基于局部相关分析法的ETM+影像修复方法研究

Research on ETM+ Image Restoration Method Based on Local Correlation Analysis

Landsat-7上搭载的专题扫描仪(ETM+)上的扫描行校正器(SLC)在2003年发生故障,导致影像出现坏行,丢失了约25%的数据。由于数据本身仍保持了良好的辐射和几何特性,且同一区域不同时相影像上坏行出现的位置具有随机性,使得使用同一区域不同时相的ETM+影像进行交叉修复成为可能。该文在总结现有修复方法的基础上,提出了一种采用相近时相、相同位置影像对ETM+影像进行修复的方法。首先利用两景影像中的准不变特征点(PIFs),以待修复影像为目标,对待填充影像进行相对辐射校正,消除不同时相影像间由于大气状况等"外源差异"导致的像元亮度值差异。然后运用局部相关分析法对待填充影像进行变换:构建一种同时考虑光谱距离和空间距离的权重系数,根据邻近光谱相似点在两个时相上的变化量对待填充影像进行变换。实验证明,该方法修复效果优于传统的局部回归法。

The Landsat-7 image data has been widely applied in environment assessment and inspection,global change detection,agriculture,geography research,mapping,archeology and also become popular RS data in China. An instrument malfunction occurred on Enhanced Thematic Mapper （ETM＋） of Landsat-7 in 2003,which was caused by failure of the scan line corrector （SLC）. Regular bad lines present on the SLC-off scenes and about 25% data is lost. Because the data itself preserves good radicalization and geometry performances, and bad lines occurred randomly in different images, it＇s possible for cross restoration by using images with the same extent and different temporal phases. At the basis of summarization of present restoration methods,a new method to restore the SLC-off data using another ETM＋ image with near time and same extent is proposed. The first step is to eliminate the affection of ＂foreign difference＂ （atmospheric condition,weather,etc） of the two images （image needs to be restored and image for filling）. Set the image needs to be restored as target, select pseudo-invariant features （PIFs） in two images, and execute local relative radiometric normalization for the image for filling. Convert the image for filling according to nearby speetral-similar pixels by constructing a weight that considers both spectral and spatial distance, Experiment results indicates that the method is an effective method for recovering ETM＋ images and is superior to traditional local regression match method.