基于高分辨一维距离像的雷达目标识别
|
摘要
雷达目标识别(RTI-radar target identification)是现代雷达的一个十分重要的发展方向,高分辨雷达技术的日趋成熟使雷达目标识别走向实用成为可能。本文应用现代模式识别技术和现代信号处理技术的方法研究了高分辨雷达一维距离像的目标识别问题,由于高分辨雷达一维距离像能够提供更加精细的目标结构和形状信息,因此,对基于一维距离像的目标识别研究具有较大的理论意义和现实意义。本文主要内容如下:
首先综述了雷达目标识别的基本概念和距离像识别的基本方法,对雷达目标识别技术现状进行分析。介绍了雷达目标散射截面(RCS-radar cross section)的概念以及宽带目标RCS特性及预估的方法。综述了目标多散射中心的概念及其距离像模型,对典型飞机的电磁散射现象做了深入分析。重点研究了高分辨率一维距离像的原理及仿真方法,从目标散射中心模型的数学描述入手,对单个距离像回波特性及对距离像交叉项进行了分析。研究了一维距离像识别的关键问题—方位敏感性、平移敏感性、幅度敏感性及其预处理方法。
在研究了线性调频、步进频率、二相编码无载波超宽带雷达成像算法基础上,分析了多目标和多散射中心分布一维距离像的回波特性。本文针对线性调频宽带数字脉冲压缩收发信号,分析了高分辨率雷达stretch处理的一维距离像目标识别问题。推导了多目标的宽带线性调频信号stretch处理的雷达回波数学模型,分析了依据一维距离像对多目标进行识别方法。讨论了目标运动所产生的距离—多普勒耦合导致距离像平移和展宽特性,并对目标速度的影响进行了补偿。
接下来,简单介绍了步进频率信号的数学模型,给出了一维距离像的外形包络表达式。针对多散射中心高距离分辨率步进频率雷达目标,分析了当目标存在相对运动时的一维距离像数学模型,按照这一模型,通过计算机对实际系统的模拟仿真,深入研究了多普勒效应对目标一维距离像的影响,最后,获得了给定高距离分辨率步进频率毫米波雷达中,用一维距离像匹配识别目标的速度不补偿条件。并对两种体制的雷达做了对比,分析了目标运动的多普勒效应对线性调频和步进频率一维距离像的影响,对运动目标的速度估计与补偿方法进行了分析,并用算法对速度估计与补偿方法进行了验证。
在研究了高分辨雷达一维成像方法的基础上,介绍了基于模式匹配识别的雷达目标结构信息的特征提取方法,同时介绍了两种常用的线性数据降维方法:主成份分析法和规范相关分析法用于对一维距离像的特征进行降维。
最后,总结了本文的研究工作,指出了需要进一步解决的问题。
Radar target identification(RTI)is an important developing field of modern radar technology. Modern radar technology become mature day by day makes it possible to develop a practical RTI system. In this paper, the problem of recognition of radar target using High Resolution Range Profiles(HRRP) is mainly studied by the means of techniques in modern signal process and in modern pattern recognition. Because High Resolution Range Profile can provide more accurate target structure and shape information,research on RTI based on one-dimensional range profile is significance for both theory and practice. In this thesis, its content and originality are synthesized as follows:
First, the basic concepts of radar target recognition and the basic method of recognition of range profiles are surveied. The status quo of radar target identification is analyzed. The paper introduce the concept of radar target cross section and broadband target RCS characteristics, RCS pre-estimating method. The concept of multi-target scattering center and Range profile model are bewrited. The characteristics of typical aircraft electromagnetic scattering that vary with aspect are analyzed deeply. In particular , the principle and simulation method of High Hesolution Range Profiles is studied. After the mathematical model of the Range Profile features which is based on the multi-scattering Center, analysis have been made on echo characteristics of a single range profile and cross-range profile. Research has also been made on key techniques of recognition of one-dimensional range profile in which aspect sensitivity,translation sensitivity and amplitude sensitivity should deal with pretreatment.
Studying of linear frequency modulation, step frequency, no-load binary phase coded ultra-wideband radar imaging algorithm based on . Analysis of the echo characteristics of multi-objective and multi-scattering centers are distributed one-dimensional range profile.
