Non-destructive Detection Of Mutton Freshness Based On Multispectral Technology

The number of microorganisms on the surface of mutton can well reflect the freshness of meat,thus knowing whether mutton is corrupt.Traditional meat detection methods usually destroy meat and extract ingredients for testing meat quality experiments.The detection method has the defects of low efficiency,long detection period,meat surface damage and the like.After the multispectral camera shoots different objects,the spectral information will be different.In this paper,multispectral technology is studied.The combination of mutton images and spectral information to comprehensively judge the meat quality is of great significance to realize accurate and rapid nondestructive testing of agricultural and livestock products.Taking Tangxian county mutton in Hebei province as the experimental research object,using multispectral camera to obtain image data,the prediction model of total bacterial count was established,and the biological chemometrics method was used as the control.At the same time,the rapid non-destructive detection research was conducted on the number of living microbial cells on the surface of cold fresh mutton in different wavelength ranges.Using multispectral and image processing techniques,a prediction model for the total number of bacteria on the surface of chilled fresh mutton was established,and the freshness was graded.In this paper,through the comparative analysis of different wavelength ranges,pretreatment methods and modeling methods,the best detection method for the total number of bacteria in cold fresh mutton was determined,which laid a foundation for the rapid nondestructive detection of cold fresh meat.The main work is as follows:(1)Pretreatment of spectral data.Multi-spectral data of cold fresh mutton samples are collected(wavelength range is 600-1000nm).Aiming at the noise and translation problems of cold fresh mutton spectral data,standard normal distribution(SNV),derivative method,vector normalization(VN),multiple scattering correction(MSC)and smoothing method are used to preprocess the spectral data.Experiments show that S-G convolution smoothing method combined with multiplicative scatter correction is the best method.(2)Prediction model of total bacteria in mutton.A kernel extreme learning machine(KELM)model is established based on an extreme learning machine(ELM),a neural network algorithm,and the KELM model is optimized by genetic algorithm.The model is verified in the range of 600-800 nm and 800-1000 nm,and compared with BP-ANN,RBF-ANN,PLSR in terms of correlation coefficient,root mean square,prediction error,etc.The improved extreme learning machine model is superior to other models.(3)Taking the total number of bacteria on the surface of cold fresh mutton as the reference value and combining with the national food safety meat standard,a freshness classification method was designed,and the cold fresh mutton was divided into three grades: fresh,sub-fresh and spoilage.(4)Implementation of testing platform.Using Java language to develop a bacterial count detection platform,the functions of spectral data pretreatment,bacterial count prediction and freshness identification are realized.