Research On The Near-ground Multi-scale Spectral Quantitative Detection Model Of Jujube Quality In Southern Xinjiang

At present,the quality detection of jujube and other fruits by spectral analysis technology is mainly concentrated on the indoor single fruit scale.The established quality prediction model is difficult to be directly applied to the quality detection of forest fruits at the canopy and regional scales.The reason is that at the canopy and regional scales,the sample spectral data will be affected by the sample orientation,illumination angle and detection angle,which makes the spectral data at different scales vary greatly.In view of this,this paper takes jujube in southern Xinjiang as the research object,studies the jujube quality detection models at different scales,collects the spectral information of jujube at three scales of single fruit,canopy and region near the ground,and explores the construction of jujube quality detection models at different scales.It provides a reference for the rapid and large-scale inversion of the quality of jujube and other fruits.The main research contents and conclusions are as follows :(1)The near-infrared hyperspectral data of jujube were collected by hyperspectral imaging system,and the prediction models of moisture content and soluble solids content of jujube were established.The effects of pretreatment method and characteristic wavelength selection method on the prediction models of moisture content and soluble solids content of jujube were compared.The best pretreatment methods for moisture content and soluble solids content detection at the single fruit scale are determined to be multivariate scattering correction and SG convolution smoothing,respectively.The correlation coefficients of the model prediction set are 0.9192 and 0.8311,respectively.On the basis of the optimal pretreatment,the characteristic wavelength selection methods of jujube moisture content and soluble solids are determined to be random frog leaping algorithm and competitive adaptive reweighted algorithm,respectively.The correlation coefficients of prediction set of jujube moisture content and soluble solids prediction model based on characteristic wavelength are 0.9658 and 0.8682,respectively,indicating that the moisture content and soluble solids content of jujube have better prediction effect at single fruit scale.(2)The near-infrared hyperspectral data of canopy jujube were collected,and the prediction models of water content and soluble solids content of jujube were established.It was found that the correlation coefficients of the prediction set of the jujube moisture content and soluble solids prediction model based on canopy spectrum are 0.8015 and 0.6452,respectively.Compared with the model established at the single fruit scale,the model performance decreased significantly,and the prediction of jujube quality could not be well realized.Therefore,this paper explores the use of spatial characteristic spectrum to correct outdoor spectrum,and uses Walthall,Shibayama,Ross-Li,Roujean and Rahman five BRDF models to invert the spatial characteristic spectrum of canopy jujube.The results show that the average inversion error of the five models is about 3.6 %,which can better invert the spatial characteristic spectrum of jujube.The spatial characteristic spectrum obtained by inversion was used to correct the canopy spectrum,and the prediction model of jujube moisture content and soluble solids content was established.The results showed that the correlation coefficient of the prediction model of jujube moisture content based on the modified spectrum obtained by Shibayama model is 0.8759,and the correlation coefficient of the prediction model of jujube soluble solids content based on the modified spectrum obtained by Walthall model is 0.7182.The performance of the model established by the modified canopy spectrum is improved,indicating that the method of using spatial characteristic spectrum to correct the canopy spectrum is feasible.It can improve the quality detection accuracy of canopy jujube.(3)At the regional scale,the correlation between multi-spectral reflectance and vegetation index of jujube and water content was analyzed,and the regression model of water content was established.The results showed that the data of 650 nm,730 nm and 840 nm in multi-spectral reflectance are highly correlated with the moisture content of jujube,and SIPI,GI,SRPI,NPCI,EXR and PSRI in vegetation index are highly correlated with the moisture content of jujube.The results of univariate regression and multiple regression models established by showed that the regression model established by multispectral reflectance and vegetation index had poor fitting effect on the moisture content of jujube,and the fitting correlation coefficients are 0.5347 and 0.5367,respectively.The multi-angle data fusion method was used to improve the prediction effect of jujube moisture content at the regional scale.The multi-spectral data at four azimuth angles of unmanned air vehicle(UAV)were extracted and fused,and the partial least squares regression model was established.The results show that the correlation coefficient of the prediction set of the regression model established by the multi-angle fusion reflectance is 0.6746,and the correlation coefficient of the prediction set of the regression model established by the multi-angle fusion vegetation index is 0.6689,which indicates that the method of multi-angle data fusion can improve the detection effect of UAV multi-spectral on the moisture content of jujube.It has certain application potential in the inversion of jujube moisture content at the regional scale of UAV.