Rapid Prediction Of Fresh Sweet Potato Quality Based On Hyperspectral Imaging Technology

Sweet potato,as an important crop,is rich in starch,sugar and other nutrients and is widely planted and produced for its drought tolerance,strong adaptability and easy planting.At present,hyperspectral imaging technology has been widely used in the rapid detection of meat,wheat,potato,fruits and vegetables,but the quality of sweet potato is seldom reported.This study was aimed to mine quantitative relationship between spectra of sweet potato and indicators based on the 900-1700 nm hyperspectral imaging technology coupled with linear algorithms,so as to realize the fast detection on sweet potato quality,providing some reference value for the analysis of postharst physiology and quality of sweet potato.The detailed research contents and results are as follows:(1)Hyperspectral imaging technology was used to quickly predict the texture characteristics of sweet potato in TPA mode.Hyperspectral images of sweet potato samples were collected and the ROI was extracted.Finally,the RAW mean reflectance spectral information of 165 sweet potato hardness samples,147 sweet potato springiness samples and 155 sweet potato cohesiveness samples were obtained,respectively.In order to improve the spectral signal-to-noise ratio,MSC,SNV,BC,GFS,and NC were adopted to preprocess RAW spectra.The full-band(900-1700 nm)models were built using PLS algorithm to predict the hardness,springiness and cohesiveness of sweet potato samples.The results showed that PLS model based on SNV spectra had a better prediction effect for hardness,and rP was0.913,RMSEP was 311.886 g;the PLS model based on BC spectra had a better prediction effect for springiness,with rP of 0.925 and RMSEP of 0.029;the PLS model based on NC spectra had a better prediction effect for cohesiveness,with rP of 0.917 and RMSEP of 0.008.Four methods,including RC,SR,SPA and CARS,were used to select the optimal wavelength(high correlation with detection index)and PLS and MLR were used to establish optimized model.With comparative analysis,the SPA-SNV-MLR based on the 7 optimal wavelengths scelected by SPA method had the best performance for predicting hardness of sweet potato(rP=0.919,RMSEP=297.268g);the SR-BC-MLR based on the 4optimal wavelengths screened by SR method had higher prediction accuracy and better stability on springiness prediction(rP=0.918,RMSEP=0.030);the CARS-NC-MLR model based on the 6 optimal wavelengths selected by CARS method was more suitable to predict cohesiveness(rP =0.911,RMSEP=0.009).F test and t test were also used to verify the applicability and data validity of SPA-SNV-MLR,SR-BC-MLR and CARS-NC-MLR models,respectively.(2)The feasibility of predicting rapidly moisture and ash content in sweet potato using hyperspectral imaging combined with MLR algorithm was explored.Different preprocessing methods were used to enhance the signal of RAW reflectance spectra of sweet potato samples(moisture,ash),and the PLS prediction model in the region of 900-1700 nm was established.By comparsion,it was found that the best full-band spectra for moisture and ash prediction models were respectively A spectra and MAS spectra.The PLS models were optimized based on optimal wavelength selected by RC,SPA and CARS methods.The double verifications of F test and t test indicated that the RC-A-MLR model based on the 11 optimal wavelengths obtained by RC method and the SPA-MAS-MLR model based on the 22 optimal wavelengths selected by SPA method were more applicable to respectively predict moisture and ash content in sweet potato and rP were 0.970 and 0.943;RMSEP were 0.0089 g/100 g and 0.042 g/100g;the residual predictive deviation(RPD)was 4.081 and 3.019,respectively.(3)Hyperspectral imaging technology(900-1700 nm)was used to quickly and quantitatively research the nutrient component of sweet potato.The PLS algorithm was applied to mine the linear relationship between the full-band spectral information of 104 sweet potato starch samples,170 reducing sugar samples,80 vitamin C samples,80 protein samples and their corresponding detection indexes.The results displayed that MSC-PLS model,RAW-PLS model,MSC-PLS model and RAW-PLS model on account of 486 wavelengths were respectively fit for predicting starch,reducing sugar,vitamin C and protein content in sweet potato.Three chemometrics methods(RC,SR and SPA)were respectively adopted to pick up the optimal wavelength originated from the optimal MSC spectra of sweet potato starch samples,the optimal RAW spectra of reducing sugar samples,the optimal MSC spectra of vitamin C samples and to the optimal RAW spectra of protein samples establish the optimization model.The study found that MLR models built with 18 optimal wavelengths selected by SR method,11 optimal wavelengths secreened by RC method and 20 optimal wavelengths obtained by SPA method had the best prediction performance for detecting starch,reducing sugar and vitamin C content in sweet potato,with rP of 0.970,0.915 and 0.915;RMSEP of 0.374 g/100 g,0.355 g/100 g and 2.492 mg/100g;RPD values of4.094,2.393 and 2.244;?E of 0.064 g/100 g,0.023 g/100 g and 0.966 mg/100 g,respectively.For the detection of protein content of sweet potato,the PLS model based on 15 optimal wavelengths extracted by SPA method had the best prediction effect,and rP?RMSEP?RPD??E was 0.911?1.029 mg/g?2.481?0.181 mg/g respectively.