Beef Meat Adulteration And Quality Index Prediction Based On NIR And NMR Technology

In this paper,beef was used as the experimental object,and the physicochemical indexes of beef were determined by chemical method.Near-infrared spectroscopy(NIR)and nuclear magnetic resonance(NMR)technique combined with powerful chemometric methods including discriminant analysis(DA),partial least squares(PLS),and support vector machine regression(SVR)were employed to evaluate the freshness,physical and chemical indicators,and adulteration level.Finally,a rapid analytical model were established to make an evaluation of beef quality based on NIR technique,and a beef adulteration model were established based on NIR and NMR technology.The main results are as follows:(1)The physical and chemical qualities of beef in major supermarkets in Nanchang,Jiangxi were analyzed,and their pH,moisture content,protein content and fat content were determined respectively.The results also show that there were differences in the pH of beef in different parts,with the highest shoulder pH and the lowest beef tenderloin.But for beef in different parts,the difference in moisture and protein content was small.The biggest difference was the fat content,the beef tenderloin fat content was about 3%,and the cow fat content was as high as 16%.(2)The prediction model of TVB-N for beef,pork,and beef-pork mixed samples was established by NIR combined with PLS and SVR.Four common methods such as no preprocessing(NONE),Savitzky-Golay(S-G),Standard Normal Variation(SNV),and derivative(1st derivative and 2nd derivative)were used to preprocess the original data to establish PLS and SVR.For the PLS model,when NONE and S-G were used as preprocessing methods,with the same Rc and RP values,and the root mean square error(RMSEC)of the correction set and the root mean square error(RMSEP)of the prediction set are the smallest,and the model was the best.In the SVR model,with Rc=0.98 and RP=0.83.This result indicate that the data was over-fitted after preprocessing.In addition,according to the VIP(variable importance in the projection)value score map in the PLS model,338 VIP>1 important variables were selected,and the performance of the PLS-VIP model based on these 338 important variables had been reduced.(3)1H NMR combined with PLS were employed to predicte the adulteration of beefl.Based on the minimum RMSECV value,the optimal number of main factors is selected,and the effects of four frequentky-used preprocessing methods(NONE,UV,Par,and Ctr)on the PLS model were analyzed.Based on the indicators such as Rc,RP,RMSEE,and RMSEP values,it can be seen that the best pretreatment method of the model was UV,and the RMSEP value was the lowest at this time.By comparing the PLS model of NIR and NMR binary adulteration,the prediction accuracy of NIR combined with PLS model was very high,Rc>0.94,RP>0.96,RMSEC=8.57 and RMSEP=7.27.On the contrary,the model of 1H NMR combined with PLS still needs to be improved,and the highest RP was only 0.67.(4)A prediction model of beef nutritional quality combined with NIR and PLS was established.Based on the minimum RMSEC value,the best principal factor was selected,and by analyzing the impact of five commonly used pretreatment techniques(NONE,MSC,SNV,first and second derivatives,etc.)on the PLS model.Based on Rc,RP,RMSEC,RMSEP and other indicators,for the prediction model of pH content,it can be seen that the best pretreatment method of the model was NONE,at this time the RMSEP value was the lowest,RP=0.9031,the PLS model had a certain prediction ability;When predicting the moisture content,the models obtained with or without pretreatment technology were not good,and the model needs to be improved.Among the PLS models that predict fat content,the best prediction model was also obtained after processing by the NONE method,with an RP value of 0.8664.Finally,when analyzing the protein content,the model with variable normalization was used to obtain the best prediction performance.The RP was 0.9434,and the model had good prediction ability.(5)NIR technology combined with chemometrics methods establishes a binary and ternary qualitative and quantitative models of beef adulteration.Through the NIR spectrum,we can see the differences in the signal strength of the main components of beef,pork,duck in protein,fat,carbohydrate and water.When using chemometrics to extract near-infrared signals to establish a DA discriminant model,pure beef,pure pork,pure duck meat and adulterated beef have a good distinction on the DA score map,and the accuracy of the model for the ternary system of adulterated beef reaches 90.2%,The discrimination rate of the binary system had reached 100%,and the accuracy of the discrimination of the ternary system had reached 91.5%after optimization by screening wavelengths.In addition,PLS was further used to predict the amount of adulteration.The PLS model of the binary system was obtained with RC>0.94,RP>0.96,RMSEC=8.57 and RMSEP=7.27.The ternary system showed higher accuracy Rc=0.96,RP=0.96,RMSEC=8.33,RMSEP=9.27.And the internal test of 1/4 samples further confirmed the rapidity and accuracy of the method.