Mixed Pixel Interpretation And Growth Monitoring Of Maize And Soybean Strip Intercropping Based On UAV Images

The maize and soybean strip compound planting system can improve the multiple cropping index and land utilization rate under the limited arable land area,which is of great significance for alleviating the contradiction between the supply and demand of maize and soybean in our country.With the development of smart agriculture,UAV-based remote sensing monitoring technology has been able to quickly and non-destructively obtain crop growth and development status at the field scale to diagnose it.and adjust for precise field management and cultivation measures.Provide important technical support and theoretical basis.It provides important technical support and theoretical basis for precise field management and adjustment of cultivation measures.In the context of the vigorous promotion and application of maize and soybean strip compound planting technology,it is of great significance to carry out multi-crop growth monitoring research based on low-cost RGB-UAV.In this study,Zhongyu 3,Denghai 605 and Qihuang 34 were selected as the research objects.Based on the maize-soybean intercropping model,three different nitrogen application levels（pure nitrogen）,namely N1:0kg/hm²;N2:120kg/hm²;N3:240kg/hm².Canopy RGB images of the two crops were acquired at the critical reproductive stage of maize,and agronomic parameters and yield were determined simultaneously.The canopy information classification model under the maize-soybean intercropping mode was established to obtain accurate canopy information of the two crops by using algorithms like random forest based on the canopy RGB features and DEM features of the two crops.Through quantitative analysis of visible light vegetation index,characteristic free combination index and agronomic parameters,linear and nonlinear regression methods were used to establish estimation models of relevant agronomic parameters and yield and perform visual inversion.This provided a theoretical basis for the rapid and accurate monitoring of multi-crop growth under the maize-soybean intercropping mode.By quantitatively analyzing the canopy RGB characteristics and DEM characteristic of maize at jointing,tasseling and filling stages and soybean at the same stage,the maximum likelihood method,support vector machine and random forest algorithm were used to construct the classification model of maize and soybean canopy.Among them,the random forest classification model performed the best in each stage,and RGB+DEM was selected as the best feature combination in the jointing stage and filling stage,and the validation accuracy and Kappa coefficient were 97.48%and 0.95,97.30%and 0.95,respectively;and RGB was selected as the best feature combination in the tasseling stage,and the verification accuracy and Kappa coefficient were 92.64%and 0.85,which laid the foundation for the accurate estimation and visual inversion of maize and soybean agronomic parameters and yield.Under different nitrogen application treatments,the plant height,biomass,leaf area index and canopy chlorophyll density of maize all increased with the increase of nitrogen application level.Among them,the plant height and LAI of Denghai 605increased the most at the jointing stage,which were 16.2%and 42.2%,respectively;the biomass and CHD increased the most at the filling stage,which were 55.4%and98.1%,respectively.The plant height and biomass of Zhongyu No.3 increased by 24.3%and 84.5%at the jointing stage,while the LAI and CHD reached the maximum at the filling stage,with an increase of 67.0%and 204.3%.The LAI of the two maize varieties was significantly different at the tasseling stage.Under different nitrogen treatments,the yield of Zhongyu No.3 increased 93.9%.For soybean,the biomass,LAI and CHD of Denghai 605 interrow soybean declined the most,which were 16.1%,19.5%and14.9%,respectively.The soybean yield of Denghai 605 with N1 treatment has a maximum value of 1.59 t/ha.The correlation analysis found that,except the agronomic parameters at the jointing and filling stage of maize had the significant correlation with the normalized red light index r,the other stage agronomic parameters had the best correlation with the texture free combination index,and the best correlation coefficient of each agronomic parameter was basically above 0.8.The linear and nonlinear estimation models of various agronomic parameters were established by using the index with the best correlation.The test effect of the quadratic function model established by r for LAI and CHD at the filling stage of maize was better,and the R² and RMSE of the validation model were 0.789 and 0.412,0.689 and 0.542,respectively.The test effect of the quadratic function model established by t RVI_1,20 for maize plant height at the tasseling stage was better（R²=0.679,RMSE=12.652）.The quadratic function models of soybean plant height,LAI and CHD were better based on t DVI_7,21,t RVI_18,6 and t NDVI_8,27 in the whole growth stage,and the validation model R² and RMSE were 0.760 and 8.867,0.617 and 0.284,0.602 and 0.279,respectively.In yield prediction models,the tasseling t DVI_8,16 and the full pod t RVI_24,2can be used to estimate the maize and soybean yield better,and the R² and RMSE were 0.709 and 1.461,0.713 and 0.140,respectively.Based on the accurate classification model of maize and soybean canopy,the optimal estimation models of various agronomic parameters and yield of maize and soybean were integrated into the classification result image for visual inversion,and the spatial distribution map of the agronomic parameters and yield predictions of the two crops was established.Through comparison,the spatial distribution of each parameter is basically consistent with the real growth situation.Rapid,non-destructive monitoring and prediction of growth levels and yields of two crops in a maize-soybean intercropping system can be achieved at the field scale using RGB-UAV.In order to further improve the prediction ability,it is necessary to consider different varieties,cultivation measures,ecological sites and other factors for comprehensive analysis in future research to improve the universality and stability of the model.