【作者】 金铭；

【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， 硕士

【摘要】 农田重金属污染是当今世界面临的重大环境问题,受到各界广泛关注。重金属污染具有复杂隐蔽性、危害持久性,尤其可能通过食物链的富集作用危及人类健康。因此,如何快速、准确地监测大范围内农田重金属污染相当迫切,具有重要的学术价值与实践意义。本研究以国家863项目为依托,使用多种方法研究大范围内重金属浓度微小变化量的遥感弱信息识别与提取,基于野外实测和Hyperion高光谱影像数据建立水稻镉污染胁迫水平评估模型,为实现大范围快速无损监测水稻镉污染提供理论基础与科学依据,具有较强的理论和现实意义。地面方法采用ASD公司的FieldSpec波谱仪获得水稻冠层光谱数据。鉴于镉对水稻各种生理特征参数的影响具有复杂隐蔽性,导致冠层光谱特征变化微小,需要采用一些方法增强和提取隐含在光谱中的污染弱信息。在水稻冠层高光谱数据和各种实测生化参数的支持下,利用主成分分析、独立分量分析、小波变换和分形分析技术从高光谱遥感数据中提取镉污染弱信息,获取水稻镉污染敏感光谱因子,包括第一主成分F1、独立分量ICA1和小波分解细节信号的光谱分维D,db系列小波函数表现良好。利用3个敏感因子分别构建水稻镉污染胁迫一维诊断模型,获得较好的诊断效果。进一步将3个光谱因子进行组合构建二维诊断模型和三维诊断模型,使污染胁迫诊断结果更稳定直观。进一步利用Hyperion高光谱影像建立诊断模型。首先对影像进行预处理,将DN值转换成地表真实反射率。利用多种小波函数对实验区影像光谱进行分解,结果表明db系列中db5小波第三层分解信号的光谱分维数最为稳定可靠。并以长春地区两个镉污染水平不同区域为例,对Hyperion影像进行小波分解获取细节信号的光谱分维图,识别大范围内水稻镉污染胁迫程度。这种方法为快速、宏观、准确地监测水稻镉污染提供了一种便捷方式。更多还原

【Abstract】 Farmland of heavy metal pollution has drawn wide attention, which is the worst serious environmental problems we are facing. The heavy metal pollution has characteristic of complex confidentiality, durable harm, especially may endanger our human health enrich through the food chain. Therefore, it’s very urgent to monitor a large area of farmland quickly and accurately, it also has important academic value and practice significance.This study based on 863 project, study on remote sensing weak information recognition and extraction through heavy metal density of small changes in wide range. Based on outdoor test and hyperspectral image data(Hyperion) this research established rice cadmium pollution stress level appraisal model, providing theoretical foundation and scientific basis for big range fast and non-destructive monitoring rice cadmium pollution, with strong theoretical and realistic significance.The ground remote sensing methods use ASD FieldSpec spectrum analyzer to obtain the rice canopy spectral data. The variation of physiological characteristic parameters in Cd -induced stress on rice is complex and unobvious, which lead to subtle changes in spectral curve shape. Some methods were used to enhance and extract subtle spectral feature information associated with heavy metal pollution. Hyperspectral data of rice canopy and biochemical parameters were measured. Three methods were used, namely principal component analysis (PCA), independent component analysis (ICA),wavelet transform fractal analysis. Spectral parameters sensitive to Cd contamination, namely F1, ICA1 and fractal dimension of spectral reflectance (D), were obtained by different methods. The above sensitive factors were used to establish one-dimensional diagnosis model for rice under Cd contamination with satisfactory results. The correlation coefficient (R2) between three sensitive spectral parameters and Cd concentration were above 0.8. In addition, stable, sensitive and visual diagnostic methods were achieved by two-dimensional and three-dimensional diagnostic models with three diagnostic indicators. Multi-dimensional spectral diagnostic models can provide an effective way to monitor rice under Cd Stress comprehensively, systematically on a large scale. Hyperion hyperspectral data were used to establish a diagnosis model. Firstly convert the DN value to the surface real reflection. Wavelet function has been used to decomposite image spectrum of the experimental area. The results showed that the 3th layer of the db5 wavelet’s to the signal decomposition spectrum fractal dimension with the most stable and reliable. And with Changchun region two cadmium pollution levels of different areas, for example, Hyperion image wavelet decomposition obtain details of fractal graph spectra signals big scope, identification of rice cadmium pollution degree. Intimidation, study on the case of Changchun’s two regions of different levels of cadmium contamination, achieved the image of Hyperion detail signal wavelet decomposition, and obtained spectra of fractal graph, distinguished big range of rice cadmium pollution degree. This method provides a convenient way for monitoring rice cadmium pollution quickly, accurately and in a large scale.更多还原