| The Lingwu long jujube is the dominant agricultural product in Ningxia.It is necessary to explore an effective method for nondestructive testing of long jujube quality,and to provide a theoretical reference for the variability,serialization,and high added value.By using hyperspectral and laser induced fluorescence hyperspectral techniques and combined with chemometrics,the main soluble sugar(sucrose,fructose and glucose)in Lingwu long jujube in different storage period was detected,and the post harvest quality detection of Lingwu jujube was realized.The main conclusions are as follows:(1)Determine the visible/near-infrared hyperspectral acquisition parameters:the exposure time is 10ms,the image acquisition speed is 200?m/s,the objective lens height is 385mm,and the scan length is 60mm.The fluorescence hyperspectral acquisition parameters:the exposure time is 20 ms,the image acquisition speed is 200 ?m/s,the scan length is 60 mm,the vertical distance between the collimator and the sample is 305 mm,and the angle between the collimator and the horizontal line is 70°.(2)The content of sucrose,fructose and glucose in jujube was determined by HPLC.The conditions for the final determination of sucrose,glucose,and fructose in jujube were determined by discussing the mobile phase and liquid phase conditions,respectively.The injection volume was 10 ?L,the mobile phase was ultrapure water,the flow rate was 0.4 mL/min,and the isocratic elution was carried out.The column temperature was 30? and the refractive index detector temperature was 35?.(3)Study on visible and near infrared hyperspectral nondestructive detection of long jujube.The mean and standard deviation of each sample were calculated by Monte Carlo sampling method,and the abnormal samples were discriminate according to the threshold value.The effects of different pretreatment methods on the model were explored.The results showed that the PLSR model of the sucrose pretreated by OSC had the best effect,and the model Rc was 0.853.The PLSR model of fructose pretreated by OSC had the best effect,and the model Rc was 0.847.The PLSR model of glucose pretreated by SNV+OSC had the best effect.The model Rc was 0.803.Using six characteristic wavelength methods to extract characteristic wavelengths and compare model effects.In the sucrose prediction model,67 characteristic wavelengths were extracted by the IRF method,and the sucrose prediction model based on IRF-LS-SVM had the best effect,the model Rc was 0.920 and Rp was 0.823.In the fructose prediction model,14 characteristic wavelengths were extracted by the GAPLS method.The fructose prediction model based on GAPLS-LS-SVM has the best effect,the model Rc was 0.850 and Rp was 0.793.In the glucose prediction model,16 characteristic wavelengths were extracted by the SPA method.The GAPLS-LS-SVM glucose prediction model had the best effect,the model Rc was 0.765 and Rp laser was used.The intensity range of the light source is 615?655nm.The laser-induced hyperspectral images of long jujube were collected to obtain the fluorescence spectra from 660 to 850 nm with a total of 40 wave points.Monte Carlo sampling method was used to detect abnormal samples,and the effects of different pretreatment,characteristic wavelength and different models were analyzed and compared.The results showed that the optimal prediction models for sucrose,fructose and glucose based on fluorescence spectra were BiPLS-LS-SVM,CARS-LS-SVM and GAPLS--LS-SVM models,respectively.The sucrose and glucose model had a poor effect,and the fructose model was relatively good.The Rc of the model was 0.670 and Rp was 0.702.Contrast the characteristic wavelengths and long jujube fluorescence peaks to analyze the model effect.This result is mainly due to a weak correlation between chlorophyll fluorescence and the three sugar contents,and there may be absorption of other substances in the fluorescence information.A variety of light signals are fused together to establish links with sugar and affect the model correlation. |
PDF Full Text Request