【作者】 王荣彬；

【导师】 李平湘；

【作者基本信息】 武汉大学 ， 摄影测量与遥感， 2014， 博士

【摘要】 遥感数据获取过程中经常受到各种因素的干扰,导致观测影像质量降低,为后续的数据应用带来了诸多困难。为了缓解以上问题,一方面需要对遥感数据进行质量评价,在现有条件下选择质量更优的遥感影像,同时过滤质量太差而无法使用的数据；另一方面,在没有高质量影像的条件下,通常需要对低质量的遥感影像进行复原处理,以提升影像的应用潜力。因此,影像评价与影像复原是遥感信息处理与应用中的重要基础环节。虽然遥感影像评价与复原问题在学术与应用领域一直广受关注,但现有研究仍然不能满足处理与应用需求。如何考虑多种辐射指标对遥感影像进行综合评价,以及如何针对不同影像的特征进行自适应复原,是当前需要解决的主要难题。本论文针对以上问题,围绕遥感影像的评价与复原展开了理论与方法研究,具体研究内容包括：(1)针对现有遥感影像噪声与调制传递函数MTF评价方法自动化程度不足、需要人工干预的问题,提出最优评价区域的自动选取与评价方法。在噪声评价方面,提出一种迭代优化的匀质区选择方法,可以解决传统方法中影像匀质区自动选择困难的难题,并通过设计基于卷积运算的评价指标实现对噪声水平的稳健估计。在影像MTF评价方面,针对传统方法往往需要人工选取刃边的问题,提出了一种自动检测影像边缘并提取最优刃边的方法,从而实现MTF的自动计算。实验结果表明,本文所提出的自动化评价方法准确有效,可以极大地提高遥感影像噪声和模糊评价的自动化程度。(2)在系统总结、改进现有遥感影像评价指标的基础上,提出一套遥感影像质量综合评价的方法。综合考虑了灰度分布、信息量、清晰度、分辨率、噪声、云量、无效像元等指标,并融入了基于参考影像的评价指标；在大量数据测试的基础上,确定了每项指标对应优、良、中、差的最优阈值；进一步从模糊理论出发,通过构建模糊评价矩阵,对多种单指标评价结果进行综合考虑,建立了对遥感影像辐射质量进行综合评定的方法。实验结果表明,本文提出的综合评价方法能准确评价遥感影像的综合质量,与人眼主观评价具有良好的一致性。(3)提出一种自适应非局部正则化的遥感影像噪声去除方法。针对传统去噪方法仅利用邻域信息、不能高效去除噪声的缺点,本文在非局部计算框架下开展去噪方法的研究,并重点解决去噪参数的自适应选取问题。为了对不同区域施加不同的去噪强度,通过计算局部噪声强度和标准差,并以此构建非局部模型的权值函数,从而建立白适应的非局部总变分去噪模型。实验结果表明该方法能针对影像的不同结构区域自适应调节去噪强度,可以在去除遥感影像噪声的同时,有效保护影像的边缘、纹理等细节信息。(4)提出一种自适应交替迭代的遥感影像盲去模糊方法。基于最大后验概率估计理论框架,建立对遥感影像与模糊函数进行联合求解的去模糊模型,并采用交替迭代的方法对影像和模糊函数进行求解；在影像求解的迭代过程中,充分利用中间复原影像求解数据一致性项和正则化项的函数值,并通过建立相应的求解准则,自适应求解影像与模糊函数的正则化参数。实验结果表明本文提出的盲复原方法,能够较为准确的估计模糊函数,并根据不同的影像特征实现自适应的影像去模糊处理,可有效提高处理精度与效率。更多还原

【Abstract】 Due to the disturbances of various factors in the acquisition process, the remotely sensed images may have lower quality, which brings a good many difficulties for the subsequent applications. In order to relieve this problem, it is needed to evaluate the image quality to select remote sensing data with better quality, and to filter the poor-quality images. On the other hand, it is needed to restore the images for potential improvement in the condition of no high-quality images existing. Therefore, image evaluation and image restoration are important steps in remote sensing information processing and application.Although the problems of image evaluation and image restoration have obtained wide attentions in the academic and application fields, the existing methods can not solve the difficulties perfectly. How to consider multiple factors for a comprehensive evaluation and how to consider the features of different images for an adaptive restoration, are the main difficulties to be solved. This thesis aims to these problems, and develop theories and methods for the evaluation and restoration of remote sensing images. The main contents are as follows:(1) The current evaluation methods of noise and Modulation Transfer Function (MTF) always have low degree of automation, and need manual intervention. Aiming at this problem, this thesis proposes automatic selection methods of the optimal evaluation regions. As for the noise assessment, an iterative optimization method for the selection of homogeneous area is proposed. This method may solve the difficulties in traditional methods and realize robust noise estimation by using convolution based evaluation index. As for the MTF assessment, in order to solve the problem of knife edge searching in MTF computing, an automatic searching and calculating method is presented. The experimental results show that the automatic evaluation methods are accurate and effective, and can greatly improve the degree of automation of remote sensing noise and MTF evaluation.(2) This thesis gives a systematically summary of the current image evaluation indices, and makes several improvements. Base on these, a comprehensive evaluation method for remote sensing images is proposed by fully consider the indices of grey distribution, information quantity, definition, resolution, noise, cloud, and invalid pixels. The reference based method is also considered. After a plenary test of remote sensing data, the optimal thresholds of different grades (Excellent, Good, Fair and Poor) are determined. In the framework of fuzzy mathematics theory, the fuzzy evaluation matrix is firstly built, and the results of the evaluation of a variety of single indicators are considered comprehensively, then a overall qualitative evaluation score is given. The experimental results show that this comprehensive evaluation method can accurately evaluate the quality of remote sensing images with good consistency with the human eye evaluation.(3) This thesis proposes an adaptive non-local regularization denoising method for remotely sensed images. Traditional denoising methods only use the neighboring information, leading to that the noise can not be effectively removed. This research develop new denoising method in the framework of non-local computation, and solve the problem of adaptive selection of filtering parameters. In this thesis, based on the non-local total variation denoising model, an adaptive filtering parameter determination approach is presented by computing the local noise intensity and standard deviation. In the iterative process, the parameters are adjusted automatically according to the noise level of the iteration results. Experimental results show that this filtering parameter determination method have the performance to adaptively adjusted the denoising strength according to different structures and can protect the edge and texture information while removing image noise.(4) This thesis proposes an adaptively alternative iteration method for blind deblurring of remotely sensed images. This method is based on the framework of maximum a posteriori (MAP), and jointly solves the image and blur function by alternative iteration procedure. During the iteration process, make full use of the partially solved image to compute the values of data consistency and regularization terms. The two regularization parameters of image and blur are adaptively solved by designing corresponding solution functions. The experimental results show that the proposed blind deblurring method has the performance to give accurate estimation of blurring function, and realize adaptive restoration according to image features. It can improve the processing accuracy and efficiency.更多还原