支持向量机在水下目标识别中的应用研究
|
摘要
论文的研究工作来源于某项目课题“XXXX基础研究”项目,主要是利用已有的水下目标特征提取和识别方法,并且采用基于支持向量机方法进行壳体目标和石头的识别分类,以提高对水下目标的识别率。
本文的研究方法是利用水下目标的回波特性来作出识别。具体地说,就是由主动声纳向待识别的水下目标发射信号,并且从回波当中提取出水下目标的有效特征信息,再将特征信息送入分类器进行识别分类。
本论文主要进行了以下工作:
1.综述了水下目标识别的背景和意义,以及国内外对水下目标的探测和识别的方法和研究现状。
2.特征提取的过程是把采集到的壳体目标和石头的回波信号变换到不同的特征空间,提取出反映样本本质特征的特征向量,并将特征向量送到分类器中。为了研究不同特征提取方法的性能,本文采用了四种特征提取方法,即小波包能量特征提取、常数Q滤波子带能量特征提取、双谱能量特征提取和主成分分析特征提取。然后对壳体目标和石头所提取的有效特征,利用K-L变换对其进行可分性判决,以分别判断四种特征提取方法的有效性。
3.叙述了统计学习理论和支持向量机理论。基于统计学习理论的支持向量机算法具有坚实的数学理论基础和严格的理论分析,有效地提高了算法泛化能力。而核函数的成功运用使大多数不可分的低维空间映射到高维空间后变为可分,并且有效的消除了维数灾难这个缺点。
4.叙述了支持向量机的SMO等三种算法并设计了三种分类器。
5.运用四种方法分别对吊放目标测量数据、掩埋定点测量数据和掩埋扫描测量数据进行特征提取,再将特征提取结果送入三种分类器,以比较分类器的识别分类性能。
目前,支持向量机方法的应用还在不断的完善,以便在更广泛的领域中得到应用。本文中支持向量机方法的应用及其性能的优越性,对水下目标识别的研究具有重要的意义。
The work of the paper derived from some subject "XXXX basic research" , which mainly use the underwater target feature extraction and recognition and support vector machine-based method to identify chitinous object and stone targets, in order to increase the recognition rate of underwater targets.
The methods in this paper that is to make identification by echo characteristics of the underwater target. Specifically, the active sonar give a signal to identify the underwater targets, and the valid characteristics information of targets were extracted from echoes, and then the characteristics information were sent to the classifier to identify and categories.
This paper mainly discussed as follows:
1. Overview the context and meaning of the underwater target recognition, as well as domestic and abroad detection methods ,and research. status on the underwater targets.
2. Feature extraction process is the collection of the target echo of the signal characteristics of the chitinous object and stone , transforming to a different space, extracted a sample to reflect the essential character of the feature vector, and send feature vector in to the classifier. In order to study the different methods of extracting characteristics, using four feature extraction methods, that is, the wavelet packet energy feature extraction, filtering constant Q sub-band energy feature extraction, feature extraction power Bispectrum principal, component analysis and feature extraction. Then the goal of the chitinous object and stone extraction by the characteristics of effective use of K-L transform, respectively, in order to determine the four feature extraction methods are effective.
3. This article describe the statistical theory and support vector machines in detail. The support vector machine based on statistical theory have theoretical foundation stability of mathematics and rigorous theoretical analysis, the theory has a complete, global optimization, adaptability, and capacity of promote, the machine learning is a new approach and new hot spots. It uses the principle by which the structure risk will be minimization, associate with statistical study, machine learning and neural networks, to minimize the risk of experience, which has effectively improved the ability of generalization algorithm. The kernel arithmetic successful use make the low points which can not be mapped to the high-dimensional space-dimensional space can be divided into, and effectively eliminates the shortcomings of the dimensions.
4. Three SVM algorithms are set forth, such as the SMO algorithm, and three categories classifier are designed
5. Using four methods to extract feature of the measure of pensile goal's data, burial site measurement data and buried scan measurement data separately, and then send the results to three categories classifier to contrast identifying and classifying performance of the classifier.
