基于支持向量机的SAR图像目标识别
|
摘要
合成孔径雷达(SAR)具有全天候、远距离、极强的穿透力和高分辨率等特点,在国民经济和军事领域中都有着广泛的应用。如何对SAR图像进行快速、准确地解译越来越引起人们的关注和重视。将机器学习领域新的研究成果应用到雷达目标识别中并构造有效的分类器具有重要的意义。建立在统计学习理论基础之上的支持向量机方法(SVM)被看作是对传统学习分类方法的一个好的替代,特别是在小样本、高维和非线性情况下,具有较好的泛化性能。
本文基于支持向量机对SAR图像的目标特征提取和分类识别方法进行了较为深入的研究,所做的工作主要有:综述了SAR图像目标识别的研究现状,总结出其中的关键技术和一般流程;研究了基于SAR图像形状特征的Hu不变矩特征提取的方法,该方法的优势是具有旋转、平移和尺度不变性;将支持向量机方法应用于SAR图像目标识别,该方法在小样本、非线性情况下能够达到较高的识别率;对传统支持向量机训练算法进行了改进:即利用SVM训练中支持向量的分布特点,采用预先提取边界向量和循环迭代的方法进行训练,减小了训练样本规模,提高了训练速度。
最后通过仿真实验证明:利用Hu不变矩特征和支持向量机相结合可以获得较高的识别精度,是一种有效的SAR图像目标识别方法;本文给出的快速支持向量机训练算法在不影响分类正确率的前提下提高了样本训练的速度。
Synthetic aperture radar (SAR) has been widely applied to national economic fields and military reconnaissance fields because of its all-weather, wide, strong penetrable ability and supplying detailed ground mapping material and images in atrocious weather with high resolution. The collection capacity for SAR images is growing rapidly, and along with that growth is the expanding need for exploitation of SAR images accurately and efficiently. Based on machine learning theory, developing the practical and effective classifier is of great importance. Support vector machines (SVM), which are based on statistical learning theory, are considered good candidates because of their high generalization performance even when the dimension of the input space is very high and the problem is nonlinear.
This thesis studies SAR image feature extraction and target recognition based on SVM. The main contents and contributions are as follows: Firstly, reviews the development of SAR ATR techniques. Secondly, the invariant moments feature extraction method is proposed. Its efficiency is proved moment invariant with scale, translation and rotating invariance. Thirdly, a SAR image target recognition algorithm using SVM is introduced. The results demonstrate the effectiveness of this method. Finally, The SVM training time is reduced dramatically owing to a fast SVM training method which gets support vectors in advance and then uses an iterative and circulative strategy for training.
In simulation experiments, we recognize the images using the invariant moments and SVM, Its successful rate is very high, therefore the results show that this method is valuable.
本文基于支持向量机对SAR图像的目标特征提取和分类识别方法进行了较为深入的研究,所做的工作主要有:综述了SAR图像目标识别的研究现状,总结出其中的关键技术和一般流程;研究了基于SAR图像形状特征的Hu不变矩特征提取的方法,该方法的优势是具有旋转、平移和尺度不变性;将支持向量机方法应用于SAR图像目标识别,该方法在小样本、非线性情况下能够达到较高的识别率;对传统支持向量机训练算法进行了改进:即利用SVM训练中支持向量的分布特点,采用预先提取边界向量和循环迭代的方法进行训练,减小了训练样本规模,提高了训练速度。
最后通过仿真实验证明:利用Hu不变矩特征和支持向量机相结合可以获得较高的识别精度,是一种有效的SAR图像目标识别方法;本文给出的快速支持向量机训练算法在不影响分类正确率的前提下提高了样本训练的速度。
Synthetic aperture radar (SAR) has been widely applied to national economic fields and military reconnaissance fields because of its all-weather, wide, strong penetrable ability and supplying detailed ground mapping material and images in atrocious weather with high resolution. The collection capacity for SAR images is growing rapidly, and along with that growth is the expanding need for exploitation of SAR images accurately and efficiently. Based on machine learning theory, developing the practical and effective classifier is of great importance. Support vector machines (SVM), which are based on statistical learning theory, are considered good candidates because of their high generalization performance even when the dimension of the input space is very high and the problem is nonlinear.
