Identification Of Mutton Origin In Ningxia And Detection Of Freshness Based On Hyperspectral Technology

Use hyperspectral technology to establish the identification model of the mutton origin in Yinchuan,Guyuan and Yanchi in Ningxia,and quantitatively analyze the conductivity,protein content and pH of the mutton to achieve the detection of the freshness by hyperspectral technology.The specific research content is as follows:(1)Visible near-infrared and near-infrared hyperspectral of samples from mutton’s hind legs in three areas were obtained.The 400-1000 nm band spectrum was pretreated with the Derivative method,and the 900-1700 nm band spectrum was pretreated with the area-normalized method.Extract the feature wavelengths from the pre-processed spectrum by the SPA,CARS and UVE methods.The PLS-DA and KNN methods are used to establish the discriminative model under the characteristic bands.The results showed that the PLS-DA model based on any spectral data is superior to the KNN model.In the 400-1000 nm band spectrum,The PLS-DA model is established in the characteristic wavelengths extracted by UVE is the best.Instead of the full spectrum is feasible.By comprehensive comparison,The UVE-PLS-DA model is optimal.It has a classification rate of 0.935,the prediction set ratio is 0.912.In the 900-1700 nm band spectrum,The PLS-DA model is established in the characteristic wavelengths extracted by CARS is the best.Therefore,The CARS-PLS-DA model is optimal,It has a classification rate of 0.905,the prediction set ratio is 0.842.It Proved that hyperspectral imaging technology could identify the origin of mutton.By comprehensively comparing the optimal models of the two bands,it is found that the origin identification model based on visible near-infrared hyperspectral is superior to the origin identification model based on near-infrared hyperspectral.(2)Establish the PLSR prediction model of mutton’s pH and conductivity with visible near-infrared hyperspectral technology.The result showed that the UVE-PLSR model was the best in the PLSR model of pH.The correlation coefficients of the calibration set and the prediction set were 0.931 and 0.891 respectively.and the RMSEC and RMSEP were 0.1067 and 0.1520 respectively.The correction and prediction performance is relatively good.In the PLSR model of conductivity,the PLSR model established by the original spectrum is the optimal model.(3)Use the model updating method and S/B method to modify the PLSR model of protein of the Guyuan Mutton sample to predict the Yinchuan Mutton sample protein content.The prediction correlation coefficient(Rp)of the model corrected by the model updating method for Yinchuan sample was increased from 0.318 to 0.771,and the prediction error(RMSEP)decreased from 0.8476 to 0.4308.The prediction correlation coefficient of the model for Guyuan sample was reduced from 0.864 to 0.756,and the prediction error was increased from 0.2062 to 0.2714,the predicted root mean square error of the model corrected by the S/B method for Yinchuan mutton was decreased from 0.8476 to 0.0008,Declined by 99.91%.It does not change the prediction performance of the model for the Guyuan mutton sample.By comprehensive comparison,the S/B method is more suitable for correcting the PLSR model of mutton’s protein to predict mutton of different origins.(4)Establish the PLSR prediction model and the LS-SVM prediction model of mutton’s pH and conductivity with near-infrared hyperspectral technology.The results showed that the validation coefficient of the calibration set in the LS-SVM model of pH was above 0.92,and the prediction correlation coefficient was above 0.85.It was more stable than the PLSR model and had a good prediction ability.The LS-SVM model was established in the spectrum preprocessed by MSC was the best.Considering comprehensively,The MSC was the optimal spectrum pretreatment method for the LS-SVM model.In the PLSR model of Conductivity,The De-Treading was the optimal spectral pretreatment method.The SNV was the optimal spectral pretreatment method for the LS-SVM model of Conductivity.Compared to the optimal PLSR model of pH and conductivity in the two bands which were 400-1000nm and 900-1700nm,we found that the mutton’s pH quantitative analysis model based on visible near-infrared hyperspectral was superior to model established based on near-infrared hyperspectral.