| As a kind of spectrometry technology,NIR spectroscopy is characterized by simple analysis,high speed and online,and is often used in the analysis of composition and function of food,process intermediates and final products.It has become a powerful technology for food quality and safety analysis..Today,the deep processing of Linde products and the increasing demand for food quality and safety,the non-destructive testing of nuts has become a hot issue for many researchers.The newly picked pine nuts contain a large amount of water,which is highly susceptible to mold and yeast during storage and causes oxidative deterioration.However,the mildewed pine nuts have a problem of being difficult to distinguish from normal pine nuts after being subjected to operation such as frying,chemical immersion,and the like.In response to this problem,this paper uses near-infrared spectroscopy to identify mildew pine nuts and carry out normal and mildew classification modeling of pine nuts.Firstly,the Fourier transform infrared spectroscopy technique is used to collect the near-infrared data of pine nuts,and the collected near-infrared spectral data is preprocessed by SNV,Norris-Williams derivation and wavelet transform to achieve spectral aggregation and wavelength curve smoothing,the result of.The near-infrared diffuse reflectance spectrum has the problems of poor precision and large error.For this reason,the improvement of the diffuse reflectance spectral index is studied to improve the correlation between the diffuse reflectance spectral data index and the chemical composition of the sample.On this basis,the Lie group geodesic metric method is used to improve the local linear embedding and equal metric mapping methods,and the data feature reduction study is carried out by means of random forest and lifting tree model,and the principal component analysis is used to improve the dimensionality.The dimensionality reduction method is compared and verified.The research results show that the improved dimensionality reduction method is more suitable for the new near-infrared spectral diffuse reflection evaluation index.Finally,combined with the Lie group geodesic measurement method and Gaussian process,pine nut classification modeling research is carried out.The corresponding kernel function-Gaussian process classification model is established for seven different kernel functions and compared.The probabilistic calibration technique is introduced into the accuracy correction of the improved Gaussian model.Improve the accuracy of the Gaussian model for correction.According to the modeling experiment,it is known that the periodic-radial basis combined kernel function can be used as the kernel function to improve the Gaussian process to obtain the best classification effect.Using the improved local linear embedding to reduce the dimension of the pine nut data,the periodic-radial basis combined kernel function is used.The model accuracy of Gaussian process modeling is 0.888;the accuracy of the model after probabilistic calibration is 0.864;the model of the periodicity-radial basis combined kernel function-Gaussian process modeling using the improved equal measure map to reduce the pine nut data The accuracy is 0.849;the accuracy of the model after calibration using the probability is 0.958.According to the classification data analysis of the model,it can be known that the improved diffuse reflection evaluation index can improve the accuracy of the final classification model;the periodic-radial basis combined kernel function has the advantages of both the periodic kernel function and the radial basis kernel function,and is the most modeled for the mildew pine.Good choice;probabilistic calibration does not guarantee that the classification effect of the model becomes higher,but the improvement of the equal measure map and the dimension reduction of the pine nuts using the probability calibration method to obtain a better classification model of the pine nutrient. |
PDF Full Text Request