联合变换学习与张量低秩的高维遥感图像复原模型与高性能算法研究

高维遥感图像是国防安全和矿藏勘探等领域的基本数据来源，具有广泛的应用价值。由于成像设备、成像环境和传输条件等限制，获取到的图像易受到云、阴影及各类噪声等污染，造成信息缺失并限制后续应用。高维遥感图像复原旨在从退化图像中估计反映真实场景的图像，是遥感图像处理中亟待解决的基础科学问题之一，具有重要的理论研究意义和应用价值。现有方法仅在固定域中挖掘图像和噪声成分的先验知识并得到次佳结果。因此，本项目拟研究联合变换学习与张量低秩的高维遥感图像复原模型与高性能算法。首先，研究新型变换保真项，用于刻画退化图像与真实图像之间的内在关系；其次，挖掘图像和结构噪声成分在变换学习框架下的潜在先验知识，并建立高维遥感图像复原的联合优化模型；最后，根据所建立模型的结构特征设计高性能优化算法求解，并分析算法的收敛性等理论性质。本项目的实施有助于解决高维遥感图像精准复原问题，并有效地提升遥感数据的实际应用潜力。

High-dimensional remote sensing images are the basic data source in the fields of national defense and security and mineral exploration. It has a wide range of application values. Due to the constraints of imaging equipment, imaging environment, and transmission conditions, the acquired images are vulnerable to the cloud, shadow, and various kinds of noises, resulting in the loss of image information and limiting subsequent applications. High-dimensional remote sensing image restoration aims to estimate underlying images from degraded images. It is one of the fundamental problems to be solved urgently in remote sensing image processing and has important academic significance and practical value. The existing methods only explore the potential prior knowledge of image and noise components in the fixed domain and achieve sub-optimal results. Therefore, this project studies the high-dimensional remote sensing image restoration model and high-performance algorithm via jointing transform learning and tensor low-rankness. First, study the novel transform fidelity term, which is used to describe the internal relationship between the degraded image and the underlying image; second, develop a joint optimization model for high-dimensional remote sensing images restoration by exploring the potential prior knowledge of the image and structure noise components under the framework of transform learning; finally, design a high-performance optimization algorithm for the developed model by considering the structural characteristics, and analyze the theoretical properties such as the convergence of the algorithm. The implementation of this project will help to solve the problem of accurate restoration of high-dimensional remote sensing images, and effectively improve the practical application potential of remote sensing data.