[Study on discrimination of varieties of corn using near-infrared spectroscopy based on GA and LDA].

A new method for the fast discrimination of varieties of corn based on near-infrared spectroscopy using genetic algorithm and linear discriminant analysis (LDA) was proposed. First, data of NIS of 37 varieties of corn was collected, second, genetic algorithm used for choosing the feature band of spectrum, then PCA and LDA were used to extract features, and finally corn seeds were classified. The result showed that GA could remove noise band effectively and improve the generalization ability of LDA. A large number of redundant data was removed to simplify the computing, which resulted in the data dimension reduction from 2075 to 233. For the 300 samples of test set one, the average correct recognition rate and average correct rejection rate attained 99.30% for both, and the average correct recognition rate of 73.33% varieties of corn attained for 100%. For the 175 samples of test set 2 (all of whose varieties had not been trained), the average correct recognition rate attained 99.65%. The run time is shorter and the correct rate is higher compared to the common method of PCA.

[DFT feature analysis of corn varieties based on near infrared spectra].

The present paper develops a new approach to the analyse of corn based on discrete Fourier transform (DFT). The experiment data is of 37 varieties of corn seed with the Fourier transform near infrared spectrometer in the wave number range from 4 000 to 12 000 cm(-1). Analyse of the origin data found that as the wave number increases, the data noise also increases. Firstly, the paper defines a calculation method of interspecific and intraspecific differences Qm to measure the effectiveness of feature selection. Secondly, Qm was used to analyse the original data and DFT-section data. Experimental results show that by choosing data of DFT with wave number range from 4 000 to 7 085 cm(-1), the mean value and the peak value of the the Qm curve markedly improved relative to the full band original data. The mean value was enhanced from the original 4.804 9 to 8.513 8, and the max of the peak value was enhanced from the original 35.924 0 to 60.821 6, while the min of the peak value was enhanced from the original 2.891 8 to 3.741 5. Data feature points (Qm value of large point) are more concentrated than the original data after DFT. Such a result is most conducive to extracting the characteristics of corn seed.

[A frequency selection method of NIR spectroscopy for discrimination of maize seed varieties].

A frequency selection method of NIR spectroscopy was proposed in the present paper for discrimination of maize seed varieties. A criterion function was defined to evaluate the discriminative ability of NIR spectroscopy at different frequencies, and then features of maize seed varieties were extracted accordingly for further processing. By eliminating correlation between features at different frequencies, the selected features are guaranteed to contain as much information of inter-variety difference as possible. Also, features with larger variances are preferred to suppress the impact of noise. Experiment results demonstrate that our frequency selection method can achieve high recognition rate with less spectroscopy features than traditional methods. Specifically, a recognition rate as high as 94.16% can be attained with NIR spectroscopy with only 30 frequencies. Simulation results show that recognition rate of NIR spectroscopy at selected frequencies is stable with small disturbance of frequencies, which verifies the robustness of the authors' method.

[Feature analysis and discrimination of varieties of corn based on near infrared spectra].

A new method for the discrimination of varieties of corn was proposed based on the data set of near-infrared spectroscopy range from 4 000 to 12 000 cm(-1) of corn seed varieties. Principal component analysis (PCA) method was used to study the feature of the data, and the authors found that the near-infrared spectroscopy of corn seed varieties has a clear feature of zonal distribution, so the correlativity between the change in the distribution of the principal component and the discrimination result was studied, according to which the normalized principal component analysis (NPCA) method was proposed. Besides, principal direction biomimetic pattern recognition (PBPR) was proposed according to the feature, which got a better discrimination result. The average correct recognition rate attained 97.67% for test set I, and the average correct rejection rate attained 98.40%, with 13 of the 30 varieties reaching the correct recognition rate of 100%; The average correct rejection rate attained 98.90% for the test set II , and 11 of the 30 varieties reached the correct rejection rate of 100%. It was proved that the method had a high correct discrimination rate.