基于神经网络的水稻双向反射模型研究

Study on Bi-directional Reflectance Model of Rice Using a Artificial Neural Network

水稻的双向反射特性与其冠层结构、各组分光谱性质以及入射光方向和观测方向之间存在着密切的、非线性的相关关系。运用人工神经网络技术 ,采用水稻田间实测数据 ,对这种关系进行拟合 ,所建立的水稻双向反射BP前向和反演模型 ,都达到了较高的拟合精度。研究表明 :采用人工神经网络技术计算水稻双向反射率和成批反演冠层结构参数是可行的。对所建模型做进一步的改进 ,可模拟水稻双向反射的实际过程 ,进而监测作物长势。

There are nonlinear relations among the bi directional reflectance of rice,its canopy architecture parameters, the spectral characteristics of the different components of rice, and the illumination and viewing geometry. This article explores the use of artificial neural network for both forward and inverse bi directional reflectance modeling of rice based on the data measured in Zhejiang University(Hangzhou,China) for field experiments from 1999 to 2000. The assumption here is that the bi directional reflectance of a rice canopy is the function of the geometry of its constituent elements,the spatial distribution,spectral features of the elements, and the illumination and viewing geometry. This implies that the bi directional reflectance of the canopy is particular sensitive to the canopy's structural parameters, the spectral characteristic of foliage,and the illumination and viewing direction. It also implies that canopies with different parameters will exhibit different bi directional reflectance. On the basis of these analysis,we decided to have 10 input parameters:model inclination angle of the canopy( θ ),eccentricity( D ),reflectance of foliage( R ),transmittance( T 1 ),sun zenith angle( θ s ),soil reflectance( sbrf ),the ratio of mean length to canopy height( L 1)and the ratio of width to length( P ),leaf area index( LAI ),diffuse to total incident radiation( Q 1). There are 17 output parameters:bi directional reflectance of the canopy in the principal plane,from -60° in the forescattering direction to +60° in the backscattering direction at increments of 7 5° in forward BP model. On the other hand,there are 3 output parameters:leaf area index,the ratio of mean length to canopy height and the ratio of width to length,and other 22 parameters mentioned above except eccentricity and model inclination angle of the canopy are input parameters in inverse BP model. After model development,the neural network model is tested against the independent data set. The Root mean square error between the bi directional reflectance of rice measured and simulated varies from 4 53×10 -6 to 3 67×10 -3 . The inversion model of artificial neural network is able to inverse the rice canopy structural parameters with 81 8% accuracy. The results of both forward and inverse modeling suggest that the model of artificial neural network is of high precise to simulate the relations of the bi directional reflectance of rice and its canopy structural parameters. Further research is needed to monitor the rice growth by the neural network model.