基于编码感知的高帧率高光谱视频获取及重建

Hyperspectral information is in high demand in various civilian and military domains. For dynamic targets and scenes, the acquisition speed of hyperspectral information should be very fast, so as to form the hyperspectral video. However, the existing imaging spectrometers mainly adopt the “time division multiplexing” scheme to capture hyperspectral images, which, as a result, cannot finish the acquisition task of hyperspectral video. Therefore, the acquisition of hyperspectral video suffers from the principle bottleneck..Based on the computational photography theory, this project will propose a new hyperspectral video acquisition system by exploiting the new mechanisms, new approaches and new technologies. Specifically, this project will first design and implement a scheme for high-dimension high-resolution hyperspectral information acquisition. Then, through exploiting the intrinsic characteristics of the high dimensional signal, this project will design a robust algorithm to finish the hyperspectral video reconstruction. The main research contents include: 1) compressive sensing based temporal-spatial-spectral joint high resolution hyperspectral information acquisition; 2) high-quality hyperspectral reconstruction algorithm based on high dimensional correlation. Through this project, we hope to achieve a breakthrough in hyperspectral video acquisition, promote the application level of the National Earth Observation System, and lay the theoretical and technical basis for designing the new generation imaging spectrometer.

高光谱信息在诸多军用和民用领域具有重要的应用需求。对于动态目标和场景，需要快速地获取其高光谱信息，形成高光谱视频。然而，传统成像技术普遍采用“时分复用”的思想进行光谱信息获取，因此无法完成光谱视频获取的任务，高分辨率光谱视频获取技术遭遇原理瓶颈。.本项目基于计算成像理论，探索光谱信息获取的新机理、新途径和新技术，提出新型的光谱视频获取系统。特别地，本项目拟设计编码感知光谱成像新模式，实现光谱信息的高维度高分辨率获取，挖掘高维信号的内在属性，设计鲁棒和快速的优化重建算法，实现高质量的光谱视频重建。研究内容包括：1）基于编码感知的时-空-谱联合高分辨率光谱信息获取；2）基于高维信息相关性的高质量光谱视频重建算法。本项目预期在高光谱视频获取技术上有突破，有助于提升我国卫星高分辨对地观测的应用水平，为新一代成像光谱仪的研制奠定理论和技术基础。

光谱信息在诸多军用和民用任务具有重要的应用需求。传统光谱成像技术采用“时分复用”的思想进行光谱信息获取，无法应用于动态场景和动态目标，高帧率光谱视频获取遭遇原理瓶颈。本项目在高帧率光谱视频获取方面取得了一定的进展。本项目针对高帧率光谱视频成像和高质量计算重建这两个关键问题展开研究，提出了基于混合相机的编码感知计算光谱视频获取新方法，开发了基于高维联合稀疏表示的重建算法，研制了原型系统，验证了方法的有效性，实现了时-空-谱联合高分辨率的光谱视频获取。本项目在TPAMI、TIP、CVPR、ICCV、Optics Express等重要期刊和会议上发表论文8篇，申请发明专利8项，培养硕士、博士研究生9人。本项目超额完成了项目预期目标，研究成果对遥感观测、目标检测、农业监测等任务具有潜在的应用价值。