基于梯度相关性的红外微弱目标检测

Detection of infrared small dim targets based on correlation of gradients

提出了一种新的基于空域梯度相关性的图像杂波自适应预测算法。该算法能显著改善微小目标的邻域信杂比(SCNR)。试验证明,本算法相对于已有的多种算法,有着更好的性能。对微弱目标SCNR的增益相对于传统算法提高3dB以上。本文还引入了一种基于统计分析的微小目标检测算法。该算法考虑了微小目标在图像集成时的重叠特性,能在不增加系统运算负荷的情况下,获得更高的检测概率,相对于不考虑重叠特性的算法,在虚警概率小于10-4时,假设重叠系数为3,SCNR为4dB,系统检测概率从小于20%,大幅提高到80%。理论分析及仿真表明,本文提出的检测系统在微小目标检测中,具有很高的实用性。系统在原始信号的邻域信杂比(SCNR)小于0dB的情况下,能有效检测出目标,在采用10帧集成检测,目标像素重叠参数为5的情况下,虚警概率小于10-7,系统检测概率大于80%,虚警概率小于10-4,系统检测概率大于95%。

This paper proposes a novel adaptive prediction algorithm of background clutter based on NACF （normalized auto correlation function） of gradients in neighborhood. The algorithm can improve the SCNR （Signal Clutter Noise Ratio） of small targets evidently. Compared with traditional methods,the gain of SCNR can be improved over 3 dB by this algorithm. And this paper proposes a detecting algorithm of small targets based on statistical analysis,The characteristic of targets overlapped is considered, and a novel method is proposed to get higher probability of detection on same load of system. If the overlapped coefident is 3 ,SCNR is 4 riB,false alarm probability Pfa〈10-4 ,the detection probability PD can be improved from less than 20% to 80%. The theoretic analysis and many simulations verify the validity of the method. The system could detect the small targets when SCNR of original signal is less than 0 dB. If 10 frames are employed to detect,over-lapped coefficient is 5,Pfa〈10^-7 ,PD〉80%;and Pfa〈10^-4 ,PD〉95%.