Research On Sweet Potato Quality Detection Method Based On Hyperspectral And Terahertz Spectroscopy

The total yield of sweet potato is the fourth place in China,next to rice,wheat and corn.It plays an important role in the development of diet and industrial economy in our country.The methods of sweet potato origin classification,postpartum online classification and rapid detection of product safety were closely concerned by consumers and food processing manufacturers.The quality of sweet potato directly affects the quality,grade and economic value of processed products,and the safety of sweet potato processed products directly threatens human health.Therefore,there is an urgent need for a fast and efficient technology to detect the quality and safety of sweet potato.In this paper,using hyperspectral and terahertz time-domain spectroscopy,combined with a variety of chemometrics methods,to study the methods of sweet potato origin identification,internal quality,and harmful additive detection.It provide the basis for the development of sweet potato industry quality and safety testing.The main research contents and conclusions are as follows:1.The identified for the origin of purple sweet potato with hyperspectral imaging technology.Taking purple sweet potatoes from Fujian,Guangxi,and Shandong as research objects,combined with principal component analysis(PCA),continuous projection algorithm(SPA),non-information variable elimination(UVE)dimension reduction and screening variables methods,there are three types of discriminant models were established,partial least squares discriminant analysis(PLS-DA),least squares support vectors(LS-SVM)and extreme learning machine(ELM).The results show that the PCA-LS-SVM and UVE-ELM were the best models for the classification of producing areas,and the misclassification rate was 4.598%.The experimental results show that the LS-SVM and ELM models established by combining suitable band screening methods can achieve rapid identification and classification of purple potato producing areas.2.The detection for the internal quality(SSC,dry matter content)of purple sweet potato with hyperspectral imaging technology.Using Guangxi purple sweet potato as the research object,PLS model was established by using a variety of spectral pretreatment methods and original spectra.The results showed that the original spectrum and the normalized pre-processed spectrum were the optimal spectral data,respectively.Then using PCA,SPA,UVE dimensionality reduction and variable selection methods,the PLS,LS-SVM,and ELM prediction models for SSC and dry matter content of Guangxi purple sweet potato were established,respectively.The results of the quantitative model established by using the method of screening variables were compared.The results showed that the LS-SVM model based on the RBF-Kernel with 20 principal component variables as input variables had the best prediction effect on SSC and dry matter content of purple sweet potato.The RMSEP of the model was 0.439~obrix and 0.010g respectively,and the R_p of the model was 0.957 and 0.953respectively.The experimental results show that PCA-LS-SVM model can improve the efficiency and prediction effect of the model,so as to speed up the speed and accuracy of SSC and dry matter content classification detection of purple sweet potato.3.Detection of harmful additives(alum)in sweet potato starch by terahertz time-domain spectroscopy.THz spectra of sweet potato starch,alum and mixture samples were collected.Analysis of the spectrum revealed that alum had obvious absorption peaks at 0.980,1.065,and 1.146 THz,and the absorption peak became more pronounced with the increase of the content of alum in the mixed sample.A variety of pretreatment methods were used to establish a LS-SVM prediction model for alum content.The results showed that the LS-SVM model established using the normalized preprocessing method has the best prediction effect,the RMSEP of model was 0.0047 and the R_p of model was 0.9972.Then,the LS-SVM model is established by using the methods of UVE,SPA and UVE-SPA to select the characteristic variables.The results showed that the normalized-SPA-LSSVM model had the best prediction ability for alum content in sweet potato starch.The RMSEP of model was 0.0046 and the Rp of model was 0.9976.Only 39 variables are needed to participate in the modeling,which can effectively improve the modeling speed and the prediction accuracy of the model.The experimental results show that THz-TDS technology can detect harmful additives quickly.To sum up,hyperspectral imaging and terahertz spectroscopy technology can be used to detect the origin,internal quality,harmful additives of sweet potato,achieve better prediction accuracy,and provide a reference basis for the sweet potato industry quality and safety real-time rapid detection,achieve better prediction accuracy,and provide reference for the quality and safety real-time and rapid detection of sweet potato industry.