高速运动条件下多因素耦合降质图像的超分辨重建

With the intensification of human activities in outer space, situational awareness of space environment is urgently needed. During the situational awareness, high resolution and high-definition space target image are supports for various applications. Therefore, super-resolution reconstruction of space target image is imperative. However, due to the high-speed dynamic and multi-factor coupling of spatial image imaging in outer space, difficulties exist as the sequence sub-pixel displacement estimation, the description of the degradation in dynamic condition as well as the exploitation of the sequence complementary information. This project is proposed for the following purpose. In view of the difficulty of sequence sub-pixel displacement estimation, a dynamic inverse filter method for implicit sub-pixel displacement expression is proposed. In view of the difficulty of degradation description in dynamic condition, a resolution mapping network for unpaired data training is proposed. In view of the difficulty of the exploitation of the sequence complementary information, an intra-frame/inter-frame visual attention mechanism based method is proposed, aiming to break through the restriction due to multi-factor coupled low-resolution degraded image under the high speed dynamic condition of the space platform. This project will give a solid support to the excellent use of the potential of the situation awareness system based on the space platform. This research will play an important role in improving the capability of space exploration and space surveillance.

随着人类在地球外层空间活动的加剧，迫切需要对空间环境进行态势感知。在态势感知过程中，高分辨率和高清晰度的空间目标图像是各种后续应用的支撑，对空间目标图像进行超分辨重建势在必行，但由于空间图像成像的高速运动和多因素耦合特性，存在序列亚像素位移估计难，动态条件下降质过程描述难，序列图像有用性和互补性利用难等问题，本项目拟针对序列亚像素位移估计难，提出动态逆滤波参数亚像素位移信息隐式表达方法；针对动态条件下降质过程描述难，提出非成对数据的分辨率映射网络建模方法；针对序列有用性和互补性利用难，提出基于帧内/帧间视觉注意机制的序列互补性模型，最终突破空间平台高速运动条件下多因素耦合降质低分辨率图像对后续处理应用的制约，充分发挥空间平台态势感知系统潜能。研究成果将对提高航天探测及空间监视能力具有重要作用。

随着人类在地球外层空间活动的加剧，迫切需要对空间环境进行态势感知。在态势感知过程中，高分辨率和高清晰度的空间目标图像是各种后续应用的支撑，对空间目标图像进行超分辨重建势在必行，但由于空间图像成像的高速运动和多因素耦合特性，存在序列亚像素位移估计难，动态条件下降质过程描述难，序列图像有用性和互补性利用难等问题，本项目拟针对序列亚像素位移估计难，动态条件下降质过程描述难，序列有用性和互补性利用难等问题，提出了基于亚像素位移运动相关性模块的序列图像超分辨方法，提高了动态条件下的亚像素位移的估计精度和利用效率。提出基于动态结构先验的耦合降质图像复原方法，提高噪声降质影响下的图像退化模糊核估计精度。提出了多尺度的非成对模糊低分辨耦合降质图像的特征融合方法，提高动态模糊条件下的图像超分辨重建效果。提出了基于域自适应模糊核估计的深度去模糊方法，提高了仿真数据训练结果在真实数据上的性能，减小了数据域鸿沟。基于通道注意力和空间注意力的有用性和互补性描述方法，提高了图像序列的多帧融合性能。并将理论应用于实践，提出了一种基于空间平台的高分辨率自适应曝光成像方法，将互补理论和方法应用于相机曝光模式指导，提高了相机的信息采集效率。项目研究成果对充分发挥空间平台态势感知系统潜能，对提高航天探测及空间监视能力具有重要作用。