Estimation Models Of Physicochemical Parameters Of Winter Rape Based On Hyperspectral Parameter

Rape is an important economic crop in the arid and semi-arid region of northwest China,the yield and quality of rape would have an importance impact on agriculture and economic development in the region.It is meaningful to the growth evaluation of rape such as chlorophyll content(SPAD),anthocyanin content(ANTH)and leaf nitrogen content(LNC)of rape,this biophysical and biochemical parameters is good indicators for yield of rape,the aim is to supply the evidence for crop management and fertilizer.The relationship between hyperspectral data and biophysical a biochemical parameters was studied,the model is established by mathematical statistics can realize the prediction of rapeseed information.The brassica napus of different grow stages and different nitrogen treatment was studied and field trials were conducted in 2016-2016,the ground non-imaging spectra data,imaging spectra data and lowaltitude UAV hyperspectral images were obtained and chlorophyll content,anthocyanin content and leaf nitrogen content of rape were measured.The correlation between the spectral characteristics and physiological and biochemical parameters was analyzed and the quantitative relationship between various hyperspectral parameters and physiological and biochemical parameters was explored,the SPAD estimation model,ANTH estimation model of different growth stages and LNC estimation model of whole growth stage were constructed based on various modeling methods and the spatial distribution map of physiological and biochemical parameters of rape budding stage was generated by the high spectral image of UAV.The main results are as follows.(1)The changes of physiological and biochemical parameters,canopy spectra and first derivative spectral were analyzed.The results showed that from seeding stage to maturity stage,chlorophyll content of rape first increased and went down.The anthocyanins showed the opposite trend,the budding stage was the lowest value in the whole stages,and the maturity stage reached the highest,the LNC kept decreasing in the whole stages.The canopy spectra and first derivative spectral of rape with SVC and SOC spectrograph showed similar changes during the whole stages,from seeding stage to budding stage,the spectral reflectance were decreased and increased in the visible waves,after the budding stage,the law of the change was the opposite;The first derivative spectra showed the characteristics of “bimodal”.There existed “red shift” phenomena for the position of red edge from seeding stage to budding stage,after budding stage,there existed “blue shift” phenomena for the position of red edge.(2)The relationship between chlorophyll content and spectral reflectance of rape was studied.The results showed that the correlation between the spectra and the first derivative spectral and SPAD showed similar changes in different growth stages.The SPAD single factor estimation model was constructed based on the original spectral and first derivative spectral,the SPAD estimation model was constructed by selecting eight spectral parameters with good correlation based on least square calculating algorithm?partial least squares algorithm?multiple stepwise regression algorithm and random forest algorithm,the results show that the accuracy of SPAD model based on random forest algorithm is better than other models.(3)The characteristic bands and optimal spectral parameters of ANTH were determined by the correlation analysis between anthocyanin content and spectral reflectance and spectral parameters in different growth stages of rape,the modeling results showed that the model based on the first derivative spectral characteristic band is superior to the original spectral characteristic band;The ANTH multispectral parameter model based on multivariate stepwise regression algorithm and partial least-squares algorithm is better than that of the least squares algorithm;The random forest model greatly improved the inversion accuracy of anthocyanin in rape,it was the optimal model for ANTH estimation of rape.(4)The correlation between the spectral characteristics of rape canopy and LNC of rape was studied,the LNC estimation model is constructed model is constructed based on characteristic band and hyperspectral parameter.The results show that the canopy spectral reflectance of rape in the visible band reduced with increasing LNC level,and the reflectance of the near-infrared band was gradually increasing.There existed “red shift” phenomena for the position of red edge with the increase of LNC.The LNC single factor estimation model was constructed based on the characteristic band of LNC;The correlation between 18 hyperspectral parameters and LNC was analyzed,the polynomial model based on NDNI was found to be the optimal single factor estimation model for rape LNC;the SPAD estimation model was constructed based on partial least squares algorithm?multiple stepwise regression algorithm and random forest algorithm,the results showed that the model based on random forest algorithm is optimal.(5)The random forest estimation model was obtained by using the training,the inversion of the spatial distribution of rape SPAD,ANTH,and LNC were achieved at regional scale based on Hyperspectral images in UHD,and the spatial distribution map of physicochemical parameters of rape was obtained.The results show that model based on random forest can accurately estimate the spatial distribution of the physical and chemical parameters of rape,its inversion results conform with practical situation fairly,it has high accuracy and reliability.