【作者】 付必涛；

【导师】 王乘；

【作者基本信息】 华中科技大学 ， 空间信息科学与技术， 2009， 博士

【摘要】 水遥感技术广泛用于水旱灾害动态监测、洪灾损失评估、水资源监测、水环境监测等研究领域。现有大量MODIS数据资源(MODerate-resolution ImagingSpectroradiometer,中分辨率成像光谱仪),它是一种可免费获取、具有高时间分辨率、高光谱分辨率和低空间分辨率的遥感影像数据,对大范围的、监测频次要求较高的水遥感应用,是一种较好的数据源。但混合像元的存在限制了这些数据的应用效果,使水体提取精度难以满足实际应用需求。所以研究水体亚像元分解重构问题,提取水体混合像元中的水体亚像元并对其进行空间定位,能有效提高水体遥感提取的精度和效费比。本文以水体为特定研究对象,以水体光谱特征和水体混合像元成像机理为基础,以遥感图像处理方法和动态规划法为手段,以MODIS为数据载体,用典型研究区域的遥感数据进行实验验证,在遥感影像预处理、水体混合像元端元选取及线性分解、水体亚像元定位等领域开展研究,建立了水体亚像元分解重构模型,设计实现了一系列相关算法,有效提高了水体提取的精度。本文的水体亚像元分解重构模型适用于所有多(高)光谱遥感影像。论文主要研究工作和创新点如下:1)设计并实现了MODIS 1B级原始影像的几何校正批处理算法,批处理软件包可在20分钟内处理完一幅36波段的影像数据,大幅提高了MODIS几何校正数据处理效率;2)提出了基于四(六)元组和面积比的特征点集匹配算法,充分利用相邻三角形面积比这一仿射不变量来加快同名特征点的搜索速度,匹配过程不需人工干预,具有计算量小,匹配速度快的特点;3)提出了移位重采样模板匹配算法,对主辅图像中的高分辨率影像进行移位重采样,使其与低分辨率影像分辨率一致,再对两者进行模板匹配,取相似性测度最高的位置作为最终匹配值,实现了不同分辨率遥感影像之间的精匹配,配准误差不大于0.1个像素;4)对MODIS影像中水体样本的高光谱特征进行统计研究,并根据统计结果建立MODIS水体提取决策树模型。5)实现了水体混合像元的端元自动选取算法,解决了水体混合像元的端元数量远大于参与解算的MODIS波段数的矛盾,实现了水体混合像元线性分解算法;6)提出了基于空间映射表的水体亚像元定位模型,并设计实现了一套完整的算法,其中包括:水体边界线走向追踪算法,建立端元坐标系及坐标转换算法,建立空间映射表的算法,路径光滑性准则及像元内最小聚类距离等评价指标体系,相邻像元水体边界线匹配算法,水体边界线动态规划搜索算法及其求解算法,像元内路径编号与水体亚像元位置的换算方法。更多还原

【Abstract】 Water remote sensing technology is widely used in dynamic monitoring of flood, drought,water resources,water environment,etc.There are a large number of free remotely sensed data resources with high temporal resolution,high spectral resolution and low spatial resolution,such as MODIS(MODerate-resolution Imaging Spectroradiometer), which is a better data source to large-scale and high frequent water remotely sensed applications.However,the water extraction accuracy can’t meet water remotely sensed applications due to the water mixed pixel.Therefore,the study on water sub-pixel unmixing and reconstruction can effectively improve water extraction accuracy and benefit-cost ratio.The research is focused on water mixed pixel based on water spectral feature,imaging mechanism,and pixel’s inner structure.It covers remotely sensed image preprocessing, water endmember selection,mixed pixel unmixing,and water sub-pixel location.In the dissertation,the author has established the model of water mixed pixel unmixing and sub-pixel mapping,and designed a series of related algorithms.They can be applied to all multi-spectral or hyperspectral remotely sensed images.The main research and findings are as follows:1) Batch algorithms of geometric correction of MODIS 1B class data is designed and implemented.Batch package is able to correct a pair of 36-band MODIS data in 20 minutes,thus significantly improving data-processing efficiency.2) The author proposes a new initial match algorithm using four-point or six-point groups and area ratio.It accelerates the counterpart points searching by area ratio of adjacent triangles.Experimental results show that it can match two images automatically without manpower intervention and other auxiliary information.3) The author proposes the algorithm of moveably resampled template matching to fine register a pair of different resolution images,which can attain 0.1 pixel precision.It resamples the high-resolution images to a sub-image by a movable template,and then uses template match method to register the sub-image and low-resoiution image. 4) A water extraction decision tree model is established by counting water spectral samples of MODIS.5) A Self-Adaptive Endmember selection algorithm is implemented to solve the problem between more endmember and less band,as well as the linear unmixing program of water mixed pixel is accomplished.6) The author proposes the model and algorithms of water sub-pixel location based on spatial look-up table:①tracking algorithm of water boundary curves;②Endmember coordinate system and its transformation algorithm;③the algorithm to establish Spatial mapping table;④evaluation index system,including curve segments’ smooth criteria and minimum clustering distance within pixel;⑤matching algorithm of curve segments between adjacent pixels alone water boundary;⑥dynamic programming searching algorithm of water boundary curves and its solution procedures;⑦conversion algorithms between the serial numbers of boundary curves and endmembers distribution within pixel.更多还原