This thesis studies SAR image feature extraction and target recognition based on SVM. The main contents and contributions are as follows: Firstly, reviews the development of SAR ATR techniques. Secondly, the invariant moments feature extraction method is proposed. Its efficiency is proved moment invariant with scale, translation and rotating invariance. Thirdly, a SAR image target recognition algorithm using SVM is introduced. The results demonstrate the effectiveness of this method. Finally, The SVM training time is reduced dramatically owing to a fast SVM training method which gets support vectors in advance and then uses an iterative and circulative strategy for training.
In simulation experiments, we recognize the images using the invariant moments and SVM, Its successful rate is very high, therefore the results show that this method is valuable.
引文
[1] 常庆瑞,蒋平安,周勇,申光荣.遥感技术导论.北京:科学出版社,2004
[2] 宋建社,袁礼海,薛文通.SAR图像处理的最新研究与应用[J].遥感技术与应用,2002,17(5):284～288
[3] 郦苏丹.SAR 图像特征提取与目标识别方法研究[D].长沙:国防科学技术大学,2001
[4] 刘迎春,宋建社,郑永安.SAR图像识别提取与选择特征[J].航空计算技术,2003,33(4):55～58
[5] Ross T.D., Bradley J.J. .SAR ATR –So What’s The Problem? –An MSTAR Perspective. Proceedings of SPIE –The International Society for Optical Engineering,1999,3721: 662～672
[6] Pham Q.H., Ezekiel A. Campbell M.T.. A New end-to-end SAR ATR System [J]. Proceedings of SPIE- The International Society for Optical Engineering,1999,3721:292～301
[7] 李补莲.自动目标识别技术发展评述[J].现代防御技术,2000,28(2):10～14
[8] Blackwell D. Feature analysis for SAR ATR [J] . Proceedings of SPIE –The International Society for Optical Engineering, 2000, 4173:75～86
[9] Schistad Solberg H., Jain A.K. Texture Fusion and Feature Selection Applied to SAR Images [J]. IEEE Transaction on Geosciences and Remote Sensing,1997,35(2):475～479
[10] Chen Chia-Tang, Chen Kun-Shan, Lee Jong-Sen. The use of fully polar metric information for the fuzzy neural classification of SAR images [J]. IEEE Transaction on Geosciences and Remote Sensing,2003,41(9):2089～2100
[11] Vapnik V. Estimation of Dependence Based on Empirical Data. New York: Springer-Verlag, 1982
[12] 许建华,张学工,李衍达.支持向量机的新发展[J].控制与决策,2004,19(5):481～484
[13] Edgar Osuna. Improved Training Algorithm for Support Vector Machines [J]. IEEE Transactions on Neural Networks,1997,10(5):1000～1017
[14] John C.Platt. Sequential Minimal Optimization: A Fast Algorithm for Training Support Vector Machines [J]. IEEE Transactions on Neural Networks,1998
[15] Zhang X G. Using Class-center Vector to Build Support Vector Machines[C]. In: Proceedings of NNSP’99,1999
[16] S.S.Keerthi. Improvements to Platt’s SMO Algorithm for SVM Classifier Design.Technical Report CD-99-14
[17] Morain S.A. Use of Radar for Vegetation Analysis,in Geosciences Application of Image Radar System. Remote Sensing of the Electromagnetic Spectrum,1976,3(3):61～78
