【作者】 杨志国；

【导师】 周智敏； 黄晓涛；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2007， 博士

【摘要】 利用超宽带合成孔径雷达(UWB SAR)探测叶簇覆盖目标具有重要的军事应用价值,对UWB SAR目标鉴别方法的研究能大大提高UWB SAR的应用价值,目前关于此方面的讨论存在角度孤立、系统性差的问题,本文力图克服上述缺点,紧紧围绕感兴趣区域(ROI)一条主线,系统全面地对UWB SAR目标鉴别中的各项关键技术展开研究。研究了叶簇杂波背景下的ROI提取问题。分析了双参数恒虚警(CFAR)检测和智能CFAR检测在叶簇杂波背景下存在的问题,提出了采用中值滤波解决上述问题的思想,从四个方面研究比较了三种二维中值滤波器的性能;分析了常规聚类算法用于叶簇杂波背景时出现的问题,针对问题,一方面通过改进聚类算法本身,另一方面通过引入形态学滤波方法,大大提高了聚类算法在叶簇杂波背景下的聚类效果;提出了一种适于叶簇杂波背景的ROI快速有效提取方法。研究了UWB SAR图像ROI的单尺度特征提取问题。分析了适于高频SAR图像的幅度特征及其提取方法用于UWB SAR图像时的有效性,提出了基于目标轮廓的幅度特征提取方法及相应的适用幅度特征;分析了已有二面角模型在建模过程中存在的问题并进行了改进建模,基于改进二面角模型结论提出了方位向特征提取方法;分析了常用变化检测方法用于提取ROI变化特征时存在的问题,提出了一种稳健可行的变化特征提取方法。研究了UWB SAR图像ROI的多尺度特征提取问题。分析了适于高频SAR图像的多分辨率特征提取方法在UWB SAR图像中的适用性,建立了树干杂波和目标的等效散射模型,分析了二者多分辨率特性的区别,分别基于UWB SAR实、复图像提出了两种有效的多分辨率特征提取方法。研究了UWB SAR目标鉴别中的分类器设计和性能评估问题。总结了目前常用的两种分类器;明确完善了适用于UWB SAR目标鉴别性能评估的参量指标;针对UWB SAR目标鉴别算法的鉴别性能,提出一种分步进行性能评估的过程评估法;针对UWB SAR目标鉴别算法的稳健性能,提出了一种基于多项式拟合的评估方法。文章最后在总结全文研究成果的基础上,提出了下一步的研究展望。更多还原

【Abstract】 Ultra-Wide Band Synthetic Aperture Radar (UWB SAR) is usually used to detect the foliage-concealed targets as is important in military application. The research of UWB SAR target discrimination can improve the application value of UWB SAR. The shortages of isolated angle and deficient systematism exist in the discussion about it in reference. This paper tries to overcome above shortages, surrounds a main clue of Region of Interest (ROI) tightly and investigates the important techniques in UWB SAR target discrimination systemically.The ROI extraction in foliage clutter is investigated. The problems, which appear when a two-parameter Constant False Alarm Rate (CFAR) detector and an intelligent CFAR detector are used in foliage clutter, are analyzed. The median filter is proposed to be used to solve above problems. The performances of three median filters are investigated and compared from four aspects. The problems, which appear when the common cluster algorithm is used in foliage clutter, are analyzed. In order to solve above problems, the first means is to improve the common cluster algorithm, and the second means is to adopt a morphological filter. Both means can improve the cluster results in foliage clutter effectively. A quick and effective extraction approach of ROI suitable in foliage clutter is proposed.The single-scale feature extraction of ROI in UWB SAR image is investigated. The validities of amplitude features and their extraction method suitable in high-frequency SAR image are analyzed when they are used in UWB SAR image. An extraction approach of amplitude feature based on target contour is presented, and the corresponding effective amplitude features are proposed. The old dihedral angle model is analyzed and improved. The approach of azimuthal feature extraction is proposed based on the conclusion of the improved dihedral angle model. The problems, which appear when the common change detection methods are used to extract the change feature based on ROI, are analyzed. And a robust and feasible extraction approach is presented.The multiscale feature extraction of ROI in UWB SAR image is investigated. The applicabilities of multiresolution feature extraction methods suitable in high-frequency SAR image are analyzed when they are used in UWB SAR image. The equivalent scattering models of trunk clutter and target are established, and the differences of multiresolution characteristic between trunk clutter and target are analyzed. Two effective multiresolution feature extraction approaches are proposed based on real and complex images respectively.The classifier design and the performance evaluation in UWB SAR target discrimination are investigated. Two kinds of classifiers in common use are summarized. The parameter system suitable for performance evaluation of UWB SAR target discrimination is consummated. A process evaluation approach is proposed to evaluate the discrimination performance of UWB SAR target discrimination algorithm step by step. An evaluation approach based on polynomial fitting is proposed to evaluate the robustness performance of UWB SAR target discrimination algorithm.Finally, the research prospect in future is proposed based on the study summary of full paper.更多还原