| Near infrared (NIR) spectroscopy technique has been applied to many fields for its high efficiency,low cost, simple preparation, and non-destructivity. Although basic researches on fruit qualitydetermination by NIR spectroscopy technique have gained some success in our country, there's still along way to apply it to practical use.The international competition for fruit export of our country is very poor due to being laggard atcommercial post-harvest process, especially at commercial grading based on internal quality.The studies on the device for non-destructive internal quality determination of fruit by NIRspectroscopy technique is still in a passing phase in our country. Based on the studies on methods anddevices for sugar and acidity content determination of fruit, a general device for determining internalquality of many kinds of fruit using NIR spectroscopy technique is designed, which is of significanttheoretical and practical value.Since real-time non-destructive determination of internal quality of fruit is in experiment phase, andexisting techniques and methods of real-time on-line determination of internal quality of fruit by NIRspectroscopy technique are immature to apply them in commercially real produces, our researchaimed to design a test-bed for real-time on-line determination of fruit internal quality by NIRspectroscopy. Citrus fruits were selected as research object and two important internal quality indexes(soluble solids content (SSC) and acidity content) were analyzed based on the test-bed. Some existingdifficulties in on-line detection and integrative knowledge of multi disciplines for resolving theseproblems were studied. The results of this work offer theory foundation and references for on-linepractices.Main contents and conclusions of this research were listed as below:1. In order to offer some theoretical foundation for on-line determination of internal quality of fruit,the thesis expatiates on physics and chemistry foundation and NIR spectral pretreatment andquantitative analysis methods.2. Methods for spectral signal extraction were studied in this research. The predictionperformances of different pretreatment methods (derivative, multipticative signal correction(MSC), standard normal variate (SNV), Savitzky-Golay filter smoothing and wavelet smoothing),different modeling methods (multiple linear regression (MLR), principal component regression (PCR) and partial least squares (PLS)), and different optimal wavelength selection methods ofmultivariable calibration model were compared. Outliers elimination methods based on therobustness and adaptability of model were studied. And the performances of different spectralobtaining modes (diffuse reflection and diffuse transmission) were also compared. For the SSCdetermination of chufa, the results for first derivative spectra were better that those for rawspectra and the others, and the determination coefficient (r~2), the root mean standard error ofcalibration (RMSEC) and the root mean standard error of validation (RMSEP) was 0.9055, 0.456and 0.491, respectively. For the SSC determination of kiwifruit, the results for MSC and SNVspectra were better than those for raw spectra, and the best were SNV spectra, the r~2, RMSECand RMSEP was 0.9712, 0.316 and 0.427, respectively. For the SSC determination of Fujiancitrus, the prediction performance of PLS was better than that of PCR, with r=0.8949,RMSEC=0.351 and RMSEP=0.409.3. The prediction performances of different wavelength selection method, such as correlationanalysis and generic algorithms etc, were compared. The results indicated that theperformance of genetic algorithms was the best. Based on the comparison results for differentoutlier elimination methods and different spectral obtaining modes (diffuse reflection and diffusetransmission), and the analysis of application surroundings of different detection modes, diffusetransmission mode was chosen as the on-line detection method for citrus internal qualitydetermination. For the SSC determination of "Huapeng" citrus, the results for diffuse reflection(r~2=0.9151, RMSEP=0.325) was slightly worse than that for diffuse transmission (r~2=0.9749,RMSEP=0.282).4. Influence parameters of real-time on-line determination by NIR spectroscopy were studiedfor the first time. Influence parameters of surrounding condition include temperature fluctuationof fruit due to environment temperature fluctuation; influence parameters of sample shapeinclude fruit size and detecting orientation etc. The robustness of on-line detecting models couldbe improved via the studies of influence parameters, and the accuracy of on-line detection couldalso be improved by establishing prediction models combined with environment influenceparameters.5. Three SSC models were established by different size (minimum axis, maximum axis andvolume of estimate). For SSC determination by minimum axis, the results for the model of71-80 mm were better than those for the models of 60-70 mm and 81-90 mm. For SSCdetermination by maximum axis, the results for the model of 71-80 mm were better than thosefor the models of 50-70 mm and 81-96 mm. For the SSC determination by volume of estimate,the results for the model of 200-250 cm~3 were better than those for the models of 150-200 cm~3and 250-300 cm~3. The results indicated that the prediction performance of models established by volume of estimate was the best and that size of fruit had influence on the prediction results. Toimprove the robustness of the models and to reduce the influence of fruit size on the models, alarge number of samples are needed.6. The influence of different stance of fruit (standing, side-lying, and being placed in arbitrary)was compared. The results indicated that the stance of fruit had slight influence on the results,and that the results for side-lying were slightly better that those for the other two modes. And itwas concluded that the influence of stance of fruit could be ignored for on-line fruit qualitydetermination.7. The influence of illumination intensity on the SSC determination was studied. The resultsshowed that as the intensity increased, r~2 increased, and RMSEC and RMSEP decreased. Theresults for 400 W were the best, r=0.8582, RMSEC=0.272, RMSEP=0.304, and the equation forreference value and predicted value was y=0.7191x+2.7442.8. An index F=(α·r~2+β(RMSEP·e~|RMSEC-RMSEP|))/t was put forward in the research, which acted asa reliable reference for the spectral pretreatment method selection. The effect of differentspectral pretreatment methods and the different combination was validated. The F value for thefirst derivative spectra was the highest, 0.6369, with the integration time is 0.05 s, r=0.9883,RMSEC=0.174, RMSEP=0.92.9. A test bed for on-line citrus fruit internal quality determination was set up in this research.The hardware system was designed according to the requirements of on-line detection, which cancomplete the real-time on-line process of sugar and acidity content detection of citrus fruit. Itprovided theory and technique foundation for development of NIR real-time on-line detectiondevice and practical application of fruit internal quality determination. Real-time on-linemethods for fruit sugar and acidity content detection were studied, including comparison andselection of real-time spectra pretreatment methods, detecting modes, and modeling methods etc.10. Software system for real-time on-line fruit internal quality determination was developed.The software comprised five main function modules—real-time spectra file management,real-time spectra display, real-time spectra signal processing (preprocessing), real-timecalibration model selection and management, prediction and classification of sugar and aciditycontent of unknown samples. The characteristics of this software system include multiplespectral formats compatibleness, general spectral analysis functions, multiple modeling methods,prediction correctness, easily maintenance, and expediently manipulation etc., which providedtechnique foundation for on-line detection of fruit internal quality. |
PDF Full Text Request