Now, the application of SVM algorithms is still consummating unceasingly, and SVM algorithms will be applied in broader area. It is of vital significance of underwater target identification that the application of SVM and advantage of its performance in this article.
本文的研究方法是利用水下目标的回波特性来作出识别。具体地说,就是由主动声纳向待识别的水下目标发射信号,并且从回波当中提取出水下目标的有效特征信息,再将特征信息送入分类器进行识别分类。
本论文主要进行了以下工作:
1.综述了水下目标识别的背景和意义,以及国内外对水下目标的探测和识别的方法和研究现状。
2.特征提取的过程是把采集到的壳体目标和石头的回波信号变换到不同的特征空间,提取出反映样本本质特征的特征向量,并将特征向量送到分类器中。为了研究不同特征提取方法的性能,本文采用了四种特征提取方法,即小波包能量特征提取、常数Q滤波子带能量特征提取、双谱能量特征提取和主成分分析特征提取。然后对壳体目标和石头所提取的有效特征,利用K-L变换对其进行可分性判决,以分别判断四种特征提取方法的有效性。
3.叙述了统计学习理论和支持向量机理论。基于统计学习理论的支持向量机算法具有坚实的数学理论基础和严格的理论分析,有效地提高了算法泛化能力。而核函数的成功运用使大多数不可分的低维空间映射到高维空间后变为可分,并且有效的消除了维数灾难这个缺点。
4.叙述了支持向量机的SMO等三种算法并设计了三种分类器。
5.运用四种方法分别对吊放目标测量数据、掩埋定点测量数据和掩埋扫描测量数据进行特征提取,再将特征提取结果送入三种分类器,以比较分类器的识别分类性能。
目前,支持向量机方法的应用还在不断的完善,以便在更广泛的领域中得到应用。本文中支持向量机方法的应用及其性能的优越性,对水下目标识别的研究具有重要的意义。
The work of the paper derived from some subject "XXXX basic research" , which mainly use the underwater target feature extraction and recognition and support vector machine-based method to identify chitinous object and stone targets, in order to increase the recognition rate of underwater targets.
The methods in this paper that is to make identification by echo characteristics of the underwater target. Specifically, the active sonar give a signal to identify the underwater targets, and the valid characteristics information of targets were extracted from echoes, and then the characteristics information were sent to the classifier to identify and categories.
This paper mainly discussed as follows:
1. Overview the context and meaning of the underwater target recognition, as well as domestic and abroad detection methods ,and research. status on the underwater targets.
2. Feature extraction process is the collection of the target echo of the signal characteristics of the chitinous object and stone , transforming to a different space, extracted a sample to reflect the essential character of the feature vector, and send feature vector in to the classifier. In order to study the different methods of extracting characteristics, using four feature extraction methods, that is, the wavelet packet energy feature extraction, filtering constant Q sub-band energy feature extraction, feature extraction power Bispectrum principal, component analysis and feature extraction. Then the goal of the chitinous object and stone extraction by the characteristics of effective use of K-L transform, respectively, in order to determine the four feature extraction methods are effective.
3. This article describe the statistical theory and support vector machines in detail. The support vector machine based on statistical theory have theoretical foundation stability of mathematics and rigorous theoretical analysis, the theory has a complete, global optimization, adaptability, and capacity of promote, the machine learning is a new approach and new hot spots. It uses the principle by which the structure risk will be minimization, associate with statistical study, machine learning and neural networks, to minimize the risk of experience, which has effectively improved the ability of generalization algorithm. The kernel arithmetic successful use make the low points which can not be mapped to the high-dimensional space-dimensional space can be divided into, and effectively eliminates the shortcomings of the dimensions.
4. Three SVM algorithms are set forth, such as the SMO algorithm, and three categories classifier are designed
5. Using four methods to extract feature of the measure of pensile goal's data, burial site measurement data and buried scan measurement data separately, and then send the results to three categories classifier to contrast identifying and classifying performance of the classifier.
Now, the application of SVM algorithms is still consummating unceasingly, and SVM algorithms will be applied in broader area. It is of vital significance of underwater target identification that the application of SVM and advantage of its performance in this article.
