Study On Monitoring Chlorophyll Model Of Rapeseed(B.napus) Leaves Based On Hyperspectral Imaging Technology

China is in its transitional stage of agriculture.The key to the transition from traditional agriculture to modern agriculture lies in the combination of agriculture and modern information technology.Agricultural remote sensing is the concrete manifestation of modern agriculture.As an important means of its monitoring,hyperspectral technology can grasp the farmland environment and crop growth conditions in a fast,nondestructive,regular and immediate way by agricultural information.Hyperspectral technology is the key basis of agricultural production guidance,and plays a very important role in the realization of agricultural informatization and precision.Rape is the main cash crop and the largest oil crop in China.How to plant rape efficiently and to adjust planting measures scientifically is the focus of current research.Chlorophyll,an important factor in crop growth,can be used to monitor crop growth in real time.This paper uses hyperspectral technology to study the quantitative relationship between chlorophyll content and spectrum,and finds out the best inversion model of chlorophyll in all growth stages,so as to provide a reference for monitoring the growth of rape by hyperspectral remote sensing technology in the future.This is of great significance not only for grasping the rape growth and development status in real time,realizing remote field management,adjusting planting measures timely and accurately,and then completing field operations efficiently,but also for improving the yield and quality of rape.This study with conventional high oleic acid rape varieties as materials,field experiments were carried out to study rape leaf the spectral response of rape leaves in seedling stage,bolting stage and initial flowering stage under different cultivation measures by field experiments.By calculating the correlation between the first derivative of the reflectance spectrum and SPAD value,the sensitive band of rape leaves was selected by combining with stepwise regression,and the spectral index and spectral characteristic parameters were calculated.Traditional regression and neural network were used to build the chlorophyll content estimation model.Main contributions of this work are as follows:The spectral curve of rape leaves was closely related to the plant growth and varied at different growth stages.At seedling stage,leaf’s ability of light reflection and absorption is relatively weak.The reflectance of the leaf spectrum in the green light region increases,the area of the absorption valley in the blue and red light range increases,and the light reflection degree in the near-infrared region increases.The spectral curve of the leaves at the early flowering stage remained high reflection in the green light region.The positions of red edge were 714nm,718nm and 718nm respectively in seedling stage,bolting stage and initial flowering stage,which showed the phenomenon of"red shift"from seedling stage to bolting stage.The amplitude and area of red edge increased from the seedling stage,and decreased gradually after reaching a peak value at the bolting stage,showing the phenomenon of"blue shift".Rape leaf spectrum has a high correlation with leaf chlorophyll content.Spectral reflectance and first derivative sensitive to SPAD values of rape leaves at three growth stages were all within the visible range.The spectral index formed by pairwise combination of common spectral characteristic parameters and sensitive wavelength enhanced the correlation with SPAD value to some extent.Among the spectral characteristic parameters,the combination parameter of area parameter has the best correlation,and the correlation coefficient is between 0.580～0.896.In the spectral index,RSI,DSI and NDSI have a good correlation with SPAD value,and most of the spectral index correlation coefficients are above 0.5.The correlation coefficients of spectral characteristic parameters and spectral index calculated at the initial flowering stage were generally higher than those calculated at seedling stage and bolting stage.The prediction model determines the estimation effect of chlorophyll in rape.The accuracy of the model constructed by spectral characteristic parameters and spectral index is significantly higher than that constructed by original spectral wavelength.The results of neural network chlorophyll estimation model and accuracy evaluation show a relatively high level,which is more stable than the traditional regression model and the accuracy has been greatly improved.Based on the spectral index,the model was constructed with the first derivative of the spectral index in the early flowering stage and the reflectance spectral index in the seedling stage,with the determination coefficient R~2 of 0.924 and root mean square error RMSE of 1.112.Based on the spectral characteristic parameters,the model constructed by the spectral characteristic parameters in the early flowering stage has the highest accuracy,the determination coefficient R~2 is 0.932,and the root mean square error RMSE is 1.142.By constructing BP neural network model in seedling stage,bolting stage and early flowering stage,the chlorophyll of rape can be retrieved better.It provides a theoretical basis for monitoring chlorophyll content in rape by hyperspectral technology.