面向雾霾天气的航空遥感观测中去雾霾及稳像方法研究

With the increasingly serious fog and haze weather in recent years, the research on fog and haze removal for the future of aviation observations has become crucial. Meanwhile, to improve the noise-signal ratio under fog and haze weather will lead to the remarkable increase of image motion, so it is of vital importance to take the image stabilization into account. The research on these two key technologies for the aforementioned fog and haze removal and image stabilization will be carried out in this project. To begin with, the existing methods lead to non-definite solution to the model for the haze image removal, which are though very efficient for the fog image. Additionally, the existing methods cannot effectively solve the problem of gray-level image restoration and so on, which is widely used in the fields of aeronautics and astronautics. Motivated by these challenges, the theory of guide filtering is employed into the research of fog and haze removal. Based on the different propagation mechanisms of light energy between fog and haze plus the distribution relationship between scene depth and fog and haze layer, the theoretical and technical difficulties in the aforementioned applications can all be solved. In the next place, although the “image type” method for image stabilization has obvious advantages in the aviation application, it is difficult to realize the real-time processing. In order to solve this problem, the concept of sparse representation is applied to image stabilization in this project. With this basic concept, the crucial sparse model of image stabilization is established by extracting the required feature information from the research of fog and haze removal, which is needed for image stabilization. In addition, an in-depth analysis of motion patterns of flight, scanning, roll, pitch, and yaw in the process of aviation remote sensing is also carried out, which solves the problem of high-speed realization of “image type” image stabilization by transforming the blind deblurring model to the non-blind one. This research is aimed to provide a theoretical basis and a technical method for the future high-resolution aviation remote sensing under fog and haze weather.

面对近年来日趋严峻的雾霾天气，去雾霾研究对未来航空观测至关重要。与此同时，雾霾天气下为提升信噪比而造成像移量显著增加，使稳像环节不可或缺。本项目拟对上述去雾霾及稳像两项关键技术开展研究。首先，针对现有方法处理“雾”图像十分有效而面向“霾”图像时模型无定解，且不能解决航空航天应用广泛的灰度影像修复等问题，将导引滤波理论应用于去雾霾研究，基于“雾”和“霾”对光能量的不同传播机理，结合场景深度与雾霾层的分布关系，解决面向上述应用的理论与技术难点；其次，针对有明显航空应用优势的“图像式”稳像难以实现实时处理，将稀疏表示思想引入稳像研究，基于去雾霾研究中对稳像所需特征信息的提取而建立关键的稀疏稳像模型，并对航空遥感中飞行、扫描、横滚、俯仰、偏航运动模式进行深入剖析，将盲去模糊模型转化为非盲去模糊模型，解决“图像式”稳像的高速实现问题。本研究可为未来雾霾天气下高分辨率航空遥感提供理论基础和技术手段。

本项目瞄准近年来不断增长的雾霾天气对航拍/遥感图像清晰化带来的挑战，研究了该清晰化技术背后涉及的三项科学问题：图像去雾霾、图像稳像、图像超分辨率。具体研究内容涉及：. （1）图像去雾霾及其DSP平台应用研究：①针对经典大气物理退化模型方法处理雾图像十分有效，而面向霾图像时模型无定解问题，开展了图像主干边缘提取、增强型去霾恢复和霾粒子散射研究，所提图像去霾方法可以较好地实现对霾图像的清晰化恢复。②针对夜晚天空光弱且存在复杂人造光源干扰，使得夜晚非统一大气光值的获取异常困难，从而导致整个大气物理退化模型无解，开展了图像颜色迁移和增强型去雾研究，在此基础上提出了一种针对夜晚雾图像的清晰化恢复方法。③针对目前研究较少的灰度图像去雾问题，融合了基于视觉增强去雾方法的简单便捷，以及大气物理退化模型方法能够更好地补偿退化过程中信息损失，从而提出了一种相应的去雾霾新方法。④开展了基于嵌入式DSP的图像去雾霾应用研究，搭建了面向DSP的图像去雾霾平台，设计了一种相应的简单可靠的图像清晰化算法，并进行了相关算法的仿真验证工作。. （2）航拍图像的DSP平台稳像研究：针对航拍运动导致的运动像移问题，开展了航拍运动模式的理论研究，并对该运动模式进行了数学建模。在已知飞行参数的情况下，可将此复杂的航拍运动像移问题转化成模糊参数已知的非盲去模糊问题。此外，搭建了面向DSP的图像稳像平台，开展了相关算法的仿真验证工作。. （3）遥感图像超分辨率研究：针对遥感图像的场景类型众多，提出了图像超分辨率的预分类选择策略，有效减少了训练集的规模以提升算法整体效率。同时由于遥感图像轮廓特征的复杂性，在组成用于匹配和分组的关键特征向量方面进行了针对性的改进，从而提出了一种适合遥感图像的小训练集超分辨率方法。. 本研究可为未来雾霾天气下高分辨率航空观测及其高速实现，提供一定的理论和技术借鉴。