[18] 徐产兴.雷达目标识别技术及其新发展[J].雷达与对抗,1994,3:1～9
[19] 张朋等.合成孔径雷达成像三维地形目标模拟方法[J].系统仿真学报,2005,17(10)
[20] 杜亚娟,潘泉.雷达目标识别方法概述[J].火控雷达技术,1998,27:1～5
[21] 傅雄军等.一种有效的聚束式合成孔径雷达图像特征提取的算法[J].北京理工大学学报,2004,24(7):14～17
[22] 韩萍.SAR 自动目标识别及相关技术研究[D].天津:天津大学,2004
[23] 鲁加国,张长耀.高分辨率机载 SAR 雷达研究与试验[C].中国合成孔径雷达会议论文集,2003,86～88
[24] Ward K.D. Compound Representation of High Resolution Sea Clutter [J].Electron lett., 1981,17:561～565
[25] Shi Z.H. A Comparison of Digital Speckle Filters [J]. Processing of IGARSS’94, 1994, 2129～2133
[26] Lee J.S. Speckle Analysis and Smoothing of Synthetic Aperture Radar Image [J]. Computer Graphic and Image Processing,1981,17:24～32
[27] Kuan D.T. Adaptive Restoration of Images with Speckle[J]. IEEE Trans. On Acoustics, Speech and Signal Processing,1987,35(3):373～383
[28] 高贵,李德仁.高分辨率 SAR 图像分割及目标特征提取[J].宇航学报,2006,27(3)
[29] 刘国庆,熊红,黄顺吉.基于小波变换和马尔可夫随机场的极化 SAR 图像自动分类[J],电子科学学刊,2000, 22(3) :359～365
[30] 杨晖,曲秀杰.图像分割方法综述[J].电脑开发与应用,2005,18(3):21～23
[31] R.C.Gonzalez and R.E.Woods. 数字图像处理(第二版).阮秋琦,阮宇智等译.北京:电子工业出版社,2003
[32] Novak L.M., Halversen Shawn D., and Hiett Margarita. Effects of polarization and resolution on the performance of a SAR automatic target recognition system[J].The Lincoln Laboratory Journal,1995,8(1):49～67
[33] Tupin Florence. Detection of Linear Feature in SAR Images: Application to Road Network Extraction[J]. IEEE Trans. On Geosciences Remote Sensing,1998,36(2): 434～453
[34] Tan Hwee H.P., Ramanathan U. Intelligent selection of useful features for optimal feature-based classification [J]. Proceedings of IEEE International Geosciences and Remote Sensing Symposium (IGARSS 2000),2000,7:3012～3014
[35] 王晓红.矩技术及其在图像处理与识别中的应用研究[D].西安:西北工业大学,2001
[36] Hu M K. Visual pattern recognition by moment invariants [J].IEEE Trans. On Information Theory,1962:170～179
[37] 薛笑荣.SAR 图像处理技术研究[D].西安:西北工业大学,2004
[38] Vapnik V. The Nature of Statistical Learning Theory. New York: Springer-Verlag,2000. (中译本,张学工译.统计学习理论的本质,北京:清华大学出版社,2000)
[39] 张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1) :32～42
[40] C Cortes,Vapnik V. Support Vector Networks[J]. Machine Learning,1995,20(3):273～297
[41] 李国正,王猛,曾华军译.支持向量机导论.北京:电子工业出版社,2000
[42] J.Weston and C.Watkins. Multi-class support vector machines. Technical Report,1998,CSD-TR-98-04: 1～9
[43] Knerr S,et al. Single layer learning revisited: A step wise procedure for building and training a neural network [A]. NATOASI. Springer, 1990
[44] C Huang,L S Davis. An assessment of support vector machines for land cover classification [J]. International Geosciences and Remote Sensing,2002,23: 725～749
[45] 姚恩瑜,何勇,陈仕平.数学规划与最优化.杭州:浙江大学出版社,2001
[46] 安金龙,王正欧等.一种新的支持向量机多类分类方法[J].信息与控制,2004,33(3) :262～267
