基于引导滤波和分块自适应阈值的单帧红外弱小目标检测  

Single-frame Infrared Dim Target Detection Based on Guided Filter and Segmented Adaptive Thresholds

作　　者：杨德振 黄静颖 喻松林[1] 冯进军[2] 李江勇[1] 刘彤 YANG Dezhen;HUANG Jingying;YU Songlin;FENG Jinjun;LI Jiangyong;LIU Tong(North China Research Institute of Electro-optics,Beijing 100015,China;Beijing Vacuum Electronics Research Institute,Beijing 100015,China)

机构地区：[1]华北光电技术研究所,北京100015 [2]北京真空电子技术研究所,北京100015

出　　处：《光子学报》2023年第4期262-272,共11页Acta Photonica Sinica

基　　金：中电科专家基金(No.N202003000390);军委科技委基础加强基金(No.2019JCJQZD33600)。

摘　　要：为了在有效地检测复杂场景下红外弱小目标的同时保持较低虚警率,在满足算法实现实时性的前提下,提出一种基于引导滤波和分块自适应阈值的单帧红外弱小目标检测。首先,为缓解边缘杂波干扰,采用具有保边特性的引导滤波对图像进行背景估计;然后,利用弱小目标具备的局部灰度最大特性,提出基于软阈值非极大值抑制的九宫格滤波计算目标的概率。通过加权的方式进一步剔除背景,抑制结果中不满足目标特性的区域;最后,针对复杂场景目标检测虚警率和漏检率高的问题,提出一种分块自适应阈值分割方法提取候选目标。实验结果表明,在公开数据集上与Top-Hat、LCM和Max-Median等经典方法相比,所提方法性能优于其他方法,恒虚警下不同复杂度场景的召回率分别达到87.97%、84.93%和86.22%,可有效抑制背景,增强目标信号,提高红外弱小目标检测的召回率,且具有更好的场景鲁棒性。A single-frame infrared dim and small target detection algorithm is proposed for the remote infrared target detection system based on the mobile platform.The system faces challenges in detecting targets due to the platform′s movement and changes in the background,leading to false alarms.To address this problem,the proposed algorithm combines guided filtering and nine-square-grid filtering for target enhancement,performing block adaptive threshold segmentation through regions of different complexity to maintain a low false alarm rate while detecting targets in different complex scenes.The background of the image is estimated using guided filtering with edge-preserving characteristics to alleviate edge clutter interference.The local grayscale maximum characteristic of dim and small targets is used to calculate the probability of the target using a nine-square filter based on soft threshold non-maximum suppression.Areas that don′t satisfy the target characteristics in the background suppression results are eliminated by weighting.A block adaptive threshold segmentation method is proposed to extract the candidate target using the sigmoid function to design the mapping curve of the standard deviation of gray value to parameter k for threshold calculation.The proposed method outperforms classical methods such as Top-Hat,LCM,and Max-Median,with Signal-to-noise Ratio(SNR)and Background Suppression Factor(BSF)indicators maintained at optimal and sub-optimal levels.The recall rates of scenes with different complexity under constant false alarm respectively reached 87.97%,84.93%,and 86.22%,improving the recall rate of infrared dim and small target detection.The algorithm is adaptable in engineering applications,as demonstrated by the addition of multi-frame target association on the basis of single-frame image detection.The hardware transplant of the infrared dim and small target detection algorithm is implemented using FPGA+DSP signal processing architecture,achieving a processing speed of over 75 frame/s for object de

关 键 词：红外弱小目标 目标检测 背景抑制 分块自适应阈值 引导滤波 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]