【作者】 李欣；

【导师】 赵亦工；

【作者基本信息】 西安电子科技大学 ， 模式识别与智能系统， 2010， 博士

【摘要】 红外弱小目标检测技术是红外预警系统中的核心技术,这类武器系统对国家安全起着极为重要的作用。为此,本文就红外弱小目标检测问题进行了深入的分析与讨论,提出了新的检测方法与算法研究思路,并从系统实现角度对武器系统的设计进行了详细的论述,提出了两种红外搜索与跟踪系统的组网构想,为红外预警系统的设计提供了相关依据。本文从不同角度对红外弱小目标图像进行分析,提出了新的弱小目标检测方法:利用图像方差加权信息熵对红外弱小目标图像进行分析,并将其作为对图像复杂度定量描述,深入讨论了不同类别区域复杂度特征值的本质成因,并基于此构造了新的图像预处理方法与自适应门限分割方法,进而实现了对红外弱小目标图像的自适应门限分割完成目标检测。利用模糊分类的思想对红外弱小目标图像进行分析,根据红外弱小目标图像中不同区域灰度分布情况,将图像分为多个类别区域,并定义了类别特征矢量,同时依据模糊分类准则定义了类别相似系数,进而对图像中不同区域进行类属性判别,最后通过对弱小目标类别进行提取从而实现弱小目标检测。对基于分类模型的模糊分类弱小目标检测方法进行扩展:重新构造分类模型,根据类别核及其类别特征矢量,结合模糊分类理论定义了类别相似系数,并定义了类别贴近度来实现不同类别区域的类别归并,从而解决了因红外弱小目标图像中包含本文建立的类别区域不全而带来的误分类问题,得到了一种全新的弱小目标检测思路。对上述两种基于模糊分类目标检测方法进行了进一步的推广,得到了基于分类算法的算法框图。对传统的空域背景抑制算法对复杂背景抑制效果较差的本质原因进行了分析,首次将区域方向直方图概念引入弱小目标检测领域,并基于此构造了背景抑制改进方法:根据背景区域的灰度分布特点针对性的构造背景抑制算法,引入了区域方向直方图作为不同区域的类别判别依据,并定义了每类类别共性模板与共性背景预测系数模板,将局部二值模式(LBP)作为类别区域的结构表达方式,利用其对不同区域的背景预测系数模板进行修正以得到适应每个区域的系数模板,由此,实现了对传统背景抑制算法的改进。最后,从系统实现角度出发,对弱小目标检测系统的组成与设计进行了较为深入的研究,给出了适用于实时弱小目标检测系统的软硬件工程化设计与开发方法,并提出了预警与火控系统组网的两种构想,为该系统在实际的应用提供了设计依据。更多还原

【Abstract】 The detection of dim target is the core technique in infrared surveillance system (IRSS), which takes a significant role in national security safeguarding. Accordingly, in this thesis, profound analysis and discussion on the detection technique above, and several novel detection approaches and research ideas are proposed; with detailed descriptions on the design of weapon system form the angles of realization. Moreover, two networking composition on infrared search and tracking system are established, providing relevant basis for the construction of infrared surveillance systems.Based on the analysis from different points of view on query images of dim targets, several novel approaches on detection of dim target are put forward.Firstly, information entropy weighted by image variance is introduced to analyze the images above and describe image complexity, and substantial causes of the complexity features in classified regions are discussed, where based a novel image preprocessing method and a self-adaptive threshold acquisition method are constructed, so that the dim targets can be finally detected with self-adaptive threshold processing. Secondly, fuzzy classification theory is proposed to analyze the infrared query images above, regions with class feature vectors are classified and defined based on the grey distribution, and class similar coefficients are defined according to the fuzzy classification, so that the target detection is achieved by reserving a dim target class.Thirdly, the fuzzy classification method described above is extended with classification models re-definition, class kernels are accordingly defined combined with class feature vectors, and class similarity degrees are defined to merge classes, so that the problems on mis-classification caused by incomplete class regions included in images of infrared dim targets are solved, where based a novel approach on dim target detection is constructed; an algorithm diagram is given based on the generalization of the two fuzzy detection approaches above.In addition, based on the traditional approaches of background suppression, analysis are made on the causes of poor suppression performance, and for the first time, regional direction histograms are introduced into the fields of dim target detection, where based an improved background suppression approach is proposed. The principle of traditional background suppression approach is discussed, and then the causes of poor suppression performance on traditional background suppression approaches are obtained. Accordingly, improvements are proposed that regional background suppression should be conducted on the basis of the regional grey features in an image. And therefore, regional direction histograms are introduced for local grey classification, while general models and general background coefficient models are defined with LBP as the structure expression for each regional class, via which background coefficient models are modified to match each region of the query images, so that the improved suppression is achieved.From the angles of system realization, research on the construction and design of the dim target detection system is profoundly conducted, with the decryptions of hardware & software engineering design and development on real-time application. Moreover, two networking composition on infrared search and tracking system are put forward, providing some basis for practical system application.更多还原