一种变信噪比条件下的恒虚警野值检测方法
收稿日期: 2015-07-20
修回日期: 2015-12-19
网络出版日期: 2015-12-28
基金资助
国家自然科学基金（61002026)
A CFAR outlier detection method under varying SNR conditions
Received date: 2015-07-20
Revised date: 2015-12-19
Online published: 2015-12-28
Supported by
National Natural Science Foundation of China (61002026)
野值检测又称异常值检测，是模式识别、机器智能和知识发现等领域经常面临的一个问题。当出现环境失配，数据信噪比（SNR）发生变化时，测试样本和训练样本所含噪声会有不同方差，以往的野值检测方法在虚警控制方面将会失效。针对这一问题，提出一种基于归一化残差（NR）的野值检测方法。该方法首先根据所需虚警概率和噪声方差变化情况确定野值检测门限，其次基于训练样本计算待考查模式的NR值，再比较NR值与检测门限的相对大小，从而判断待考查模式是否为野值。这一方法所依赖的检测门限对所需虚警率和噪声方差变化具有适应能力，因此可以在变信噪比条件下实现恒虚警（CFAR）野值检测。仿真实验验证了所提方法在虚警控制和野值检测方面的优越性能。
汝小虎 , 柳征 , 姜文利 , 黄知涛 . 一种变信噪比条件下的恒虚警野值检测方法[J]. 航空学报, 2016 , 37(7) : 2259 -2268 . DOI: 10.7527/S1000-6893.2015.0348
Outlier detection, also called anomaly detection, is a commonly encountered problem in areas such as pattern recognition, machine intelligence and knowledge discovery. When environmental mismatch occurs and the signal-to-noise ratio (SNR) of data changes, the noise variances in testing and training instances will be different, resulting in the ineffectiveness of previous outlier detection methods which aim to control the false-alarm probability. To solve the problem, an outlier detection method based on normalized residual (NR) is proposed in this paper. This method first calculates the outlier detection threshold according to the desired false-alarm probability and the change of noise variance; second measures the NR value of the query pattern using training instances; and then compares this NR value with the predefined detection threshold to determine whether the query pattern is an outlier or not. The detection threshold defined in this paper is adaptive to the desired false-alarm probability and the varying noise variance, thus can realize constant false-alarm rate (CFAR) outlier detection under varying SNR conditions. Simulation experiments validate the superior performance of the proposed method on false-alarm probability controlling and outlier detection.
[1] HAWKINS D. Identification of outliers[M]. London:Chapman and Hall, 1980:1-25.
[2] DANESHPAZHOUH A, SAMI A. Entropy-based outlier detection using semi-supervised approach with few positive examples[J]. Pattern Recognition Letters, 2014, 49:77-84.
[3] ALBANESE A, PAL S K, PETROSINO A. Rough sets, kernel set, and spatiotemporal outlier detection[J]. IEEE Transactions on Knowledge and Data Engineering, 2014, 26(1):194-207.
[4] WESTERWEEL J, SCARANO F. Universal outlier detection for PIV data[J]. Experiments in Fluids, 2005, 39(6):1096-1100.
[5] DUNCAN J, DABIRI D, HOVE J, et al. Universal outlier detection for particle image velocimetry (PIV) and particle tracking velocimetry (PTV) data[J]. Measurement Science and Technology, 2010, 21(5):57002-57006.
[6] LIU J, WAN J, ZHENG H, et al. A method of specific emitter verification based on CSDA and SVDD[C]//Proceedings of the IEEE 2nd International Conference on Computer Science and Network Technology. Piscataway, NJ:IEEE Press, 2012:562-565.
[7] RU X H, LIU Z, JIANG W L. Normalized residual-based outlier detection[C]//Proceedings of the IEEE International Conference on Signal Processing, Communications and Computing (ICSPCC). Piscataway, NJ:IEEE Press, 2014:190-193.
[8] NATTORN B, ARTHORN L, KRUNG S. Outlier detection score based on ordered distance difference[C]//Proceedings of the IEEE International Computer Science and Engineering Conference (ICSEC). Piscataway, NJ:IEEE Press, 2013:157-162.
[9] ZHAO M, SALIGRAMA V. Anomaly detection with score functions based on nearest neighbor graphs[J]. Advances in Neural Information Processing Systems, 2009, 22(1):2250-2258.
[10] QIAN J, SALIGRAMA V. New statistic in p-value estimation for anomaly detection[C]//Proceedings of the IEEE Statistical Signal Processing Workshop (SSP). Piscataway, NJ:IEEE Press, 2012:393-396.
[11] CHEN Y T, QIAN J, SALIGRAMA V. A new one-class SVM for anomaly detection[C]//Proceedings of the IEEE International Conference on Acoustics, Speech, and Signal Processing (ICASSP). Piscataway, NJ:IEEE Press, 2013:3567-3571.
[12] SCHÖLKOPF B, PLATT J C, SHAWE-TAYLOR J C, et al. Estimating the support of a high-dimensional distribution[J]. Neural Computation, 2001, 13(7):1443-1471.
[13] FURLANI M, TUIA D, MUNOZ-MARI J, et al. Discovering single classes in remote sensing images with active learning[C]//Proceedings of the IEEE International Geoscience and Remote Sensing Symposium (IGARSS). Piscataway, NJ:IEEE Press, 2012:7341-7344.
[14] JUMUTC V, SUYKENS J. Multi-class supervised novelty detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 36(12):2510-2523.
[15] XUE Z, SHANG Y, FENG A. Semi-supervised outlier detection based on fuzzy rough C-means clustering[J]. Mathematics and Computers in Simulation, 2010, 80(9):1911-1921.
[16] HUNG J W, FAN H T. Subband feature statistics normalization techniques based on a discrete wavelet transform for robust speech recognition[J]. IEEE Signal Processing Letters, 2009, 16(9):806-809.
[17] SQUARTINI S, PRINCIPI E, ROTILI R, et al. Environmental robust speech and speaker recognition through multi-channel histogram equalization[J]. Neurocomputing, 2012, 78(1):111-120.
[18] DAI P, SOON I Y. An improved model of masking effects for robust speech recognition system[J]. Speech Communication, 2013, 55(3):387-396.
[19] VALENZUELA O, PASADAS M. Fuzzy data approximation using smoothing cubic splines:Similarity and error analysis[J]. Applied Mathematical Modelling, 2011, 35(5):2122-2144.
[20] FREI M G, OSORIO I. Intrinsic time-scale decomposition:Time-frequency-energy analysis and real-time filtering of non-stationary signals[J]. Royal Society of London Proceedings, 2007, 463(2078):321-342.
[21] ZENG J X, WANG G F, ZHANG F Q, et al. The de-noising algorithm based on intrinsic time-scale decomposition[J]. Advanced Materials Research, 2011, 422:347-352.
/
| 〈 |
|
〉 |
CJA