Non-destructive Detection Of Pepper Quality Using Fourier Transform Near Infrared Spectroscopy

Peppers rich in nutrients, especially with the highest vitamin C (Vc) content in all kindsof vegetables. It has an important economic value and therapeutic health effects, for it can befresh, seasonings and can also be used as medicine. With increasing consumer concern ofpeppers, it becomes a demand of sealers and consumers to find a rapid technology or methodto detect pepper quality. With the fresh peppers and pepper powder as the objects in thisresearch, Fourier transform near-infrared spectroscopy (FT-NIR) and chemometric analysiswere used to carry out the quantitative detection of soluble solids contents (SSC) and vitaminC (Vc) in fresh peppers, classification identification of pepper powder and commonadulterants, and rapid detection of adulterants content in the mixture powders.Mainly research results were as follows:â‘ Spectral pretreatment methods, effective wavelength selection methods andcalibration modeling methods were studied for nondestructive measurement of SSC and Vc infresh peppers. The diffuse reflectance spectra of fresh pepper samples were used to comparethe prediction results with different spectral pretreatment methods. The results showed thatcalibration model with first derivative (1st D) combine with multiplicative scatter correction(MSC) pretreatment was better for SSC, and model with standard normal variables (SNV)pretreatment was the best for Vc. The spectra, pretreated by the chosen methods, were used tocompare the prediction results with different effective wavelength selection methods. Theresults showed that MC-UVE method was better than the other three (iPLS, SPA, GA) forboth SSC and Vc. The principal components (PC) selected by PLS and effective wavelengthsselected by MC-UVE were used as the inputs of least square-support vector machine(LS-SVM) to develop PC-LS-SVM and MC-UVE-LS-SVM models. Better results wereachieved by MC-UVE-PLS model for SSC with the correlation coefficient (rp) of0.987andRMSEP of0.274oBrix in validation set. The best results were achieved byMC-UVE-LS-SVM model for Vc with the rp of0.911and RMSEP of19.271mg/100g invalidation set.â‘¡The optimization models were evaluate by predicting the unknown samples.27unknown fresh pepper samples were used as the prediction set. The results show that the rpand RMSEP were0.971and0.382oBrix,0.899and21.022mg/100g for SSC and Vc,respectively. The results indicate that FT-NIR technique can be used for the quantitativeanalysis of SSC and Vc in fresh peppers.â‘¢The effects of different classification methods were explored for the Classificationand identification of pepper powder, common adulterants and the adulterated mixtures.Discriminant analysis (DA), independent soft-mode cluster class classification (SIMCA) andpartial least squares discriminant analysis (PLS-DA) methods were used to establishdiscriminant model. And the effects of different bands and different spectral pretreatment methods were also compared for each classification method. The results show that the threeclassification methods allowed us to make a good discrimination between pepper andcommon adulterants. However, tests on adulterated mixtures showed that only PLS-DA andSIMCA were really useful for detecting counterfeits and adulterations. PLS-DA does providethe best results and is able to discover adulterated mixtures with lower amounts of adulterantthan those obtained with SIMCA, with the rc of0.989and only two mixtures weremisclassifed as belong to pepper class.â‘£Quantitative models were established for the rapid detection of adulterants content inadulterated mixtures. The ratios of the mixtures were used as the Y value. The effect ofdifferent bands, different spectral pretreatment methods and different modeling methods to theaccuracy of the quantitative models were compared. The results show that the bestorange-peel content model was achieved by using the SNV pretreated spectra in the range of10600~7500cm-1and PC-LS-SVM modeling method with rp of0.997and RMSEP of0.486%.The best saw-dust content model was achieved by using the full SNV pretreated spectra andPLS modeling method with rp of0.977and RMSEP of0.872%. The best red-brick contentmodel was achieved by using the full SNV pretreated spectra and PLS modeling method withrp of0.988and RMSEP of0.631%.