【作者】 曲笑江；

【导师】 宋千；

【作者基本信息】 国防科学技术大学 ， 电子科学与技术， 2011， 硕士

【摘要】 通常情况下,由于地雷目标较小且所处环境复杂,基于能量的简单检测方法虚警率过高。本文根据实际雷达系统几何模型和地雷物理特性,分析了地雷散射模型并将其参数化,得到地雷的有效特征并将其用于鉴别。本文针对国防科学技术大学研制的车载前视超宽带地表穿透雷达,首先分析其等效单站模型及全孔径和各子孔径成像特点;然后对地雷电磁散射特性进行建模,说明双峰特征的形成机理及其方位一致性。预筛选是目标检测重要环节,能够有效降低运算量并获取疑似目标区域,是特征提取的必要准备。本文采用常规的检测流程,包括耦合信号抑制、图像增益均衡、恒虚警检测、形态学滤波和加权聚类等。在地雷鉴别过程中,本文选取四个与成像几何及地雷结构相关的特征,分别是方位视角偏离度,双峰特征模型中的双峰间距、双峰峰值比以及主峰宽度。根据特征确定性的特点,分别采用简单门限比较和超球面支持向量机一类分类器进行目标鉴别,在较低虚警率的情况下得到了较高的检测率。通过实测数据验证,基于物理模型提取的特征能够有效抑制杂波,并获得较高检测率。同时,由于成像分辨率受到带宽和孔径长度影响,而其对于目标特征提取密切相关,本文又研究了APES超分辨成像和压缩感知稀疏成像以及基于超分辨图像的特征提取和鉴别方法,这也为地雷目标鉴别研究探索了一条新途径。更多还原

【Abstract】 Generally, landmine detection method based on energy has a high false alarm rate because of the complex environment and small size of landmine itself. Based on geometric model of real radar system and physic characteristic of landmine, scattering model of landmine is analyzed and established, at the same time, efficient landmine features are extracted for discrimination.For the real system, Stepped-frequency Forward-looking Ground-penetrating Virtual Aperture Radar developed by the National University of Defense Technology, the equal single station model together with the characteristics of sub-aperture image and full aperture image are analyzed firstly. Following that, a model of landmine electromagnetic scattering is established, and then the mechanism of double-hump characteristics together with its aspect consistency are introduced detailedly.Filtering in advance is an important step in target detection and is essential to characteristic extraction, which can efficiently reduce operation size and help to get region of interest (ROI). In this paper, conventional flow of detection is used for extracting the target’s ROI, which include coupling signal depression, noise depression, image gain proportion, constant false alarm rate (CFAR) detection, morphological filter and weighted clustering.In the process of landmine discrimination, four characteristics related to the system and landmine structure are adopted, including angle of view offset degree, distance between the double-hump, double-hump peaks value ratio and width of main peak in double-hump characteristic model. According to the characteristics which are decided by the physics model, target discriminators are adopted easy threshold and Hyper Sphere Support Vector Machine respectively.Features extracted from the physics-model can be used for reducing clutter efficiently and get higher detection rate, and that is proved to be valid by the real data. Because the image revolution is limited by the bandwidth of signal and length of aperture, which makes features extraction difficultly. So new features extraction from super-revolution images using APES and Congress Sensing are studied, which is a new try in landmine detection.更多还原