Research On Rapid Analytical Methods For Pork Quality Assessment Using Near Infrared Spectroscopy

In order to protect the health,safety and legal rights of consumers,the quality of pork must be strictly controlled in the process of processing,transportation and sales of pork.The traditional methods of pork inspection include sensory inspection,physical and chemical inspection methods,which are typically characterized by subjective results,complex analysis,high cost-time and requiring well-skilled operator.In view of this,this paper studies the rapid detection method of pork quality based on near-infrared spectroscopy,including the preprocessing and dimensionality reduction methods of the pork spectra,and the construction of qualitative analysis models.First,in view of the deviation of the absorption rate caused by spectral data acquisition,the internal structure of the spectrometer and other human factors,several preprocessing methods(e.g.,SNV,MSC,moving average smoothing,Savitzky-Golay convolution smoothing method and first derivative spectral method)are applied to recover and denoise the spectral data.Considering that the high-dimensional data has the features of information redundancy and collinearity,principal component analysis and Fourier transform are used to extract the features and reduce dimension.Secondly,on the basis of denoising and dimensionality reduction of pork spectra data,the qualitative analysis models of pork quality are built based on support vector machine,BP neural network,random forest and broad learning system.Combined with the theory of modeling method,the performance of qualitative analysis models is comprehensively evaluated on the conditions of the same preprocessing methods and sample set partition.The best 5-fold cross-valid prediction accuracy of those four qualitative analysis models are 88.48%,88.57%,89.05% and 79.62%,respectively.Then,on the basis of the theoretical methods above,the rapid measurement technology of pork freshness using a portable near infrared spectroscopy is studied.The experimental results including accuracy and time show that the combination model,standard normal variable + discrete Fourier transform(DFT)+ support vector machine(SVM),yields better results than the other methods.Consequently,it is applied to build qualitative analysis model of the pork quality and detect fresh pork and chilled pork on-line and real-time.Finally,based on the research theory above,the user interface of pork quality near-infrared spectroscopy system is designed,to speed up the process of using near-infrared spectroscopy to detect the pork quality on-line and real-time.