×
【作者】 刘波
【导师】 童庆禧;郑兰芬;张立福
【 学位年度 】 2010
【论文级别】 博士
【关键词】 成像光谱,植物信息,荧光,杂草识别,叶绿素,光谱指数,光谱响应,支持向量机,FISS
【Key words】 the vegetation index time series of remotely sensed data, Harmonic Analysis, HANTS, number of frequency, outlier detection
【中文摘要】
地面成像光谱系统有着极其广阔的应用前景,而相关硬件研制、系统开发和应用研究在我国尚处于起步阶段。基于这种系统研发与应用研究都严重不足的现状,本文立足于我国第一套地面成像光谱系统(Field Imaging Spectrometer System ,FISS),主要完成两项系统性的工作:其一,在系统研发与评价方面,完成了对地面成像光谱辐射仪的数据定标、数据预处理以及测量规范的初步探讨,系统地评价了其各项性能和特点,为促使其向系统化、集成化方向发展开展了基础性的工作;其二,在系统应用与推广方面,围绕植物信息提取这个主题进行了探索性研究工作,特别是对植物“精准”信息的提取,主要从植物精细识别(类别信息)、植物生化参量反演以及植物对环境因素的生理反应状况三个方面开展了详细的研究,为植物信息的快速获取提供了新技术和新方法,同时也为地面成像光谱系统的推广使用提供了示范研究。论文的主要研究内容和成果如下:
1.介绍了自主研发的我国第一套地面成像光谱系统(FISS)的系统原理、系统组成以及技术指标,并对其进行了光谱定标与辐射定标,开展了噪声评估、噪声去除、反射率转换等数据预处理研究,并且在大量的野外和室内试验基础之上,建立了FISS系统的测量规范。
2.在自然环境下,利用FISS系统进行了杂草与作物的识别研究,对比了不同模型输入特征和不同识别方法的结果。结果表明,利用少量的光谱波段(8个)即可有效地实现对杂草之间以及杂草与作物的多类识别,位于红边区域的波段对杂草与作物有着显著的区分能力。对于不同识别方法,具有非线性分类能力的支持向量机表现更优,特别是在作物与杂草的多类识别问题上。对于不同分类特征,少量变量情况下小波系数表现优于原始波段,但变量增加到一定程度之后二者差别微弱。
3.基于FISS数据对生化参量进行了反演研究,对比了线性回归、多元线性回归、偏最小二乘法与支持向量机等方法的结果。不同模型方法之间的精度整体上差别不是很大,模型方法对反演精度的影响低于模型输入特征对精度的影响。反演叶绿素仅需要少量的重要波段信息即可,并且适应于多元线性模型。
4.图像信息对叶绿素含量有着很好的指示作用,进一步的图信息与谱信息的融合可以显著提升叶绿素含量反演精度。FISS数据由于其独特的测量方式和图谱合一的特点,可以更加准确地用于叶绿素含量反演,反演误差比非成像数据降低约30%-45%。
5. 基于FISS数据研究了植物在不同光照条件下的光谱响应变化,并尝试建立与光合作用参数的关系,用以反映植物的生理状况。PRI日变化呈现“早晚高中午低”的特点,与其指示的植物热耗能机制(叶黄素循环)正好对应,并且与光能利用率和光合有效辐射的变化规律吻合,表明利用FISS数据,可以从光谱响应变化中成功地探测到植物光能利用热耗散变化信息。
6.利用氧气吸收的760nm通道,基于夫琅和费暗线法(FLD)成功地从FISS辐射数据提取了太阳光诱导的光合作用荧光定量信息,获取了超高空间分辨率成像荧光图像,荧光强度呈现出理想的“早晚低中午高”日变化规律,并且展现了丰富的空间细节信息,与植物生理变化特点一致。与光谱指数相比,荧光对光能利用率的预测能力更强,说明荧光作为光合作用的“探针”对光能利用率等参数变化的指示作用更加灵敏。本项研究结果不仅对地面植株群体尺度上的获取荧光信息用于植物生理状况以及逆境指示研究有直接应用价值,有望可以广泛地应用于农作物估产、中药材种植、植物环境胁迫、植物病虫害等应用领域,同时也为我国尝试从航空航天被动遥感角度获取植物成像荧光信息提供了有益的参考。
7.利用FISS数据成功地探测到了植物叶片脱水过程中的“红边蓝移”现象,并且对这一现象以图谱可视化形式展示。
8.本文围绕的植物信息提取的三个方面,即植物类别、生化参量、生理变化的研究实现了对FISS系统性能的全面评价,表明FISS系统具有优越的光谱辐射性能,可用于定量化研究以及精准信息制图。
【Abstract】
Field imaging spectrometer system spans a very broad range of applications. However, sensor design, system development and application researches are all at the very beginning stage in China. To promote both of development and application of this kind of system, based on Field Imaging Spectrometer System (FISS) which was firstly self-developed in China, this dissertation has a pioneer exlporation of this system. The study concentrates mainly on two aspects: for the development and evaluation of FISS, the research discusses data calibration, data pre-processing, operation specifications and instruction for use of FISS, evaluates the overall performance of this novel system, lays the first stone for its progressing towards systematization and integration. For the application and popularization of FISS, the research mainly focuses on plant information extraction especially at the respect of precise information, namely, plant precise identification and classification, biochemical parameters estimation and physiological reaction adapting to change of environmental factors, presents new technology and methods for quick acquisition of plant information, and thus provides a demonstration research for the wide-use of this system in future. Main contents and results are summarized as follows:
1. The first self-developed field imaging spectrometer system (FISS) is introduced in detail, including its imaging principle, structural design and technical parameters. Spectral calibration, radiometric calibration and data pre-process such as noise estimation, noise reduction, reflectance calculation and so on, are also discussed. Based on numerous field and indoor experiments, operation specifications and instruction for use of this novel system are summarized as well.
