| Today consumers pay more attention to quality of horticultural products as their living standard continue to improve.Currently,several nondestructive techniques have been developed for assessing quality and maturity of horticultural products,among which optical techniques are most commonly used.Conventional visible and near infrared(Vis/NIR)technique only measures an aggregate amount of the light that is reflected or transitted from a specific area(or point)of samples without providing spatially resolved information.The technique is thus unable to directly measure the optical absorption and scattering properties of horticultural and food products.Moreover,conventional Vis/NIR technique may not adequately assess horticultural and food samples whose properties vary spatially or with depth,because it lacks the capability of providing spatially-resolved measurements.Thus,acquisition of spatial information and understanding of light interaction with plant tissues should be considered when analyzing samples by optical techniques,so as to improve quality assessment.In this research,two spatially resolved systems were developed,i.e.,spatially resolved spectroscopy system and noncontact semi-transmittance multichannel system.Tomato quality and maturity was evaluated by using the spatially resolved spectroscopy system,while the semi-transmittance multichannel system was used for detecting internal defect of apples.The spatially resolved spectroscopy(SRS)system is capable of-acquiring 30 spatially resolved reflectance spectra simultaneously at the source-detector distances of 1.5-36 mm from horticultural samples with either flat or curved surface.It enables to measuring the optical absorption and scattering properties,while it can also be used for assessing the tissues properties of horticultural and food samples at varying depths based on individual spatially resolved spectra.The noncontact semi-transmittance multichannel system was made up of six detecting fibers arranged at six different locations allowing to measure samples in 360 degrees.The major accomplishments of the research are summarized as follows:(1)Three calibration methods and procedures were proposed for calibrating the SRS system.Linearity calibrations would ensure consistent linear responses of individual fibers of the same size and between the three sizes of fiber.Optical property calibrations were conducted using three reference liquid samples so as to accurately measure optical absorption and scattering properties of samples.The calibration results showed that all 30 fibers had high linearity with exposure time with the coefficient of determination being greater than 0.99 for wavelengths greater than 700 nm;however,linearity responses,as measured by the slope and intercept,for the individual fibers differed greatly and also varied with wavelength.After the calibrations,the probe was able to measure the scattering coefficient of the reference liquid samples with the relative errors less than 11%,whereas higher measurement errors were obtained for the absorption coefficient due to low absolute absorption values for the samples.Furthermore,the SRS system was demonstrated for being able to measure the absorption and reduced scattering coefficients of tomato fruit at different maturity stages for the wavelengths of 550-1,350 nm.(2)Spatially resolved diffuse reflectance spectra of 600 'Sun Bright' tomato samples harvested at six maturity stages were acquired using the SRS system,from which the reduced scattering(?s')and absorption(?a)coefficients were estimated using an inverse algorithm for the diffusion theory model.Distinct changes in the absorption coefficient spectra for tomatoes at different maturity stages were observed.The prominent peak at 675 nm,due to chlorophylls absorption,became less pronounced as the tomato maturity advanced from the green to red stage.On the other hand,the absorption peak around 560 nm became more pronounced as the tomato fruit turned from green to red,which was likely caused by the absorption of anthocyanin and turmeric pigments in the tissue.In addition,a small peak around 750 nm was caused by water absorption.Two other major absorption peaks were observed at 970 nm and 1,180 nm,which were due to the combined effect of water and the absorption bands of C-H,N-H and O-H in the tomato tissues.Values of the reduced scattering coefficient decreased steadily with the increasing wavelength for all tested tomato samples over the spectral region of 550-1,300 nm.While both ?s' and ?a were correlated with tomato firmness,soluble solids content(SSC)and pH,better prediction results were obtained for the multiplication of ?s' and ?a(i.e.,?aŚ?s')based on partial least squares(PLS)models.Compared with linear PLS models,?s' spectra based on nonlinear support vector machine(SVM)models resulted in better predictions for quality attributes of tomato.The best correlations for prediction of acoustic firmness,impact firmness,compression area and puncture peak,slope and flesh firmness were 0.695,0.881,0.900,0.915,0.930,0.835,respectively,and the models predicted tomato SSC and pH with the best correlation coefficients of 0.655 and 0.769,respectively,which indicated absorption and scattering properties were capable for assessing quality attributes of tomato.(3)Results for tomato quality assessment indicated that tomato quality predictions varied with individual spatial resolved(SR)spectra and the optimal single SR spectra were different for prediction of different quality parameters.Overall,those SR spectra for the light source-detector distances of 6-24 mm had better prediction results for firmness,while SR spectra acquired for small light source-detector distances overall had better predictions of SSC5 and those spectra for the light source-detector distances of 6.0-12.5 mm gave better results for pH prediction.Combinations of SR spectra gave consistently better predictions for the quality parameters than single SR spectra.PLS models using the optimal combinations of SR spectra yielded better predictions for the tomato quality parameters compared with the optimal single SR spectra,with the correlation coefficients(rp)of 0.760 and 0.911 for acoustic and impact measurement,rp=0.935 for compression,and rp=0.917,0.948 and 0.859 for puncture maximum force,slope and flesh firmness,rp=0.801 and 0.819 for SSC and pH.Compared with conventional Vis/NIR spectroscopy,SR spectra provided better predictions in tomato firmness parameters and pH,while SR predictions of SSC were comparable to conventional near-infrared(NIR,900-1,700 nm)predictions(rp=0.801 versus 0.810)but better than conventional visible and shortwave near-infrared(Vis/SWNIR,400-1,000 nm)(rp=0.729).(4)The 600 tomato samples were grouped into six maturity grades(i.e.,green,breaker,turning,pink,light red,red),based on their surface color and internal color distributions,respectively.PLS discriminant analysis(PLSDA)and SVM discriminant analysis(SVMDA)models for the 15 spatially resolved spectra were developed and compared to determine the optimal models for classification of tomato maturity.The results showed that different mathematical models had influenced the maturity classification.Overall,the SVMDA models had better performance than PLSDA models for tomato maturity classification.Moreover,SR spectra and conventional Vis/NIR spectra provided better classification accuracy for tomato maturity based on surface color grading than that based on internal color grading,while absorption coefficient spectra had advantages for tomato maturity classification based on internal color.The three types of spectra had different performance in tomato maturity classification;SR spectra overall gave better classification accuracy than that of conventional Vis/NIR spectra,while absorption coefficient spectra were comparable to SR spectra for maturity classification based on internal color.However,for maturity classification based on surface color,absorption coefficient spectra did not achieve as good results as that by SR spectra and conventional Vis/NIR spectra.(5)The noncontact semi-transmittance multichannel system was applied for assessing internal defect of apples.The six channels allowed to detect apple samples in 360 degrees for better evaluating internal defect in apples.Each individual channel provided different recognition performance for defective apple classification,and the optimal channel depend on apple orientation.Spectra acquired for apple orientation C(when the stem-calyx axis was in horizontal direction)offered overall better classification with about 90%accuracy based on PLSDA models.Besides,channel combination further improved the results with 91.5%,89.2%and 93.1%accuracy for apple orientations A(when the stem-calyx axis was in vertical direction,stem was in the top),B(when the stem-calyx axis was in vertical direction,calyx was in the top)and C,respectively.In addition,the optimal channel combinations had 92.9%correct classification rate for good apples and 90.0%for defective apples for orientation A;91.4%for good apples and 86.7%for defective apples for orientation B;and 95.7%for good apples and 90%for defective apples for orientation C. |
PDF Full Text Request