搜索
北京工业大学毕业设计（论文）I摘要近年来，作为一种重要的模式识别应用，遥感图像目标识别技术在各种领域中，如军事、地质、灾害预报、社会治安管理等，均得到了广泛的应用。其逐渐成为当今社会中不可缺少的应用技术。由于遥感图像背景和目标特征的复杂性，如何更好、更快实现对特定目标的识别成为研究的难点和热点。本文的工作是对遥感图像目标识别技术进行研究，使用数学形态学和离散型Hopfield神经网络对遥感图像目标进行识别，识别的主要目标为水上桥梁。分析该方法的可行性、有效性，及研究、应用前景等，并提出改进意见，以便后人参考。本文首先利用数学形态学去除噪声和干扰，并通过水域分割找出遥感图像中可能存在水上桥梁的感兴趣区域，之后对感兴趣区域中的子图像进行特征提取，随后利用离散型Hopfield神经网络对感兴趣区域进行分类，最终实现水上桥梁目标的识别。结果表明，该方法可以较好地实现遥感图像中水上桥梁的识别。有一定的应用价值。关键词：遥感；图像分割；目标识别；特征提取；Hopfield神经网络；北京工业大学毕业设计（论文）IIAbstractInrecentyears,asanimportantapplicationofPatternRecognition,targetrecognitioninremotesensinghasbeenwidelyutilizedinvarioussphere,suchasmilitary,geology,disasterforecast,socialsecurityadministrationandsoforth.Ithasgraduallybecomeanindispensableapplicationtechnologyinsociety.Duetothecomplexityofboththebackgroundandfeatureoftargetinremotesensingimage,itisadifficultandinterestingdomainofrelatedresearchhowtorecognizeaspecifictargetinabetterandfasterway.Inthisdissertation,thetechnologyoftargetrecognitioninremotesensingimagewillbestudied,andalsosometarget,especiallythebridgeoverwater,willberecognisedbyusingMathematicalMorphologyandDiscreteHopfieldNeuralNetwork.Andthen,thefeasibilityandtheeffectivenessofthismethod,andalso,theprospectofresearchandapplicationwillbeanalysed.Inaddition,someadviceswillbeproposedforreference.First,eliminatethenoiseusingMathematicalMorphology,andthroughsegmentationofwaterarea,findtheRegionofInterest(ROI)wheremayexistbridgeoverwaterintheremotesensingimage.Second,extractthefeatureofROIs.Afterthat,classifytheROIsbyusingDiscreteHopfieldNeuralNetworksoastocompletetherecognitionofthetarget.Theresultshowsthatthismethodperformswellintherecognitionofthebridgeoverwaterinremotesensingimage.Ithas,tosomeextent,valueforapplication.Keywords:Remotesensing;imagesegmentation;targetrecognition;featureextraction;Hopfieldneuralnetwork;北京工业大学毕业设计（论文）III目录摘要......................................................................IAbstract.................................................................II1.绪论.................................................................11.1研究背景及意义......................................................11.2目标识别系统.......................................................11.2.1目标识别的关键技术..............................................11.2.2目标识别的一般过程..............................................21.3本文的主要工作及结构安排...........................................31.4本文的主要创新点...................................................32.基于数学形态学的图像预处理及水域分割算法的研究及实现....................42.1水上桥梁目标特征及先验知识梳理......................................42.2水上桥梁识别算法及流程展示..........................................52.3图像预处理及水域分割的实现..........................................62.3.1灰度图像转换...................................................62.3.2水域分割.......................................................62.4基于数学形态学的水域分割后续处理研究................................92.4.1算法流程展示...................................................92.4.2数学形态学介绍.................................................92.4.3开运算处理....................................................112.4.4闭运算处理....................................................112.4.5腐蚀运算处理..................................................122.4.6膨胀运算处理..................................................122.5感兴趣区域定位与提取实现...........................................132.5.1感兴趣区域定位流程展示........................................132.5.2连通性分析....................................................142.5.3水域选择......................................................152.5.4水域边缘统计及感兴趣区域的定位................................163.特征提取技术的研究及其算法的实现......................................183.1特征提取流程展示...................................................183.2颜色特征提取.......................................................193.3纹理特征提取.......................................................193.4形状特征提取.......................................................20北京工业大学毕业设计（论文）IV3.5特征向量的生成.....................................................204.离散型Hopfield神经网络的研究及目标识别的实现..........................224.1Hopfield神经网络基本原理..........................................224.2Hopfield神经网络算法流程展示......................................224.3特征向量的二进制编码的实现.........................................234.3.1颜色与纹理特征编码............................................254.3.2形状特征编码..................................................254.4模板的设置与存储...................................................274.5分类识别功能的实现及改进...........................................274.5.1分类识别效果..................................................274.5.2误差分析......................................................284.5.3算法改进......................................................294.6Hopfield神经网络的通用性分析......................................305.目标识别算法的编程实现................................................31结论.....................................................................34致谢.....................................................................35参考文献..................................................................36北京工业大学毕业设计（论文）11.绪论1．1研究背景及意义随着人类社会的发展，遥感技术已经成为高科技领域中的一个重要分支，其