基于稀疏表示的高分辨率遥感影像人工目标识别算法研究

发布时间：2018-11-14 16:26

【摘要】：随着航天技术和传感器技术的迅猛发展,卫星遥感影像得到了广泛的应用,尤其是高空间分辨率遥感影像已经成为了民用和军用等方面必不可少的数据来源。目前,高分辨率遥感影像在众多领域都得到了广泛应用并取得了一定的应用成果,如城市规划与环境监测、自然灾害监测与应急处理、对地观测系统、军事监控与侦查、土地利用及适应性评价等。高分辨率遥感影像中地物的空间信息较为完整、清晰,能够为上述应用提供海量的数据信息。但是,其海量的数据信息、地物形状、空间结构及复杂的结构特征给遥感数据的自动化、智能化处理带来了新的挑战,传统的遥感图像处理技术已经远远不能满足当前高分遥感影像的应用需求,因此,探求新的研究方法尤为重要。针对此问题,本文引入信号的稀疏表示理论,对稀疏表示理论与低秩表示理论进行了系统深入地研究和探讨。本文的主要贡献如下:(1)深入学习和探讨稀疏表示理论,根据生物视觉的研究,对高分辨率影像的稀疏性进行了探究和讨论,将稀疏表示理论应用到高空间分辨率遥感影像上,提出了基于稀疏表示的高分辨率遥感影像人工目标识别方法,并通过Uc Merced和Wh U两个数据集的实验验证了所提出方法的有效性。(2)考虑遥感影像中云雾的干扰,导致图像质量退化,影响后续图像处理的应用效果。根据遥感图像中被薄云污染的数据具有过度缓慢、分布均匀、空间纹理结构自相关性较强的特点,据此对薄云信息做出低秩性的假设。在此基础上提出了一种基于低秩矩阵分解的遥感影像薄云去除方法,对受薄云污染的遥感影像进行了有效地去云预处理,为后续的目标识别做准备,实验验证了该方法的有效性。(3)综合考虑人工目标的稀疏性和低秩性,联合低秩矩阵分解理论,提出了一种新的目标识别方法,该方法首先通过对遥感图像进行低秩矩阵分解,得到具有低秩性和稀疏性的两部分信息,然后通过K-SVD算法分别对这两部分信息进行字典学习,学习后将其联合共同构建稀疏表示的过完备字典,最后通过稀疏表示的求解算法求解出待分类的目标在过完备字典上稀疏系数,并根据稀疏系数最大准则对目标进行分类识别。同理,在Uc Merced和Wh U两个数据集上进行了相应的实验,实验结果显示本文所提出方法的有效性。
[Abstract]:With the rapid development of space technology and sensor technology, satellite remote sensing image has been widely used, especially high spatial resolution remote sensing image has become a necessary data source for civil and military. At present, high resolution remote sensing images have been widely used in many fields, such as urban planning and environmental monitoring, natural disaster monitoring and emergency treatment, Earth observation system, military monitoring and reconnaissance. Land use and adaptability evaluation. The spatial information of ground objects in high resolution remote sensing images is relatively complete and clear, which can provide massive data information for the above applications. However, its massive data information, ground object shape, spatial structure and complex structural features bring new challenges to the automation and intelligent processing of remote sensing data. The traditional remote sensing image processing technology can not meet the needs of the application of high score remote sensing image. Therefore, it is very important to explore new research methods. To solve this problem, the sparse representation theory of signals is introduced in this paper, and the sparse representation theory and the low rank representation theory are systematically studied and discussed. The main contributions of this paper are as follows: (1) the sparse representation theory is deeply studied and discussed. According to the research of biological vision, the sparsity of high-resolution images is discussed. The sparse representation theory is applied to high spatial resolution remote sensing images, and a method of artificial target recognition for high resolution remote sensing images based on sparse representation is proposed. The validity of the proposed method is verified by the experiments of Uc Merced and Wh U data sets. (2) considering the cloud interference in remote sensing images, the image quality is degraded and the application effect of subsequent image processing is affected. According to the characteristics of the data contaminated by thin cloud in remote sensing images such as slow distribution uniform distribution and strong autocorrelation of spatial texture structure the assumption of low rank of thin cloud information is made. On this basis, a method of thin-cloud removal from remote sensing image based on low-rank matrix decomposition is proposed, and the cloud removal of remote sensing image contaminated by thin cloud is carried out effectively, so as to prepare for the subsequent target recognition. Experimental results show that the method is effective. (3) considering the sparsity and low rank of artificial targets, a new target recognition method is proposed by combining the theory of low rank matrix decomposition. In this method, two parts of information with low rank and sparsity are obtained by decomposing the low rank matrix of remote sensing image, and then the two parts of information are studied by K-SVD algorithm. After learning, the over-complete dictionary of sparse representation is constructed jointly. Finally, the sparse coefficients of the target to be classified in the over-complete dictionary are solved by the algorithm of sparse representation, and the targets are classified and identified according to the maximum criterion of sparse coefficient. In the same way, experiments are carried out on two data sets, Uc Merced and Wh U, and the experimental results show the effectiveness of the proposed method.
【学位授予单位】：云南师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【相似文献】