The Establishment And Maintenance Of Prediction Model For Duck Meat Freshness Quality Based On Near-infrared Spectroscopy

Duck meat is an important part of the dietary structure for residents,as rich in nutrients with unique flavor and easy to digest and absorb.With the rapid economic development,food demand structure and consumption concept of the public have been changed,and meat quality and safety have become the focus of increasing attention.The implementation of rapid prediction and assessment for duck meat quality indicator,is beneficial to control the quality of livestock and poultry meat and ensure the meat safety for consumers.In this study,the measurements of duck color(L*,a*,b*),pH values and total volatile basic nitrogen(TVB-N)content were performed using visible and near-infrared spectroscopy(Vis/NIR)technique,chemometrics methods and calibration maintenance methods.After the model establishment based on partial least square regressions(PLSR),the calibration maintenance and transfer were carried out to improve the applicability and stability of precdtion model for duck meat quality.The main research results are as follows:(1)The study of prediction models for duck meat color(L*,a*,b*)and pH values.The original spectral information analysis of duck meat was performed.Then three different mathematic pre-processing methods including multiplicative scatter correction(MSC),standard normalized variate(SNV)and Savitzky-Golay(S-G)smoothing were adopted prior to data modeling.The PLSR algorithm was used to construct prediction models of L*,a*,b*parameters and pH values.After contrastion of model prediction performance,the PLSR model after SNV pre-processing for L*parameter and pH values obtained the best prediction results with Rc of 0.98 and 0.90,Rp of 0.92 and 0.87,RMSEC of 0.71 and 0.05,RMSEP of 1.20 and 0.07,respectively.For a*and b*parameters,the PLSR model after S-G smoothing acquired the optimum results with R,of 0.97 and 0.98,Rp of 0.97 and 0.97,RMSEC of 0.43 and 0.31,RMSEP of 0.60 and 0.41,respectively.(2)The study of prediction model for TVB-N content of duck meat.After MSC,SNV and S-G smoothing pretreatments,the PLSR model was established for duck meat TVB-N prediction.Then the synergy interval partial least squares(Si-PLS)and principal component analysis(PCA)methods were used to select important wavelengths for model input variables.In comparison,prediction model based on the full spectra after MSC pre-processing yielded the best prediction effects with Rc and Rp of 0.91 and 0.86,RMSEC and RMSEP of 1.01 and 1.06.(3)The study of applicability for duck meat quality prediction models.PCs scores spatial distribution,mahalanobis distance and model verification methods were used to test the model applicability for duck L*,a*,b*parameters,pH values and TVB-N contents.The results indicated that the prediction performances of duck L*,a*and b*parameters were not declined obviously.However,the prediction accuracy for pH values and TVB-N contents were influenced,with the obviously decreased prediction correlation coefficient.That is to say,the applicability of prediction models for duck pH values and TVB-N contents were declined when predicting the duck samples of new variety,which were necessary to carry out calibration maintenance and transfer.(4)The study of calibration maintenance for prediction model of duck meat pH values and TVB-N contents.Three model maintenance methods based on different principles were applied for prediction model maintenance of duck pH and TVB-N,including model updating correction(MU),slope/bias correction(SBC)and direct standardization(DS).And the numbers of standardization set samples for transfer effect were investigated.The results demonstated that the maintenance effect of duck meat pH values after SBC and DS transfer methods were unsatisfactory,because that the small pH range resulted in inhomogeneous distribution of spectral signal data and pH reference values to influence the optimal transformation parameters.The best maintenance results were obtained by MU method for prediction model of duck meat pH values,with RMSEP reduced from 0.33 to 0.17 and absolute value of mean bias declined from 0.25 to 0.03.In addition,the results showed that univariate SBC method exhibited a more significant maintenance effect on TVB-N prediction model than multivariate DS transfer method and MU,with RMSEP reduced from 1.90 to 0.93 and absolute value of mean bias declined from 1.49 to 0.13.