结合视觉注意力机制基于尺度自适应局部对比度增强的红外弱小目标检测算法

Infrared Dim-small Object Detection Algorithm Based on Adaptive Scale Local Contrast Enhancement Combined with Visual Attention Mechanism

如何在没有先验信息的情况下从复杂噪声背景下快速检测到远距离进入的弱小目标,提高整个装备系统的响应能力,是目前IRST热门研究课题。本文通过引入视觉注意机制,提出了一种结合尺度自适应的局部对比度测量的红外弱小目标检测方法。本文首先采用拉普拉斯金字塔尺度空间理论对所有像素点局部对比度进行分析,获得对应的自适应尺度信息;然后在跳出效应的基础上设计了一种基于改进的局部对比度测量模型,最终生成一个显著图来突出目标特性,该方法能够在增强目标对比度同时,抑制背景杂波。定性定量实验结果表明,本文提出的方法相比于对比算法具有较高的红外小目标检测性能,能够对对比度不低于5%的目标稳定检测,适合防空武器装备工程应用。

Infrared search and track(IRST) is one of the indispensable key systems in the field of modern warfare.It is widely used in precision guidance,photoelectric warning,air defense,and anti-missile.Methods for quick detection of dim small object entering from a complex background without a priori information and improvement of the responsiveness of the entire equipment system is currently a trending research topic in IRST.In this paper,the visual attention mechanism is introduced,and an infrared dim small object detection method based on local scale-perception contrast measurement is proposed.First,the local contrast of all the pixels is analyzed using Laplacian pyramid scale-space theory,wherein the corresponding adaptive scale information is obtained.Next,according to the pop-out effect in visual system,an improved local contrast measurement model is designed to obtain the optimal saliency map.Finally,a weight map is generated to highlight the object features,thereby suppressing the background clutter while enhancing the object contrast.Qualitative and quantitative experiments show that our proposed method has better performance in terms of superior object detection than the algorithms it was compared with and can detect object with better stability and a contrast of at least 5%,thus ensuring the method’s suitability for engineering application.