摘要
为了有效控制红外弱小目标检测过程中的虚警率,提高复杂云背景下的目标检测准确度,提出了基于空域-频域映射与虚警抑制的弱小目标检测算法。根据红外中心像素不同方向上的强度值,构建了方向最大中值滤波器,有效消除噪声;并利用中心像素与其邻域像素的强度差异,形成背景抑制滤波,充分增强弱小目标;考虑云区域的特有属性,联合非线性滤波,定义了云区域识别机制,提取空域映射;引入Butterworth差异低通滤波器,对去噪图像中的显著目标完成初步识别;基于其幅度信息,进行显著目标的精细检测;再利用细显著性检测结果,计算阈值,利用二值分割方法来获取去噪红外图像的频域映射;联合空域映射与频域映射,提取红外图像中的候选目标;根据真实动目标与虚警之间的运动特征差异,利用多尺度改进的管道滤波来抑制虚警,准确识别出真实目标。实验数据表明,相对于已有的弱小目标识别方案,所提方案能够准确地识别出真实目标,拥有更好的ROC特性曲线。
In order to effectively control the false alarm rate in the process of infrared dim small target detection and improve the accuracy of target detection in complex cloud background,A small target detection algorithm based on spatial-frequency domain mapping and false alarm suppression was proposed in this paper. A directional maximum median filter is constructed according to the intensity values in different directions of the infrared central pixel to effectively eliminate noise. And a background suppression filter is formed by using the intensity difference between the center pixel and its neighboring pixels to fully enhance the dim target. Considering the unique attributes of cloud region,a cloud region recognition mechanism was defined to extract spatial mapping by combining with nonlinear filtering. The Butterworth differential low-pass filter is introduced to complete the coarse saliency detection of the denoised image. And the fine saliency detection was completed according to the amplitude information of the coarse saliency detection result. Then,the threshold was calculated by using the coarse and fine saliency detection results,so that the frequency domain mapping of denoised infrared image was obtained by using the adaptive binary segmentation method. The candidate targets in infrared images were extracted by jointing spatial map and frequency domain map. According to the difference of motion characteristics between real moving target and false alarm,these false alarms in candidate targets were suppressed based on improved pipeline filtering to accurately detect the real infrared dim targets.Test results show that this algorithm can accurately identify the real target,which has a better ROC characteristic curve compared with current infrared dim and small target detection technology.
引文
[1] HAN J H,LIANG K,ZHOU B. Infrared small targetdetection utilizing the multiscale relative local contrastmeasure[J]. IEEE Geoscience and Remote SensingLetters,2018,15(4):612-616.
[2] WANG X Y,PENG Z M,KONG D H. Infrared dim andsmall target detection based on stable multisubspacelearning in heterogeneous scene[J]. IEEE Transactionson Geoscience and Remote Sensing,2017,55(10):5481-5493.
[3]张智丰,裴志利,张亚荣.加权融合特征耦合Top-Hat变换的红外目标检测算法[J].计算机工程与设计,2017,38(10):2838-2844.ZHANG ZH G,PEI ZH L,ZHANG Y R. Infrared targetdetection algorithm based on weighted fusion featurecoupled Top-Hat transform[J]. Computer Engineeringand Design,2017,38(10):2838-2844.
[4]黄莉,王文波.基于差异结构描述符与自适应侧抑制的红外弱小目标检测算法[J].电子测量与仪器学报,2018,32(7):68-75.HUANG L,WANG W B. Infrared dim target detectionalgorithm based on differential structure descriptor andadaptive lateral suppression[J]. Journal of ElectronicMeasurement and Instrumentation,2018,32(7):68-75.
[5] ZHANG H Y,BAI J J,LI Z J,et al. Scale invariantSURF detector and automatic clustering segmentation forinfrared small targets detection[J]. Infrared Physics andTechnology,2017,83(10):7-16.
[6] QI H,MO B,LIU F X. Small infrared target detectionutilizing local region similarity difference map[J].Infrared Physics and Technology, 2017, 72(2):131-139.
[7]常亮亮,王广龙.基于中值滤波和提升小波分析的图像去噪方法研究[J].应用光学,2012,33(5):894-898.CHANG L L, WANG G L. Image denoising methodbased on median filter and lifting wavelet analysis[J].Applied Optics,2012,33(5):894-898.
[8]谭永前,曾凡菊,吴位巍,等.基于非线性滤波和边缘检测的纹理传输图像风格化处理[J].激光与光电子学进展,2018,55(6):061012-1-061012-6.TAN Y Q,ZENG F J,WEI W W,et al. Texture transferimage stylization based on nonlinear filtering and edgedetection[J]. Progress in Laser and Optoelectronics,2018,55(6):061012-1-061012-6.
[9] MA W W,ZHAO Y Q,ZHANG G H. Infrared dim targetdetection based on multi-structural elementmorphological filter combined with adaptive thresholdsegmentation[J]. Acta Photonica Sinica,2011,40(7):1020-1024.
[10]田雯,李素若.基于频域映射与多尺度Top-Hat变换的红外弱小目标检测算法[J].光学技术,2018,44(3):325-332.TIAN W,LI S R. Infrared dim target detection algorithmbased on frequency domain mapping and multi-scaleTop-Hat transform[J]. Optical Technology,2018,44(3):325-332.
[11] CUI Z, YANG J L, JIANG S D. Target detectionalgorithm based on two layers human visual system[J].Algorithms,2015,8(3):541-551.
[12] FENG K,WANG K S,NI Q,et al. A phase angle baseddiagnostic scheme to planetary gear faults diagnostics undernon-stationary operational conditions[J]. Journal of Soundand Vibration,2017,408(12):190-209.
[13]刘刚,梁晓庚.基于小波变换和管道滤波的红外空中小目标检测[J].计算机工程与应用,2011,47(30):198-201.LIU G,LIANG X G. Infrared aerial small target detectionbased on wavelet transform and pipeline filtering[J].Computer Engineering and Application,2011,47(30):198-201.
[14] ZHAO L,YANG H J. Small-target leak detection for aclosed vessel via infrared image sequences[J]. InfraredPhysics and Technology,2017,81(6):109-116.
[15] WANG B,XU W H,ZHAO M,et al. Antivibrationpipeline-filtering algorithm for maritime small targetdetection[J]. Optical Engineering,2014,53(11):1-11.
[16]穆柯楠,赵祥模,惠飞.基于非采样高斯差分金字塔的多尺度融合边缘检测[J].四川大学学报(工程科学版),2015,47(5):130-138.MU K N,ZHAO X M,HUI F. Multiscale fusion edgedetection in pyramid based on nonsubsampled Gaussdifference[J]. Journal of Sichuan University(Engineering Science Edition),2015,47(5):130-138.
[17] ZHAO Y,PAN H B,DU C P. Principal Curvature forinfrared small target detection[J]. Infrared Physics andTechnology,2015,69(7):36-43.
[18]何玉杰,李敏,张金利,等.基于低秩三分解的红外图像杂波抑制[J].光学精密工程,2015,23(7):2069-2078.HE Y J,LI M,ZHANG J L,et al. Infrared image cluttersuppression based on low rank three decomposition[J].Optical Precision Engineering, 2015, 23(7):2069-2078.