基于主被动遥感数据协同处理的地表环境监测与分析

发布时间：2018-05-07 08:21

本文选题：主动SAR数据 + 被动光学数据　；参考：《中国矿业大学》2013年博士论文

【摘要】：地表环境监测需要从土地利用/覆盖分类、地表温度反演、土壤湿度反演、地表形变提取等方面进行研究,实现定性、定量、几何、物理一体化综合分析的目的。现有的遥感数据源中光学遥感影像,具有丰富的光谱色调变化,目标检测和识别都相对简单,但受天气状况和观测时刻影响较大。相比之下主动微波遥感影像具有全天候、全天时、多极化、穿透性好,纹理结构信息丰富等特点,但受频率、极化方式、目标几何信息、介电特性等影响,而且受到SAR成像系统固有的斑点噪声影响,使图像的解译能力降低,导致单独使用雷达图像来进行分类和信息提取变得非常困难。因此主被动遥感数据各有应用优势,为挖掘主被动遥感信息在环境监测方面的优势,满足地表环境分类、定性、定量、几何、物理一体化监测需求,本论文深入研究了基于主被动数据协同处理的地表环境监测体系结构与技术方法,从信息融合与分类、协同目标识别、协同参数反演与关联分析四个层次开展了研究,以具有代表性的城区、矿区、湿地等为应用领域,综合信息融合、特征提取与目标识别、分类器集成、地表参数反演、地表形变监测等多种技术手段,多方面、多角度研究了主被动多源遥感信息协同处理在环境监测方面的应用。论文主要内容和成果如下： (1)构建了多源信息融合和多分类器集成的地表环境主被动遥感数据协同分类方法。结果表明,改进的小波信息融合和Bagging完全样本集的SMO分类器集成方法具有稳健的提高分类精度的能力。通过IHS算法对小波融合方法进行改进,通过样本筛选策略对分类器集成方法进行优选,将数据信息融合和分类器技术集成应用于土地利用覆盖分类。相比其他融合方法,改进的融合方法获取最优融合结果；融合后分类结果验证了基于多分类器集成方法能够在一定程度上提高分类精度。 (2)提出了基于主被动遥感数据多特征组合的多分类器集成分类策略。通过等值权重对主被动遥感数据的光谱特征、纹理特征、极化特征进行组合,通过并联、串联策略对多种分类器进行集成。结果表明光谱特征和SAR强度特征组合在使用串联的分类器集成策略时获取最高分类精度,光谱特征和SAR极化特征组合在使用并联的分类器集成策略时获取最优分类精度。提出的基于主被动遥感数据多特征组合的多分类器集成分类方法对不同地物类型的提取精度均有不同程度的提高,对难以区分的复杂研究区域改进明显,极化、光谱、纹理特征集成策略适用于多分类器串联集成策略,极化特征和光谱特征集成策略在多分类器并联协同中获得最高分类精度。 (3)改进了基于典型地类目标识别基础的决策分类方法。利用特征因子、纹理特征、投票决策改进基于空间关联度指数的人类居住地识别算法,实现基于协同主被动遥感数据多源特征的目标识别与分类。实验结果表明协同主被动遥感数据多特征目标识别基础上的决策分类方法不仅能够提高单一地物类别识别精度,对整体分类精度也有明显提高。 (4)设计了基于主被动遥感数据的地表环境几何、物理、定性、定量一体化监测方法。利用热红外数据反演地表温度参数,利用主动SAR数据和光学数据协同反演土壤水分,利用两轨差分方法提取地表形变信息,实现地表覆盖、地表温度、浅层土壤水分和地下变形的‘地空一体化’协同分析的基本条件。通过关联分析,初步得出地表覆盖类型与地表温度、地表形变、土壤湿度的关系；高温热场与地表形变、地表覆盖类型的关系。最后结合实例,应用CA_Markov模型、RUSLE模型对土地利用／覆盖类型、地表温度覆盖等级的变化趋势进行模拟与分析。结果表明协同主被动多源数据研究地表覆盖状态和地表参数关系,是充分利用主被动数据协同优势,实现‘地、空一体化监测’和快速地表环境集成监测的有效方法,体现了主被动遥感信息协同处理地表环境监测体系在实际应用处理中的优势。
[Abstract]:Surface environmental monitoring needs to be studied in aspects of land use / cover classification, surface temperature inversion, soil moisture inversion, surface deformation extraction and so on. The purpose of integrated analysis of qualitative, quantitative, geometric and physical integration can be achieved. The existing optical remote sensing images in the remote sensing data source have rich spectral tone changes, target detection and recognition. Relatively simple, but influenced by weather condition and observation time, the active microwave remote sensing image has the characteristics of all-weather, multi polarization, good penetration, and rich texture information, but influenced by frequency, polarization, target geometry and dielectric properties, and is influenced by speckle noise inherent in SAR imaging system. As a result, the ability of image interpretation can be reduced, and it is very difficult to use radar images to classify and extract information alone. Therefore, the main and passive remote sensing data have their advantages to mine the advantages of the passive remote sensing information in environmental monitoring and meet the integrated monitoring requirements of the surface environment classification, qualitative, quantitative, geometric, and physical. In this paper, the structure and technology of surface environment monitoring system based on CO processing of active and passive data are studied. The research is carried out from four levels, including information fusion and classification, cooperative target recognition, collaborative parameter inversion and association analysis, which are representative area, mining area, wet land and so on as application fields, integrated information fusion, feature extraction and extraction. Target recognition, classifier integration, surface parameter inversion, surface deformation monitoring and so on, many aspects and multi angles are used to study the application of cooperative processing of passive multi-source remote sensing information in environmental monitoring. The main contents and achievements of this paper are as follows:
