Hyperspectral Characteristics And Growth Monitoring Of Rice Canopy Under The Changing Ratio Of Scattered Radiation

In recent years,the weakening of solar radiation and the increasing proportion of scattered radiation in China have had an important impact on rice.This paper combines the hyperspectral remote sensing technology to study the characteristics of the hyperspectral reflectance curve of the rice canopy under different scattered radiation ratios,respectively analyzes the band reflectance,many vegetation indices and based on the ensemble empirical mode decomposition(EEMD),fast Fourier transform(FFT)and the relationship between the absorption valley parameters of the envelope elimination method and rice LAI,dry matter,yield.Principal component analysis(PCA)and least squares method are used to extract the spectral parameters that are significantly related to rice physiology.Based on these spectral parameters,three types of machine learning methods:ridge regression,decision tree regression,and support vector machine regression are used to establish predictive model of LAI,dry matter,yield.This article provides technical support for the subsequent hyperspectral research and a certain theoretical basis for the subsequent research on the effect of the scattered radiation ratio change on rice.The main conclusions are as follows:1)The spectral reflectance of the original rice canopy is in the visible light range(400?650nm)T3<T2<T1,and in the near-infrared part(750?1350nm)T1<T2<T3,that is,the spectral reflectance follows the scattered radiation in the visible light the ratio increases and decreases.In the near-infrared part,it increased with the increase of the scattered radiation ratio,and the reflectivity difference of each treatment in the range of 750?950nm was the most significant and decreased with the delay of the growth period.The"red edge"(680?760nm)first derivative spectral reflectance increases with the increase in the proportion of scattered radiation.In the red edge parameter,the red edge position ? moves to the long wave direction(red shift)as the proportion of scattered radiation increases.The maximum slope of the red edgeincreases as the proportion of scattered radiation increases.The area of the red border increases significantly as the proportion of scattered radiation increases.The red valley widthbased on the IG red edge model does not change significantly with the proportion of scattered radiation.2)The experimental results show that the rice LAI increases significantly with the increase in the proportion of scattered radiation.LAI has the highest and extremely significant correlation with the spectral reflectance of 771?872nm(R=0.54).In the vegetation index,MSAV,TVI-3,SAVI,VARIgreen,OSAVI and EVI are extremely significantly correlated with LAI.The difference vegetation index(DVI)in the two regions of 500-600nm and 750-1150nm,730-750nm and 750-1150nm is extremely significantly correlated with LAI.The Normalized Vegetation Index(NDVI)in the 540-570nm and 740-1200nm regions is extremely significantly correlated with LAI.In the absorption valley parameters,A_553-673 and A_673-788 have significant positive correlations with LAI(0.56,0.55),D₉₆₉ and_D(1192) have significant negative correlations with LAI(-0.69,-0.61).The linear model shows that D₉₆₉ is best.The LAI prediction models show that theof ridge regression has the best simulation effect in0.5?10(RMSE=0.182,R²=0.7).the 3 maximum depth decision tree regression model is best(RMSE=0.177,R²=0.726).The model that uses Gaussian kernel function in support vector regression works best(RMSE=0.169,R²=0.751).3)The dry matter of rice above ground decreases significantly with the decrease of solar radiation but does not change significantly with the proportion of scattered radiation.The dry matter is significantly related to the spectral reflectance of 765?813nm and 823?864nm.The vegetation indices VARIgreen and MSAVI are significantly correlated with the dry matter.The DVI in the rectangular regions of 750-950nm and 750-1350nm was significantly correlated with it.The discrete NDVI centered on(450nm,450nm),(500nm,500nm),(700nm,700nm),(760nm,760nm)is significantly correlated with it.In the absorption valley parameters,_553-673 has a significant negative correlation with the dry matter(-0.65).Prediction models show that the model performs best when the regularization parameter?is 0.8?1 in ridge regression(R² is 0.6,and RMSE is 0.26kg/m²).When the maximum depth of the decision tree is 3 layers,the RMSE is 0.235kg/m²,and R² is 0.663.The model using Gaussian kernel function in support vector regression has the best simulation effect,with RMSE of 0.201 kg/m² and R²of 0.778.4)The rice yield of different treatments is significantly different.The rice yield decreases significantly with the decrease of light transmittance and increases significantly with the increase of the scattered radiation ratio.The spectral parameters related to rice are different in different growth periods.When flowering 14 days,the spectral reflectance of 613?712nm,vegetation index m ND705,350?540nm and 560?740nm rectangular area and diagonal DVI,540?560nm and 550?650nm,690?740nm and 750?1350nm area The NDVI and the absorption parameter D₆₇₃ are significantly correlated with rice yield.When flowering 21 days,the original spectral reflectance of 518?617nm,redeage,MCARI,350?550nm and 400?600nm,540?730nm and 550?700nm in the triangle area of DVI,546?555nm and 750?1350nm,745?760nm and 750?1350nm area of NDVI was significantly correlated with rice yield.At 28days after flowering,the spectral reflectance of 521?614nm,MCARI,DVI in the regions of350?540nm and 400?600nm,530?700nm and 550?700nm,NDVI in the regions of360?530nm and 400?700nm,and_969-1072 is significantly correlated with rice yield.The14-day flowering prediction model shows that the ridge regression with?between 0.1 and 0.5has the best effect(RMSE=0.04kg/m²,R²=0.58),and the decision tree with a maximum depth of 3 has the best regression effect(RMSE=0.035kg/m²,R²=0.82),the Gaussian kernel function support vector regression model has the best effect(RMSE=0.037kg/m²,R²=0.72).The prediction model shows that the regression effect of?is the best in the range of 0.1?0.5(RMSE=0.038kg/m^2,R²=0.67)at the 21st day of flowering,and the regression effect of the decision tree with the maximum depth of 3 is the best(RMSE=0.032kg/m²,R²=0.83),the Gaussian kernel function support vector regression model has the best effect(RMSE=0.026kg/m²,R²=0.87).At 28 days after flowering,the prediction model shows that the regression effect of?is the best in the range of 0.1?1(RMSE is 0.038kg/m²,R² is 0.72),and the regression effect of the decision tree with the maximum depth of 2 is better(RMSE is0.033kg/m²,R² is 0.75),the Gaussian kernel function support vector regression model has the best effect(RMSE is 0.032kg/m²,R²is 0.80).The support vector regression model has the best effect during the same reproductive period;the simulation effect of the ridge regression model gradually increases with birth.Decision tree and support vector regression models work best at21 days of flowering.The Gaussian kernel function support vector regression model for 21 days of flowering is the best for all yield prediction models.