Modeling of spectral data characteristics of healthy and Fusarium diseased corn kernels

Ts. Draganova
Abstract. Two approaches for identification of infected with Fusarium maize kernels using spectral characteristics in the near infrared region are described in the paper. They are used for spectral data reduction and for corn classification. First approach is based on analysis of discrete linear parametric models coefficients. Second approach is based on principal component analysis. Maximum percentage of correct rate achieve 74,67% for Fusarium infected corn kernels when linear discrete model coefficients are used as classification features. Principal components show overlapped classes but in combination with appropriate classifier as classification tree percentage of correct rate achieve 100% for Fusarium infected corn kernels. The main advantage for the classifier which is used (classification tree) is that it is created using simple procedure without any additional parameters accept training data set and target for each input variable.