Under the wide-band digital pulse compression of linear frequency modulation (LFM) signal both for receiving and transmitting, the target recognition problem forming the one-dimension range profile in high-resolution radar stretch processing range is analyzed. The radar-echo model for multi-target of wideband linear frequency modulated signal Stretch processing is derived, and the multi-target identification according to one-dimension range profile is analyzed. The received pulse width expanding and center frequency shifting due to range-doppler coupling of moving target are discussed, and a radial velocity compensation method is proposed.
Secondly, this paper firstly introduces the mathematical model of stepped frequency and one dimensional range imaging technique. The mathematical model of one dimensional range imaging for high range resolution radar target when there is relative motion between the radar and the target is given. Using this mathematical model, the influence of target relative motion on one dimensional range imaging is investigated by computer simulation.Finally, the value of the compensation error of this relative motion between the radar and the target is derived. when one dimensional range imaging technique is used for target identification. By comparison with two kinds of radar systems, the received pulse width expanding and center frequency shifting due to range-doppler coupling, the impact of the doppler effect on the stepped frequency and linear frequency from one-dimensional range profile of moving target are discussed,in which analyze of velocity estimation and compensation methods . Computer simulation shows that the method of speed estimation and compensation has been validated。
In the paper, a method is presented which gets radar target high dimensional structure information from one dimensional range Profiles based on pattern-matching recognition. When doing dimension reduction of the Range Profile features, two kinds of common linear data dimensional reduction methods is principle component analysis and canonical component analysis would contributed greatly to it.
In conclusion, the main works of the dissertation are summarized and the future research areas are pointed out.
首先综述了雷达目标识别的基本概念和距离像识别的基本方法,对雷达目标识别技术现状进行分析。介绍了雷达目标散射截面(RCS-radar cross section)的概念以及宽带目标RCS特性及预估的方法。综述了目标多散射中心的概念及其距离像模型,对典型飞机的电磁散射现象做了深入分析。重点研究了高分辨率一维距离像的原理及仿真方法,从目标散射中心模型的数学描述入手,对单个距离像回波特性及对距离像交叉项进行了分析。研究了一维距离像识别的关键问题—方位敏感性、平移敏感性、幅度敏感性及其预处理方法。
在研究了线性调频、步进频率、二相编码无载波超宽带雷达成像算法基础上,分析了多目标和多散射中心分布一维距离像的回波特性。本文针对线性调频宽带数字脉冲压缩收发信号,分析了高分辨率雷达stretch处理的一维距离像目标识别问题。推导了多目标的宽带线性调频信号stretch处理的雷达回波数学模型,分析了依据一维距离像对多目标进行识别方法。讨论了目标运动所产生的距离—多普勒耦合导致距离像平移和展宽特性,并对目标速度的影响进行了补偿。
接下来,简单介绍了步进频率信号的数学模型,给出了一维距离像的外形包络表达式。针对多散射中心高距离分辨率步进频率雷达目标,分析了当目标存在相对运动时的一维距离像数学模型,按照这一模型,通过计算机对实际系统的模拟仿真,深入研究了多普勒效应对目标一维距离像的影响,最后,获得了给定高距离分辨率步进频率毫米波雷达中,用一维距离像匹配识别目标的速度不补偿条件。并对两种体制的雷达做了对比,分析了目标运动的多普勒效应对线性调频和步进频率一维距离像的影响,对运动目标的速度估计与补偿方法进行了分析,并用算法对速度估计与补偿方法进行了验证。
在研究了高分辨雷达一维成像方法的基础上,介绍了基于模式匹配识别的雷达目标结构信息的特征提取方法,同时介绍了两种常用的线性数据降维方法:主成份分析法和规范相关分析法用于对一维距离像的特征进行降维。
最后,总结了本文的研究工作,指出了需要进一步解决的问题。
Radar target identification(RTI)is an important developing field of modern radar technology. Modern radar technology become mature day by day makes it possible to develop a practical RTI system. In this paper, the problem of recognition of radar target using High Resolution Range Profiles(HRRP) is mainly studied by the means of techniques in modern signal process and in modern pattern recognition. Because High Resolution Range Profile can provide more accurate target structure and shape information,research on RTI based on one-dimensional range profile is significance for both theory and practice. In this thesis, its content and originality are synthesized as follows:
First, the basic concepts of radar target recognition and the basic method of recognition of range profiles are surveied. The status quo of radar target identification is analyzed. The paper introduce the concept of radar target cross section and broadband target RCS characteristics, RCS pre-estimating method. The concept of multi-target scattering center and Range profile model are bewrited. The characteristics of typical aircraft electromagnetic scattering that vary with aspect are analyzed deeply. In particular , the principle and simulation method of High Hesolution Range Profiles is studied. After the mathematical model of the Range Profile features which is based on the multi-scattering Center, analysis have been made on echo characteristics of a single range profile and cross-range profile. Research has also been made on key techniques of recognition of one-dimensional range profile in which aspect sensitivity,translation sensitivity and amplitude sensitivity should deal with pretreatment.