[47] Chih-Chung Chang and Chih-Jen Lin. LIBSVM: a Library for Support Vector Machines.Department of Computer Science and Information Engineering, National Taiwan University, Taipei 106, Taiwan. April,17,2005
[48] 何壸，刘宏伟等. 基于 MSTAR 数据的 SAR 图像自动目标识别研究[C],中国合成孔径雷达会议论文集,2005(10) :278～282
[49] Qun Zhao, Jose C. Support Vector Machines for SAR Automatic Target Recognition [J]. IEEE Transactions on Aerospace and Electronic Systems,2001,37 (2):643～654
[2] 宋建社,袁礼海,薛文通.SAR图像处理的最新研究与应用[J].遥感技术与应用,2002,17(5):284～288
[3] 郦苏丹.SAR 图像特征提取与目标识别方法研究[D].长沙:国防科学技术大学,2001
[4] 刘迎春,宋建社,郑永安.SAR图像识别提取与选择特征[J].航空计算技术,2003,33(4):55～58
[5] Ross T.D., Bradley J.J. .SAR ATR –So What’s The Problem? –An MSTAR Perspective. Proceedings of SPIE –The International Society for Optical Engineering,1999,3721: 662～672
[6] Pham Q.H., Ezekiel A. Campbell M.T.. A New end-to-end SAR ATR System [J]. Proceedings of SPIE- The International Society for Optical Engineering,1999,3721:292～301
[7] 李补莲.自动目标识别技术发展评述[J].现代防御技术,2000,28(2):10～14
[8] Blackwell D. Feature analysis for SAR ATR [J] . Proceedings of SPIE –The International Society for Optical Engineering, 2000, 4173:75～86
[9] Schistad Solberg H., Jain A.K. Texture Fusion and Feature Selection Applied to SAR Images [J]. IEEE Transaction on Geosciences and Remote Sensing,1997,35(2):475～479
[10] Chen Chia-Tang, Chen Kun-Shan, Lee Jong-Sen. The use of fully polar metric information for the fuzzy neural classification of SAR images [J]. IEEE Transaction on Geosciences and Remote Sensing,2003,41(9):2089～2100
[11] Vapnik V. Estimation of Dependence Based on Empirical Data. New York: Springer-Verlag, 1982
[12] 许建华,张学工,李衍达.支持向量机的新发展[J].控制与决策,2004,19(5):481～484
[13] Edgar Osuna. Improved Training Algorithm for Support Vector Machines [J]. IEEE Transactions on Neural Networks,1997,10(5):1000～1017
[14] John C.Platt. Sequential Minimal Optimization: A Fast Algorithm for Training Support Vector Machines [J]. IEEE Transactions on Neural Networks,1998
[15] Zhang X G. Using Class-center Vector to Build Support Vector Machines[C]. In: Proceedings of NNSP’99,1999
[16] S.S.Keerthi. Improvements to Platt’s SMO Algorithm for SVM Classifier Design.Technical Report CD-99-14
[17] Morain S.A. Use of Radar for Vegetation Analysis,in Geosciences Application of Image Radar System. Remote Sensing of the Electromagnetic Spectrum,1976,3(3):61～78
[18] 徐产兴.雷达目标识别技术及其新发展[J].雷达与对抗,1994,3:1～9
[19] 张朋等.合成孔径雷达成像三维地形目标模拟方法[J].系统仿真学报,2005,17(10)
[20] 杜亚娟,潘泉.雷达目标识别方法概述[J].火控雷达技术,1998,27:1～5
[21] 傅雄军等.一种有效的聚束式合成孔径雷达图像特征提取的算法[J].北京理工大学学报,2004,24(7):14～17
[22] 韩萍.SAR 自动目标识别及相关技术研究[D].天津:天津大学,2004
[23] 鲁加国,张长耀.高分辨率机载 SAR 雷达研究与试验[C].中国合成孔径雷达会议论文集,2003,86～88
[24] Ward K.D. Compound Representation of High Resolution Sea Clutter [J].Electron lett., 1981,17:561～565
