The Research Of Sugar Characterizationand Non-destructive Determination Of Mini-watermelon

In this paper， the distribution of sugar content level within themini-watermelon is studied， and a new characterization method ofmini-watermelon using the average sugar, the highest sugar and the lowestsugar index is proposed. PLS models measuring the three sugar levels areestablished. The near infrared reflectance spectra of44ZC and42QL areobtained，and the calibration models of three kinds of sugar are established byPLS and PCA. The results not only proves the feasibility of non-destructivedetection of sugar content of mini watermelon by diffuse reflectancespectroscopy but shows that there are good correlation between three sugarcontent and the near-infrared diffuse reflectance spectra of two kinds ofsamples. Specially, the maximum of SEP of all the initial calibration models isless than0.8Brix%. The result is acceptable comparing with the similarfindings of other researchers.Considering that the impact of noise is fairly obvious in the course of mo-del building, which leads the SEP and correlation of initial models to be not satisfactory. After that, the spectra of samples are pretreated by OSC，and thenthe effective wavelength ranges are selected scientifically by CARS and thenew calibration models are established. From the results of the models, it canbe seen that the correlation tends to homogenization and the absolute value ofcorrelation coefficients are generally larger. That is to say the correlation hasbeen significantly improved by OSC and CARS treatment. In addition，all theabsolute value of SEP are generally smaller, in another way，the predictioneffects of the models get better.In addition，the models are established by PLS for three sugar indicatorsof mini-watermelon samples which are mixed, whose pretreat ways are thesame with those of the separate models. The result shows that the predictioneffects of the mixed models are not as good as those of the separate models.But the Rcand SEP of all the mixed models are about0.8and0.7-0.9,respectively. The results above prove that there are general characteristics in acertain extent between the sugar indicators of two kinds of samples. The resultis meaningful to broadening the application scope of mini-watermeloncalibration models.Finally, classified identification qualitatively is applied to two kinds ofmini-watermelon by PCA and LS-SVM. It can be seen that the classifiedeffects of space lattice of two kinds of samples in the PCA scoring diagram ofthree-dimensions is better. In addition, the correction rate of LS-SVM modelwhich is pretreated by PCA to two kinds of samples is high. All results indicate that there are some differences in the original spectra of ZC and QL.The result has great practical meanings for on–site detection ofmini-watermelon to preventing shoddy phenomenon and removingcounterfeits in mini-watermelon market.