投影模板耦合遗传算法的红外图像点目标检测  

Infrared image point objects detection based on projection coefficient templates coupled genetic algorithm

作　　者：吴琼飞 朱勇[2] 张志强 

机构地区：[1]武汉设计工程学院信息工程学院,湖北武汉430205 [2]武汉纺织大学数学与计算机学院,湖北武汉430200

出　　处：《计算机工程与设计》2016年第7期1885-1891,共7页Computer Engineering and Design

基　　金：国家自然科学基金项目(61142006);湖北省教育科研基金项目(hklse2010005)

摘　　要：当前基于模板匹配的红外图像点目标定位检测技术难以扑获图像的高阶统计特性,无法描述图像的非线性结构,且此类算法直接将图像作为模板,使算法稳定性不佳,为此提出非线性模板匹配耦合遗传算法的红外图像点目标定位检测技术。定义参数约束边界,改进2D高斯强度函数,获取目标训练图像;引入主成分分析(principal component analysis,PCA),利用其特征提取函数,通过预设阈值,将目标训练矢量投影到该函数的特征向量中,设计PCA加权投影系数模板;考虑图像的高阶统计特性,嵌入内核函数,联合非线性映射与相关系数,构造特征空间内的归一化非线性相关系数计算模型,估算投影系数模板与测试模板之间的相似度;引入遗传算法,快速求解相关匹配。实验结果表明,与当前红外图像微小目标检测技术相比,该算法拥有最佳的稳定性与定位性能,在复杂背景下,成功检测出了点目标,且其信杂比值最高,时耗最短。Current infrared image point obj ects location detection algorithm based on template matching can not capture the higher order statistics and represent the nonlinear structures of images,and it has poor stability when directly using the image as tem-plates,the infrared image point objects location algorithm based on projection coefficient templates coupled non-linear correlation coefficient was proposed to solve these problems.The target training images were obtained by defining the boundary constraints to improved 2D Gaussian intensity function.The projection coefficient template was designed for reference patterns by projecting the target training vector to eigenvectors of feature extraction function based on the principal component analysis.And the non-linear correlation coefficient computation model was constructed for detecting objects by embedding kernel functions,combining nonlinear mapping with linear correlation coefficient,as well as considering higher-order statistical properties of the image.Ex-perimental results show that this algorithm has best stability and location performance for successfully detecting the point objects in complex background,and signal to clutter ratio of this algorithm is the highest.

关 键 词：红外图像 点目标定位检测 非线性模板匹配 加权投影系数 遗传算法 非线性相关系数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]