基于结构张量分析的弱小目标单帧检测  

Dim Small Target Single-frame Detection Based on Structure Tensor Analysis

作　　者：赵高鹏[1] 李磊[1] 王建宇[1] ZHAO Gao-peng;LI Lei;WANG Jian-yu(Department of Automation, Nanjing University of Science and Technology, Nanjing 210094, China)

机构地区：[1]南京理工大学自动化学院,南京210094

出　　处：《光子学报》2019年第1期135-145,共11页Acta Photonica Sinica

基　　金：国家自然科学基金(No.61473153)~~

摘　　要：针对复杂场景图像由于背景边缘干扰和噪声导致弱小目标检测困难的问题,提出了一种基于结构张量分析的弱小目标单帧检测方法.利用结构张量对不同局部结构的表示特性,通过计算结构张量特征值矩阵和均值滤波得到点状和矩形状目标的结构张量响应图;采用高斯差分带通滤波器计算灰度差分图;通过归一化融合处理得到最终响应图;采用自适应阈值分割得到目标位置.采用该方法对天空、海面等多种场景的红外图像和可见光图像进行实验,并与典型方法对比,结果表明该方法能够有效地抑制背景干扰和噪声、快速且准确地检测目标.The challenge of detecting the dim small target in complex scene is to suppress the edge interference and the noise.A dim small target single-frame detection method based on structure tensor analysis is proposed.By adopting the structure tensor,which can measure the different local structure information,the structure tensor response map is calculated by employing the eigenvalue matrix of the structure tensor and the mean filter.A gray difference map is computed by adopting difference of Gaussians band-pass filter.By normalizing and fusing the two maps,the final response map is obtained.The target can be detected by segmenting the final response map with an adaptive threshold.Experiments with the standard infrared and visible datasets are performed including the different scenes such as sky,sea etc.Results demonstrate that the proposed algorithm can suppress background and nosie effectively,and detect targets efficiently and accurately in comparison with several typical methods.

关 键 词：机器视觉 目标检测 单帧检测 结构张量 弱小目标 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]