光学卫星遥感影像自动云检测是卫星产品生产系统的一个重要环节。利用资源三号卫星编目生成的浏览图，采用树状判别结构进行云检测，对浏览图进行分块，提取子块图像的特征进行云地分类。由于云类和地物类过于繁杂，且浏览图的分辨率较低，传统通过图像特征对云地进行分类的算法有很大的局限性，针对这一问题，本文提出了在分类之前对原始的子块图像进行增强处理，扩大云和地物的纹理差异，然后以二阶矩、一阶差分等作为云地分类的图像特征，并在多尺度空间内进行特征延拓，经过综合分析估计云在影像中的比例。该云检测算法应用于资源三号卫星应用系统工程，实际测试结果表明，该算法能够较好地提升云量检测的准确率。%Automatic cloud detection for optical satellite remote sensing images is a significant step in the production system of satellite products .For the browse images cataloged by ZY‐3 satellite , the tree discriminate structure is adopted to carry out cloud detection .The image was divided into sub‐images and their features were extracted to perform cl assification between clouds and grounds .However ,due to the highcomplexityofcloudsandsurfacesandthelowresolutionofbrowseimages,thetraditionalclassifica‐tion algorithms based on image features are of great limitations .In view of the problem ,a prior enhance‐ment processing to original sub‐images before classification was put forward in this paper to widen the texture difference between clouds and surfaces .Afterwards ,with the secondary moment and first difference of the images ,the feature vectors were extended in multi‐scale space ,and then the cloud proportion in the image was estimated through comprehensive analysis .The presented cloud detection algorithmhas already been applied to the ZY‐3 application system project ,and the practical experiment results indicate that this algorithm is capable of promoting the accuracy of cloud detection significantly .
展开▼
