Rice Yield Estimation Methods Based On Unmanned Aerial Vehicle (UAV) Imaging Hyperspectral Remote Sensing Data

As the most important food crop in China,the importance of rice is self-evident.The timely and accurate monitoring of rice output is crucial to China’s food security.Traditional rice yield estimation research is mainly carried out with a series of remote sensing satellites,but due to the constraints of a series of factors such as temporal and spatial resolution,weather conditions and cloud cover,it is difficult to achieve a more accurate and timely yield estimation effect.With the rapid development in the field of unmanned aerial vehicles(UAVs)in recent years,the high convenience,high flexibility,high stability and high spatial and temporal resolution of the unmanned aerial vehicle platform have made it possible to accurately estimate the yield of rice in near-ground areas.In this paper,the three-year experimental hyperspectral remote sensing data of rice were acquired by the UAV equipped with imaging hyperspectral camera.On the basis of the traditional yield estimation method,the advantage of the hyperspectral image is fully utilized.Refined growth stage length information,precise stage information and rice spatial texture information have further improved the accuracy of yield estimation,and finally reached a more accurate and more generalized and more generalized result of yield estimation.The specific research results are as follows:(1)This article first uses the advantage of the rich spectral information of the imaging hyperspectral camera to establish a more accurate rice multi-period vegetation index estimation model based on the empirical statistical model estimation method.The best yield estimation model is based on the empirical yield estimation model of NDVI [824,728] at booting stage,NDVI [784,740] at heading stage,and NDVI [776,724] at ripening stage.The estimated yield R-square reaches 0.75,and the average absolute percentage error MAPE reaches 5.26%.Subsequent research will use this as a reference and consider introducing more information to further improve the accuracy of production estimation.(2)The longer the period of rice growth,the more photosynthetic products,the greater the yield.To this end,by introducing the effective growth length variable into the yield estimation model,this study improves the accuracy of yield estimation of the best multi-stage vegetation index,alleviates the systematic errors of rice growth inconsistency caused by different varieties and different years,and improves the interannual robustness of the estimated production model,the estimated production result R-square increased from 0.75 to 0.78,and the MAPE decreased from 5.26% to 4.84%.The consideration of the growth stage length information will improve the regional applicability of the yield estimation model and reduce the impact of rice varieties and growth conditions in different regions on the growth period of rice.(3)The traditional production estimation method uses the vegetation index of different growth stage to establish the production estimation model,but the growth stage has a certain length.The selection of remote sensing images in the early and late growth periods will cause greater uncertainty in the production estimation model.Therefore,by fitting the full growth period curve of the vegetation index to determine the precise growth stage of yield estimation,a rice yield estimation method based on the precise growth stage vegetation index was established to improve the accuracy of the yield estimation model.The R-square was increased from 0.75 to 0.76.MAPE was reduced from 5.26% to 5.06%.The rice yield estimation method based on the precise growth stage vegetation index can alleviate the systematic errors in the inconsistency of the actual growth period of rice caused by the differences in rice varieties and growth conditions.(4)The spatial resolution of the imaging hyperspectral data reaches the centimeter level and has rich spatial texture information.Therefore,this study introduced a spatial texture index that can characterize the internal growth difference of the field on the basis of the multi-stage vegetation index rice estimation.Through the coupled yield estimation study of vegetation index and texture index,the texture vegetation index coupled yield estimation model at booting stage and heading stage has achieved a higher accuracy of yield estimation than the multi-stage vegetation index,and the R-square increased from 0.75 to 0.76.MAPE was reduced from 5.26% to 5.07%,which achieved the early rapid yield estimation of rice;at the same time,compare with the multi-stage vegetation index estimation model,the introduction of the texture index made the R-square increased from 0.75 to 0.80 and MAPE reduced from 5.26% to 4.66%,which further proves that the introduction of texture index makes up for the lack of spatial information of vegetation index and greatly improves the accuracy of the vegetation index estimation model.