【作者】 刘美玲；

【导师】 刘湘南；

【作者基本信息】 中国地质大学（北京） ， 地图制图学与地理信息工程， 2011， 博士

【摘要】 运用遥感技术进行农作物重金属污染识别与监测对于农业生产、粮食安全、人类生存环境具有重要的意义。然而“自然农田生态系统”中农作物重金属污染胁迫光谱特征信息微弱且不稳定,直接利用原始光谱信号评价农作物重金属污染状况其可信度不高。因而,如何增强农作物重金属污染胁迫光谱弱信息,如何建立农作物重金属污染胁迫水平遥感计算模型,是实现农作物重金属污染遥感快速、准确监测需解决的关键问题,是遥感技术应用向定量化、精细化方向发展需解决的科学问题。本文选取长春、苏州的受重金属污染(Cu和Cd)的水稻农田样地作为实验区,采集水稻关键生长期的ASD光谱数据、生化数据、环境数据,并获取准同步的Hyperion数据。分析土壤重金属对水稻胁迫的影响机理,提取水稻污染胁迫敏感光谱参数,提出水稻污染胁迫光谱弱信息增强与分离方法,建立水稻污染胁迫遥感估算模型。本文的研究工作与主要结论如下:(1)利用多尺度小波变换实现了水稻重金属污染胁迫弱信息分离与增强,并提取了水稻重金属污染胁迫3类敏感特征光谱参数:①反映光谱“整体”变化的小波-分形维数;②反映光谱“突变”信息的小波系数;③反映光谱“奇异性”的光谱指数。(2)运用多元逐步回归筛选出对水稻重金属污染胁迫响应敏感的光谱参数,利用灰色关联度分析方法筛选出对水稻重金属污染吸收累积关系密切的环境参数(土壤属性、气象因子);以光谱参数和环境参数为输入参数,以水稻重金属污染胁迫水平为输出参数,运用动态模糊人工神经网络算法,建立了“自然农田生态系统”中水稻重金属污染胁迫水平光谱分析模型,模型精度高(R2为0.69～0.98)、结构紧凑(模糊规则为5～13个),相比其它算法(如BP、ANFIS)构建的水稻重金属污染胁迫光谱分析模型,具有评价结果更加精确、稳定,更加符合实际,原理更易解释的优点。(3)利用Hyperion数据,结合空间插值技术,将所建立的水稻重金属污染胁迫光谱分析模型进行区域外推,建立了大尺度水稻重金属污染胁迫水平卫星遥感评价模型,模型取得了满意的结果(R2为0.69～0.72)。实现了水稻重金属污染胁迫光谱分析模型的尺度转换。本文的创新点:①提出了多尺度小波变换的光谱弱信息分离与增强的方法,可以推广应用到各种环境胁迫下所引起的农作物异常探测和弱信息增强;②基于动态模糊人工神经网络算法,建立了融合光谱参数和环境参数的水稻重金属污染胁迫水平光谱分析模型,为遥感地学应用中如何融合遥感因子与非遥感因子建模提供了一种新方法。更多还原

【Abstract】 It is important for agricultural production, food security and human survival environment using remote sensing technology to identify and monitor crops heavy metal contamination. However, the pollution level in the‘real world’agro-ecosystems is relatively low, which means there are subtle and unstable characteristic in leaf reflectance spectra. And therefore it is incredible in assessing stress levels of crop under heavy metal pollution using leaf reflectance spectra directly. So, how to enhance subtle spectral characteristic information associated with heavy metal pollution, how to establish effective model for assessing stress levels of crop with heavy metal pollution, they are key issues for applying remote sensing technology to achieve fast and accurate identification of crop heavy metal contamination. At the same time, they are also scientific problems to be solved for remote sensing technology in quantitative and fine application.Several experiment paddies located in Changchun, Jilin Province, and Suzhou, Jiangsu Province, China with different pollution levels were selected. We collected various data from experimental farms in rice during the typical growth stages, including ASD data, Hyperion hyperspectral data, biochemical data, heavy metal content data (i.e.soil, rice), soil properties and meteorological factors, and other basic data. Based on ground measured data, soil heavy metal effect on stress mechanisms in rice were analyzed, spectral parameters sensitive to rice under heavy metal pollution stress were calculated, and the theory and technical methods about enhancing and deriving subtle spectral characteristic information of rice under heavy metal pollution were proposed, and the models for assessing stress levels of rice under heavy metal pollution were constructed. The most important findings and conclusions drawn from this study include:(1) Wavelet transform was adopted to enhance and derive subtle spectral characteristic information of rice under heavy metal pollution. Three categories of sensitive spectral parameters were extracted,①the fractal dimension of reflectance with wavelet transform(FDWT) as a quantitative and comprehensive indicator by capturing‘global variation’of spectrum curve,②wavelet coefficients (WC)by capturing spectral reflectance singularity information,③vegetation indices based on singularity points.(2)According to different characteristic in various types’parameters, Firstly, the stepwise regression and gray correlation analysis were used to choose the sensitive spectral parameters and environmental factors which are closely correlated to heavy metal diffusion in rice, respectively. Secondly, based on dynamic fuzzy neural-network model, spectral analysis model for assessing heavy metal stress levels of rice in‘real world’agro-ecosystems were constructed by integrating spectral parameters with environmental factors. In addition, the constructed models were verified using training and validation sets, they got satisfactory results with a high level of accuracy (R2: 0.69-0.98), a compact structure (fuzzy rules: 5-13). As compared with other model algorithm (such as BP, ANFIS) for constructing local spectral model to assess rice with heavy metal stress, it had more accurate, stable results and higher reliability. In addition, it can obtain rules with a physical meaning. The models were characterized by strong stability and better effects in evaluation of stress levels of rice under heavy metal pollution.(3) Hyperion data was used, and spatial interpolation technology was adopted. The established spectral analysis models were applied to the polluted area in a large scale, and satellite remote sensing model for assessing rice with heavy metal stress level were constructed successfully. The established spectral analysis model for assessing rice with heavy metal stress level succeeded in completing scale transformation.In this paper, it is believed that the wavelet technique will play an important role in detecting other environmental aspects of crop stress in the future. The method, the spectral analysis models were established by integrating spectral parameters with environmental factors on the basis of dynamic fuzzy neural-network algorithm, can provide important reference and theory basis for modeling in the various geosciences application.更多还原