Monitoring And Modeling Of Chlorophyll Content In Northeast Japonica Based On Hyperspectral Remote Sensing Based On Unmanned Aerial Vehicle

Chlorophyll is the most important pigment for photosynthesis of crops,and its content changes directly indicate the photosynthetic capacity and health status of the crops,and then determine the final yield.Timely and accurate,rapid crop chlorophyll content can be estimated to provide a valid data source for agricultural decision-making.In this paper,the northeast japonica rice was used as an example to study the chlorophyll content of japonica rice by hyperspectral remote sensing data of experimental area.The canopy spectral reflectance and the chlorophyll content of canopy leaf were studied in four different growth stages of japonica rice.The correlation between chlorophyll content and spectral reflectance of canopy leaf during growth and development was studied.At the same time,the sensitive band variables of chlorophyll content in canopy leaf were extracted to construct a single variable linear,logarithmic and cubic regression equation of chlorophyll content.The accuracy of the model with high coefficient of determination(R2)in the constructed model is evaluated to determine the optimal model for estimating the growth information of japonica rice.In order to further and accurately use the effective information of different spectral bands,And the four vegetation indices NDVI,DVI,RVI and TCARI were selected according to the previous research results.The linear and non-linearity of the chlorophyll content of canopy leaves at tillering stage,jointing and booting stage,heading grain filling stage and ripening stage was established.linear regression model,multiple linear regression model,BP neural network operator Model.The results of the study are as follows:(1)Under the condition of canopy spectrum of japonica rice,the curve characteristic of the spectrum is similar to that of japonica rice at different growth stages,and there are extremely obvious reflection and absorption valleys.The chlorophyll content of the canopy leaf was closely related to the characteristics of the spectral curve.The higher the chlorophyll content in the visible and near infrared regions,the higher the reflectance of the corresponding spectral curve.In the "red edge" In the case of first order differential spectroscopy,the spectral curves of the first order differential of the "red edge" of the japonica rice were similar to those of the different growth stages of the japonica rice and the chlorophyll content of different canopy leaves at the same growth period.Under the condition of chlorophyll content of different canopy leaves,with the increase of chlorophyll content in japonica rice leaves,the red edge was redistributed.(2)In the model of estimating the chlorophyll content of the canopy leaf using the original spectral reflectance and the first order differential spectral reflectance of the japonica canopy,The results of the inversion model of chlorophyll content in 452nm sensitive zone based on the original spectrum were the best,The establishment of the cubic curve model y = 26118.168x2-2722.032x + 122.252 the highest coefficient of determination for the R2 = 0.798,RMSE = 2.381,MAPE%=4.768;The effect of 715nm chlorophyll content inversion model based on first order differential spectroscopy was also the best,and the established cubic curve model y =15623618.191x3-979.064x + 85.840 has the highest coefficient R2 = 0.666,RMSE ?2.561,MAPE%= 3.798.(3)Using the single variable vegetation index to estimate the model,the establishment of univariate linear model,logarithmic model,cubic curve regression model,the estimation effect is not very good,in the japonica rice grain filling stage RVI established by the cubic curve Estimation coefficient y = 0.100x3-2.923x2 +25.778x-28.219 The highest coefficient of determination R2 = 0.560,RMSE = 2.144.MAPE%= 3.778.The multivariate linear regression model was established by using four different vegetation indices NDVI,DVI,RVI and TCARI as independent variables of multiple linear models,and the advantages of four different types of vegetation indices were fully utilized.The fitting effect was obvious Which is better than the one-dimensional regression model established during the same growth period.The four-line linear estimation of chlorophyll model established at heading stage is y?-9.055x1 + 8.869x2-0.075x3-6.469x4 + 51.566 The highest coefficient R2 = 0.790,RMSE = 1.773,MAPE%= 2.756.Using the BP neural network to establish the chlorophyll content estimation model,four different vegetation indices were BP neural network independent variables,one hidden layer(including 12 neuron nodes),the output variables were japonica rice leaf chlorophyll content data,The selection of the model 4-12-1,the study of chlorophyll content modeling,the full use of four different types of vegetation index advantage,making the estimated model of goodness and accuracy have been significantly improved,especially in the heading grouting BPE model has the highest R2 = 0.864,RMSE = 1.427 and MAPE%= 2.369,which proves that it is feasible to establish the estimation model of chlorophyll content vegetation index of canopy leaf using BP neural network.