| The output of fruit in China is very large, but the volume of exports is a few. Low degree of fruit commercialized treatment is one of the reasons that caused low competitiveness in the international market and few volume of fruits exports in our country. Increasement of fruit commercialized treatment level can enhance the competitiveness of fruits in the international market, and expand the volume of fruit exports. Quick and nondestructive detection of fruit external and internal qualities are the basis of fruit commercialized treatment, so it is important and necessary to detect fruit external and internal qualities quickly and nondestructively.'Cuiguan'pear,'Shuijing' pear and 'Gong' pear were used as objects in our research project, and the research project was to use the visible/near infrared spectroscopy technique, optical fiber sensing technique and chemometrics method to detect soluble solids content (SSC) and acidity of pears statically and on-line, and use the different calibration methods combined with different pretreatment methods to develop static calibration model and on-line calibration model for SSC and acidity of pears. Some factors including angle of light source-detector, fruit moving speed, and fruit orientation that affected the performance of calibration model were studied, also different variable selection methods were investigated.Research contents and conclusions were listed as below:(1) Parameter optimization of static and on-line spectra acquirement for USB4000 miniature spectrometer was investigated. For SSC detection of'Cuiguan'pears, the optimal parameters of Boxcar and Average for static spectra acquirement were 6 and 4, and the optimal parameter of Boxcar for on-line spectra acquirement was 6.(2) Effect of Some factors including angle of light source-detector, fruit moving speed, and fruit orientation on performance of SSC calibration model of'Cuiguan'pears was investigated. The light source-detector angle of 120°was better than that of 180°for SSC detection of'Cuiguan'pears. The performance of PLS model for SSC was decreased, but not so much at the three fruit moving speeds (0.3 m/s,0.5 m/s,0.7 m/s). The fruit orientations of HH and HV were more suitable than that of VU and VD for SSC detection, and there were little difference between fruit orientations HH and HV, VU and VD.(3) Static detection of SSC and pH of'Cuiguan'pears from Cixi and Henggang orchards in year of 2009 were investigated in Vis/NIR detection system AI. Calibration methods such as partial least squares (PLS), principal component regression (PCR), stepwise multiple linear regression (SMLR) and least squared support vector machine (LS-SVM) combined with different pretreatment methods were used to develop SSC calibration models of'Cuiguan' pears for single orchard and two orchards, and PLS combined with different pretreatment methods was used to develop pH calibration models of'Cuiguan'pears for single orchard and two orchards. The correlation coefficient (r), RMSEC, RMSEP, RMSECV and RPD of the best static SSC calibration model of'Cuiguan'pears for Cixi orchard, Henggang orchard and two orchards in year of 2009 were 0.971,0.203,0.328,0.322 and 2.59,0.941,0.289,0.322, 0.355 and 2.58,0.946,0.277,0.354,0.347 and 2.37. The correlation coefficient, RMSEC, RMSEP, RMSECV and RPD of the best static pH calibration model of'Cuiguan'pears for Cixi orchard, Henggang orchard and two orchards in year of 2009 were 0.765,0.113,0.112, 0.128 and 1.60,0.885,0.095,0.123,0.117 and 1.63,0.835,0.105,0.115,0.123 and 1.65. PLS method was better than other methods such as PCR, SMLR and LS-SVM.(4) On-line detection of SSC and pH of'Cuiguan'pears from Cixi and Henggang orchards in year of 2009 were investigated in Vis/NIR detection system AI. Calibration methods such as PLS, PCR, SMLR and LS-SVM combined with different pretreatment methods were used to develop SSC calibration models of'Cuiguan'pears for single orchard and two orchards, and PLS combined with different pretreatment methods was used to develop pH calibration models of'Cuiguan'pears for single orchard and two orchards. The correlation coefficient, RMSEC, RMSEP, RMSECV and RPD of the best on-line SSC calibration model'Cuiguan'pears for Cixi orchard, Henggang orchard and two orchards in year of 2009 were 0.966,0.222,0.253,0.350 and 3.36,0.953,0.259,0.326,0.344 and 2.55, 0.923,0.306,0.331,0.376 and 2.54. The correlation coefficient, RMSEC, RMSEP, RMSECV and RPD of the best on-line pH calibration model of'Cuiguan'pears for Cixi orchard, Henggang orchard and two orchards in year of 2009 were 0.814,0.105,0.127,0.123 and 1.41, 0.881,0.098,0.104,0.117 and 1.92,0.854,0.102,0.120,0.121 and 1.58. PLS method was better than other methods such as PCR, SMLR and LS-SVM. The method system and prediction models of SSC and pH on-line detection of'Cuiguan'pears based on visible/near infrared (Vis/NIR) spectroscopy technique and chemometrics method were established at fruit moving speed of 0.5 m/s.(5) On the grading machine, on-line detection of SSC of'Gong'pears was investigated in the Vis/NIR detection system AI. PLS method combined with different pretreatment methods was used to develop SSC calibration models. The correlation coefficient, RMSEC, RMSEP, RMSECV and RPD of the best SSC on-line calibration model of'Gong'pears were 0.945,0.285,0.290,0.344 and 2.86.(6) A new variable selection methods UVE-GA-PLS was proposed, and was compared with other variable selection methods including BiPLS, SiPLS, UVE-PLS, GA-PLS, SPA-MLR and UVE-SPA-MLR. For SSC detection, UVE-GA-PLS was the best variable selection method among these methods. The performance of SSC calibration model developed using 50 selected wavelengths was comparative to that developed using full wavelengths. The correlation coefficient, RMSEC, RMSEP, RMSECV and RPD of the best SSC calibration models of'Cuiguan'pears developed by UVE-GA-PLS were 0.958,0.248, 0.290,0.292 and 2.97. |
PDF Full Text Request