遥感影像是获取地表覆盖信息的重要数据源，高光谱遥感图像在成像过程中易受云层遮挡的影响，如何高效地进行云层去除是高光谱遥感图像处理领域的巨大挑战。传统去云方法不能充分挖掘高光谱图像信息，效率低下，生成式对抗网络是深度学习领域最具潜力的算法之一，以生成和判别相对抗的思路提取影像低层次到高层次潜在信息的表达，且具有较好的泛化能力，为高光谱遥感图像去云提供了新的解决方案。本项目首先针对高光谱图像云层覆盖的数据特性构建基于超像素分割的高光谱遥感图像云标签数据集，解决云层覆盖高光谱标注数据不足的问题；其次，针对高光谱遥感成像的特点，设计适用高光谱多波段生成和对抗的网络模型；最后，引入云类别条件变量优化网络模型，实现高光谱遥感图像云层去除。通过本项目的研究，能够增强遥感数据的有效性，解决高光谱图像云层遮挡造成图像利用率下降的问题，为高光谱图像的实际应用提供重要的数据基础和技术支撑。

Remote sensing images are the important data source for obtaining land ground cover information. Hyperspectral remote sensing images are easily affected by cloud cover. It is a great challenge that how to effectively remove clouds in the field of hyperspectral remote sensing image processing. Traditional cloud removal methods cannot fully extract hyperspectral image information and are with low efficiency. Generative Adversarial Network (GAN) is one of the most potential algorithms in the field of deep learning. It can extract the low-level to high-level potential information by the way of generating and discriminating process and has high generalization ability. It provides us a new solution for removing clouds from hyperspectral remote sensing images. In this project, a cloud labeled dataset for hyperspectral images is firstly created via superpixel segmentation based on the features of cloud cover from hyperspectral remote sensing images, which is to address the issue of only limited cloud labeled data for hyperspectral images. Further, a GAN model is established for hyperspectral images based on the characteristics of hyperspectral remote sensing imaging. At last, combining the spectral and spatial distribution characteristics of clouds, a cloud labeled conditioning variable is proposed in the model to achieve cloud removal. The research of the project can increase the availability of remote sensing data, and solve the issue of degraded hyperspectral image utilization caused by cloud cover. It provides an important data basis and technical support for the practical application of hyperspectral images.