引文
[1]田坦,刘国枝.声纳技术.哈尔滨工程大学出版社,2000:2-5页,237-243页
[2]侯自强,李贵斌.声纳信号处理一原理与设备.海洋出版社,1986:10-58页,441-464页
[3]Tang Xiaoou.Multiple competitive learning network fusion for object classification.IEEE Trans on Sustems,Man and Cybernetics part B:Cybernetics.1998,28(4):532-543P
[4]G.L.FORESTI,S.GENTILI.A vision based system for object detection in underwater images.International journal of Pattern Reeognition and Artificial Intelligence.2001,14(2):167-188P
[5]K.S.Thyagarajan,T.Nguyen,C.E.Persons.Maximum likelihood approach to image texture and acoustic signal classification.IEE Proceedings:Vision,Image and Signal Processing.1999,146(1):34 - 39P
[6]Sutton Jeffrey et al.Enhancing mine signatures in sonar images using nested neural networks.Proceedings of SPIE-The Inetrnational Society for Optical Engineering.1999,37(1):570 - 577P
[7]B.R.Calder,L.M.Linnett,D.R.Carmichael.Bayesian approach to object detection in sidescan sonar.IEE Proceedings:Vision,Image and Signal Processing.1998,145(3):221-228P
[8]M.Vittorio,T.Andrea.Confidence-based approach to enhancing underwater acoustic image formation.IEEE Trans.on Image Processing.1999,8(2):270-285P
[9]M.A.Hodgetts,A.R.Greig,A.Fairweather.Underwater imaging using Markov Random Fields with feed forward prediction,Underwater Technology.1999,23(4):157 - 167P
[10]M.Mignotte et al.Bayesian inference and optimization strategies for some detection and classification problems in sonar imagery.Proceedings of SPIE-The International Society for Optical Engineering.1999,36(2):14-27P
[11]L.M.Linnett et al.Texture classification using a spatial-point Process model.IEE Proceedings:Vision,Image and Signal Processing.1995,142(1):1-6P
[12]V.Murino.Acoustic image reconstruction by Markov random fields.Electronics Letters.1996,32(7):697-698P
[13]M.Mignotte et al.Markov Random Field and Fuzzy Logic Modeling in Sonar Imagery:Application to the Classification of Underwater Floor.Computer Vision and Image Understanding.2000,79(1):4-24P
[14]郭海涛.高分辨率成像声呐后置图像处理.哈尔滨工程大学博士学位论文.2002:13,74-82页
[15]S.Danlel.et al.Side-scan sonar image matching.IEEE Journal of Oceanic Engineering.1998,23(3):245-259P
[16]R.Slater,C.Robinson,S.Lingsch.Object detection and discrimination in side-scan sonar by means of intensity contouring.Proceedings of SPIE-The International Society for Optical Engineering.1999,10(1):496-503P
[17]刘晨晨.数学形态学图像处理算法应用研究.哈尔滨工程大学硕士学位论文.2002:2-3,58-61页
[18]Liu chen-chen,Sang en-fang.Application of morphological filter in underwater acoustic-image processing system.Papers of the 3~(rd)International Workshop on Acoustical Engineering and Technology.2002:207-210P
[19]严学强.数学形态学方法在水声图象预处理中的应用研究.哈尔滨船舶工程学院硕士学位论文.1995:4页
[20]Vapnik V.The nature of statistical learning theory[M].New York:Springer-Verlag,1995.
[21]Cortes C,Vapnik V.Support vector networks[J].Machine learning,1995,20(1):273-297P
[22]刘向东,陈兆乾.基于支持向量机方法的人脸识别研究[J].小型微型计算机系统,2004,25(12):2261-2263页
[23]张爱丽,刘广利,刘长宇.基于SVM的多类文本分类研究[J].情报杂志, 2004,23(9):6-10页
[24]魏新,冯兴杰,刘山.基于支持向量机的多元文本分类研究[J].海军工程大学学报.2004,16(5):30-32页
[25]Smola A,Scholkopf B.On a kernel based method for pattern recognition,regression,approximation and operator inversion[J].Algo- rithmica,1998,22(1):211-231P
[26]朱埜.主动声纳检测信息原理[M].北京:海洋出版社,1990.
