基于多时相无人机遥感植被指数的夏玉米产量估算

为建立夏玉米无人机遥感估产模型,正确评价规模化农业经营管理和用水效率,以内蒙古自治区规模化种植的夏玉米为研究对象,设置了5个不同水分处理的实验区域,每个实验区域布置了3个样区,利用自主研发的多旋翼无人机多光谱遥感平台,对夏玉米进行多时相的遥感监测。采用牛顿-梯形积分和最小二乘法,构建了基于多种植被指数和多种生育期对应的夏玉米实测产量的6种线性模型,并采用阈值滤波法减少土壤噪声对模型精度的影响。结果显示,不同生育期的玉米估产模型精度存在显著差异。单一生育期中,精度由高到低依次为:抽雄期、吐丝期、蜡熟期、拔节期,最优植被指数为EVI2(决定系数R^2=0.72,均方根误差RMSE为485.46 kg/hm^2);多生育期的最优植被指数为GNDVI(R^2=0.89,RMSE为299.35 kg/hm^2)。经过土壤滤波后,拔节期和多生育期的R^2提升显著,其中基于植被指数GNDVI、MASVI2、EVI2的多生育期估产模型的决定系数R2提升到0.87以上。多生育期的无人机遥感估产优于单生育期,最优估产植被指数为GNDVI,阈值滤波法可以有效提升估产精度,优化后基于植被指数的无人机遥感估产模型可以快速有效诊断和评估作物长势和产量。

Summer Maize Yield Estimation Based on Vegetation Index Derived from Multi-temporal UAV Remote Sensing

The remote sensing of unmanned aerial vehicle (UAV) is accurate, flexible and fast. It is of great significance for large-scale agricultural management and water efficiency evaluation to establish yield estimation model of summer maize based on drone remote sensing. It was reported such an effort for summer maize in Inner Mongolia by using UAV multi-spectral platform. Six kinds of linear models for the measured summer maize yield maize as function of various vegetation indices derived at various growth stages were constructed by using Newton-trapezoidal integral and least squares method. And the threshold filtering method was used to reduce the influence of soil noise on the accuracy of the model. The results showed that there were significant differences in the accuracy of the models at different growth stages. In single growth period, the model precision from high to low was ordered as tasseling silking, wax maturity, and jointing, and the optimal vegetation index was EVI2 (R2=0.72,RMSE was 485.46kg/hm2).For most growth periods the superior vegetation index was GNDVI (R2=0.89,RMSE was 299.35kg/hm2). After soil filtration, the increase of R2 in jointing stage and multiple growth stages was significant. The correlation coefficient R2 was increased to above 0.87 for the multi fertility estimation model based on vegetation indices GNDVI, MASVI2 and EVI2. In summary, the UAV yield estimation model can quickly and effectively diagnose and assess crop growth and yield. The estimation accuracy of the model in multiple growth periods was better than that in a single one, and GNDVI was the optimal model parameter. The threshold filtering method can effectively improve the estimation accuracy.