基于频域映射与多尺度Top-Hat变换的红外弱小目标检测算法  

Infrared dim target detection algorithm based on frequency-space domain mapping and multi-scale top-hat transform

作　　者：田雯[1] 李素若[1] TIAN Wen;LI Suruo(College of Computer Engineering, Jingchu University of Technology, Jingmen 448000, Chin)

机构地区：[1]荆楚理工学院计算机工程学院

出　　处：《光学技术》2018年第3期325-332,共8页Optical Technique

基　　金：湖北省教育厅科学重点研究项目(B2015241);荆楚理工学院科学研究重点基金项目(ZR201402);荆楚理工学院科学研究引进人才科研启动金项目(QDB201605)

摘　　要：提出了基于频域映射与多尺度Top-Hat变换的红外弱小目标检测算法。通过分割经典Top-Hat的单一结构元素,获得多尺度膨胀结构元素,对红外弱小目标进行增强,有效抑制杂波与噪声背景;基于Butterworth低通滤波与截止频率,构建Butterworth差异带通滤波,联合Fourier变换,建立粗显著性检测机制,通过提取其幅度与相位频谱,基于2D高斯平滑滤波,定义细显著性检测机制,在频域中凸显弱小目标,并将红外目标的空间与强度相关性作为识别标准,精确定位候选目标;根据红外目标运动与虚警的速度差异特征,定义弱小目标连续帧速度模型,在帧间充分抑制候选区域中的虚假目标,检测出完整的弱小目标。实验结果显示：与当前红外弱小目标检测技术相比,面对复杂背景干扰,提出的算法具有更高的检测精度,可精确定位出完整的弱小目标,呈现出更好的ROC特性曲线。The infrared dim target detection algorithm based on frequency-space domain mapping and multi-scale Top-Hat transform is proposed. Firstly,the multiple scale expansion structure elements are obtained by dividing the single structure elements of classical Top-Hat transform to enhance the infrared dim target,which can suppress the background clutter and noise.Then Butterworth differences band-pass filter is constructed based on Butterworth low-pass filter and cut-off frequency,and joint the Fourier transform to establish the coarse saliency detection mechanism,the fine saliency detection mechanism is defined by extracting the amplitude and phase spectrum of Butterworth differences band-pass filter for highlighting the dim target in frequency domain. Finally,the successive frame velocity model of the small target is defined according to the difference of the velocity between the infrared target and the false alarm for fully suppressing the false targets inter-frame of the candidate regions,which fully detect the small targets. Experiment results show that this algorithm has higher detection accuracy for completely precise positioning of the small target,and showing a better ROC curve.

关 键 词：红外弱小目标检测 多尺度Top-Hat变换 截止频率 Butterworth差异带通滤波 显著性检测 连续帧速度模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]