[27]余秋星等.舰船噪声的状态空间重构与特征压缩,系统仿真学报,Vol.18,No.8,2003:1079-1080.
[28]范海宁等.基于小波包分解的声信号特征提取方法.《现代电子技术》2004.195(4)
[29]李秀坤.水雷目标特征提取与识别研究[D].哈尔滨工程大学博士论文.2000:58-66页
[30]Vapnik著.许建华,张学工 译.统计学习理论[M].电子工业出版社,2004.06
[31]Vapnik V N,Levin E,Le Cun Y.Measuring the VC-dirrension of a learning machine.Netrel Computation,1994,6:851-876P
[32]邓乃扬,田英杰.数据挖掘中的新方法-支持向量机[M].北京:科学出版社,2004.
[33]K.R.Muller,S.Mika,G.Ratsch,et al.An Introduction to kernel-Based Learning Algorithms[J].IEEE Tansactions on Neural Networks,2000.12(3):181-201
[34]陈丹,李京华,许俊峰,等.基于主分量分析的声信号特征提取及识别研究[J].声学技术,2005,54(1):39-41.
[35]Vapnik V N.统计学习理论的本质[M].张学工译.清华大学出版社,2000.
[36]边肇琪,张学工,模式识别[M],北京:清华大学出版社,2000,第二版.
[37]JANAK,ROBERT PWDUN,MAO J.Statistical Pattem Recognition:A Review[J].IEEE Transactions on Pattem Analysis and Machine Intelligence, 2000,22(1):4-37.
[38]张贤达著.现代信号处理,北京:清华大学出版社,2000,第二版.274-281.
[39]张贤达,保铮.非平稳信号的分析与处理.北京:国防工业出版社,1998.224-238.
[40]程正兴.小波分析极其在信号处理中的应用.北京:电子工业出版社,2006.24-30,162-170.
[41]张小云,刘允才,高斯核支持向量机的性能分析,计算机工程,29(8),2003.
[42]吴涛,核函数的性质方法及其在障碍检测中的应用,博士学位论文,国防科学技术大学,2003.
[2]侯自强,李贵斌.声纳信号处理一原理与设备.海洋出版社,1986:10-58页,441-464页
[3]Tang Xiaoou.Multiple competitive learning network fusion for object classification.IEEE Trans on Sustems,Man and Cybernetics part B:Cybernetics.1998,28(4):532-543P
[4]G.L.FORESTI,S.GENTILI.A vision based system for object detection in underwater images.International journal of Pattern Reeognition and Artificial Intelligence.2001,14(2):167-188P
[5]K.S.Thyagarajan,T.Nguyen,C.E.Persons.Maximum likelihood approach to image texture and acoustic signal classification.IEE Proceedings:Vision,Image and Signal Processing.1999,146(1):34 - 39P
[6]Sutton Jeffrey et al.Enhancing mine signatures in sonar images using nested neural networks.Proceedings of SPIE-The Inetrnational Society for Optical Engineering.1999,37(1):570 - 577P
[7]B.R.Calder,L.M.Linnett,D.R.Carmichael.Bayesian approach to object detection in sidescan sonar.IEE Proceedings:Vision,Image and Signal Processing.1998,145(3):221-228P
[8]M.Vittorio,T.Andrea.Confidence-based approach to enhancing underwater acoustic image formation.IEEE Trans.on Image Processing.1999,8(2):270-285P
[9]M.A.Hodgetts,A.R.Greig,A.Fairweather.Underwater imaging using Markov Random Fields with feed forward prediction,Underwater Technology.1999,23(4):157 - 167P
[10]M.Mignotte et al.Bayesian inference and optimization strategies for some detection and classification problems in sonar imagery.Proceedings of SPIE-The International Society for Optical Engineering.1999,36(2):14-27P
[11]L.M.Linnett et al.Texture classification using a spatial-point Process model.IEE Proceedings:Vision,Image and Signal Processing.1995,142(1):1-6P
