| Mutton has the advantages of delicious taste and rich nutrition,so it is very popular with consumers.With the enhancement of people’s awareness of food safety,the mutton quality is required by consumers.As an important evaluation index of mutton quality,the freshness determines the economic value and edible nature of mutton.The accurate detection of mutton freshness can protect consumers’ rights and interests and food safety,and strengthen supervision of meat market by food supervision department at the same time.Because the traditional meat freshness detection method can not meet the requirements of meat processing enterprises to achieve the large-scale,efficient,nondestructive and online detection for meat products,it is very necessary to find a rapid,accurate and nondestructive detection method for mutton freshness.Through test determination,the correlation analysis of freshness indexes of vacuum packaged chilled mutton was carried out and the comprehensive evaluation criteria were established.The spectral information of 400~1000 nm and 900~2500 nm was obtained by the hyperspectral image technology with visible short wave and the near infrared spectroscopy technology with long wave.The optimal spectral preprocessing method was determined,the feature variables were screened to establish a better discriminant model,and a better discriminant model was established.Fused discriminant model of mutton freshness based on the feature layer fusion was established through information fusion to discuss the improvement of information fusion on the classification accuracy of the discriminant model of mutton freshness,which could realize the rapid and accurate evaluation and provide a theoretical basis for developing the miniaturized online detection device of mutton freshness.The specific research contents and conclusions are as follows:(1)Correlation analysis and comprehensive evaluation among mutton freshness indexesThrough test determination,the correlation analysis of freshness indexes of vacuum packaged chilled mutton was carried out and the comprehensive evaluation criteria were established,which were as follows: for fresh samples,the storage time is on day 1~11,the content of TVB-N is less than15 mg/100 g,the value of TVC is less than 5×106 CFU/g,the value of L* is greater than 40.0,and the value of a* is greater than 19.3;for sub-fresh samples,the storage time is on day 12~16,the content of TVB-N is between 15 and 25 mg/100 g,the value of TVC is between 5×106 and7×106 CFU/g,the value of L* is between 38.9 and 40.0,and the value of a* is between 18.9 and19.3;for corrupt samples,the storage time begins in seventeenth days,the content of TVB-N is greater than 25 mg/100 g,the value of TVC is greater than 7×106 CFU/g,the value of L* is less than 38.9,and the value of a* is less than 18.9.(2)Nondestructive detection of mutton freshness based on hyperspectral image technologyFor the spectral information of mutton in the range of 400~1000nm,the optimal spectral preprocessing method was 1D+S-G(3).CARS and GA algorithm were used in preliminary screening variable for data reduction,and the number of spectral variables was reduced to 15%and 17% of the all variables,respectively.For the established prediction models of mutton freshness,the discriminant accuracy of the calibration set and prediction set was 99% and 98%,respectively,and the number of variables selected was still larger.SPA algorithm was further select variables to solve the problem of spectral data redundancy effectively.The number of spectral variables was reduced to 8 and 10,respectively.But the accuracy of the prediction model for mutton freshness was reduced,the accuracy of prediction set was 91% and 89%,respectively.(3)Nondestructive detection of mutton freshness based on near infrared spectroscopy technologyFor the spectral information of mutton in the range of 900~2500 nm,the optimal spectral preprocessing method was 1D+S-G(3).CARS and GA algorithm were used in preliminary screening variable for data reduction,and the number of spectral variables was reduced to 25%and 22% of the all variables,respectively.For the established prediction models of mutton freshness,the discriminant accuracy of the calibration set and prediction set was 96% and 91%,respectively,and the number of variables selected was still larger.SPA algorithm was further select variables to solve the problem of spectral data redundancy effectively.The number of spectral variables was reduced to 14 and 15,respectively.But the accuracy of the prediction model for mutton freshness was reduced,the accuracy of prediction set was 80% and 82%,respectively.(4)Nondestructive detection of mutton freshness based on information fusionIn this study,the fused discriminant model of mutton freshness was established by using information fusion method,and the data layer fusion and feature layer fusion were compared.The data layer fusion model had a better discriminant performance,but the whole variable information led to its low modeling efficiency.The feature layer fusion model had good discriminant performance and could realize the rapid and accurate discrimination of mutton freshness.Compared with the single detection model,the accuracy of the prediction set was significantly improved,which was 96%. |
PDF Full Text Request