【作者】 张志龙；

【导师】 沈振康；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2005， 博士

【摘要】 遥感图像在军事侦察、精确打击和民用方面都有重要的应用,因此开展遥感图像的特征提取和目标识别工作具有实际意义和应用前景。本文以团块目标、阵列目标和港口目标作为研究对象,以空间关系作为切入点,系统研究了以上目标的特征提取和目标识别的方法。 论文的第一章是绪论,介绍了课题的研究背景、遥感图像特征提取和目标识别的主要内容和发展现状以及本文的主要工作和创新点。 第二章研究了纹理特征的提取和纹理鉴别性能的评价问题。本章的工作主要包括:(1)回顾了常用的八种纹理特征提取方法以及前人在纹理鉴别性能评价方面的主要工作;(2)提出一种新的基于局部沃尔什变换(LWT)的纹理特征提取方法,给出了LWT变换的定义并对其进行了空域推广,分析了LWT系数的统计特性及其各阶矩的纹理鉴别性能,进一步选取具有较好鉴别性能的二阶矩作为纹理特征;(3)从纹理鉴别性能、纹理分割效果和计算量三个方面,将本文提出的方法与其它八种方法进行综合比较,证实了本文方法优于其它方法;(4)结合全色遥感图像中海域的纹理和结构特性,提出了一种基于LWT变换的海域分割算法。 第三章研究了团块目标的检测问题,提出了一种基于视觉注意模型的团块目标检测方法。该方法根据团块目标与背景在多种特征、多个尺度上存在的差异,利用视觉注意模型确定目标位置,并根据尺度显著性准则提取目标区域。改进了视觉注意模型中显著图的计算过程,提高了视觉显著图的计算速度和空间分辨率,使之更适合于团块目标检测的实际需要。实验结果表明,基于视觉注意模型的团块目标检测方法对图像畸变和目标变化具有较强的适应性,对较复杂背景中出现的各种团块目标取得了较好的检测效果。 第四章研究了阵列目标的特征提取和目标识别问题,本章的工作主要包括:(1)根据阵列目标的结构特性,确定了进行特征提取和目标识别的具体思路;(2)提出一种结合线度和长度约束的局部化空间关系基元选取算法;(3)提出了基于模糊理论的空间关系基元之间的规则性测度;(4)建立了以局部化空间关系基元为顶点的场景结构图,证明了场景结构图的邻接矩阵的若干特性,根据这些特性设计出一种快速的谱图划分算法——“迭代谱图划分算法”;(5)研究了油库、导弹阵地、高炮阵地等阵列目标的识别问题,取得了满意的识别效果。 第五章沿着海岸线形状分析的思路研究了港口的特征提取和识别问题,本章的工作主要包括:(1)引入新的张力和外力计算方法来改善活动轮廓模型的性能,在此基础上研究了基于活动轮廓模型的海岸线高精度提取技术;(2)提出一种基于特征聚类的内港岸线分割算法;(4)提出一种基于特征点松弛匹配的特定港口识别算法。更多还原

【Abstract】 Remote sensing imagery has great importance for military reconnaissance, precision attack and civil activities, so it has good application prospect to study feature extraction and target recognition methods of remote sensing imagery. This dissertation investigate the feature extraction and target recognition methods for blob targets, array targets and ports, and focus our research work mainly on their characters in structure and spatial relationship.Chapter 1 is the preface of this dissertation, which introduces the background knowledge, reviews the main content and the state of arts development of the feature extraction and target recognition in remote sensing imagery, and summarizes the central research work and innovative points in the dissertation.Chapter 2 studies the extraction and selection of texture features and the evaluation of texture feature discrimination performance (TFDP). The central work of this chapter includes: (1)reviewing the eight kinds of texture feature extraction methods currently used and the previous work in the evaluation of TFDP; (2)presenting a new texture feature extraction method using Local Walsh Transform (LWT), giving the definition of LWT and generalizing it in spatial domain, analyzing the statistic property of LWT coefficients, examining the TFDP of the central moments of LWT coefficients, and selecting the 2nd,4th,6th order moments which have better TFDP as texture features; (3) comparing our texture feature extraction method with the other eight methods in TFDP, texture image segmentation effect and computational complexity, which indicates that our method has the best performance; (4)presenting a new method based on LWT to segment sea area in optical remote sensing imagery, which integrates the characters of sea area in texture and structure.Chapter 3 studies the detection of blob targets. This chapter presents a novel algorithm based on visual attention model to detect blob targets in optical and infrared images. The blob targets have multi-feature and multi-scale difference as compared with their backgrounds, which is used by the visual attention model to locate the blob targets in the scenes. Farther on, scale salience is used to extract the region of the blob targets. The salience map computation procedure is modified. As a result, the salience map has higher spatial resolution and lower computation complexity, which make it more suitable to detection blob targets. Experiments reveal that our algorithm is immune to the distortion of images and targets, it can detect several classes of blob targets from scenes with considerable clutters.Chapter 4 studies the feature extraction and target recognition methods of array targets. The central work of this chapter includes: (1)presenting a practical scheme for feature extraction and target recognition based on the characters of array targets; (2) giving the definition of spatial relationship primitive(SRP), and presenting an effectiveSRP selection algorithm to make the full graph sparse; (3) presenting a spatial relationship regularity measure for arbitrary two SRPs based on Fuzzy theory; (4) establishing a full graph, namely Scene Structure Graph (SSG), and testifying some important properties of the adjacency matrix of the SSG, designing a fast spectral graph partitioning algorithm, namely Iterative Spectral Graph Partitioning Algorithm; (5) studying the recognition of oil tanks, missile positions and flack positions in remote sensing imagery using the algorithms presented above, which acquires promising recognition results.Chapter 5 studies the feature extraction and target recognition methods of port. This chapter treats port as a segment of coastline with special structure, detects and recognizes it by the way of analyzing the shape of coastline. The central works of this chapter includes: (1)bringing forward a new elasticity force computing method and a new extent force computing method to improve the performance of the traditional active contour model, and designing a high accurate coastline detection method using the improved active contour model; (2)giving out an effective inner port coastline extraction algorithm based on eigencluster technology; (3)presenting a port recognition algorithm using feature points relaxation matching method.Chapter 6 summarizes the dissertation and brings forward some problems which need further investigating.更多还原