Quantitative Analysis the Protein of Millet by Artificial Neural Network and Fourier Coefficients of Near Infrared Diffuse Reflectance Spectroscopy

For biomaterial, the signal energy is concentrated at low frequencies, and then the first few Fourier coefficients can represent the whole spectrum. Fourier coefficients are useful wavelength reducing method. The first few Fourier coefficients of millet's near infrared diffuse reflectance spectroscopy were used as the input nodes of artificial neural network, to build the quantitative analysis calibration model of protein in millet. The advantages of this method are that Fourier coefficient can reduce spectrum, filter the high frequency noise with an ideal filter of unity gain and zero phase shift. Better results were obtained from artificial neural network quantitative analysis model, the correlation coefficient and relative standard deviation of protein is 0.971 and 2.40% in calibration set, 0.955 and 2.96% in prediction set respectively. These results were satisfactory.