中国科学院遥感与数字地球研究所-高光谱遥感应用技术研究室

【作者】              刘海霞                                                                                                                                           

【导师】              童庆禧；张霞

【 学位年度 】    2010

【论文级别】      硕士    

【关键词】          多时相影像，辐射归一化，变化检测，北京一号小卫星 

【Key words】    Multi-temporal images, Relative radiometric normalization, Change detection, Beijing-1  

【中文摘要】

变化检测是通过分析多时相遥感图像实现土地利用动态监测的一种有效方法，但在变化检测分析前，需要经过辐射校正消除光照等因素对地物光谱辐射的影响，使同一地物在不同时相影像中具有相同的辐射量。辐射校正可以分为绝对和相对辐射校正，其中相对辐射校正又称为辐射归一化。

        本文主要介绍了辐射归一化的基本概念、数学模型和处理流程，对现有的辐射归一化方法进行了较为全面的归纳和评述。本文在传统伪不变特征(PIF)算法的基础上，根据地物在不同时相遥感图像中的光谱特性满足线性关系的特点，提出一种半自动实现多时遥感图像辐射归一化的稳健方法。首先通过阈值法与城镇掩模找出未变化地物，然后利用主成分分析优先选取占主体信息量的未变化像元，并用相关系数进行质量控制；再利用这些未变化像元样本点在多时相遥感图像中的辐射关系，计算各波段的辐射归一化系数；最后根据各波段的辐射归一化系数实现图像间的辐射归一化。本文方法利用统计特征客观地选取样本点，克服了常用方法选取样本点具有主观性的缺陷，提高了校正精度，从而为进一步的土地利用/覆盖变化信息的有效提取提供保证。

        本文选择北京市某一区域的两时相影像，首先对图像进行几何配准作为试验所需的数据影像。然后基于matlab语言编程，对上述图像实现图像回归法、直方图匹配法、传统的PIF法、MAD法和基于PIF改进的辐射归一化方法进行辐射归一化实验。实验数据采用北京一号小卫星和Landsat ETM数据，我们分别进行了同一传感器不同时相影像之间的辐射归一化和不同传感器之间的辐射归一化。实验结果表明，利用本文方法对图像进行辐射归一化比其它方法更能准确可靠地选取未变化像元样本点，可以较好地消除不同时相的差异及不同传感器之间的差异，说明该方法具有一定的准确性和可靠性。

        最后将本文提出的方法应用于多时相北京地区的北京一号小卫星影像作辐射归一化后，进行了相应的变化检测与分析，验证了辐射归一化对于变化检测的必要性，并比较了各种辐射归一化方法对变化检测结果的影响。 

【Abstract】

Detecting changes in land cover through time using remotely sensed imagery is a powerful application that has seen increased use, as imagery has become more widely available and inexpensive. Before a time series of remotely sensed imagery can be used for change detection, images must first be standardized for effects outside of real surface change.

      This thesis introduced the concept, mathematical mode, and the work flow of relative radiometric normalization. This thesis established an improved pseudo-invariant features method to normalize temporally separate but spatially coincident imagery. Using the concept of pseudo-invariant features between master-slave image pairs, spatially coincident urban features with difference thresholds are identified from images; then these features were filtered using principal component analysis, their quality control is through correlation coefficients; finally a regression equation is calculated using robust orthogonal regression to normalize slave images to a master. This improved method uses objective statistical characteristics of sampling points, overcomes the defect of subjective sampling, and improves the normalization accuracy, so as to guarantee further effective extraction of land use/cover change information.

      This thesis used two sets of imagery to test the performance of the standardization process, a temporally variable image pair of the same sensor, and a temporally variable image pair of different sensors. This thesis calculate RMSE, statistical characteristics, slope of the major axis in PCA, and performed simple NDVI image subtraction, to validate the reduction of master-slave differences using invariant locations. As a result of the standardization process, RMSE showed decreases in master-slave differences, statistical characteristics including mean and standard deviation is more similar to the master image, the slope of major axis in PCA is closer to unit, and NDVI image subtraction showed decreases in master-slave differences. Also, this thesis compared the result of this improved method with other methods including image regression, histogram match, Schott-PIF, and multivariate alteration detection. According to visual inspection and quantitative evaluation parameters, this improved method has accuracy and reliability; it can sample pseudo invariant points with better quality, and eliminate the multi-temporal differences and multi-sensor differences. After relative radiometric normalization, we perform change detection using three change detection algorithms and make land cover change detection analysis.