基于BP和GRNN神经网络的冬小麦冠层叶绿素高光谱反演建模研究

根据高光谱遥感获得的冬小麦冠层数据,把由逐步回归方法和基于遗传算法(GA)的广义回归神经网络(GRNN)筛选到的光谱参数作为网络输入,冠层叶绿素含量作为网络输出,采用线性逐步回归方法、反向传播神经网络(BPNN)和GRNN来构建反演模型,模拟结果表明,GRNN和BPNN的预测精度要高于逐步回归方法,其RMSE分别为0.36 mg/g、0.52 mg/g和0.98 mg/g。由于GRNN可应用于小样本问题的学习,比BPNN对叶绿素具有更好的预测和泛化能力。

Contrastive Analysis Based on Neural Network of Winter Wheat's Chlorophyll Concentration Inversion with Hyperspectral Data

Based on winter wheat' s canopy spectra data,the spectral parameters which are selected by linear regression method and Generalization Regression Neural Network(GRNN) method is as network input, canpoy chlorophyll Concentration as network output,we use three models,linear regression model, Back Propagation neural network(BPNN)and GRNN to inverse chlorophyll concentration. The result shows BPNN and GRNN have higher estimation precision than linear regression model, the RMSEP are 0.52、0.36 and 0.98 respectively. Due to adapation to small sample study and needless iterate repeatly, GRNN' s froecast abality,generalization and study speed is better than BPNN.