基于相关滤波器的红外弱小目标检测算法  

Infrared Small Target Detection Method Based on Correlation Filter

作　　者：何玉杰[1] 李敏[1] 张金利[1,2] 邢宇航[1] 

机构地区：[1]中国人民解放军第二炮兵工程大学908教研室,陕西西安710025 [2]中国人民武装警察部队工程大学信息工程系,陕西西安710086

出　　处：《光学学报》2016年第5期88-96,共9页Acta Optica Sinica

基　　金：国家自然科学基金(61102170)

摘　　要：针对复杂背景下红外图像中低信噪比弱小目标实时检测问题,提出一种基于相关滤波器的红外弱小目标检测算法。该算法将红外目标检测转化为模式分类问题,在离线训练阶段,利用二维高斯模型构造红外小目标训练集,在此基础上训练得到对目标背景具有区分能力的相关滤波器,在线检测阶段,利用滤波器对图像分块进行滤波操作,目标和背景的滤波响应有着显著的差异,最后生成整幅图像的滤波响应置信图以此来判断图像中是否包含目标及其具体位置。在单帧单目标图像、序列图像多目标检测实验结果表明,与经典检测算法相比,所提方法不仅具有更高检测性能,有效降低了虚警概率,而且具有较好的实时性,适用于复杂背景条件下弱小目标的实时检测。To solve the infrared target real-time detection problems caused by low signal to noise ratio with complex background,an infrared small target detection method based on correlation filter is proposed.The small target detection problem is transformed into pattern classification task,which consists of two stages：off-line training and on-line detection.In training stage,a correlation filter is obtained using the training dataset produced by twodimensional Gaussian model.It has ability to distinguish target and background.In detection stage,the sub-image blocks of the infrared image are extracted successively and the filtering response confidence map which indicates the target location is computed.The experiments under two conditions demonstrate that the proposed method not only has better detection performance with effective reduction of the false alarm rate,but also has better real-time performance.

关 键 词：测量 机器视觉 红外弱小目标检测 相关滤波器 置信图 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]