(1) construction of multi source information fusion and multi classifier integration of surface environment and passive remote sensing data synergetic classification method. The results show that the improved wavelet information fusion and the Bagging complete sample set SMO classifier integration method has a robust ability to improve the classification accuracy. Through the IHS algorithm to improve the wavelet fusion method, through the sample This filtering strategy optimizes the classifier ensemble method, and applies the data information fusion and classifier technology to the land use coverage classification. Compared with other fusion methods, the improved fusion method is used to obtain the optimal fusion results, and the fusion results verify that the multi classifier integration method can improve the score to a certain extent. Class precision.
(2) a multi classifier ensemble classification strategy based on the multi feature combination of the passive remote sensing data is proposed, which combines the spectral features, the texture features and the polarization characteristics of the passive remote sensing data by the equivalent weight, and integrates various classifiers through parallel and series strategy. The results show that the spectral features and the SAR intensity features are combined in use. In series classifier ensemble strategies, the highest classification accuracy is obtained. Spectral features and SAR polarization features are combined to obtain the optimal classification accuracy when using a parallel classifier ensemble strategy. The proposed multi classifier ensemble classification method based on the multi feature combination of the passive remote sensing data has different degrees of extraction accuracy for different terrain types. The integration strategy of polarization, spectrum and texture feature is suitable for the multi classifier series integration strategy. The polarization feature and spectral feature integration strategy can obtain the highest classification precision in the parallel collaboration of multiple classifiers.
(3) the decision classification method based on typical target recognition base is improved. Using feature factor, texture feature and voting decision, the human residence recognition algorithm based on spatial correlation index is improved to realize target recognition and classification based on multi source features of cooperative passive remote sensing data. Experimental results show cooperative passive remote sensing data. The decision classification method based on multi feature target recognition can not only improve the accuracy of single object classification, but also improve the overall classification accuracy.
(4) the surface environment geometry, physical, qualitative and quantitative integrated monitoring methods based on the passive remote sensing data are designed. Using the thermal infrared data to invert the surface temperature parameters, use the active SAR data and the optical data to coordinate the soil moisture, and use the two track differential method to extract the surface deformation information, and realize the surface cover, surface temperature and shallow layer. The basic conditions for the synergistic analysis of soil water and underground deformation are "ground space integration". Through correlation analysis, the relationship between surface cover type and surface temperature, surface deformation and soil moisture; the relationship between high temperature field and surface deformation and surface cover type. Finally, the application of CA_Markov model and RUSLE model to land The results show that the relationship between the surface coverage and the surface parameters of the cooperative principal and passive multi source data is an effective method to make full use of the synergistic advantages of the active and passive data to realize the integrated monitoring of ground, air integration and rapid surface environment. The advantages of active passive remote sensing information in collaborative processing of surface environmental monitoring system in practical application are presented.

【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：P237

【参考文献】