Studying of linear frequency modulation, step frequency, no-load binary phase coded ultra-wideband radar imaging algorithm based on . Analysis of the echo characteristics of multi-objective and multi-scattering centers are distributed one-dimensional range profile.
Under the wide-band digital pulse compression of linear frequency modulation (LFM) signal both for receiving and transmitting, the target recognition problem forming the one-dimension range profile in high-resolution radar stretch processing range is analyzed. The radar-echo model for multi-target of wideband linear frequency modulated signal Stretch processing is derived, and the multi-target identification according to one-dimension range profile is analyzed. The received pulse width expanding and center frequency shifting due to range-doppler coupling of moving target are discussed, and a radial velocity compensation method is proposed.
Secondly, this paper firstly introduces the mathematical model of stepped frequency and one dimensional range imaging technique. The mathematical model of one dimensional range imaging for high range resolution radar target when there is relative motion between the radar and the target is given. Using this mathematical model, the influence of target relative motion on one dimensional range imaging is investigated by computer simulation.Finally, the value of the compensation error of this relative motion between the radar and the target is derived. when one dimensional range imaging technique is used for target identification. By comparison with two kinds of radar systems, the received pulse width expanding and center frequency shifting due to range-doppler coupling, the impact of the doppler effect on the stepped frequency and linear frequency from one-dimensional range profile of moving target are discussed,in which analyze of velocity estimation and compensation methods . Computer simulation shows that the method of speed estimation and compensation has been validated。
In the paper, a method is presented which gets radar target high dimensional structure information from one dimensional range Profiles based on pattern-matching recognition. When doing dimension reduction of the Range Profile features, two kinds of common linear data dimensional reduction methods is principle component analysis and canonical component analysis would contributed greatly to it.
In conclusion, the main works of the dissertation are summarized and the future research areas are pointed out.
引文
[1]Barton D.K.Sputink II as observed by C band radar[J].IRE Nat.Conf.Rec.1959,7(Parts):67-73.
[2]孙文峰.雷达目标识别技术评述[J].雷达与对抗,2001,3:1-8.
[3]王晓丹,王积勤.雷达目标识别技术综述[J].现代雷达,2003,25(5):22-26.
[4]廖学军.基于高分辨距离像的雷达目标识别[D].西安:西安电子科技大学博士论文,1999.
[5]袁莉,刘宏伟,保铮. MUSIC超分辨距离像在雷达目标识别中的问题[J].现代雷达,2005,27(5):45-48.
[6]黄培康.雷达目标特征信号[M].北京:宇航出版社,1993,2-12.
[7]黄培康.反导弹系统中的目标识别技术[J].北京:战略防御,1981,1-14.
[8]V.K.Joni,T.K.Sarker,D.D.Weiner.Rational modeling by Pencil of functions method[J].IEEE Trans.A.S.S.P,1983,31(3):564-573.
[9]BerniA J.Target identification by natural resource estimation[J].IEEE TransAES 1975,11(2):147-154.
[10]K.M.chen.Radar waveform synthesis method—a new radar detection scheme[J].IEEE Trans.A.P,1981,29(4):553-565.
[11] E.J.Rothwell , D.P.Nyqist,K.M.chen.Radar target discrimination using the extinction-pulse tehnique[J].IEEETrans.A.P,1985,33(9):929-936.
[12]何松华,高距离分辨率毫米波雷达目标识别的理论与应用[D].长沙:国防科技大学博士论文,1993:5-40.
[13]闫锦,黄培康.高距离分辨像雷达目标识别[J].航天电子对抗,2004(2):36-41.
[14]Blaricum MLVan,Mittra R.A technique for extracting the Poles and residues of a system dircetly from its translent response[J]. IEEE Trans.on AP,1975,23(6):77-781.
