Research On Quality Detection Of Hetao Muskmelon By Visible And Near Infrared Spectra

This paper makes use of high-sensitivity back-illuminated2D FFT-CCD portable spectrometer(made in Ocean Optics, USA), as the research object to three varieties "xizhoumi","jinhongbao","baiyu" of hetao muskmelon, collecting static and on-line(0.07m/s、0.09m/s、0.11m/s)diffuse transmittance spectra, in order to research the detection method of melon internal quality based on the on-line detection system of visible near infrared diffuse transmission spectrum and image fusion by the research group. Using PR-101alpha portable digital refractometer and GY-3hardness to determine sugar content and hardness of hetao muskmelon. The diffuse transmittance spectra of melons collected was pretreated(smooth、derivative、MSC、SNV) and the models was established by related softwares. The main contents of the study are as follows:(1) Research on non-edible tissue influence on model accuracy of spectra and sugar content and hardness of "baiyu" melons using PLS method. The results show:the model results of the first derivative specra and sugar conten and hardness with non-edible tissue melons (Sugar Content:correlation coefficient(r) of0.875, root mean square errors of calibration (RMSEC) of0.589, root mean square errors of prediction (RMSEP) of1.09; Hardness:r=0.840, RMSEC=0.254, RMSEP=0.313) are better than the model results of the first derivative specra and sugar content and hardness with intact melons (Sugar Content:r=0.867, RMSEC=0.628, RMSEP=0.926; Hardness:r=0.816, RMSEC=0.258, RMSEP=0.304).(2) Research on extracting the quality characteristic wavelength variables of visible near infrared spectra of "jinhongbao" melon using individually SMLR、iPLS、biPLS and siPLS methods. The results show:extracting the12wavelengths of original spectra of sugar content based on SMLR method, the model is best, r=0.813, RMSEC=0.897, RMSEP=1.17, RPD=1.56; extracting the14wavelengths of original spectra of hardness based on SMLR method, r=0.819, RMSEC=0.484, RMSEP=0.653, RPD=1.13.(3) Using the traditional modeling methods(PL、PCR、SMLR)and the new network algorithms(BP-ANN、LS-SVM) to establish the model of "jinhongbao"melon spectra and sugar content and hardness, studying on the effects of different methods on the model accuracy and the prediction results. The best modeling result of the first derivative spectra and sugar content afer multiplicative scatter correction (MSC) is r=0.844, RMSEC=0.844, RMSEP=0.978using PLS method.