基于稀疏性和局部显著性的气动光学效应图像校正方法研究

As an aircraft with an infrared (IR) imaging detection system flies at high speed, both aerothermal radiation effects and aero-optical transmission effects can lower the signal-to-noise ratio and image quality of the IR imaging detection system, and seriously affect the sensitivity and operating range of the system, which have become bottleneck problems in the development of imaging terminal guidance, navigation guidance, and high-speed reconnaissance systems. In this project, we explore the sparse characteristics of real IR scene and experimental images, design the non-parametric correction model based on high-order regularization and residue correction model based on weighted nuclear norm of aerothermal radiation, which effectively remove multi-scale aerothermal radiation noise and system noise of IR images from the aero-heating environment. To effectively solve the spatial-invariant and variant blur induced by aero-optical effects, we also investigate the blur estimation method based on sparse representation, use the salient structure characteristics to select the key image patch, efficiently and effectively estimate the point spread function and build an optimization correction model to tackle the error of the point spread function under the guidance of the mechanism model of aero-optical effects. Meanwhile, we propose the feature-based methods of image quality assessment of aero-optical effects to measure the effectiveness of correction models. Finally, this project can provide the technical support for the application of aero-optical correction on the IR imaging detection systems of high-speed aircrafts.

带有红外成像探测系统的飞行器在大气层内高速飞行时，气动热辐射效应和气动光学传输效应会降低成像探测系统的图像质量和信噪比，严重影响其灵敏度和作用距离，已成为高速飞行器成像末制导、导航制导与高速侦察系统的瓶颈问题。本项目从真实场景红外图像和风洞试验图像中挖掘先验知识，设计基于高阶正则化的气动热辐射图像非参数化校正模型和基于加权核范数最小化的残差校正模型，以去除气动热环境中不同尺度热辐射噪声与随机噪声。在气动光学传输效应机理模型的指导下研究基于稀疏表示的图像模糊度估计方法，基于显著边缘特征选择利于点扩展函数估计的最佳区域，实现点扩展函数的快速有效估计，构建点扩展函数误差条件下的优化校正模型，能有效解决气动光学传输效应所导致的空不变与空变模糊问题。研究基于图像特征的气动光学效应图像质量评价方法，以验证校正模型的有效性。本项目为气动光学效应图像校正技术在高速飞行器成像探测系统上的应用提供技术支撑。

气动光学效应图像校正是高速飞行环境中光学成像系统面临的重要问题，其研究具有重要的理论和应用价值。本项目围绕气动光学效应图像校正模型与方法开展研究。首先,在特性先验知识的指导下构建了气动热辐射效应图像非参数化校正模型，有效降低校正算法的计算复杂度；并提出了基于深度卷积神经网络的气动热辐射效应图像校正模型，引入残差学习策略有效提升训练效率，进一步改善算法的计算效率；其次，针对复杂环境中红外图像的背景强度偏移、条带噪声与随机噪声等多重干扰，提出了基于全局和局部先验约束的红外图像非均性校正方法，与传统的序贯执行处理相比，所提出的红外非均性校正方法具有更低的残余误差；最后，构建基于L2-Lp范数最小化的红外图像与可见光图像融合与去噪统一模型，显著提升融合图像质量，为气动光学效应图像校正提供新的途径。本项目的创新性在于通过特性知识建模和深度学习构建复杂环境中红外图像背景干扰抑制模型，结合多谱图像融合技术，全面提升复杂环境中信息获取。本项目的研究成果可用于改善复杂环境中光学成像系统的性能。