航天探测多谱图像的特征级自主统一复原方法研究

The astronomical detection enters the time of multi-spectral, during the astronomical multi-spectral detection, target imaging will be affected by many factors, imaging environments is dynamic, and not be selected, image degradation caused by point spread function which is random, varying, complicated, and it is difficult to be described with mathematical models. The fast restoration method of astronomical images has been studied for long time by many domestic and foreign researchers, but up to now this scientific issue has not been attacked effectively. The project changes the existing idea of image restoration and tries to explore a new direction, in which the degradation process is deduced according to the change of visual significant features instead of the information of all pixels. With making research on new thinking of autonomous restoration on real images, an universal self-determination restoration method of real astronomical detecting images is proposed. The research contents include: the analysis of the change of visual significant features of image transition regions and the extraction method of image zero-cross point feature in real degraded images; the self-determination of support domain of point spread functions and the unified calculation model; the research of minimization constraints of second-order gradient with spatial correlation and its fast matrix construction method; some key technologies of self-determination restoration of astronomical detecting images and multi-DSP real-time processing method. Universal validity experiments are performed to test the effectiveness of the proposed algorithm. The universal self-determination restoration method will be applied in astronomical detection systems. The clearness degree and resolutions of real multi-spectral images will be enhanced, which will promote the high point development of astronomical detection technology in our country.

航天探测进入了多光谱时代，在航天多谱探测过程中，目标成像会受到多种因素的干扰，成像环境是动态的、不可选择的，引起图像退化的点扩散函数是随机的、变化的、复杂的，难以用数学来描述。航天探测图像的快速复原是国内外许多学者长期致力而未能有效解决的科学难题。本项目转变图像复原现有思想，探索新方向，依据图像视觉显著性特征的变化而不再是利用所有像素点信息来反演退化过程，开展不依赖于退化模式的统一复原新思维研究，提出航天探测图像的自主统一复原方法。研究内容包括：实际退化图像过渡区显著性特征的变化分析与零交叉点特征提取方法；点扩散函数域的自主确定及点扩散函数的统一计算模型；基于空间相关性二阶梯度极小约束及矩阵快速构造方法；航天探测图像自主复原的几个关键技术及多DSP实时处理方法研究。完成算法实验验证和测试，将算法集成在航天成像探测系统中，提高航天探测过程的图像清晰度和分辨率，推动我国航天探测技术的高端发展。

在航天多谱探测过程中，目标成像会受到多种因素的干扰，引起图像退化的点扩散函数是随机的、变化的、复杂的，很难用数学模型建模。航天探测图像的快速复原是国内外许多学者长期致力而未能有效解决的科学难题。本项目转变图像复原思想，探索新方向，依据图像视觉显著性特征的变化而不再是所有象素点信息来反演退化过程，开展不依赖于退化模式的统一复原新思维研究。对实际退化图像过渡区视觉显著性特征的变化进行了分析，提出了图像过渡区零交叉点特征提取方法，提出了基于模糊与畸变可区分性特征构造的湍流退化图像复原方法。研究了图像过渡区像素灰度值空间分布预测原理，建立了点扩散函数离散值计算的统一公式。提出了基于大梯度特征选择的点扩散函数估计方法和空变模糊图像的多点扩散函数估计方法，。提出了基于点扩散函数相关性偏差自主补偿的快速反卷积方法。提出了基于少量帧的交替迭代最大似然估计算法，研究了自适应的Richardson-Lucy自主复原方法，利用少量帧（3-5帧）湍流退化效应图像数据信息互补，可以估计各帧点扩散函数及目标图像。研究了基于微分连续的边界效应自适应消除方法，提出了基于超拉普拉斯先验特征的FFTW反卷积快速复原方法，消除了图像复原时潜在的的振铃效应，同时在速度上提升了几个数量级。分析了基于误差计算的评价参数存在不足和缺陷，提出了符合视觉的基于图像特征的梯度能量、能量集中度、图像模糊度等评价参数。研究了统一复原算法的DSP多核并行处理方法，完成了空中非合作目标模糊图像定位检测的多核DSP实时实现，采用TMS320C6678 DSP，实现了检测帧频为200帧/秒,正确检测率为95%以上。本项目针对航天多谱探测目标成像过程中所遇到的图像退化问题，提出了适合于航天探测过程的自主统一复原方法，为我国航天探测技术的高端发展提供了关键技术。发表高水平学术论文49篇，本项目有关内容作为相关成果主要创新点，获湖北省自然科学奖二等奖1项，教育部科技进步奖二等奖1项，获中国自动化学会科技进步二等奖奖1项。