针对韦布尔分布的循环剔除TLM-CFAR检测器

韩东娟, 谭小敏, 史平彦. 针对韦布尔分布的循环剔除TLM-CFAR检测器[J]. 光电工程, 2018, 45(5): 170593. doi: 10.12086/oee.2018.170593
引用本文: 韩东娟, 谭小敏, 史平彦. 针对韦布尔分布的循环剔除TLM-CFAR检测器[J]. 光电工程, 2018, 45(5): 170593. doi: 10.12086/oee.2018.170593
Han Dongjuan, Tan Xiaomin, Shi Pingyan. A cycle elimination TLM-CFAR detector for Weibull clutter[J]. Opto-Electronic Engineering, 2018, 45(5): 170593. doi: 10.12086/oee.2018.170593
Citation: Han Dongjuan, Tan Xiaomin, Shi Pingyan. A cycle elimination TLM-CFAR detector for Weibull clutter[J]. Opto-Electronic Engineering, 2018, 45(5): 170593. doi: 10.12086/oee.2018.170593

针对韦布尔分布的循环剔除TLM-CFAR检测器

详细信息
    作者简介:
    通讯作者: 韩东娟, E-mail: HDJ1525454790@163.com
  • 中图分类号: TN911.7

A cycle elimination TLM-CFAR detector for Weibull clutter

More Information
  • 针对韦布尔分布杂波背景中的目标检测问题,本文提出了一种CFAR检测器—循环剔除TLM-CFAR检测器。该检测器基于TL矩估计方法,首先获得韦布尔分布的两参数估计值,进而确定CFAR检测门限,然后利用循环剔除法剔除干扰目标和强散射点的影响。文中证明了TLM-CFAR检测器具有恒虚警性,利用Monte Carlo仿真方法研究了这种检测器的性能,并与MLH-CFAR检测器进行了比较。仿真结果和实验结果表明,这种检测器不仅可以达到MLH-CFAR检测器的性能,同时避免了最大似然估计需要迭代计算的繁琐性,提高了检测算法的效率和适用性。

  • Overview: Target detection in a changing clutter and keeping the probability of false alarm constant is one of the important issues which cannot be avoided by each system and designer. CFAR technology is the most important tool in the control of false alarm rate in automatic radar target detection system, which plays an important role in the process of automatic radar detection. CFAR processing refers to adjusting the detection threshold by estimating the average clutter power of reference cells near cell under test under the condition of a constant false alarm rate. The classic CFAR detectors include ML-CFAR detectors and OS-CFAR detectors, and others are combination of the two detectors. However ML-CFAR detectors and OS-CFAR detectors are single-parameter CFAR detectors, while the Weibull PDF is a two-parameter distribution. If both the shape parameter and the scale parameter are taken into account, the two-parameter estimation CFAR detection method should be used.

    The shape parameter and scale parameter of Weibull distribution are estimated by using the reference samples in the reference sliding window. Werber and Haykin propose a method to set the detection threshold using two ordered statistical samples which is known as Werber-Haykin algorithm internationally, referred to as WH algorithm. Levanon and Ravid proposed the maximum likelihood CFAR algorithm (MLH-CFAR). Levanon has pointed out that CFAR loss is related to the accuracy of shape parameter estimation. MLH is an option when estimating the two parameters, but MLH estimator does not have a closed-form expression for estimation of shape and scale parameters and needs to be computed by the iterative process, thus reducing usability. In order to reduce the estimated variance of shape parameters and improve the efficiency of the algorithm, we estimate the two parameters of Weibull distribution by TL-moment estimation (TLME), and propose a cyclic elimination TLM-CFAR detection device.

    For the problem of constant false alarm rate (CFAR) detection in Weibull clutter background, a CFAR detector—cycle elimination TLME-CFAR detector is proposed. The detector calculates its detection threshold through the estimation of two parameters of Weibull distribution, which is based on TL-moment estimation. The two parameters (shape and scale) of the background statistics are estimated using a TL-moment estimation algorithm. A CFAR threshold based on parameters estimated in this way exhibits a smaller variance, and hence a smaller CFAR loss, than thresholds based on other estimation algorithms such as moments. The effect of the interference target and the strong scattering point are then eliminated by the cyclic elimination method which analyzes detector performance in comparison with MLH-CFAR detector, and the cyclic elimination TLM-CFAR detector has very nearly the same performance with MLH-CFAR detector. The detector avoids iterated operation of maximum likelihood estimation, and improves the efficiency and applicability of detection algorithm.

  • 加载中
  • 图 1  TLM-CFAR检测器结构图

    Figure 1.  Structure of TLM-CFAR detector

    图 2  虚警概率与门限系数的关系(c=1.452, n=32)

    Figure 2.  The relationship between false alarm proba-bility and threshold coefficient (c=1.452, n=32)

    图 3  形状参数未知时虚警概率和门限因子的关系曲线

    Figure 3.  The relationship between Pfa and α when shape parameter is unknown

    图 4  循环剔除TLM-CFAR检测流程图

    Figure 4.  The flow chart of cycle elimination TLME-CFAR detector

    图 5  循环剔除过程中的检测结果。(a)原图像;(b)一次剔除结果;(c)二次剔除结果;(d)三次剔除结果;(e)四次剔除结果;(f)最终结果

    Figure 5.  The detection results of cycle elimination process. (a) Original SAR image; (b) First elimination result; (c) Second elimination result; (d) Third elimination result; (e) Fourth elimination result; (f) Fifth elimination result

