【摘要】 遥感影像的几何配准是影像后续处理的重要前提和遥感农情监测等应用的重要保障。不同的自动几何配准算法在配准效果上存在差异,单一配准算法难以满足所有类型数据的配准要求。根据不同地形特征和不同时相特征,选择了平原和山地、夏季和冬季4个实验区,以现有的基于区域的互相关法、互信息法和基于特征的SIFT算法为基础,分别对上述4个实验区的高分一号影像数据进行自动配准实验,对比3种算法的配准精度、配准效率和稳定性。实验结果表明:应用SIFT算法进行配准,4组实验结果均目视接边效果良好且均方根误差达到10-5数量级,满足精度要求。该方法简单、高效,可以应用于农情遥感监测等日常业务。更多还原

【Abstract】 The geometrical registration of remote sensing image is an important premise for the subsequent processing of image. And it’s also an important security for the application,such as agricultural condition monitoring. Different algorithms of automatic geometry registration lead to various registration effects. It’s hard to meet the registration requirements of all images. Four testing types of plains,mountains,summer and winter were selected based the features of terrain and time. The main three registration methods were: cross correlation algorithm based on region gray,mutual information algorithm based on region gray and SIFT algorithm based on features. SIFT feature is the partial feature of the image,which can keep the invariance in rotating, scale-zooming and brightness changing. Then the automatic geometric registration was made for four classes of GF-1 remote sensing image using the above three algorithms. Two kinds of experiments were conducted for GF-1 remote sensing image under various conditions such as different terrains and different imaging time. The comparison of different geometric registration algorithms were made in the aspects of accuracy,efficiency and stability. The results show that the SIFT algorithm is the most appropriate one. The visual edge effect is good and the root mean square error reaches the magnitude of 10-5,which can satisfy the demand of precision. This method is simple and efficient,and it can be applied into agricultural condition monitoring and other business efficiently.更多还原