| Xinjiang is known as " the town of fruits ", Hami melon is one of the characteristics of traditionalproducts of Xinjiang, because it has unique flavor and rich nutritional value, it’s popular for the majority ofconsumers, so it become the main way to increase revenue for farmers. But in the process of the Hamimelon’ quality grading, which mainly rely on mechanical grading or classification by artificial vision,increasing labor costs greatly, at the same time, it will result in the low productivity, uneven quality of thebreed, disunion of grading standards, mechanical damage severity and other defects, If some are serious, itwill cause bad effect on the market’s competitiveness, and caused great losses on the economic interests offarmers. Utilization of spectral imaging technology combined with chemometrics methods and testingstandards can be realized Hami melon’s nondestructive detection, in order to enhance the added value ofmelon, improve brand position of melon and enhance market competitiveness. The main research contentsand methods include the following:(1)Analysis of the morphology of each set of samples which were used in this study,including verticaland horizontal parts of the equatorial diameter, weight and other detection results, as well as analysis ofthe different varieties of melon firmness values, analysis of different maturity and firmness values,analysis of firmness of different detection sites and analysis of different spectral;Analysis of thedifferences between the different test site (the night side of equator, the sun side of equator, the scar ofequator) of the spectral information of Hami melon, test modeling results show that the spectral modelingresults of the equatorial part collection are better than the scar, comparing the sun side of equator withthe night side of equator, the results of forecasting model which is built by the spectral informationcollected Hami melon surface position of the sun side is the best;(2) In order to establish an accurate model of Hami melon firmness, abnormal samples of participatedmodeling were identified and removed. In this study, it were used abnormal spectra removed, Dixon (Dixon)detection, Leverage and student T-test guidelines and the residual principal component score to judge Hamimelon samples comprehensively. In accordance with changes of model performance, And by recycling thesuspected sample analysis, By comparing the final decision as outliers and removed them.(3) In this study, do quantitative analysis of melon’s firmness, modeling different ranges of spectralbands, and comparing the results effectively and ultimately determine the optimal combination band; Whilecomparative analysis the predictive accuracy of the model for firmness of Hami melon by comparing withthose at different spectral pretreatment methods, with different optical correction methods and differentquantitative calibration modeling. Using PLS, SMLR and PCR correction algorithms method forcomparative analysis of No.16Jinmi and No.17Jinmi firmness detection model melon. The experimentalresults showed that: No.16Jinmi in819.82-843.3nm,899.14-882.64nm and996.22-930.23nmcombination of wavelength range, the application of PLS method for the first derivative spectra processingof snv processing to establish the prediction model of firmness of the results is better. Correction samplecorrelation coefficient R is0.9085, the lower root mean square errors of correction (RMSEC) is2.16N,with a prediction sample correlation coefficient (Rp) is0.8139and root mean square errors of prediction(RMSEP) is3.24. No.17Jinmi in the range of757.01-1000nm spectral bands, using PLS method afterstandard normal variety, MSC pre-processing and Savitzky-Golay smoothing spectral of to model, theresult is better. Correction sample correlation coefficient R is0.8927, the lower root mean square errors ofcorrection (RMSEC) is1.24N,with a prediction sample correlation coefficient (Rp) is0.7030and root meansquare errors of prediction (RMSEP) is1.70N. |
PDF Full Text Request