Research On Non-destructive Testing Of Chilled Mutton Freshness Based On Near-infrared Spectroscopy

As the main edible meat,mutton has the advantages of unique flavor,fresh and tender meat,rich in a variety of nutrients,and is deeply loved by consumers.Domestic mutton consumption demand is strong.At the same time,the traditional quality inspection of mutton meat has high requirements,which are mainly reflected in the experimental environment,equipment and facilities,time and professional staff.Overly refined operations cannot meet the needs of the public for meat quality testing.Near-infrared spectroscopy is a simple,fast,efficient,and easy-to-use detection technology,which can meet the needs of fast,nondestructive and real-time mutton quality detection technology.In this study,using near-infrared diffuse reflectance spectroscopy non-destructive testing technology,combined with two types of TVB-N and p H physicochemical values,140 chilled mutton samples were selected for the experiment.The experimental period was 10 days,and the storage temperature of mutton was the two boundary values of 0°C and 4°C in a cold environment.The prediction model of TVB-N and p H was established,and the data analysis process was shown,and the changes of indicators were analyzed.And based on this,developed and realized the freshness grading system of chilled mutton.The main results of this research are as follows:(1)Due to the influence of baseline drift,experimental environment,instrument accuracy and other factors,the spectrum needs to be preprocessed to reduce noise,using multiple scattering correction(MSC),average smoothing,standard normal transformation(SNV),SG Convolution smoothing(SGS),SavitzkyGolay first derivative(SG-1st)and second derivative(SG-2nd).After SGS preprocessing,the R2 c of the partial least squares(PLS)model is as high as 0.9973,and the R2 v is 0.8892.The best preprocessing method is determined to be SGS.Outliers were removed using principal component analysis(PCA)combined with Mahalanobis distance(MD)for outliers and triple standard deviation method.Select the common exclusion part of PCA+MD and three times the standard deviation,which is 3.By eliminating abnormal data samples,the stability of the model is effectively improved.(2)The two types of physical and chemical values are divided into characteristic wavelength selection and three methods of PLS,BPNN and SVM are used for modeling prediction;for TVB-N content,SGS preprocessing is used,and the competitive adaptive algorithm is the characteristic wavelength The selection method,combined with the support vector machine model(SGS-CARS-SVM),the predicted value and the actual value are basically linearly correlated,the calibration set R2 c is 0.9776,and the validation set R2 v is0.9497.For p H value,using SGS preprocessing,the competitive adaptive algorithm is the characteristic wavelength selection method,and the BP neural network model(SGS-CARS-BPNN)model has the best effect.The training set R2 train is 0.9423,and the test set R2 test is 0.8736.It can be seen that CARS is the best preprocessing method.(3)Based on the Matlab App Designer platform,the freshness grading system of chilled mutton was researched and developed.According to two physical and chemical values,the freshness of chilled mutton is divided into three grades: fresh,sub-fresh and spoiled.The mutton freshness grading system was designed and developed,which effectively enhanced the automation of mutton quality and safety testing.It meets the needs of all links in the prediction of key parameters of chilled mutton quality and freshness grading.