云覆盖光学遥感影像的时-空-谱融合方法研究

The high spatio–temporal–spectral resolution remote sensing images are highly desirable and widely used in various areas. However, the satellite sensors have to make a fundamental tradeoff between the high spatial resolution, the high temporal resolution, and the high spectral resolution. Unfortunately, the existing spatio–temporal–spectral fusion methods can only be applied to the fusion of cloud-free optical remote sensing images, and they fail to fuse the cloud-contaminated images. This project focuses on the following scientific issues: 1) the spatio-temporal-spectral relations are difficult to be expressed synergistically; 2) the high-fidelity spatio-temporal-spectral fusion model is difficult to be constructed; 3) the spatio-temporal-spectral fusion model is difficult to be verified quantitatively due to lack of the real reference images. To solve the technical bottlenecks of the spatio–temporal–spectral fusion under the cloud conditions, we carry out the research work on the following aspects: the relationship expression, the fusion model construction, and the fusion model verification. Accordingly, we propose the spatio-temporal-spectral relationship model based on the tensor by considering the cloud contaminations; the integrated fusion model for the cloud removal and the improvement of all the spatial, temporal, spectral resolutions; and the non-reference quality assessment method for the proposed fusion model. By merging the complementary advantage of the optical remote sensing images under cloud conditions, the high-fidelity cloud-free images with all the high spatial resolution, the high temporal resolution, and the high spectral resolution will be obtained, and this has a great potential applications and significance in science.

高时–空–谱分辨率遥感影像在诸多领域具有迫切需求和广泛应用价值。但卫星成像受系统硬件限制，获取影像在高空间、高时间、高光谱分辨率方面相互制约，而现有时–空–谱融合方法仅适用于无云覆盖光学影像，无法满足云覆盖影像的融合需求。本项目针对云影响下时–空–谱融合的技术瓶颈，以“时–空–谱融合模型”为研究主题，围绕以下科学问题展开研究工作：1）云影响下时–空–谱关系难以协同表达；2）云影响下时–空–谱融合高保真建模困难；3）云影响下无真实参考影像进行融合模型定量验证。本项目紧紧围绕以上科学问题，以“关系表达–融合建模–模型验证”为研究主线，提出顾及云覆盖的时–空–谱张量关系表达、“去云–时空谱分辨率提升”一体化融合模型、面向云影响下时–空–谱融合的无参考质量评价方法。本项目研究成果可集成云覆盖影像数据的互补优势，生成高保真的无云覆盖高时–空–谱分辨率遥感影像，具有重要科学研究意义和应用前景。

受卫星传感器成像系统限制，获取影像在空间、光谱、时间分辨率方面相互制约，此外，光学影像获取过程中易受云覆盖的影响，限制了其应用价值与潜力。为此，本项目开展了云覆盖光学影像的时空谱融合方法研究，围绕项目研究内容，在遥感影像融合数据集构建、多源影像时空谱融合、融合影像质量评价等方面展开了研究工作。1）遥感影像融合数据集构建：针对缺少大型公开影像融合数据集的现状，一定程度上限制了本领域尤其数据驱动融合方法的发展，提出了大型遥感影像融合数据集，并免费公开供研究人员使用。2）多源影像时空谱融合方法：提出了全色-多光谱、多光谱-高光谱、时空融合等系列影像融合方法，构建了云雾影响条件下的一体化融合模型，实现了云雾覆盖下影像信息复原与重建，以及分辨率提升的协同处理。3）融合影质量评价：提出了融合影像无参考质量评价方法，实现了融合影像在无真实参考影像下的有效定量验证。此外，本项目在雷达与光学影像融合等方面进行了拓展研究。在本项目资助下，课题组在IEEE Transactions on Geoscience and Remote Sensing、IEEE Geoscience and Remote Sensing Magazine、遥感学报等期刊正式发表论文10篇，此外，接收并在线发表论文6篇；申请发明专利5项，其中授权2项，授权软件著作权2项。本项目提出的理论与方法在卫星影像质量改善与增值服务等方面具有较好的应用价值与前景。