[A New Discrimination Method of Maize Seed Varieties Based on Near-Infrared Spectroscopy].

Guo Ting-ting,Wang Shou-jue,Wang Hong-wu,Hu Hai-xiao,An Dong,Wu Wen-jin,Xia Wei,Zhai Ya-feng

DOI: https://doi.org/10.3964/j.issn.1000-0593(2010)09-2372-05

2010-01-01

Abstract:A new discrimination method for the maize seed varieties based on the near-infrared spectroscopy was proposed. The reflectance spectra of maize seeds were obtained by a FT-NIR spectrometer (12000-4000 cm(-1)). The original spectra data were preprocessed by first derivative method. Then the principal component analysis (PCA) was used to compress the spectra data. The principal components with the cumulate reliabilities more than 80% were used to build the discrimination models. The model was established by psi-3 neuron based on biomimetic pattern recognition (BPR). Especially, the parameter of the covering index was proposed to assist to discriminating the variety of a seed sample. The authors tested the discrimination capability of the model through four groups of experiments. There were 10, 18, 26 and 34 varieties training the discrimination models in these experiments, respectively. Additionally, another seven maize varieties and nine wheat varieties were used to test the capability of the models to reject the varieties not participating in training the models. Each group of the experiment was repeated three times by selecting different training samples at random. The correct classification rates of the models in the four-group experiments were above 91.8%. The correct rejection rates for the varieties not participating in training the models all attained above 95%. Furthermore, the performance of the discrimination models did not change obviously when using the different training samples. The results showed that this discrimination method can not only effectively recognize the maize seed varieties, but also reject the varieties not participating in training the model. It may be practical in the discrimination of maize seed varieties.