[15]H.Li,S.Yang.Using range Profiles as feature vectors to identify aerospace object[J].IEEE Trans.on AP,1993,41(3):261-280.
[16]S.Hudos,Psalltis.Correlation filters for aircraft identification from range Profiles[J]. IEEE Trans.on AES,1993,29(3):741-748.
[17]苗雨,姜文利等.基于高分辨距离像的雷达目标识别特征分析[J].国防科技大学学报,2001(5):93-97.
[18]R.Carriere, R.L.Moses.High resolution radar modeling using a modified Prony estimator[J].IEEE Trans.on AP,1992,40:13-8.
[19]姜卫东.光学区雷达目标结构成像的理论及其在雷达目标识别中的应用[D].长沙:国防科技大学博士论文,2000.6-30.
[20]K.T.Kim,D.K.Seo,et.al.Effieient radar target recognition using the MUSIC algorithm and invariant features[J ].IEEE Trans.on AP,2002,50(3):325-337.
[21]Bowman J J,et al.Electromagnetic and Acoustic Scattering by Simple Shapes[M].Amster-dam:North-HollandPublishing,1969,45-63.
[22]Kennaugh E M,Moffatt D L.Transient and Impulse Response Approximations[J].Proc.of IEEE,1965(8):893-901.
[23]Burnside W D,et al.A Technique to Combine the Geometrical Theory of Diffraction and the Moment Method[J].IEEE Trans.on Antenas Propagant,1974(4):551-558.
[24]Ruck G T.Radar Cross Section Handbook[M].New York:Plenum Press,1970.28-125.
[25]J.W.Crispin.Method of Radar Section Analysis[M].ed.New York:Academic Press.1968.12-25.
[26]Fritch P C.Special Issue on Radar Reflectivity[J].Proc.IEEE,1965,53(8).
[27]刘国岁,孙光民等.连续波雷达调频波接收机性能分析[J].现代雷达.1995,12(6):20-36.
[28]赵霞.三毫米线性调频连续波雷达信号处理技术的研究[M].南京:南京理工大学硕士论文,2004,7-30.
[29] Caputi W J . St retch : a time t ransformation technique.[J ] . IEEE Trans on AES , 1971 , 7 (2) : 26-29.
[30]王俊,张守宏.微弱目标积累检测的包络移动补偿方法[J].电子学报2000,28(12):56-59.
[31]Wang Z,Willett P,Streit P.Detection of long_duration narrowband processes.IEEE Trans AES,2002,38(1):211-277.
[32]Farina A.Detection with high resolution radar:Great promise big challge.Microwave[J],1999,23(5):263-265.
[33]沈吉,向锦武,祁载康.目标运动对步进频率毫米波雷达图像匹配识别的影响研究[J].电子学报,2003,31(3):1-4.
[34] Wu X Q , Zhu Z D. Simultaneous imaging of multiple targets in an inverse synthetic aperture radar [A] . Proc of IEEE NAECON[ C]. USA: IEEE , 1990. 210– 214.
[35]杨军,王民胜,保铮.一种ISAR多目标实时成像方法[J] .电子学报, 1995 , 23(4) :1-5.
[36]Y.H.Cao,Y.B.Zhang,S.H.Wang,S.H.Zhang.Wideband digital beamforming when using LFM waveform [C]. Beijing:Microwave,Antenna,Propagation and EMC Technologies for Wireless Communications8-12 Aug. 2005 :1275– 1278.
[37] Wehner D R. High Resolution Radar. Norwood[J],MA:Artech house,1987.
[38]王瑜,秦忠宇,文树梁.高分辨雷达去斜处理一维距离像速度补偿技术[J].系统工程与电子技术,2006,26(12):1757-759.
[39]黄德双.高分辨率雷达智能信号处理技术[M]北京:机械工业出版社,1996.3-45.
[40]P Granell.Guided Weapon Control System[M]UK:Pergamon press.1980.2-18.
[41]龙腾.频率步进雷达信号的多谱勒性能分析[J].现代雷达,1996(2):31-37.
[42]刘静,李兴国,顾玉辉.频率步进雷达关键技术的研究[J].制导与引信,2003,24(4),24-28.
[43]沈吉,等.IFFT步长对步进频率毫米波导引头动目标辨识的影响研究[J].弹箭与制导学报,2001,21(4):21-26.
[44]王华力,李兴国.高距离分辨率脉间频率步进系统的探讨[J].现代雷达,1995,17(l):91-99.
