Premium
Strain selection of Metarrhizium anisopliae by image analysis of colony morphology for consistency of steroid biotransformation
Author(s) -
Yang Kui,
Wang Jifen,
Li Xiaojing,
Feng Xia,
Duan Shiduo
Publication year - 2001
Publication title -
biotechnology and bioengineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.136
H-Index - 189
eISSN - 1097-0290
pISSN - 0006-3592
DOI - 10.1002/bit.1164
Subject(s) - pattern recognition (psychology) , artificial intelligence , feature selection , classifier (uml) , multifractal system , mathematics , fractal analysis , linear discriminant analysis , fractal , feature vector , support vector machine , fractal dimension , computer science , mathematical analysis
Abstract Quantitative description of colony morphological images of the fungus Metarrhizium anisopliae was done by means of computer image analysis based on fractal analysis. Compared with simple fractal analysis, the differences in the morphological features of colonies were expressed more accurately by multifractal analysis. The use of multifractal analysis markedly improved the reliability of colony image analysis, which could be extended to morphological characterization for colony recognition and classification. First, the multifractal spectrum was calculated from each morphological image and 14 features were defined on the spectrum to describe the spectral line profile. Second, three features, α‐right, α‐width, and f‐start, were selected from the 14 features in two‐dimensional feature vector spaces and a feature augmented vector ( α‐right, α‐width, and f‐start, 1) T was generated and used for classifier design. Third, a statistical least mean square error (LMSE) algorithm was applied to design a piecewise‐linear classifier, which consists of two linear classifiers, d 1 (x) and d 2 (x). Representative colony samples were used as training and test sample sets, which were previously classified into a high‐activity class and another class by a training person. After passing the training phase, the piecewise‐linear classifier could be used for automated classification of unknown colony samples. A correct recognition rate of 96% was achieved in this manner. Compared with conventional strain selection methods, this new method has the advantage of less subjectivity, reduced labor and time, and improved ability to deal with a large number of colonies. © 2001 John Wiley & Sons, Inc. Biotechnol Bioeng 75: 53–62, 2001.