[12]V.Murino.Acoustic image reconstruction by Markov random fields.Electronics Letters.1996,32(7):697-698P
[13]M.Mignotte et al.Markov Random Field and Fuzzy Logic Modeling in Sonar Imagery:Application to the Classification of Underwater Floor.Computer Vision and Image Understanding.2000,79(1):4-24P
[14]郭海涛.高分辨率成像声呐后置图像处理.哈尔滨工程大学博士学位论文.2002:13,74-82页
[15]S.Danlel.et al.Side-scan sonar image matching.IEEE Journal of Oceanic Engineering.1998,23(3):245-259P
[16]R.Slater,C.Robinson,S.Lingsch.Object detection and discrimination in side-scan sonar by means of intensity contouring.Proceedings of SPIE-The International Society for Optical Engineering.1999,10(1):496-503P
[17]刘晨晨.数学形态学图像处理算法应用研究.哈尔滨工程大学硕士学位论文.2002:2-3,58-61页
[18]Liu chen-chen,Sang en-fang.Application of morphological filter in underwater acoustic-image processing system.Papers of the 3~(rd)International Workshop on Acoustical Engineering and Technology.2002:207-210P
[19]严学强.数学形态学方法在水声图象预处理中的应用研究.哈尔滨船舶工程学院硕士学位论文.1995:4页
[20]Vapnik V.The nature of statistical learning theory[M].New York:Springer-Verlag,1995.
[21]Cortes C,Vapnik V.Support vector networks[J].Machine learning,1995,20(1):273-297P
[22]刘向东,陈兆乾.基于支持向量机方法的人脸识别研究[J].小型微型计算机系统,2004,25(12):2261-2263页
[23]张爱丽,刘广利,刘长宇.基于SVM的多类文本分类研究[J].情报杂志, 2004,23(9):6-10页
[24]魏新,冯兴杰,刘山.基于支持向量机的多元文本分类研究[J].海军工程大学学报.2004,16(5):30-32页
[25]Smola A,Scholkopf B.On a kernel based method for pattern recognition,regression,approximation and operator inversion[J].Algo- rithmica,1998,22(1):211-231P
[26]朱埜.主动声纳检测信息原理[M].北京:海洋出版社,1990.
[27]余秋星等.舰船噪声的状态空间重构与特征压缩,系统仿真学报,Vol.18,No.8,2003:1079-1080.
[28]范海宁等.基于小波包分解的声信号特征提取方法.《现代电子技术》2004.195(4)
[29]李秀坤.水雷目标特征提取与识别研究[D].哈尔滨工程大学博士论文.2000:58-66页
[30]Vapnik著.许建华,张学工 译.统计学习理论[M].电子工业出版社,2004.06
[31]Vapnik V N,Levin E,Le Cun Y.Measuring the VC-dirrension of a learning machine.Netrel Computation,1994,6:851-876P
[32]邓乃扬,田英杰.数据挖掘中的新方法-支持向量机[M].北京:科学出版社,2004.
[33]K.R.Muller,S.Mika,G.Ratsch,et al.An Introduction to kernel-Based Learning Algorithms[J].IEEE Tansactions on Neural Networks,2000.12(3):181-201
[34]陈丹,李京华,许俊峰,等.基于主分量分析的声信号特征提取及识别研究[J].声学技术,2005,54(1):39-41.
[35]Vapnik V N.统计学习理论的本质[M].张学工译.清华大学出版社,2000.
[36]边肇琪,张学工,模式识别[M],北京:清华大学出版社,2000,第二版.
[37]JANAK,ROBERT PWDUN,MAO J.Statistical Pattem Recognition:A Review[J].IEEE Transactions on Pattem Analysis and Machine Intelligence, 2000,22(1):4-37.
[38]张贤达著.现代信号处理,北京:清华大学出版社,2000,第二版.274-281.
[39]张贤达,保铮.非平稳信号的分析与处理.北京:国防工业出版社,1998.224-238.
[40]程正兴.小波分析极其在信号处理中的应用.北京:电子工业出版社,2006.24-30,162-170.
[41]张小云,刘允才,高斯核支持向量机的性能分析,计算机工程,29(8),2003.
[42]吴涛,核函数的性质方法及其在障碍检测中的应用,博士学位论文,国防科学技术大学,2003.