    图 6  两种参数估计方法性能对比图。(a), (b)形状参数;(c), (d)尺度参数

    Figure 6.  Performance comparison of two parameter estimation methods. (a), (b) Shape parameter; (c), (d) Scale parameter

    图 7  形状参数已知时检测性能曲线

    Figure 7.  Detection performance curve when shape parameter is known

    图 8  形状参数未知时检测性能曲线

    Figure 8.  Detection performance curve when shape parameter is unknown

    图 9  单目标检测结果。(a)原始SAR图像;(b) MLH-CFAR检测结果;(c)循环剔除TLM-CFAR检测结果

    Figure 9.  Single target detection results. (a) Original SAR image; (b) The results of MLH-CFAR algorithm; (c) The results of cycle elimination TLM-CFAR

    图 10  多目标检测结果。(a)原始SAR图像;(b) MLH-CFAR检测结果;(c)循环剔除TLM-CFAR检测结果

    Figure 10.  Multi-targets detection results. (a) Original SAR image; (b) The results of MLH-CFAR algorithm; (c) The results of cycle elimination TLM-CFAR

    表 1  循环剔除过程中的计算结果

    Table 1.  The calculation results of cycle elimination process

    剔除次数 尺度参数 形状参数 剩余目标像素个数/个 检测门限
    1 0.2231 1.0184 156 2.9395
    2 0.2028 1.3443 137 1.4300
    3 0.1955 1.4701 47 1.1666
    4 0.1939 1.4976 10 1.1197
    5 0.1936 1.5027 2 1.1113
    下载: 导出CSV

    表 2  两种估计方法用时对比

    Table 2.  Comparison of two estimation methods

    最大似然估计 TL矩估计 倍数
    100次估计平均用时/s 1.8e-3 5.1e-5 36.1
    200次估计平均用时/s 1.4e-3 3.6e-5 37.6
    500次估计平均用时/s 1.2e-3 2.8e-5 42.8
    1000次估计平均用时/s 1.1e-3 2.3e-5 48.5
    下载: 导出CSV

    表 3  两种检测器的恒虚警损失对比

    Table 3.  The comparison of CFAR loss of two detectors

    MLH-CFAR TLM-CFAR
    CFAR loss (Pd=0.9)/dB 8.79 8.49
    CFAR loss (Pd=0.5)/dB 6.66 6.36
    CFAR loss (Pd=0.2)/dB 5.68 5.38
    下载: 导出CSV

    表 4  两种检测算法的效率对比

    Table 4.  Comparison of the efficiency of the two detection algorithms

    MLH-CFAR 本文算法
    100次平均时间/s 0.2505 0.0894
    200次平均时间/s 0.2498 0.0880
    500次平均时间/s 0.2507 0.0883
    1000次平均时间/s 0.2584 0.0891
    下载: 导出CSV
  • [1]

    何友, 关键, 孟祥伟, 等.雷达目标检测与恒虚警处理[M].北京:清华大学出版社, 2011.

    He Y, Guan J, Meng X W, et al. Radar target detection and CFAR processing[M]. Beijing: Tsinghua University Press, 2011.

    [2]

    Weber P, Haykin S. Ordered statistic CFAR processing for two-parameter distributions with variable skewness[J]. IEEE Transactions on Aerospace and Electronic Systems, 1985, AES–21(6): 819–821. doi: 10.1109/TAES.1985.310668

    [3]

    Ravid R, Levanon N. Maximum-likelihood CFAR for Weibull background[J]. IEE Proceedings F-Radar and Signal Processing, 1992, 139(3): 256–264. doi: 10.1049/ip-f-2.1992.0033

    [4]

    Levanon N, Shor M. Order statistics CFAR for Weibull background[J]. IEE Proceedings F-Radar and Signal Processing, 1990, 137(3): 157–162. doi: 10.1049/ip-f-2.1990.0023

    [5]

    Teimouri M, Hoseini S M, Nadarajah S. Comparison of estimation methods for the Weibull distribution[J]. Statistics, 2013, 47(1): 93–109. doi: 10.1080/02331888.2011.559657

    [6]

    Akram M, Hayat A. Comparison of estimators of the Weibull distribution[J]. Journal of Statistical Theory and Practice, 2014, 8(2): 238–259. doi: 10.1080/15598608.2014.847771

    [7]

    Hosking J R M. L-moments: Analysis and estimation of distributions using linear combinations of order statistics[J]. Journal of the Royal Statistical Society. Series B, 1990, 52(1): 105–124. https://www.jstor.org/stable/2345653

    [8]

    Elamir E A H, Seheult A H. Trimmed L-moments[J]. Computational Statistics & Data Analysis, 2003, 43(3): 299–314. https://www.sciencedirect.com/science/article/pii/S0167947302002505

    [9]

    李军, 王雪松, 王涛.基于分数阶矩估计的非参量CFAR检测[J].电子与信息学报, 2011, 33(3): 642–645. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkxxk201103023

    Li J, Wang X S, Wang T. Nonparametric CFAR detection based on fractional moment estimations[J]. Journal of Electronics & Information Technology, 2011, 33(3): 642–645. http://www.wanfangdata.com.cn/details/detail.do?_type=perio&id=dzkxxk201103023

  • 加载中

(10)

(4)

计量
  • 文章访问数:  6688
  • PDF下载数:  2600
  • 施引文献:  0
出版历程
收稿日期:  2017-11-01
修回日期:  2018-02-09
刊出日期:  2018-05-01

目录

/

返回文章
返回