多时相遥感影像薄云-厚云联合去除方法研究

Optical remote sensing images are inevitably influenced by clouds cover, and clouds removal is not only an effective approach to improve the usability of remote sensing images, but also a hot issue of concern. This project innovatively proposes a joint removal method of thin and thick clouds, and fundamentally solves the problem of incompletely and inefficiently removing clouds, which is resulted from the independent processing for thin and thick clouds by traditional methods. Concretely, this project proposes a multi-temporal and multi-feature clouds detection method, which unites multiple features, such as spectral, texture and geometric, etc., and the multi-temporal reference information to solve the leak detection of clouds and the fault detection of highlight features. In the framework of variational theory, this project introduces the local sparse constraint term and the nonlocal spatial constraint term, to make the best of the spatial-temporal complementary information, and to build the robust and unified model of thin and thick clouds removal. This project develops the fast optimization method to realize the high-efficiency model solution, and lays the foundation for the widespread applications of optical remote sensing images.

光学遥感影像不可避免地受到云层覆盖的影响，云去除是提高数据可用性的有效途径，也是备受关注的热点问题。本项目创新性地提出薄云-厚云联合去除方法，从根本上解决传统方法对薄云、厚云分开处理造成的去云不彻底、效率不高的问题。具体地，提出多时相多特征联合的云检测方法，联合光学遥感影像的光谱、纹理、几何等多种特征以及多时相参考信息，避免云层漏检、高亮地物错检等问题；在变分理论支持下，引入局部稀疏约束先验项和非局部空间约束先验项，充分利用影像的时空互补信息，构建稳健的薄云-厚云统一去除模型；进一步发展快速的最优化方法，实现去云模型的高效求解，为光学遥感影像的广泛应用奠定基础。

针对光学遥感影像云去除这一热点问题，本项研究创新性地提出了多特征联合的云检测方法、基于非负矩阵分解的云去除方法，形成了系统的遥感影像缺失信息重建体系，并将该体系成功引入到其他定量遥感产品、基础地理信息产品的重建中，取得了较好的效果。研究成果可以有效提升影像的综合潜力，具有重要的应用前景。项目圆满地完成了课题设定的研究计划，并在研究深度和广度上都进行了拓展。在项目资助下，项目组发表科研论文14篇，其中10篇SCI检索，3篇EI检索；出版著作1部；申请、授权、转让国家发明专利各1项；在参加、举办学术会议方面，举办国际遥感影像质量研讨会1次，累计参加国际学术会议5次、国内学术会议6次；在人才培养方面，项目负责人作为骨干完成人获得测绘科技进步一等奖1项，培养/联合培养研究生13人，其中毕业博士4人，毕业硕士5人，在读博士2人，在读硕士2人。