Monitoring Rice Growth Parameters And Estimating Yield Based On UAV Images

Food security has become one of the focal issues in the world.In recent years,hybrid rice planting area accounts for 50%to 60%of the total rice growing areas in China,yield and quality of hybrid rice have made outstanding contributions to ensuring national food security.Rapid and accurate acquisition of key growth parameters of hybrid rice is of great significance for precise breeding,scientific fertilization and efficient management at field scale.Unmanned Aerial Vehicle(UAV)remote sensing can obtain data rapidly,accurately and non-destructively,which provides great convenience for monitoring crop growth parameters in large area.At present,there are three problems to be solved in the research of monitoring rice growth parameters and yield estimation based on UAV images:(1)Heading stage is the most critical stage for hybrid rice pollination.How to use UAV remote sensing technology to accurately monitor rice heading phenology at field scale;(2)How to reduce the interference of rice panicle in growth parameters estimation,and improve the estimation accuracy of growth parameters in the whole growth period under the normal environment of field;(3)There are few studies on yield estimation of different hybrid rice varieties.In addition,when estimating yield,there is a lack of comprehensive consideration of growth stages and growth parameters of different varieties of rice,and the importance of growth parameters are also rarely evaluated.In order to solve these scientific problems,this research takes the widely cultivated hybrid rice as the research object.UAVs equipped with visible light sensor and multispectral sensor are applied to obtain time series images of rice canopy.Combining with field survey of rice phenology,growth parameters and yield,spectral and geometric features of time series images were extracted for monitoring heading stage at field scale.A new parameter was proposed to improve the estimation accuracy of growth parameters of different rice varieties in the whole growth period.Growth parameters of different stages were used to estimate rice yield.The prospective results would provide reliable data and technical support for hybrid rice breeding,cultivation and field management.The main research issues and conclusions are as follows.(1)In order to monitor the initial heading stage(IHS)of rice.Three kinds of functions(double logistic function,asymmetric Gaussian function and symmetric Gaussian function)were applied to reconstruct CIrededge for reducing the abnormal points,and the maximum curvature(MC)of the reconstructed CIrededge was extracted to monitor IHS.The results indicated that the MC of CIrededge reconstructed by symmetric Gaussian function had stronger robustness and higher accuracy(R~2=0.86,RMSE=3.31 days,RPD=2.68)for monitoring IHS of different rice varieties planted in different regions and different years.(2)In order to estimate heading ratio of different rice varieties.The high spatial and temporal resolution images were obtained by UAV equipped with RGB camera,then serval RGB vegetation indices were extracted from images to estimate rice heading ratio.The results showed that R and VARI performed well in the estimation of rice heading ratio,R~2 was 0.82 and 0.81,RMSE was 13.12%and 13.47%,RPD was 2.38 and 2.30,respectively.These results indicated that R and VARI were sensitive to rice panicle,and could be used to estimate heading ratio of different rice varieties.(3)In order to monitor the complete heading stage(CHS)of rice.The rice canopy height extracted from RGB images,proved to have good linear correlation with the actual measured plant height,R~2 was 0.92,RMSE was 0.06 metrer,RPD was3.64.The rice canopy height could accurately present the changes of plant height at field scale.By analyzing the change of rice canopy height,the maximum value of rice canopy height was used to monitor CHS,the model accuracy was R~2=0.96,RMSE=1.65 days,RPD=5.46.The maximum value of rice canopy height could accurately monitor CHS of different rice varieties.Three patterns of rice canopy height change from heading stage to maturity stage were also summarized.(4)In order to monitor rice growth parameters at different growth stages.The results indicated that the commonly used vegetation indices were applied to monitor rice growth parameters in the whole growth period and after IHS,with very low accuracy,but performed relatively good before IHS.Therefore,it was necessary to use vegetation index to monitor growth parameters at two growth stages(tillering stage to IHS,IHS to maturity stage).The differences of growth stages,canopy structure and leaf pigment content of different rice varieties were taken into account,learning from the principle of crop photosynthesis,a new parameter,the sum of cumulative vegetation index(SC?VI)was proposed to reduce the interference of rice panicle on growth parameters estimation.The results indicated that SC?VI improved the estimation accuracy of rice growth parameters(aboveground biomass,plant nitrogen accumulation,plant carbon accumulation)in the whole growth period.(5)The growth parameters at five growth stages of different rice varieties were applied for rice yield estimation using partial least square(PLS).The results showed that single growth parameter models performed poorly,and it was difficult to estimate the yield of different varieties of rice accurately.The accuracy of multi growth parameter models were higher than that of single growth parameter models,but the multi parameter models were still lack of robustness and accuracy.Growth parameters proved to have weak correlation with yield.