2. Under outdoor conditions, crop-weed discrimination is explored and studied using FISS, two discrimination models Linear Discrimination Analysis (LDA) and Support Vector Machine (SVM) with different features (spectral bands and wavelet coefficients) are compared. The results show that good discrimination accuracies could be obtained using several spectral bands and those bands located in ‘red edge’ range have prominent discrimination performance. For different models, SVM with V nonlinear trait is superior to LDA especially in weed-crop multi-classes discrimination. For different features, wavelet coefficients show better performance when the amount of variables used is small (e.g. less than 10), but the difference becomes negligible when amount of variables increases.
3. Biochemical parameters are estimated based on FISS, four models, namely, Linear Regression (LR), Multiple Linear Regression (MLR), Partial Least Squares Regression (PLSR) and Support Vector Regression (SVR), were compared. No prominent difference was observed between estimation accuracies using different models. Features impose much influence on estimation accuracy than models. To estimate chlorophyll contents of leaves, only several key spectral bands are needed and a multiple linear model could be competent.
4. Image information could be a good indicator of chlorophyll contents of leaves. Fusion of image information and spectral information could improve precision of chlorophyll contents estimation further. Due to its unique mode of data acquisition and trait of combination of image and spectral information, FISS could be employed to estimate chlorophyll contents more accurately. Estimation error using data from FISS is 30%-40% lower than error using counterpart data from non-imaging spectrometer.
5. This dissertation studies spectral changes of plant responding to natural illumination changes. This kind of changes could be used as indicators of plant physiological status. Photochemical Reflectance Index (PRI) shows a pattern of lower values around noon and higher values during morning and dusk. This corresponds to mechanism of plant heat dissipated (xanthophyll cycle), and is consistent with Light Use Efficiency (LUE) and Photosynthetically Active Radiation (PAR) changing patterns. This result indicates that plant heat dissipation information could be detected from spectral changes using FISS data.
6. Solar induced fluorescence information was successfully extracted from FISS data based on Fraunhofer Line Depth (FLD) method using 760nm O − A 2 absorption band. Fluorescence image spectrum with high spatial resolution was obtained and showed abundant information of spatial details. Fluorescence presents an ideal pattern of higher values around noon and lower values during morning and dusk. Both spatial distribution and temporal variation of Fluorescence are in consistent with physiological changes of plants. Compared to spectral indices, Fluorescence is much sensitive to LUE and thus could be a better indicator of LUE. This result may be directly used for researches of plant physiological status and its change responding to environmental stresses, may be widely used for crop yield estimation, Chinese medicinal materials production, environmental stress detection and monitoring, plant pests and diseases and many other application fields. This study is also beneficial to the development of passive remote sensing on airborne or spaceborne platforms to detect fluorescence of plants.
7. The phenomenon of blue shift of red edge of plant leaves during dehydration process is successfully detected using FISS data and visualized in the form of image series.
8. This dissertation evaluated and assessed the whole performance of FISS from researches of three respects on plant identification, biochemical parameter and physiological change. All shows that FISS has good spectral and radiometric properties and could be used in quantitative researches and precise information mapping.