DETECTION OF CONTAMINATED AGAR MEDIA BY MULTISPECTRAL IMAGING AND LINEAR DISCRIMINANT ANALYSIS

摘要

This study applied a multispectral image system to detect the contaminated agar media. Among dozens spectral, four bands of wavelength were chosen and applied in this research. They were 300～400, 560～620, 400～770, and 850-870 nm. Images of normal and contaminated agar medium in petri dishes were taken at the four bands from day 2 to day 7 after samples were prepared. Images were also extracted to form 9 texture features after they were processed by gray level transform, threshold and edge removal. Classification methods named Linear Discriminant Analysis (LDA) were also applied in couple with 9 texture features to separate normal and contaminated agar medium. The results showed that LDA calibrated by 200 calibration samples and tested by 100 testing samples at 300～400 nm and 400～770 nm wavelength image taken on day 3, could obtain the highest classification rate of 92%. This result was also analyzed by confusion matrix and scatter diagram to show type of classification error and samples numbers of misclassification.