摘要
随着海洋资源的开发和使用,海洋受到了严重的污染,其中石油污染表现得尤为突出。溢油事故往往造成大面积海域污染,危害十分巨大。因此实时地、正确地监测溢油具有重大意义。
本文认真对比了国内外已有的溢油识别方法,分析了以往的人工神经网络激光荧光光谱识别速度较慢、精度不是很高的原因。提出以光谱形状作为溢油识别的关键特征,并引入基于统计学习理论的支持向量机(SVM)方法来识别溢油荧光光谱的种类。
在人工神经网络众多的模型中,选择了应用比较广泛的误差反向传播(BP)网络和径向基函数(RBF)网络。本文的核心工作是建立了BP、RBF网络和SVM激光荧光光谱分析识别模型。并利用已有的激光遥感设备获取光谱对建立的几种模型进行测试实验,得到了实验结果。测试实验主要考察的是这3种模型的识别精度和速度。
测试实验得出,BP网络激光荧光光谱分析识别模型的平均训练时间16.0280s,正确率为86.7%;RBF网络激光荧光光谱分析识别模型的平均训练时间0.6064s,正确率为86.7%;SVM激光荧光光谱分析识别模型的平均训练时间0.0184s,正确率为96.7%。RBF网络和SVM两种识别模型的训练耗费时间较少。BP网络识别模型的训练所要耗费的时间显然大些。在识别的精度方面,BP网络和RBF网络模型的没有差别,而SVM模型的更好。
结果表明,SVM激光荧光光谱分析识别模型的绩效是最好的。它是一种很有前途的方法。
With the development and use of marine resources, sea is suffered serious pollution, and oil pollution is especially obvious. Oil spills often cause extensive sea pollution and tremendous harm. Therefore, real-time and correct monitoring oil spill is great significance.
The paper carefully compares the domestic and foreign oil spills identification method, and analyses the reason of previous neural network laser fluorescence spectrum identification with slower speed and lower accuracy rate. The paper advances that the shape of the fluorescence spectral signatures is the crucial character of identification oil spills, and introduces the Support Vector Machine (SVM) which is based on the statistical learning theory, to identify the types of fluorescence spectrum.
The paper chooses the abroad application neural network model for back-propagation (BP) network and Radial Basis Function (RBF) network in many of the artifiacal neural network models. The core work of the paper is the establishment of BP, RBF networks and SVM laser fluorescence spectrum identification model. The use of laser remote sensing equipment obtained spectrums to test the several establishment model and obtain the results of the experiments. The main research of experimental testing is the identification speed and identification accuracy of three models .
Testing experiments, the average training time of BP network laser fluorescence spectrum identification model is 16.0280s, with an accuracy rate of 86.7%. The average training time of RBF network laser fluorescence spectrum to identification model is 0.6064s, with an accuracy rate of 86.7%.The average training time of SVM laser fluorescence spectrum identification model is 0.0184s, with an accuracy rate of 96.7%. RBF network and SVM identification model spend less time training. BP network identification model of training to be time-consuming obviously much bigger. The accuracy of BP and RBF network identification model is the same, and the accuracy of SVM identification model is the best.
The results show that performance of SVM laser fluorescence spectrum identification model is the best. It is a very promising approach.
引文
[1]闫峰.海洋石油污染的人文思考[J].中学地理教学参考,2003,(10):21-22.
[2]http://www.oceancentury.cn/youqi/ShowArticle.asp?ArticleID-713.
[3]王飞.开发与保护并重保障海洋经济的可持续发展[J].海洋开发与管理,2005,22(4):3-6.
[4]http://www.sdinfo.net.cn/acca21/pc6-8.html.
[5]安居白.航空遥感探测海上溢油的技术[J].交通环保,2002,23(1):24-26.
[6]张永宁,丁倩,李栖筠.海上溢油污染遥感监测的研究[J].大连海事大学学报,1999,25(3):1-5.
[7]Jose Bava,Tristan Osan,Lavier Yasnikoushi.Earth Observation Application through Systems and Instruments of High Performance Computer[R].ASI/CONAE Training Course,Italy,2002:20-22.
[8]Jianfeng Li,Steven Fuller,Julie Cattle et al.Matching fluorescence spectra of oil spills with spectra from suspect sources[J].Analytica Chimica Acta,2004,514(1):51-56.
[9]马永安,赵云英.荧光光谱向量空间距离模式识别法鉴别海面溢油源[J].海洋学报,1998,20(4):1-7.
[10]何晓媛,石金辉,辛海虹.南海区平台石油的特征荧光光谱分析[J].海洋环境科学,2004,23(4):60-63.
[11]阎吉祥,龚顺生,刘智深.环境监测激光雷达[M].北京:科学出版社,2001.
[12]Carl E.Brown,Mervin F.Review of the development of laser fluorosensors for oil spill application Fingas[J].Marine Pollution Bulletin,2003,47(10-12):477-484.
[13]李四海.海上溢油遥感探测技术及其应用进展[J].遥感信息,2004,(2):54-58.
[14]王斌永.海洋和水环境航空光学遥感系统[J].红外,2002,(10):16-19.
[15]于欣洋.人工神经网络在识别海面溢油中的应用研究[D].大连:大连海事大学,2005.
[16]高隽.人工神经网络原理及仿真实例[M].北京:机械工业出版社,2003.
[17]飞思科技产品研发中心.神经网络理论与MATLAB7实现[M].北京:电子工业出版社,2006.
[18]程相君,王春宁,陈生潭.神经网络原理及其应用[M].北京:国防工业出版社,1995.
[19]侯北平,卢佩.基丁MATLAB的BP神经网络建模及系统仿真[J].控制系统,2001,(1):34-36.
[20]周俊武,孙传尧,王福利.径向基函数(RBF)网络的研究及实现[J].矿冶,2001,10(4):71-75.
[21]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.
[22]唐发明.基于统计学习理论的支持向量机算法研究[M].武汉:华中科技大学,2005.
[23]Vapnik V N.The Nature of Statistical Learning Theory[M].New-York:Springer-Verlag,1995.
[24]Bertsekas D P.Nonlinear Programming[M],Belmont:Athena Scientific,1995.
[25]祁亨年.支持向量机及应用综述[J].计算机工程,2004,30(10):11-14.
[26]钟珞,潘吴,封筠,等.模式识别[M].武汉:武汉大学出版社,2006.
[27]Vapnik V N.Statistical Learning Theory[M].New-York:John Wiley,1998.
[28]KrBel U.Pairwise classification and support vector machines[M].Cambridge:The MIT Press,1999.
[29]Khaled Mohamed Almhdi.Oil Classification with Fluorescence Spectroscopy[D].Oldenburg and Emden,2005.
[30]李庆扬,王能超,易大易.数值分析[M].武汉:华中科技大学出版社,2003.
[31]苏金明,阮沈勇.MATLAB 6.1实用指南[M].北京:电子工业出版社,2002.
[32]http://blog.csdn.net/qschiang/archive/2006/09/19/1243538.aspx.
[33]C.W.Hsu,C.J.Lin.A comparison of methods for multi-class support vector machines[J].IEEE Transactions on Neural Networks,2002,(13):415-425.
[34]朱立松.应用纹理分析和神经网络方法分类海上SAR溢油图像[D].大连:大连海事大学,2005.
[35]http://sandytxp.blog.163.com/blog/static/91420782007112775615482/