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】             刘伟东                                                                                                                                             

【导师】             童庆禧；郑兰芬

【 学位年度 】    2002

【论文级别】      博士    

【关键词】          航空遥感，多光谱，数字相机，图像采集，辐射纠正，波段配准  

【Key words】   Airborne remote sensing; Multispectral; Digital camera; Imageacquire; radiometric correct; band register 

【中文摘要】

精准农业的发展迫切要求遥感技术提供给其快速与准确的地表信息。对于土壤来说，土壤湿度、土壤的有机质含量、土壤粗糙度、土壤质地等特性是精准农业中所需要的重要信息。高光谱遥感技术作为国际遥感科学的研究前沿和热点，除具备常规遥感对农作物监测的大面积、适时、非破坏性等优点，它能够克服常规遥感的不足，通过其精细光谱优势提高农业分类的精度和准确性，动态地监测和分析作物的健康状况与影响作物产量的环境因素，具有定量反演地物特性的潜力。高光谱遥感正是凭借其极高的光谱分辨率在农业土壤和植被特性的研究中表现出非凡的研究潜力。

        本论文围绕高光谱土壤信息的提取为中心，着重研究了土壤的光谱特性以及土壤特性的实验室反演研究。在论文的第一章主要介绍了高光谱遥感的概况与精准农业对高光谱遥感的需求，及高光谱遥感在精准农业中的广泛前景。在论文的第二章，主要介绍了实验室土壤光谱数据的采集与相应土壤特性信息的实验室测试方法，对土壤光谱数据进行预处理，对土壤的光谱进行特征参数提取与特征分析。第三章是本论文的重点，主要探讨了土壤的光谱特性，包括土壤光谱特征与土壤矿物成分的关系；土壤颜色与土壤反射率的关系及其土壤颜色的反演；土壤表面湿度与土壤反射率的关系及其土壤表面湿度的几种反射率反演方法的评价；土壤有机质含与土壤反射率的关系及其有机质含量的反演；氧化铁与土壤质地与土壤反射率的关系。论文的第四章主要对土壤的二向反射特性进行了研究，并且通过两个已有的模型对土壤的模型参数进行反演，探讨这些模型参数与土壤特性的关系。论文的第五章主要介绍了一些高光谱遥感图像的预处理的基本知识，并且对北京市精准农业示范基地的航空高光谱遥感图像进行了土壤的一些特性填图。论文的第六章，主要是对全文进行了概括总结，列举了作者的主要研究进展和在高光谱遥感图像中精确反演土壤特性参数的地难点及其改进之处。

        主要成果与结论如下：

1. 通过对大量的土壤实验室光谱进行特征分析除在明显的吸收峰外发现波长在400、600、800、1350、1800、2100、2400 nm 位置的控制点的连线与土壤的光谱曲线吻合很好这对于波段选择与土壤光谱数据压缩与波段选择都具有重要意义。

2. 由土壤光谱反射率与土壤孟塞尔颜色属性的相关分析知，在可见光光谱波段土壤光谱反射率与土壤色调和色度的相关不明显，而与土壤的明度值相关显著，能够通过土壤的反射率直接预测土壤的明度值。能够通过多元预测方程提高预测土壤明度值与土壤色度的精度，而多元方程对土壤的色调预测结果不好。通常，由于土壤的孟赛尔颜色属性是通过比对孟赛尔颜色卡获得，因此土壤孟赛尔颜色具有一定的主观性，而且孟赛尔颜色属性的量化比较粗略，这些都影响了通过反射率预测土壤颜色属性的精度。

3. 分析了土壤湿度与土壤光谱反射率的关系。在高土壤湿度水平时，土壤的光谱反射率随土壤湿度的增加而增加，在低土壤湿度水平时，土壤的光谱反射率随土壤湿度的增加而最小。这种增加或减小的幅度与土壤的类型有关，也与波长有关。通过分析土壤湿度与土壤相对反射率的关系，建立了利用相对反射率对土壤表面湿度的预测方法，我们发现土壤湿度水平不高时，使用近红外波段（如1998nm）预测土壤水分含量的效果好于使用可见光波段(如574nm)的效果，然而当土壤湿度水较高时，使用可见光波段对土壤水分的预测效果好于使用近红外部分的。

