【摘要】 为了提高大面积冬小麦农田产量快速估算的准确率,选取Landsat 8 OLI卫星遥感数据,计算归一化植被指数NDVI、比值植被指数RVI、绿度植被指数GVI、增强植被指数EVI,分别建立4种植被指数组合与地面冬小麦实测产量的回归方程或神经网络和SVM模型。结果表明:单植被指数的非线性回归方程估产精度高于线性回归方程,冬小麦实测产量与遥感植被指数表现为非线性关系;线性回归方程估产时多植被指数组合精度高于单植被指数,多植被指数组合可实现信息互补,提高遥感估产精度;建立多植被指数组合与实测产量的非线性遥感估产模型时,SVM模型的均方根误差RMSE为339.6kg·hm^-2,决定系数R²为0.7852,估产精度高于BP神经网络模型、RBF神经网络模型,可应用于冬小麦遥感估产的快速、准确实现。更多还原

【Abstract】 In order to improve yield estimation accuracy of winter wheat in a large area of farmland,normalized difference vegetation index（NDVI）,ratio vegetation index（RVI）,greenness vegetation index（GVI） and Enhanced vegetation index（EVI） were got from Landsat 8 OLI data. Then regression equations,neural network and SVM models were set up with measured yield against the four vegetation indices combination. Result shows yield estimation accuracy with nonlinear regression equation is higher than that with linear regression equation for single vegetation index because the relationships between winter wheat measuring yield and vegetation index is non-linear. Yield estimation accuracy with multiply vegetation index combination is higher than that with single vegetation index for linear equation. Multiply vegetation index combination can complement each other and remote sensing yield estimation accuracy is improved. While remote sensing nonlinear yield estimation model is established with vegetation index combination input and measured yield output,yield estimation accuracy of SVM model with root mean square error（RMSE） of 339. 6kg·hm^-2and the coefficient of determination（R²） of 0.7852 is higher than BP and RBF neural network models. SVM modal can be used to estimate winter wheat yield accurately and quickly with remote sensing data.更多还原