Systematic approach for precise correction and landcover classification of satellite images

在崎岖地形的卫星图像中，我们可以发现一些地形效应与大气效应纠缠在一起。如果不对这些影响进行校正，就很难对用于土地覆盖分类和自然环境多时相监测的卫星图像进行评价。为了在计算机上再现这些地形效应，需要具有与卫星影像相同的地图投影和空间分辨率的数字高程模型（DEM）。本课题在遥感原理的基础上，系统地研究了卫星影像的正射纠正方法，并将其应用于Landsat TM影像。提出了一种利用模拟太阳直射辐射图像进行几何校正的方法，将卫星图像自动叠加到DEM上。基于卫星辐射物理模型的地形影响校正方法也被发展到高太阳高度卫星图像。我们还提出了土地覆盖模型作为一个镶嵌的空间结构，并应用空间分割的图像第一。然后，混合像素被确定为边界像素，在邻域中找到确定的端元。

In satellite images over rugged terrain, we can find some topographic effects entangled with atmospheric effects. Without correcting these effects, it is difficult to evaluate the satellite images for land cover classification and multi-temporal monitoring of natural environment. In order to reproduce these topographic effects in computer, we need digital elevation model(DEM) with the same map projection and spatial resolution as the satellite image concerned. The DEM is also required for accurate ortho-rectification of the satellite Image.In this research project, we systemize satellite image processing based on the principles of remote sensing, which is applied to Landsat TM images. The satellite images are automatically laid over the DEM by using the proposing geometric correction method which utilizes the simulated direct solar irradiance image. Advanced methods for topographic effect correction based on a physical model of satellite level radiance is also developed to a satellite image with high solar elevation.We also propose to model land cover as a tesselated spatial structure and apply the spatial segmentation to the image first. Then, mixed pixels are determined as the boundary pixels with definite end members found in neighborhood.