基于模糊分类的弱小目标检测方法  

Approach to dim and small target detection based on fuzzy classification

作　　者：李欣[1] 赵亦工[1] 陈冰[1] 薛晶[2] 

机构地区：[1]西安电子科技大学模式识别与智能控制研究所,陕西西安710071 [2]西安电力高等专科学校电力工程系,陕西西安710032

出　　处：《光学精密工程》2009年第9期2311-2320,共10页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.60572151);教育部科学技术研究重点项目(No.03154)

摘　　要：为了实现对红外云层背景下的弱小目标检测,提出了一种新的基于模糊分类的红外弱小目标检测方法。该方法直接从待分类图像中提取出不同的类别区域,使得分类模板准确地体现当前图像的不同类别,从而得到图像的准确类别以实现弱小目标检测。首先,对红外天空背景弱小目标图像进行分析,将图像中的3类物体:净空、云及弱小目标细分为11个类别区域;其次,定义了类别特征矢量并基于此提出了类别核的定义,然后,根据类别核的定义从待检测图像中提取出11类区域的类别核;最后,根据模糊分类理论,定义了类别相似系数和类别贴近度,通过类别核对图像进行分类和类别归并,保留弱小目标类别完成检测。实验结果表明,该方法可对信噪比大于1.0的天空背景红外弱小目标图像中不同类型的区域进行准确分类,实现了对低信噪比的复杂云层背景图像中的弱小目标检测。In order to achieve robust dim and small target detection in a infrared cloud clutter, a new approach based on fuzzy classification is proposed. Different kinds of class regions are extracted from the query image to get several classification models to describe different classifications in the image exactly. The classification based on such models will classify the image effectively and achieve robust dim and small target detection. Firstly, the dim and small infrared target image are analyzed and eleven kinds of class regions are proposed to describe the sky, cloud and the target in the image. Then, class feature vectors and class kernels are defined, and the class kernels of eleven class regions are extracted from the query image. Finally, the class similar coefficient and class similarity degree are defined ac- cording to the fuzzy classification theory. The target detection is achieved by reserving a dim and small target class, after image classification and class merge are performed. Experimental results show that the proposed method can classifly different kinds of regions in the dim and small infrared target image with a Signal Noise Ratio（SNR） larger than 1.0 exactly and can realize the robust dim and small infrared target detection in heavy background clutter.

关 键 词：目标检测 模糊分类 红外弱小目标 类别核 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]