piotr Rapiejko,
Zbigniew M. Wawrzyniak,
Ryszard Jachowicz,
Dariusz Jurkiewicz
Publication year - 2012
Publication title -
acta agrobotanica
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.4
H-Index - 9
eISSN - 2300-357X
pISSN - 0065-0951
DOI - 10.5586/aa.2006.040
Subject(s) - pollen , feature vector , identification (biology) , artificial intelligence , pattern recognition (psychology) , computer science , feature (linguistics) , segmentation , object (grammar) , computer vision , dimension (graph theory) , feature extraction , mathematics , ecology , linguistics , philosophy , botany , pure mathematics , biology
In allergology practice and research, it would be convenient to receive pollen identification and monitoring results in much shorter time than it comes from human identification. Image based analysis is one of the approaches to an automated identification scheme for pollen grain and pattern recognition on such images is widely used as a powerful tool. The goal of such attempt is to provide accurate, fast recognition and classification and counting of pollen grains by computer system for monitoring. The isolated pollen grain are objects extracted from microscopic image by CCD camera and PC computer under proper conditions for further analysis. The algorithms are based on the knowledge from feature vector analysis of estimated parameters calculated from grain characteristics, including morphological features, surface features and other applicable estimated characteristics. Segmentation algorithms specially tailored to pollen object characteristics provide exact descriptions of pollen characteristics (border and internal features) already used by human expert. The specific characteristics and its measures are statistically estimated for each object. Some low level statistics for estimated local and global measures of the features establish the feature space. Some special care should be paid on choosing these feature and on constructing the feature space to optimize the number of subspaces for higher recognition rates in low-level classification for type differentiation of pollen grains.The results of estimated parameters of feature vector in low dimension space for some typical pollen types are presented, as well as some effective and fast recognition results of performed experiments for different pollens. The findings show the ewidence of using proper chosen estimators of central and invariant moments (M21, NM2, NM3, NM8 NM9), of tailored characteristics for good enough classification measures (efficiency > 95%), even for low dimensional classifiers (≥ 3) for type differentiation of pollens grain