工程科学学报，第38卷，第11期：1652-1658,2016年11月 Chinese Journal of Engineering,Vol.38,No.11:1652-1658,November 2016 D0l:10.13374/j.issn2095-9389.2016.11.020:http://journals.ustb.edu.cn 基于非线性局部滤波的红外小目标检测方法 赵爱罡2)四，王宏力》，杨小冈2》，陆敬辉》，黄鹏杰) 1)火箭军工程大学士官学院，青州2625002)火箭军工程大学控制工程系，西安710025 ☒通信作者，E-mail:zhaoaigang1986120@163.com 摘要为提高复杂环境下红外小目标的检测效率，将图像分为平坦区域、边缘区域和小目标区域三种区域，并针对三种成 分的特点，提出基于拉普拉斯金字塔的非线性局部滤波检测方法.首先将图像进行高斯金字塔分解，将高斯低通金字塔与原 图像尺寸匹配后，相减并进行阈值操作，抑制平坦区域：其次将标记像素灰度值与其周围环域均值的最小差作为指标，滤除边 界区域：最后将非线性局部滤波结果生成相应的拉普拉斯金字塔各层系数，重构得到高对比度的检测图像，利用邻域特点剔 除孤立噪声点并通过简单阈值标记红外小目标.实验结果表明：与现有其他算法相比，该检测算法能够对复杂背景有效抑 制，检测速度快 关键词红外图像处理：目标探测：拉普拉斯金字塔：非线性滤波：局部滤波 分类号T765.3 Infrared small target detection method based on nonlinear local filter ZHAO Ai-gang,WANG Hong-i,YANG Xiao-gang?,LU Jing-hui,HUANG Pengjie 1)School of Sergeaney,Rocket Force University of Engineering,Qingzhou 262500,China 2)Department of Control and Engineering,Rocket Force University of Engineering,Xi'an 710025,China Corresponding author,E-mail:zhaoaigang1986120@163.com ABSTRACT In order to improve the efficiency of infrared small target detection against complex background,the image was decom- posed into three regions:flat region,edge region and small target region.A method of nonlinear local filter detection using the Lapla- cian pyramid was presented based on each character of the three components.Firstly,Gaussian pyramids were built for the image, each level was subtracted from the original image with matching size,and the flat region was restrained by simple threshold operation. Secondly,the minimum difference between the marked pixel gray value and the mean value of the hollow annular region was used as quota to filter out the edge region.At last,each layer coefficient of the Laplacian pyramid was generated from the results of nonlinear local filtering and then a high-contrast detection image was reconstructed.The isolated noise points were removed based on the charac- ter of the neighborhood and the infrared small target was marked by simple threshold operation.Compared with other existing methods, the experimental results show that this method can effectively restrain complex background and the detection speed is fast. KEY WORDS infrared image processing:target detection:Laplacian pyramid:nonlinear filtering:local filter 随着制导技术的突飞猛进，红外探测以其具有的 背景中，检测比较困难-，所以小目标检测作为制导 隐蔽性好、抗干扰能力强、探测距离远、可全天候工作 领域的关键技术，成为目前学者研究的重点 等优点，成为精确制导技术发展的一个重要方向.但 红外小目标检测算法主要分为两种：第一种是 因测探距离比较远，弱小目标在图像中仅占几个像素， 滤波算法，如形态学滤波(Tophat)、最大均值(Max- 信息相对匮乏，并且经常淹没在海杂波和云杂波复杂 Mean)和最大中值(MaxMedian)同检测算法，这些滤波 收稿日期：2015-09-16 基金项目：国家自然科学基金资助项目(61203189,61374054)工程科学学报，第 38 卷，第 11 期: 1652--1658，2016 年 11 月 Chinese Journal of Engineering，Vol． 38，No． 11: 1652--1658，November 2016 DOI: 10． 13374 /j． issn2095--9389． 2016． 11． 020; http: / /journals． ustb． edu． cn 基于非线性局部滤波的红外小目标检测方法 赵爱罡1，2) ，王宏力2) ，杨小冈2) ，陆敬辉2) ，黄鹏杰2) 1) 火箭军工程大学士官学院，青州 262500 2) 火箭军工程大学控制工程系，西安 710025  通信作者，E-mail: zhaoaigang1986120@ 163． com 摘 要 为提高复杂环境下红外小目标的检测效率，将图像分为平坦区域、边缘区域和小目标区域三种区域，并针对三种成 分的特点，提出基于拉普拉斯金字塔的非线性局部滤波检测方法． 首先将图像进行高斯金字塔分解，将高斯低通金字塔与原 图像尺寸匹配后，相减并进行阈值操作，抑制平坦区域; 其次将标记像素灰度值与其周围环域均值的最小差作为指标，滤除边 界区域; 最后将非线性局部滤波结果生成相应的拉普拉斯金字塔各层系数，重构得到高对比度的检测图像，利用邻域特点剔 除孤立噪声点并通过简单阈值标记红外小目标． 实验结果表明: 与现有其他算法相比，该检测算法能够对复杂背景有效抑 制，检测速度快． 关键词 红外图像处理; 目标探测; 拉普拉斯金字塔; 非线性滤波; 局部滤波 分类号 TJ765. 3 Infrared small target detection method based on nonlinear local filter ZHAO Ai-gang1，2)  ，WANG Hong-li 2) ，YANG Xiao-gang2) ，LU Jing-hui 2) ，HUANG Peng-jie 2) 1) School of Sergeancy，Ｒocket Force University of Engineering，Qingzhou 262500，China 2) Department of Control and Engineering，Ｒocket Force University of Engineering，Xi'an 710025，China  Corresponding author，E-mail: zhaoaigang1986120@ 163． com ABSTＲACT In order to improve the efficiency of infrared small target detection against complex background，the image was decom￾posed into three regions: flat region，edge region and small target region． A method of nonlinear local filter detection using the Lapla￾cian pyramid was presented based on each character of the three components． Firstly，Gaussian pyramids were built for the image， each level was subtracted from the original image with matching size，and the flat region was restrained by simple threshold operation． Secondly，the minimum difference between the marked pixel gray value and the mean value of the hollow annular region was used as quota to filter out the edge region． At last，each layer coefficient of the Laplacian pyramid was generated from the results of nonlinear local filtering and then a high-contrast detection image was reconstructed． The isolated noise points were removed based on the charac￾ter of the neighborhood and the infrared small target was marked by simple threshold operation． Compared with other existing methods， the experimental results show that this method can effectively restrain complex background and the detection speed is fast． KEY WOＲDS infrared image processing; target detection; Laplacian pyramid; nonlinear filtering; local filter 收稿日期: 2015--09--16 基金项目: 国家自然科学基金资助项目( 61203189，61374054) 随着制导技术的突飞猛进，红外探测以其具有的 隐蔽性好、抗干扰能力强、探测距离远、可全天候工作 等优点，成为精确制导技术发展的一个重要方向． 但 因测探距离比较远，弱小目标在图像中仅占几个像素， 信息相对匮乏，并且经常淹没在海杂波和云杂波复杂 背景中，检测比较困难［1 － 2］，所以小目标检测作为制导 领域的关键技术，成为目前学者研究的重点． 红外小目标检测算法主要分为两种［3］: 第一种是 滤波算法，如形态学滤波( Tophat) ［4］、最大均值( Max￾Mean) 和最大中值( MaxMedian) ［5］检测算法，这些滤波