摘要
为了去除厚云及其阴影对光学遥感图像的影响,以TM图像为实验数据提出一种基于光谱及几何信息的去厚云算法。在分析单图像云区多光谱特征及对比多时相图像光谱特征的基础上,首先检测光谱特征明显的厚云,依据云和云影成对出现的成像几何关系,按确定的方位和距离搜索云影;然后采用数学形态学的侵蚀与膨胀算法对云区边缘进行碎片去除及填补、合并处理,使其准确反映图像中受云影响的数据总量;最后利用光谱线性回归匹配后的参考图像替换目标图像中的云区,达到去云目的。实验结果表明,上述算法去厚云效果显著,能够有效排除水体及地形阴影对云影识别的影响,具有快速、简单、实用性强的特点。
A new cloud removal algorithm based on spectral and geometric information is proposed for generating cloud free and cloud-shadow free mosaic image from multi-temporal TM images.At first,single image and multi-temporal images of thick cloud and cloud-shadow multi-spectral detection models are built based on TM spectral characteristics analysis.Secondly,cloud is detected according to spectral characteristics,and then a technique is applied based on coupled geometrical relationship between the cloud and its shadow by using sun azimuth angle,sun elevation angle and statistical distance between cloud and its shadow so as to automatically predict the approximate location of cloud-shadow.After that,erosion filtering and dilation filtering are used sequentially in the cloud and cloud-shadow fraction image to eliminate the small bits and generate the exact zones contaminated by cloud and cloud-shadow.At last,the cloud and cloud-shadow zones of the target image are replaced by the same-location cloud-free zones on reference images whose spectral information is matched with the target image by the linear regression method.The results show that this algorithm is capable of eliminating the cloud influence from TM image significantly.Moreover,it can effectively eliminate the influence of water body and hill shadow on cloud-shadow.This method can therefore support producing cloud removal images in a quick,simple and applicable way.
出处
《国土资源遥感》
CSCD
北大核心
2012年第4期55-61,共7页
Remote Sensing for Land & Resources
基金
水利部公益性行业科研专项(编号:200901067-02)
国家自然科学基金项目(编号:41071230)
水利部948项目(编号:200820)共同资助
关键词
云检测
云影检测
去云
TM图像
cloud detection
cloud-shadow detection
cloud removal
TM image
