Study Of The Nondestructive Test Model Of Apple Quality By CT And Image Transformation

China is a big agricultural country in apple production and planting, but not the strong. One of the reasons is the poor quality detection and separation technology. Therefore the advantages have made the non-destructive test become an important study subject among abroad and domestic scholars. This article employs the computed tomography (CT) scanning technology and image processing method, conducts a series of basic researches on the internal quality of apple during storage time. In order to predict the internal quality, CT scanning was applied at first, and then titrable acidity, soluble solid. pH and moisture content were measured and regarded as the index to evaluate the quality, at last the prediction model was established. The research and results are as follows:(1)Fushi apples in three producing areas(Shandong, Shanxi and Xinjiang) were chosen as the material and the linear models between the CT number and internal quality were built by the Least Square method(LS), and the surface models of internal quality were set up by the way of quadratic fit.The conclusion that there was strong linear relationship statistically between the CT number and the main components was verified, furthermore, it can be discovered that the variation tendency and the specific component contents were calculated through the models.(2)In order to establish the integrated model by the way of quadratic fit to predict internal quality of apple, the producing area, CT number and storage time were chosen as parameters. Among the four measurement contents, the models of titrable acidity, pH and moisture content have a good predictive ability, and the correlation coefficient of each model was0.9375、0.9634、0.9029respectively, while the root-mean-square error was0.2286、0.1183、0.4949. However, the predictive ability of soluble solid mode was a little weaker with0.7891in correlation coefficient and0.6671in root-mean-square error.(3) With the purpose of changing the time-domain space picture into frequency domain, the apple CT gray scale images was transformed by Fourier transform. One image with a summit which was enclosed by flat area was acquired by the transform, and the peak height was between14.0244and14.9589. After several experiments, sixteen numbers was selected from each frequency image to compose the characteristic parameters, and then two sets of models were established by the principal components regression and partial least squares regression respectively. The results show that the models have good predictive ability. In the principal components regression, since the accumulating contribution rate of ten principal components reaches99.4376%, four contents models were built according to them. The models showed good prediction results by the F-criterion with all the p value lower than0.05. In partial least squares regression, each content model had the lowest sum of squared errors when the number of latent variable was12. The sum of squared errors after data standardization is14.93of soluble solid,10.017of pH,16.9126of moisture content,9.9252of titrable acidity.