Studies On The Test Method Of Camellia Oleifera Seed Quality Inspection Based On Hyperspectral

Abstract:30different varieties of camellia seed are purchased from Hunan. Hubei, Jiangxi and other places with different quality for testing. They are randomly divided into calibration set samples (20samples) and prediction set samples (10samples) in this study. Firstly, the hyperspectral reflectance image of camellia seed is obtained using high spectrometer FieldSpec HH2. The content of oleic acid, linoleic acid and hexadecanoic in fatty acids of camellia seed is determined using gas chromatography. Based on the reflectance spectra from camellia seed we got, the Savitzky-Golay smooth processing, spectral preprocessing, the evaluation and build-up of the spectra model and a series of operations are carried on. Multiple scattering correction, first-order differential equation and second-order differential equation are used for the spectral preprocessing, and the influence of the three pretreatment methods for modeling is discussed. For the build-up of the spectra model, the principal component regression, partial least-squares regression and radial basis neural network method are used, also with the evaluation of the effects of three kinds of modeling method. In the end, the high spectral detection system software of camellia seed is developed, which realized the fast detection of fatty acid content in camellia seed.The results showed the following:1. The spectral characteristic of camellia seed showed that it had strong absorption for the blue light and red orange light, and had strong reflection for the near infrared spectra of around900nm. Including the reflectivity of visible light wavelengths up to0.35and the reflectance of near infrared spectral range up to0.58. It showed good correlation between spectral characteristics change and the chemical measured component content in fatty acid of camellia seed, the sensitive wavelengths range of400-500nm,600-630nm and900nm. Through stepwise regression, the4sensitive wave bands with strongest correlation are used for modeling. 2. The influence of three different kinds of pretreatment methods on modeling was also compared, and the showed that the optimal pretreatment method for oleic acid and linoleic acid is multiple scattering corrections. For palmitic acid is second order differential equation. The sensitive wavelengths of oleic acid, linoleic acid and palmitic acid in camellia seed are modeled by using principal component regression, the partial least-squares regression and radial basis function neural network. The experiment showed that the cross-validation correlation coefficient R of principal component regression modeling of oleic acid, linoleic acid, palmitic acid in camellia seed are0.8948,0.8955and0.8265respectively. The corrected root mean square error and the root mean square prediction error are0.1091,0.0085,0.0288and1.0811、0.9122、0.5288. For the partial least squares regression model, the cross-validation correlation coefficient R of oleic acid, linoleic acid, palmitic acid in camellia seed are0.9176、0.8941and0.8499respectively, the corrected root mean square error and the root mean square prediction error are0.4847、0.14,0.0797and0.8298、0.9227、0.4665. respectively. For the radial basis function neural network model, the cross-validation correlation coefficient R of oleic acid, linoleic acid, palmitic acid in camellia seed are0.9409、0.9022�'�0.8598. The corrected root mean square error and the root mean square prediction error are0.402、0.1064.0.0254and0.6163、0.8519、0.4357, respectively. Results showed that when modeling the composition content of high spectral camellia seed, the radial basis neural network is better than that of partial least-squares regression and principal component regression. It showed that the use of hyperspectral method would well predict the content of the oleic acid linoleic acid palmitic acid in camellia seed.