| In this study, red Fuji apple and Gannan navel orange were studied as test subjects, andfruit sugar content was studied as detection index. Portable device for nondestructivedetermination of fruit sugar content by near infrared spectroscopy was designed andconstructed, which provided the hardware platform for the follow-up studies. Differentaffecting factors, such as light type, light angle, spectrometer, integration time and placementof fruits, were contrasted and studied for the nondestructive determination of fuit sugarcontent by near infrared spectroscopy. Developed and optimized the prediction model of thedevice. Genetic algorithm (GA), backward interval partial least squares (BiPLS), intervalpartial least squares (iPLS), and competitive adaptive reweighted sampling (CARS) were usedfor the selection of effective spectral variables from the full spectrum variables.Main contents and conclusions of this research were listed as follows:(1) According to the theory of near infrared spectroscopy, portable device fornondestructive determination of fruit sugar content based on near infrared spectroscopy wasdesigned and built. This device was divided into three modules: optical module, dataprocessing module and accessory module. In the design of optical module, the works arecomposed of the choice of light source model and arrangement of optical path. The dataprocessing module was composed of microprocessors, software systems and touchscreendisplays. The accessory module was composed of the various accessories, and the design ofthe device box was needed to meet the requirement of compact-size and appearance.(2) Different affecting factors, such as light type, light angle, spectrometer, integrationtime and placement of fruits, were contrasted and studied for the determination of fuit sugarcontent by near infrared spectroscopy. The comparisons were listed as follows:When there were not bezels between the light sources and the fruit samples, theperformance of the model was superior to another light type. When the angle between thelights direction and the vertical direction was45°, the performance of the model was superiorto the other two models which the light angle was30°and60°. When the spectrometerUSB4000was used, the performance of the model was superior to USB2000+. When theintegration time was set as200ms, the performance of the model was superior to theintegration time of50ms,100ms, and150ms. When the placement of fruits was cdpx (theequatorial plane of fruits was parallel to the direction of the light sources), the performance ofthe model was superior to cdcz (the equatorial plane of fruits was perpendicular tothe direction of the light sources). (3) Developed and optimized the prediction model of the device. Characteristicwavelengths were chosen by the method of GA, BiPLS, iPLS and CARS from the full spectraof fruits, and the partial least squares (PLS) model were developed based on the selectedspectral variables respectively, and then the comparisons were made between them. At last,the comparison were made between PLS model based on the selected spectral variables andPLS model based on the full spectra respectively. By comparison, all the performance of PLSmodel based on the selected spectral variables was superior to that based on the full spectra.Because the number of spectral variables was substantially reduced, all the predictionefficiency of the models based on the selected spectral variables was improved. Theperformance of the PLS model by the method of CARS was superior to the performance ofthe PLS model by the the other spectral variables selection method, which the Rc, RMSEC, Rpand RMSEP were0.99,0.19,0.93and0.38, and the number of selected spectral variables was107. Therefore, the PLS model which was optimized by CARS can be used as the predictionmodel of the device. |
PDF Full Text Request