中国科学院遥感与数字地球研究所-高光谱遥感应用技术研究室

【作者】             焦全军                                                                                                                                              

【导师】             童庆禧；张兵；郑兰芬

【 学位年度 】    2008

【论文级别】      博士    

【关键词】          遥感，植被信息提取，光谱指数，成像过程，模型优化 

【Key words】   Remote sensing; vegetation information extraction; spectral index; imaging process; model optimization

【中文摘要】   

植被信息提取是遥感应用的一个重要内容，通过遥感手段提取的植被信息已被应用到生态模型运转、植被覆盖类型填图、植被生长状况诊断、作物估产、灾害监测等多个领域。植被信息提取模型构建大多借助于遥感地面实验，为了提高模型的信息提取精度和普适性，仍需更多的地面实验数据进行植被信息提取模型的检验和完善。新的遥感器不断出现，针对新遥感器的植被信息提取建模也需要借鉴已有的地面实验数据和植被信息提取模型。遥感数据的多元化一定程度上加大了植被遥感信息提取模型推广和改进的难度，而遥感数据特征与遥感成像过程的特点是分不开的。因此，研究遥感的成像过程对植被信息提取的影响，是发掘遥感图像特点、优化植被信息提取模型的重要途径。

        本文将光学遥感成像模型与植被遥感信息提取模型相结合，在植被信息提取常规建模、特定遥感成像参数下的遥感数据模拟、成像因子对植被信息提取影响分析、植被信息提取模型优化等方面开展了系统的研究工作。

        论文首先分析和总结了目前常用的植被遥感信息提取建模方法，针对植被理化参量之间以及光谱波段之间的强相关的问题，引入偏最小二乘方法（PLS）和基于PROSAIL模型的植被参量反演的统计方法，构建了普适性较好的植被信息提取模型，并利用实测数据进行模型精度验证。在此基础上，研究了光谱指数参与树种分类中的特征提取和特征选择，并将特征提取和选择结果用于森林树种分类。光谱指数能够更有效地表征植被结构和理化属性信息，提高了树种分类精度。

        通过分析遥感成像的主要过程，完成了遥感成像过程数学模型建模，论文基于遥感成像过程模型实现了特定成像条件下的遥感数据模拟。利用特定成像参数的遥感模拟数据以及对应的植被属性，分析了空间尺度、光谱采样、成像几何等成像关键因子对植被参量反演和地物分类的影响。

        在对成像因子的植被信息提取影响的分析基础上，研究了不同波段宽度下的遥感特征之间的相关关系，建立了不同波段宽度的遥感特征之间的转换关系，并将转换关系耦合到已有的植被参量反演模型中，获得了适应新遥感数据、鲁棒性更强的植被参量反演优化模型；论文研究了成像因子修正对植被参量提取的影响，提出了基于遥感尺度因子的多参量协同反演模型。通过考虑像元内的植被混合状况，构建基于植被盖度的LAI、叶绿素含量协同反演模型；提出了使用以成像几何参数匹配植被参量反演模型的方法，建立了环境与减灾小卫星HJ-1A多光谱相机的成像几何模型，构建了成像几何参数匹配的植被参量反演模型查找表，并利用地面成像光谱仪实验数据、高光谱卫星CHRIS图像数据进行了模型验证和推广。

        本篇博士论文力图在降低植被遥感不确定性、提高植被信息提取精度方面做出尝试。其研究成果可以为新遥感器的植被信息提取模型建模提供有效思路和方法，同时有助于扩大已有遥感应用模型的数据适用范围和提高已有模型、地面数据的应用效率。

【Abstract】 

Vegetation information extraction is an important content of remote sensing application. Vegetation information extracted through remote sensing image has been used in the application of many fields, such as running bio-ecological model, vegetation sort mapping, monitoring disaster, and so on. Vegetation information extraction model is usually built by remote sensing ground synchronous experiment, in order to improve the veracity and applicability, more ground experiment data are needed to check out and adjust some existing models of vegetation information extraction. Because of new remote sensors’ continual invention, the vegetation information extraction model prepared with new remote sensors should also draw the existing ground experiment data and the existing model. The diversification of remote sensing data enlardges difficulty in model generalizing and model modificatio, while, the features of remote sensing data are inseparable to those of remote sensor’s imaging process. Then, study on the influence of remote sensor imaging process to vegetation information extraction is an important approach of mining remote sensing image‘s features of and optimizing the existing remote sensing model.

      This paper couples remote sensing imaging process model with vegetation information extraction model and studies systemically on the vegetation information extraction general modeling, data simulation with the remote sensing image features, analyzing the influence of imaging factor to vegetation information extraction and optimizing vegetation information extraction model.

      At first, this paper analyzes and summarizes the methods commonly used of vegetation information extraction modeling. Because of the problem that there is strong correlation between vegetation parameters and also between vegetation remote sensing signals, Partial Least-squares Regression (PLS) method and the PROSAIL-based inversion algorithm are introduced to retrieve vegetation biophysical and biochemistrical parameters. Vegetation spectral freatures extraction including spectral indices and feature selection method are also used in forest species classification. Those spectral indices can represent forest canopy structure characteristics and leaf biochemical characteristics and the classification accuracies of forest spieces in this paper are indeed improved.

      Through analyzing imaging processes of optical remote sensing, mathematical modeling of those processes is accomplished. Based on the imaging process model, this paper accomplishes remote sensing data under some imaging conditions. Utilizing the simulated spectra data and the corresponding vegetation attribute, we analyze the influences of some key imaging factors including spatial scales, spectrum sampling and image geometry to vegetation parameter prediction and land cover classification.

      On the basis of analyzing the imaging factors’ influence to vegetation information extraction，the conversion relation of the remote sensing features between different band widths is set up. Coupling the conversion relation with the existing model of retieving vegetation parameters, vegetation parameter predicted models can be adjusted and those adjusted models are fit to new remote sensing data. This paper studies the influences of imaging factors to the vegetation parameter extraction and brings forward the Multi-Parameters Synergy Inversion model based on the remote sensing scale factor. Through thinking about the mixed pixel involving vgeation and non-vegetation subpixel and setting up the LAI, chlorophyll content synergy inversion model based on the vegetation fraction, we bring forward the method of matching vegetation parameter inversion model according to image geometry parameter. The image geometrical model of HJ-1A multi-spectral camera is build. Model validation is carried out with ground experiment data and CHRIS hyperspectral images.

      This papermakes great efforts in reducing the indetermination of vegetation remote sensing and improving the precision of vegetation classification. The study production might supply the vegetation information extraction model modeling of new remote sensor with available thoughtway and method, at the same time, it is also redound to enlarge the application scope of the existing remote sensing model and to improve the application efficiency of the existing model and ground data.