【作者】 吴文宇；

【导师】 尹东；

【作者基本信息】 中国科学技术大学 ， 信号与信息处理， 2009， 硕士

【摘要】 道路和桥梁是当今世界上主要的陆地交通设施,同时也是军事领域中的重要目标。合成孔径雷达(SAR)可以有效获得大面积的高分辨率的雷达图像,且具有全天候、全天时的成像特点,这些优点使得其在民用和军事的各个领域应用十分广泛。随着合成孔径雷达技术的发展以及交通、战争对SAR图像中的陆地目标信息需求的增多,人们对SAR图像中道路和桥梁的目标识别研究也日益增多。本文以道路和桥梁目标的特征为基础,以对SAR图像的理解为背景,重点围绕SAR图像中道路和桥梁的特征选择与目标识别进行了较为深入的研究。论文综合讨论了各种传统滤波和基于相干斑噪声滤波的方法,在此基础上,比较并选取了适合SAR图像滤波并能较好保留图像边缘信息的Lee滤波作为预处理的工具。同时讨论了多种可行的图像分割方法,为下面的目标识别提供充足的理论准备。在对道路目标的几种特征进行分析的基础上,选择了较易提取和使用的形状特征和辐射特征,研究了一种基于动态规划的SAR图像道路目标识别的方法。实验证明该方法能够比较准确对道路目标进行识别与定位,同时能够较好的解决时间开销大、虚警率高的问题。但是由于道路阴影和噪声干扰的影响,识别结果仍然存在一定的断裂和虚警情况,需要进一步的研究解决。在对桥梁目标的几种特征进行分析的基础上,选择了较易提取和使用的形状特征和辐射特征。首先利用桥梁的形态特征,提出了一种利用桥梁的长宽比、位置信息、平行边缘的信息进行识别的方法。实验证明该方法简单、快速、虚警率较低、识别率较高。但是由于SAR图像的噪声干扰,存在目标外形失真的情况,本文针对这个问题,提出了一种基于Pun熵的统计特征的识别方法。该方法在不明显提高算法复杂度和时间开销的前提下,提高了识别率,降低了虚警率,并具有一定的算法鲁棒性。更多还原

【Abstract】 Road and Bridge are the main land traffic facilities in the world today, while they are also the important targets in the military field. Synthetic Aperture Radar (SAR) has been widely applied to gain large-area and high-resolution images, all-day and all-weather. These make SAR widely used in aerospace, ground reconnaissance, remote sensing or resource census etc. As the recognition of road and bridge is very useful for image registration, precision-guidance, map drawing and target detection and so on, the research on this field becomes more and more important.Based on the features of road and bridge, this paper mainly researched on the choosing of features and target recognition of road and bridge with the background of SAR image understanding.This paper firstly study on several traditional filters and some filters based on coherent speckle, and choose Lee filter to pre-process the original images, for Lee filter can de-noise well and reserve the edge information effectively. Also we talk about some image segmentation algorithms to get the theoretical basis of the recognition steps.Based on analyzing kinds of features, we choose shape and radiate features to recognize road targets, as these features are easy to extract and use. We advised a road recognition algorithm based on dynamic programming, which is proved to be effective on recognition and orientation, with high rate of recognition and low rate of false target. However, the result still has some problems about rupture and false target because of the noise and shadow.Based on analyzing kinds of features, we choose shape and radiate features to recognize bridge targets, as these features are easy to extract and use. First, we advised an algorithm based on the shape features, such as parallels, length-width ratio, position information and so on. Experiment results show the good qualities of the algorithm, such as fast speed, high rate of recognition, and low rate of false target. However, since the shape may be distorted in SAR images by the noise, we proposed a novel algorithm for bridge recognition using Histogram Entropy presented by Pun which is stable on shape distortion. This algorithm increased the recognition rate, decreased the rate of false target, and showed robustness in several conditions, while the time cost and algorithm complexity are not increased evidently.更多还原