线状目标特征提取及其机场目标识别中的应用

【作者】 眭新光；

【导师】 平西建；

【作者基本信息】 中国人民解放军信息工程大学 ， 信号与信息处理， 2002， 硕士

【摘要】 随着遥感技术在资源勘探、全球定位系统等民用领域以及机场、港口目标识别等军事领域的广泛应用，对遥感影像目标的有效提取与识别手段逐渐受到了人们的重视，这对遥感影像的应用以及遥感技术本身的发展都很重要。作为遥感技术的一个重要部分，遥感影像目标的计算机自动识别是一个热门研究课题，也是计算机视觉中的一项关键技术，一直受到了广大研究人员的重视，并吸引了许多人对其进行广泛而深入的研究。然而，作为一个兴起不久的研究课题和计算机视觉的难点问题，遥感影像目标的计算机自动识别目前还存在着相当一些问题有待于进一步的研究和解决。 形状是目标的一种本质特征，基于形状的目标识别是遥感影像目标的计算机自动识别中一种重要的技术方法。在各种形状的目标中，线状目标占有很大一部分比例，因而研究线状目标的自动识别具有很重要的理论意义和实用价值。 本文分析了形状作为目标的一种固有特征的各种表达方法，并从基于区域和基于边界两个方面讨论了形状的各种描述方法。同时根据目标识别的特征选择原则，从遥感影像许多重要目标表现为线性形状的特点及这些目标识别的实际需求出发，提出了以目标边缘信息作为反映图象内容的一个特征，研究了基于线状目标的特征提取和识别的方法。 本文研究遥感影像线状目标的特征提取和机场目标计算机自动识别方法。在讨论遥感影像的基础知识和分析遥感影像数据中的干扰因素的基础上，从目标识别的要求研究了适合于进行形状分析和描述以及边缘特征提取的各种预处理技术。在形状分析中，通过曲线分割、分段识别以及相同特征的合并来提取形状特征基元，即弧线段和直线段，提出一种基于直线段的目标描述和识别方法。由于直线段反映了目标的直线状结构特性，同时满足平移、旋转和尺度不变性的三个属性，这用来描述机场跑道的形状特征有效。通过用直线段来描述机场跑道的形状，以直线段的长度来区分不同的目标，可以达到识别机场目标的目的。 文章最后根据形状分析的成果和遥感影像实际应用的要求设计了一个基于直线性形状的识别试验系统。经过直方图变换、图象增强、图象平滑、图象预分割、边缘提取和细化后，得到了清晰的边缘信息图象，通过特征基元的提取和识别，用直线段来描述直线状目标，而用直线段的长度来作为目标定量特征的描述，可以实现遥感影像机场目标的计算机识别。实验结果证明了系统的有效性。更多还原

【Abstract】 Along with the widely using of remote sensing technology in civil and military fields, such as resource reconnoitering, the GPS, recognition of airports and naval ports, the efficient extraction and recognition of the remote sensing image objects has been attached great importance to, which is very important for the application of remote sensing images as well as for the development of remote sensing technology itself. As plays an important part in the remote sensing technology, the auto-recognition of remote sensing image objects by computer is a hot field and a key technology in computer vision. It has attracted many researchers in different fields to work over it broadly and deeply. However, as a new and difficult field in computer vision, there are still many problems to be studied and resolved ulteriorly.Shape is an essential character of objects and recognition through shape is an effective method in the computer-aided auto-recognition of remote sensing image objects. Among various objects with different shapes, there are linear objects in large quantity, so the research on the auto-recognition of linear objects has significant academic sense and practical value.This paper analyzes several expressions about shape as an inhere character of objects, and discusses each description of shape based on area and edge. At the same time, based on the character-selecting principles of recognition and the fact that many important objects in remote sensing images behave linear and the practical need of recognition of these objects, we point out that the edge information represents the image content and raise the method of character extraction and recognition through linear shape.This paper studies the feature extraction of linear objects in remote sensing images and the computer-aided auto-recognition of airports. After discussing the basic knowledge of remote sensing and the interferential factors in remote sensing images, the paper analyzes the pre-processing technology that is fit for shape analysis and expression and edge feature extraction based on the requirement of object recognition. In shape analysis, we get feature units of shape, that is, arc segments and straight-line segments through segmentation and identification of digital curves and the following combination of the same features, and present a new method of object expression and recognition based on straight-line segments. Because straight-line segments represent the edge structure of the straight linear objects and meet the three properties as moving, rotating and scale inflexibility, it’s effective for describing the shape of runways of the airport. We can achieve the recognition of airport through describing the shape of it with straight-line segments and distinguishing different objects with the length of the lines.Finally, we design a recognition system through straight linear shape based on the result of shape analysis and the practical requirement on application of remote sensing images. After histogram transform, enhancement, smoothing, pre-segmentation, edge extraction and thinning, we get an image with clear edges. After extracting of feature units, we describe the shape of airports with straight-line segments and distinguish different objects with the length of the lines, which can achieve the computer-aided auto-recognition of airports. The experiment results prove its validity.更多还原