光学遥感图像质量提升及评价技术研究

发布时间：2018-09-17 13:14

【摘要】：光学遥感成像是一种重要的信息探测手段,它依靠接收目标自身辐射来进行信息获取,在城市规划、环境监测、资源勘探等民用领域以及侦察预警、目标识别等军事领域有着重要的应用价值。在遥感成像链路过程中,遥感图像会受到各种退化因素的影响,包括目标辐射、大气扰动、光学成像系统、电子信号转换、卫星平台颤振等,这些退化因素都会不同程度的引起遥感图像的退化,导致图像质量的下降,影响了图像的后续处理,严重制约了遥感图像的应用。利用在轨设备优化手段耗资巨大、周期长,并且效果有限。在已有硬件条件下,如何通过软件处理的手段,提出遥感图像质量改善提升的技术方法,一直是人们关注的重点。此外,对于遥感图像建立客观完善的图像质量定量评价标准,为卫星参数的调整和提升算法提供指导,也具重要的意义。本文通过对遥感成像链路的退化因素进行分析,结合视觉显著特性,重点研究了遥感图噪声水平估计和噪声图像质量评价方法,研究并提出了多种遥感图像质量改善提升的处理算法,并针对月面模糊图像质量提升的任务,设计了针对性的图像复原方案和评价方法,满足了任务需求。分析了光学遥感图像成像链路模型,建立了地物模型、大气传输辐射模型、相机模型、轨道姿态模型。讨论和分析了光学遥感成像主要的退化因素,针对每种类型的退化因素,分析了图像退化产生的原因,建立了对应的数学模型。在遥感图像质量提升技术方面开展了深入研究。针对单幅图像复原问题,提出了一种基于自适应参考图的改进RL非盲复原方法,通过引入两个阶段的自适应参考图,更为准确地计算图像边缘信息,在振铃抑制和细节保持上有很好的表现；利用同一目标场景的长曝光模糊图像和短曝光噪声图像,提出了一种基于长短曝光图像对的图像复原算法,引入局部约束掩膜矩阵和显著性权重图,有效增强了复原结果的细节,同时抑制了噪声和振铃效应；针对遥感图像去噪,提出了一种基于非下采样轮廓波变换和相对总变分约束的遥感图像去噪方法；为了能够综合利用不同波段的遥感图像信息,提出了一种基于多源图像融合的遥感图像信息量提升方法,结合显著性提取和多尺度分解技术开展了多波段图像融合研究,融合结果有效保留和增强了不同波段图像的信息,显著提升了图像信息量。总结了典型的图像质量评价方法,介绍了针对光学遥感图像的几种特定评价参数,分析了几种遥感图像信噪比计算方法的特点和适用范围。为了能够更好的实现图像信号成分和噪声成分的有效分离,构建了图像信号仿射重建模型,将噪声图分割为若干大小相当的图像块,求解仿射重建模型得到信号图。在此基础上,提出了一种基于噪声水平累积的图像噪声和信噪比估计方法,计算出各图像分块的强度-噪声散点分布图,并将图像强度各区间噪声标准差值加权求和得到噪声水平累计值和信噪比累积指标数值。结合人眼视觉对比敏感函数和仿射重建模型,提出了一种针对噪声图像的无参考质量评价方法,评价算法对不同数据库都具有很好的主客观一致性和准确性。开展了大像移月面模糊图像复原技术研究,分析了大像移月面成像的任务需求和图像特点,针对性地设计了两种图像复原算法,制定了整体的实验方案。分别开展了理论仿真实验和地面实拍仿真实验,对仿真实验复原前后的图像进行了质量评价比较,验证了两种复原算法的有效性。针对不同像移量的实拍月面模糊图,进行了复原实验。分析了原始模糊图像中的压缩块效应,验证了复原结果能够有效的去除块效应现象。设计了一种无参考的综合图像质量提升评价方法,对于实拍月面图复原结果进行了图像质量的评价比较,验证了复原算法能够显著提升图像质量。
[Abstract]:Optical remote sensing imaging is an important means of information detection. It depends on receiving the radiation of the target itself for information acquisition. It has important application value in civil areas such as urban planning, environmental monitoring, resource exploration, reconnaissance and early warning, target recognition and other military fields. Degradation factors, including target radiation, atmospheric disturbance, optical imaging system, electronic signal conversion, satellite platform flutter, etc., will cause the degradation of remote sensing images to varying degrees, resulting in the degradation of image quality, affecting the subsequent processing of images, seriously restricting the application of remote sensing images. Under the existing hardware conditions, how to improve the quality of remote sensing images by means of software processing has always been the focus of attention. In addition, the establishment of objective and perfect quantitative evaluation criteria for remote sensing images, and the adjustment of satellite parameters have been the focus of attention. In this paper, the deterioration factors of remote sensing imaging links are analyzed, and combined with visual saliency characteristics, the noise level estimation and noise image quality evaluation methods of remote sensing images are studied. Various processing algorithms for improving remote sensing image quality are studied and put forward, and the lunar model is considered. In order to improve the quality of blurred image, a pertinent image restoration scheme and evaluation method are designed to meet the task requirements. The imaging link model of optical remote sensing image is analyzed, and the terrain model, atmospheric transmission radiation model, camera model and orbit attitude model are established. Each type of degradation factor is analyzed, the causes of image degradation are analyzed, and the corresponding mathematical model is established. The technology of remote sensing image quality improvement is studied in depth. Aiming at the problem of single image restoration, an improved RL non-blind restoration method based on adaptive reference is proposed, which introduces two stages of adaptive reference. Image edge information is calculated more accurately, which has good performance in ringing suppression and detail preservation. Using the long exposure blurred image and the short exposure noise image of the same target scene, an image restoration algorithm based on the long exposure image pair and the short exposure image pair is proposed. Local constraint mask matrix and saliency weight map are introduced to enhance the image effectively. For remotely sensed image denoising, a method of remotely sensed image denoising based on non-downsampled contour wave transform and relative total variational constraints is proposed, and a remote sensing image fusion method based on multi-source image fusion is proposed to synthetically utilize the remote sensing image information of different bands. Multi-band image fusion based on saliency extraction and multi-scale decomposition is studied. The fusion results effectively preserve and enhance the information of different bands of images, and significantly improve the image information. Typical image quality evaluation methods are summarized. Several special evaluation methods for optical remote sensing images are introduced. In order to realize the effective separation of image signal components and noise components, the affine reconstruction model of image signal is constructed. The noise image is divided into several image blocks of the same size, and the signal graph is obtained by solving the affine reconstruction model. On the basis of this, an image noise and signal-to-noise ratio estimation method based on noise level accumulation is proposed, and the intensity-to-noise scatter distribution maps of each image block are calculated. The cumulative value of noise level and the cumulative index value of signal-to-noise ratio are obtained by weighting the standard deviation of noise in each region of image intensity. Affine reconstruction model is used to propose a non-reference quality evaluation method for noisy images. The evaluation algorithm has good subjective and objective consistency and accuracy for different databases. Two kinds of image restoration algorithms are proposed and the whole experiment scheme is worked out. The theoretical simulation experiment and the ground real-time simulation experiment are carried out respectively. The quality of the images before and after restoration is evaluated and compared. The effectiveness of the two restoration algorithms is verified. The compressed blocking effect in the original blurred image verifies that the restoration results can effectively remove the blocking effect. A comprehensive image quality improvement evaluation method without reference is designed. The image quality of the reconstructed results from the real lunar images is evaluated and compared. The results show that the restoration algorithm can significantly improve the image quality.
【学位授予单位】：浙江大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP751

【相似文献】