[45]王卫江,沈庭芝.步进频信号处理的多普勒补偿新方法[J]北京理工大学学报,2006,126(4):353-355.
[46]孙晓乐.相位编码信号脉冲压缩研究与实现[D].西安:西安电子科技大学硕士文.2008,3-42.
[47]田黎育,高梅国.一种二相码信号多普勒补偿方法的研究与实现[J].北京理工大学学报.2002,22(6).757-760.
[48]周正.一维相扫雷达混合编码脉冲压缩技术[D].西安:西安电子科技大学硕士学位论文.2008,6-25.
[49]廖学军,保铮.雷达目标距离像分类机理的研究[J].西安电子科技大学学报,1997,24 (增刊):68-74.
[50]郭桂蓉,庄钊文,陈曾平.电磁特征提取与目标识别[M].长沙:国防科技大学出版社,1996:12-35.
[51]J.Shawe Taylor&N.Cristianini,Kernel Methods for Pattern Analysis[M], Cambridge University Press,2004.20-42.
[52]B.Scholkopf,A.Smola,and K.R.Müller,“Nonlinear component analysis as a kernel eigenvalue problem,”Neural Compute,vol.10,pp. 1998,1299–1319.
[2]孙文峰.雷达目标识别技术评述[J].雷达与对抗,2001,3:1-8.
[3]王晓丹,王积勤.雷达目标识别技术综述[J].现代雷达,2003,25(5):22-26.
[4]廖学军.基于高分辨距离像的雷达目标识别[D].西安:西安电子科技大学博士论文,1999.
[5]袁莉,刘宏伟,保铮. MUSIC超分辨距离像在雷达目标识别中的问题[J].现代雷达,2005,27(5):45-48.
[6]黄培康.雷达目标特征信号[M].北京:宇航出版社,1993,2-12.
[7]黄培康.反导弹系统中的目标识别技术[J].北京:战略防御,1981,1-14.
[8]V.K.Joni,T.K.Sarker,D.D.Weiner.Rational modeling by Pencil of functions method[J].IEEE Trans.A.S.S.P,1983,31(3):564-573.
[9]BerniA J.Target identification by natural resource estimation[J].IEEE TransAES 1975,11(2):147-154.
[10]K.M.chen.Radar waveform synthesis method—a new radar detection scheme[J].IEEE Trans.A.P,1981,29(4):553-565.
[11] E.J.Rothwell , D.P.Nyqist,K.M.chen.Radar target discrimination using the extinction-pulse tehnique[J].IEEETrans.A.P,1985,33(9):929-936.
[12]何松华,高距离分辨率毫米波雷达目标识别的理论与应用[D].长沙:国防科技大学博士论文,1993:5-40.
[13]闫锦,黄培康.高距离分辨像雷达目标识别[J].航天电子对抗,2004(2):36-41.
[14]Blaricum MLVan,Mittra R.A technique for extracting the Poles and residues of a system dircetly from its translent response[J]. IEEE Trans.on AP,1975,23(6):77-781.
[15]H.Li,S.Yang.Using range Profiles as feature vectors to identify aerospace object[J].IEEE Trans.on AP,1993,41(3):261-280.
[16]S.Hudos,Psalltis.Correlation filters for aircraft identification from range Profiles[J]. IEEE Trans.on AES,1993,29(3):741-748.
[17]苗雨,姜文利等.基于高分辨距离像的雷达目标识别特征分析[J].国防科技大学学报,2001(5):93-97.
[18]R.Carriere, R.L.Moses.High resolution radar modeling using a modified Prony estimator[J].IEEE Trans.on AP,1992,40:13-8.
[19]姜卫东.光学区雷达目标结构成像的理论及其在雷达目标识别中的应用[D].长沙:国防科技大学博士论文,2000.6-30.
[20]K.T.Kim,D.K.Seo,et.al.Effieient radar target recognition using the MUSIC algorithm and invariant features[J ].IEEE Trans.on AP,2002,50(3):325-337.
[21]Bowman J J,et al.Electromagnetic and Acoustic Scattering by Simple Shapes[M].Amster-dam:North-HollandPublishing,1969,45-63.
[22]Kennaugh E M,Moffatt D L.Transient and Impulse Response Approximations[J].Proc.of IEEE,1965(8):893-901.