4. 通过使用相对反射率方法、一阶微分方法、差分方法对土壤表面湿度进行预测并且进行验证，结果表明，从总体上看，反射率倒数的对数的一阶微分与差分方法对土壤水分的预测能力较强。

5. 本文分析了土壤光谱反射率与土壤有机质含量的关系，建立了预测土壤有机质含量的模型，结果表明，由反射率倒数的对数的一阶微分建立的多元回归方程预测结果较好。

6. 本文通过已有的几何光学模型与辐射传输模型，对土壤的光谱二向反射特性，进行了研究，分析了不同土壤质地土壤的二向反射特性，相同土壤不同湿度的二向反射特性。

        通过实验室光谱所建立的土壤特性参数的反演模型，尝试了对高光谱遥感图像进行了土壤部分特性的填图，建立了较为精细的土壤参数空间分布图.

【Abstract】

The development of precision farming urgently requests that remote sensing technique offers to timely and accurate ground information. Soil water content, soil organic matter content, soil roughness and soil texture etc. are very important information in precision farming. As hot point and frontier in remote sensing, hyperspectral remote sensing technique not only has the advantages of traditional remote sensing that can timely and undisturbedly be used to detect large area crop, but also has special advantages. It has very high spectral resolution. More delicate spectral difference of crops can help us to precisely classify crops types and to monitor and analyze crops’ vigor and the environment factors that affect crops’ product. Hyperspectral remote sensing has great potential of quantitatively retrieving for objects’ characteristics.

      This thesis focuses on extracting soil information from hyperspectral data, and puts great emphasis on the study of retrieving soil characteristics from laboratory spectra. The first chapter mainly introduced the background of hyperspectral remote sensing and precision farming, and then, introduced the applications and perspectives of hyperspectral remote sensing in precision farming. In the second chapter, we primarily introduced the measurement of soil characteristics and soil spectra in laboratory, and analyzed feature of soil spectra. The third chapter is the most important part of this thesis. We discussed soil spectral properties. It included: 1) The relationship between soil minerals and soil spectral reflectance; 2) The relationship between soil color and soil spectra as well as inversion of soil color from spectral reflectance; 3) The relationship between soil surface moisture and soil spectral reflectance as well as evaluation of several inversion method of soil surface moisture from reflectance; 4) The relationship between soil organic matter and soil spectral reflectance as well as inversion of soil organic matter and soil spectral reflectance; 5) The relationship between soil texture, soil ferric oxide and soil spectral reflectance. The fourth part studied the BRDF properties of soil and with two models inverse models’ parameter of soils. The fifth part introduced the imaging mechanism of remote sensing and the spectra and radiance calibration methods for remote sensing images, as well as inversion of soil characteristics from airborne remote sensing image. The sixth chapter summarized the whole thesis and listed the achievement of this study, as same as, pointed out the difficulties in precise inversion of soil characteristics from hyperspectral image.Main development and conclusion as follows:

(1) By analyzing a large number of soil spectra, we found except at the obvious absorption position, the line of these points’ reflectance at the wavelengths 400, 600, 800, 1350, 1800, 2100 and 2400 nm are fitted well with spectral curve. This is useful for soil spectral data compressing and band selecting.

(2) From the correlation between soil spectral reflectance and soil color, we utilized regression model to forecast soil Munsel properties.

(3) The relationship between normalized soil reflectance and moisture was investigated.

For all the wavelengths and all the soils, results show that for low soil moisture levels, the reflectance decreased when the moisture increased. Conversely, after a critical point, soil reflectance increased with soil moisture. For some soils, the reflectance of the wettest conditions can overpass that of the driest conditions. For both low and high soil moisture levels, and the seven wavelengths selected, the relative reflectance was strongly correlated with moisture. Adjustment of the relationships over individual soil types provides better soil moisture retrieval performances.

(4) The normalization of reflectance approach, derivative approaches and the difference approaches were used to forecast soil surface moisture. And the best overall retrieval performances were achieved with the absorbance derivatives and the difference of absorbance.

(5) By analyzing the relationship between soil organic matter and soil reflectance, we forecasted soil organic matter and verified the performances of models.

(6) By using BRDF model we analyzed the BRDF property of different soil at different moisture.

(7) Retrieved soil characters from hyperspectral image and developed the soil characteristics map for precision farming.