Study On The Texture Detection Of Valencia Orange Based On Hyperspectral Image Technology

The study took 9 kinds of valencia orange of which were picked in May 2013 from the Zigui breeding demonstration in yichang city as the research object. At first, the hyperspectral reflectance images of valencia orange were collected. This paper study the texture quality and sensory properties of valencia orange, as well as the differences of quality indexes between different varieties and the correlation among them. The main texture indices affecting the quality of summer orange were selected, and the model of valencia orange mastication degress was constructed. Then, based on fact that the texture parameters and the data of hyperspectral data were significantly correlated, the prediction model of main texture parameters of summer orange based on hyperspectral data was conducted. At last, the nondestructive detection model of sensory quality of valencia orange was conducted.By studying the texture characteristic(texture profile TPA test parameters, parameters of the shear test) and reflectance spectra data of the nine kinds of summer orange, the model of the the sensory mastication of valencia orange quantitatively and the regression model of the main texture parameters of Valencia orange based on hyperspectral information data were established. The research could be summarized as follows:(1) Through the analysis of the significance, the hardness, chewing ability, maximum shear force and cut off power of the TPA texture parameters and parameters in the shear test were significantly different among the varieties of valencia orange. There was a significant correlation between texture parameters and sensory score, which mean the hardness, elasticity, and cut off work and other indicators and sensory pulp texture, residue, easy to chew and juicy showing a strong negative correlation, and the maximum shear strength was negatively correlated with the sensory score at a certain degree. What’s more, these parameters like hardness, elasticity, maximum shear force and cutting off work and other indicators also showed a strong correlation with the total score of the senses. The model of the degree of summer orange slag based on texture characteristics was constructed by using the co linear diagnosis and principal component regression. The decision coefficient of the model was 0.88, so the forecasting model was highly accurate, which indicates that the model could be used for the prediction of the extent of summer orange slag in practical application.(2) The reflectance spectra of summer orange were processed by multivariate scatter correction and S-G smoothing filtering, then with the parameters of summer orange texture parameters simple correlation analysis was done. The results showed that the correlation coefficient among the hardness, the maximum shear force and the cut-off work and the spectral reflectance were between-0.3~0.3 when the sample of the summer orange were under different wavelengths. The curve trend of the maximum shear force and the cut off power were almost the same, that is, the correlation of the maximum shear force and cut off power of the fruit of summer orange was relatively high.(3) There were some correlations between the parameters of summer orange texture and the spectral reflectance. A prediction model of the valencia orange texture parameters based on hyperspectral image technology was established by using partial least squares regression. Methods such as first order derivative, multiple linear regression and vector normalization were used in spectral preprocessing. The results of the model showed that the prediction model of the summer orange hardness, chewing, maximum shear force and cut off power, which was established both by the of the first derivative and partial least squares was the best. The R2 and RMSECV of the prediction model of summer orange were 0.856 and 28.582 respectively. Hardness prediction model of R2, RMSECV were 0.899, 3.569. The maximum shear force prediction model of R2 and RMSECV were 0.728 and 6.102 respectively. The R2 and RMSECV of the forecasting model of cutting off power were 0.671 and 15.574 respectively.(4) Using principal component analysis and multivariate linear regression method to establish the sensory characteristics prediction models of valencia orange based on reflection spectra. The results showed that the indices of residue, mastication and total score were significantly correlated with the reflectance of valencia orange(P<0.05). The partial least squares model of sensory characteristics of valencia orange was conducted. Performance parameters of PLS models was optimized, relatived to multivariate linear regression models. The determination coefficient of the PLS models of sensory characteristics wer above 0.64,namely correlation coefficients were above 0.8. The root mean square error of the correction and the root mean square error of the test models are closer, so the reflectance spectrum can explain its sensory properties to a certain extent.