模型树和支持向量回归在高光谱遥感中的应用

利用连续植被辐射传输模型（SAIL模型）模拟生成小麦冠层反射率数据，比较了数据挖掘中的新方法模型树、支持向量回归与传统的逐步回归用于高光谱数据定量预测的效果．结果表明：支持向量回归和模型树的预测精度都要远远高于逐步回归，在训练样本数量减少时，它们的优势更加明显；支持向量回归在高维空间中有很好的泛化能力，其预测精度随维数的增加呈持续上升的趋势；模型树的预测精度在低维条件下和支持向量回归相仿，但在高维条件下则比支持向量回归差很多，通过逐步回归的特征选择预处理，可以提高模型树的预测精度，缩小其与支持向量回归之间的差距．

Application of Model Tree and Support Vector Regression in the Hyperspectral Remote Sensing

The prediction performances of model tree (MT), support vector regression (SVR) and stepwise regression (SR) were compared using the wheat canopy hyperspectral reflectance data which were simulated by a radiative transfer model of SAIL (scattering by arbitrarily inclined leaves). The results show that the prediction accuracy of SVR and MT is much higher than that of SR. When the training sample size is reduced, the superiority of SVR and MT is more significant. The prediction accuracy of SVR improves continually when dimensionality increases because of its excellent generalization ability in high dimensional space. The prediction accuracy of MT is close to that of SVR when dimensionality is low, but much lower than that of SVR when dimensionality is high. Feature pre-selection through stepwise regression can enhance the prediction accuracy of MT and reduce the accuracy gap between MT and SVR.