[25] Shi Z.H. A Comparison of Digital Speckle Filters [J]. Processing of IGARSS’94, 1994, 2129～2133
[26] Lee J.S. Speckle Analysis and Smoothing of Synthetic Aperture Radar Image [J]. Computer Graphic and Image Processing,1981,17:24～32
[27] Kuan D.T. Adaptive Restoration of Images with Speckle[J]. IEEE Trans. On Acoustics, Speech and Signal Processing,1987,35(3):373～383
[28] 高贵,李德仁.高分辨率 SAR 图像分割及目标特征提取[J].宇航学报,2006,27(3)
[29] 刘国庆,熊红,黄顺吉.基于小波变换和马尔可夫随机场的极化 SAR 图像自动分类[J],电子科学学刊,2000, 22(3) :359～365
[30] 杨晖,曲秀杰.图像分割方法综述[J].电脑开发与应用,2005,18(3):21～23
[31] R.C.Gonzalez and R.E.Woods. 数字图像处理(第二版).阮秋琦,阮宇智等译.北京:电子工业出版社,2003
[32] Novak L.M., Halversen Shawn D., and Hiett Margarita. Effects of polarization and resolution on the performance of a SAR automatic target recognition system[J].The Lincoln Laboratory Journal,1995,8(1):49～67
[33] Tupin Florence. Detection of Linear Feature in SAR Images: Application to Road Network Extraction[J]. IEEE Trans. On Geosciences Remote Sensing,1998,36(2): 434～453
[34] Tan Hwee H.P., Ramanathan U. Intelligent selection of useful features for optimal feature-based classification [J]. Proceedings of IEEE International Geosciences and Remote Sensing Symposium (IGARSS 2000),2000,7:3012～3014
[35] 王晓红.矩技术及其在图像处理与识别中的应用研究[D].西安:西北工业大学,2001
[36] Hu M K. Visual pattern recognition by moment invariants [J].IEEE Trans. On Information Theory,1962:170～179
[37] 薛笑荣.SAR 图像处理技术研究[D].西安:西北工业大学,2004
[38] Vapnik V. The Nature of Statistical Learning Theory. New York: Springer-Verlag,2000. (中译本,张学工译.统计学习理论的本质,北京:清华大学出版社,2000)
[39] 张学工.关于统计学习理论与支持向量机[J].自动化学报,2000,26(1) :32～42
[40] C Cortes,Vapnik V. Support Vector Networks[J]. Machine Learning,1995,20(3):273～297
[41] 李国正,王猛,曾华军译.支持向量机导论.北京:电子工业出版社,2000
[42] J.Weston and C.Watkins. Multi-class support vector machines. Technical Report,1998,CSD-TR-98-04: 1～9
[43] Knerr S,et al. Single layer learning revisited: A step wise procedure for building and training a neural network [A]. NATOASI. Springer, 1990
[44] C Huang,L S Davis. An assessment of support vector machines for land cover classification [J]. International Geosciences and Remote Sensing,2002,23: 725～749
[45] 姚恩瑜,何勇,陈仕平.数学规划与最优化.杭州:浙江大学出版社,2001
[46] 安金龙,王正欧等.一种新的支持向量机多类分类方法[J].信息与控制,2004,33(3) :262～267
[47] Chih-Chung Chang and Chih-Jen Lin. LIBSVM: a Library for Support Vector Machines.Department of Computer Science and Information Engineering, National Taiwan University, Taipei 106, Taiwan. April,17,2005
[48] 何壸，刘宏伟等. 基于 MSTAR 数据的 SAR 图像自动目标识别研究[C],中国合成孔径雷达会议论文集,2005(10) :278～282
[49] Qun Zhao, Jose C. Support Vector Machines for SAR Automatic Target Recognition [J]. IEEE Transactions on Aerospace and Electronic Systems,2001,37 (2):643～654