Uav Remote Sensing Diagnosis Models Of Water Stress Of Winter Wheat Based On Canopy Spectrum Information

Monitoring of crop water stress is the basic condition to ensure the normal growth and development of crops,and accurately detection of crop water stress is the basis to achieving precise irrigation.UAV multispectral remote sensing has great advantages in diagnosis of crop water stress by obtained high resolution spectrum information,but the soil background on crop water stress has a great impact on the accuracy of the diagnosis model,at the same time the canopy spectral information such as image texture and vegetation indices（VIs）in the diagnosis of crop water stress performance needs further research.Therefore,this study took the winter wheat with 4 different water treatments as the research object,and uses the DJI M600 drone equipped with a Micro-MCA multispectral camera.Multispectral images at key growth stages of winter wheat were obtained at 9:00,11:00,13:00,15:00 and 17:00,respectively,and stomatal conductance（Gs）and soil water content（SWC）at different depths of winter wheat were collected simultaneously.UAV multispectral images were interpreted to accurately obtain winter wheat canopy spectrum information（canopy reflectance,vegetation indices and image texture）,analyze the change characteristics of winter wheat canopy spectrum before and after removing the soil background,and the influences of soil background on the inversion of SWC by vegetation indices（VIs）.Combining image texture and VIs to constructed water stress diagnosis models based on Gs,and finally explored the best coupling model of different screening methods and machine learning algorithms in inverting Gs to improve the accuracy of winter wheat water stress diagnosis.The main results obtained are as follows:（1）An improved vegetation index threshold method was proposed to eliminate the soil background,and we revealed the influence of the soil background on the vegetation index inversion of soil moisture content.The improved vegetation index threshold method could effectively remove the soil background from the multispectral images,which improved the extraction accuracy of spectral information of winter wheat,and the vegetation index-RDVI had the highest accuracy,with the overall accuracy of more than91.32%.Soil background has a greater effect on the reflectance of the near-infrared band,followed by the red-edge band,and a smaller effect on the reflectance of the visible band.The VIs and SWC before and after removing the soil background are linearly related,but removing the soil background significantly improved the accuracy of inverting the SWC.Among them,NGRDI has the highest accuracy in inverting SWC in the root zone of winter wheat at a depth of 10～20 cm.The calibration set R_c²and RMSE_cwere 0.739 and 2.0%,and the verification set R_v²and RMSE_vwere 0.787 and 2.1%.（2）Gs estimation models combining UAV multi-spectral image texture and VIs were constructed to improve the accuracy of water stress diagnosis models of the winter wheat.The image texture obtained from the high-resolution multispectral images had a high correlation with Gs,and the image texture（VAR,HOM,CON,DIS,ENT and SEC）at550nm had the most significant correlation.The higher the ground resolution,the lower the correlation between the Gs and the image texture,the vegetation indices,respectively.The image texture with a ground resolution of 0.008m combined with VIs and Gs had the highest correlation,and combining image texture and VIs can significantly improved the estimation accuracy of winter wheat Gs.Among the three estimation models,the BPNN model constructed by combining the image texture and VIs（MEA,VAR,ENT,DWSI and EXG）had the best estimation performance,and an accurate estimation could even be achieved at a lower Gs value.The_vR²,RMSEv and MAEv of the validation set were 0.834,0.018 mol H₂Om^-2s^-1and 0.014 mol H₂Om^-2s^-1,respectively,and the R_v²/R_c²were 0.928,indicating that the model had good robustness.（3）Coupling models of different screening methods and machine learning algorithms were constructed,and a more robust water stress diagnosis model of winter wheat was obtained.The screening methods could effectively screen out the VIs that has significant correlation with the Gs.The Gs of winter wheat at different stages were estimated and compared,through the coupling models of different screening methods and machine learning algorithms.Results showed that the accuracy of the coupling models between the IVSO,VISSA,and GA screening methods and the machine learning algorithms were poor,while the coupling model between the full subset screening method and the machine learning algorithms has the highest accuracy,the determination coefficients of the modeling and verification sets are both above 0.676.Comparing the coupling accuracy of the full subset selection method and the BPNN,ELM and Cubist in different growth stages of winter wheat.The R_v²/R_c²of the full subset and BPNN coupling model at jointing stage,flowering stage and full growth stage are 0.949,0.954 and 0.949,respectively,which are all very close to 1,indicating that the full subset and BPNN coupling model has the best robustness.