[23]Burnside W D,et al.A Technique to Combine the Geometrical Theory of Diffraction and the Moment Method[J].IEEE Trans.on Antenas Propagant,1974(4):551-558.
[24]Ruck G T.Radar Cross Section Handbook[M].New York:Plenum Press,1970.28-125.
[25]J.W.Crispin.Method of Radar Section Analysis[M].ed.New York:Academic Press.1968.12-25.
[26]Fritch P C.Special Issue on Radar Reflectivity[J].Proc.IEEE,1965,53(8).
[27]刘国岁,孙光民等.连续波雷达调频波接收机性能分析[J].现代雷达.1995,12(6):20-36.
[28]赵霞.三毫米线性调频连续波雷达信号处理技术的研究[M].南京:南京理工大学硕士论文,2004,7-30.
[29] Caputi W J . St retch : a time t ransformation technique.[J ] . IEEE Trans on AES , 1971 , 7 (2) : 26-29.
[30]王俊,张守宏.微弱目标积累检测的包络移动补偿方法[J].电子学报2000,28(12):56-59.
[31]Wang Z,Willett P,Streit P.Detection of long_duration narrowband processes.IEEE Trans AES,2002,38(1):211-277.
[32]Farina A.Detection with high resolution radar:Great promise big challge.Microwave[J],1999,23(5):263-265.
[33]沈吉,向锦武,祁载康.目标运动对步进频率毫米波雷达图像匹配识别的影响研究[J].电子学报,2003,31(3):1-4.
[34] Wu X Q , Zhu Z D. Simultaneous imaging of multiple targets in an inverse synthetic aperture radar [A] . Proc of IEEE NAECON[ C]. USA: IEEE , 1990. 210– 214.
[35]杨军,王民胜,保铮.一种ISAR多目标实时成像方法[J] .电子学报, 1995 , 23(4) :1-5.
[36]Y.H.Cao,Y.B.Zhang,S.H.Wang,S.H.Zhang.Wideband digital beamforming when using LFM waveform [C]. Beijing:Microwave,Antenna,Propagation and EMC Technologies for Wireless Communications8-12 Aug. 2005 :1275– 1278.
[37] Wehner D R. High Resolution Radar. Norwood[J],MA:Artech house,1987.
[38]王瑜,秦忠宇,文树梁.高分辨雷达去斜处理一维距离像速度补偿技术[J].系统工程与电子技术,2006,26(12):1757-759.
[39]黄德双.高分辨率雷达智能信号处理技术[M]北京:机械工业出版社,1996.3-45.
[40]P Granell.Guided Weapon Control System[M]UK:Pergamon press.1980.2-18.
[41]龙腾.频率步进雷达信号的多谱勒性能分析[J].现代雷达,1996(2):31-37.
[42]刘静,李兴国,顾玉辉.频率步进雷达关键技术的研究[J].制导与引信,2003,24(4),24-28.
[43]沈吉,等.IFFT步长对步进频率毫米波导引头动目标辨识的影响研究[J].弹箭与制导学报,2001,21(4):21-26.
[44]王华力,李兴国.高距离分辨率脉间频率步进系统的探讨[J].现代雷达,1995,17(l):91-99.
[45]王卫江,沈庭芝.步进频信号处理的多普勒补偿新方法[J]北京理工大学学报,2006,126(4):353-355.
[46]孙晓乐.相位编码信号脉冲压缩研究与实现[D].西安:西安电子科技大学硕士文.2008,3-42.
[47]田黎育,高梅国.一种二相码信号多普勒补偿方法的研究与实现[J].北京理工大学学报.2002,22(6).757-760.
[48]周正.一维相扫雷达混合编码脉冲压缩技术[D].西安:西安电子科技大学硕士学位论文.2008,6-25.
[49]廖学军,保铮.雷达目标距离像分类机理的研究[J].西安电子科技大学学报,1997,24 (增刊):68-74.
[50]郭桂蓉,庄钊文,陈曾平.电磁特征提取与目标识别[M].长沙:国防科技大学出版社,1996:12-35.
[51]J.Shawe Taylor&N.Cristianini,Kernel Methods for Pattern Analysis[M], Cambridge University Press,2004.20-42.
[52]B.Scholkopf,A.Smola,and K.R.Müller,“Nonlinear component analysis as a kernel eigenvalue problem,”Neural Compute,vol.10,pp. 1998,1299–1319.