基于虚警识别与空-频域显著性映射的红外弱小目标检测算法

为了提高复杂背景与低信杂比率环境下的弱小目标检测准确度,有效控制虚警的干扰,考虑真实目标与背景的差异,设计了虚警识别耦合空-频域显著性映射的红外弱小目标检测算法。首先,根据红外中心像素在不同方向的强度,基于中值滤波器,构建了新的噪声滤波方法,充分抑制红外背景中的噪声干扰。随后,考虑中心像素与其邻域像素间的强度差别,设计背景抑制滤波机制,消除背景信息。根据初始红外图像与背景抑制结果,在空域内计算灰度映射。基于Fourier变换的相位谱,在频域内提取红外目标的显著性映射。利用背景的均值与方差,通过一个滑动窗口,建立候选目标检测方法,从灰度映射与显著性映射中确定候选目标。最后,利用真实目标位置的相关性,建立虚警识别方法,从候选目标中消除虚警,以保留真实弱小目标。实验数据表明:较已有的弱小目标识别技术而言,在复杂干扰背景下,所提方案可准确定位出真实目标,拥有更大的信杂比增益值与背景抑制因子,以及更好的ROC(receiver operating characteristic curve)特性曲线。

Infrared Dim Target Detection Algorithm Based on False Alarm Recognition and Spatial-Frequency Domain Saliency Mapping

In order to improve the detection accuracy of dim small target in the environment of complex background and low signal-to-clutter ratio, as well as control the interference of false alarm, considering the difference between real target and background, an infrared dim small target detection algorithm based on false alarm recognition coupled with spatial-frequency saliency mapping is designed in this paper. Firstly, according to the intensity of the infrared central pixel in different directions, a new noise filtering method based on the median filter was proposed to suppress the noise interference in the infrared background. Then, considering the intensity difference between the center pixel and its neighboring pixels, a background suppression filtering mechanism was designed to eliminate the background information. According to the initial infrared image and background suppression results, the weighted gray mapping was calculated in the airspace. The saliency map was extracted from the infrared target in the airspace based on the phase spectrum of Fourier transform. A sliding window was defined to determine candidate targets from weighted gray map and saliency map. Finally, a false alarm recognition method was established to eliminate false alarm from candidate targets by using the correlation of real target locations so as to preserve the real small and weak targets. Experimental results show that this proposed scheme can accurately locate the real target in complex interference background, and it has a larger signal-to-clutter ratio gain and background suppression factor, as well as a better ROC characteristic curve compared with the existing dim and small target recognition techniques.