资源描述:
《自适应脉冲耦合神经网络在图像处理中应用》由会员上传分享,免费在线阅读,更多相关内容在工程资料-天天文库。
1、Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/274890135StudyonSelf-adaptivePulseCoupledNeuralNetworkandItsApplicationinFieldsofImageProcessingArticle·June2008CITATIONSREADS5413authors,including:YideMaLanzhouUniversity245PUBLICATIONS1,952CITAT
2、IONSSEEPROFILESomeoftheauthorsofthispublicationarealsoworkingontheserelatedprojects:DEEPNEURO-CORTICALNEURALNETWORKSViewprojectneuralnetworksprinciple&applicationViewprojectAllcontentfollowingthispagewasuploadedbyYideMaon13April2015.Theuserhasrequestedenhancementofthedownloadedfile.第20卷第11期系统仿真学
3、报©Vol.20No.112008年6月JournalofSystemSimulationJun.,2008自适应脉冲耦合神经网络在图像处理中应用111,2马义德,绽琨,齐春亮(1.兰州大学信息科学与工程学院,甘肃兰州730000,2.中国酒泉卫星发射中心,甘肃兰州732750)摘要:尽管Johnson提出的PCNN模型具有强大的图像处理功能,以时间序列进行特征提取时具有旋转、尺度、平移、扭曲不变性,可实践中发现依然存在着不足,特别对图像亮度、对比度比较敏感。添加了误差反向传播(ErrorBackPropagation,EBP)学习准则的自适应脉冲耦合神经网络模型能自适应设定模型参数,
4、是脉冲耦合神经网络模型研究的主要内容。特别地,应用这种自适应模型进行特征提取时,能弥补原来PCNN模型对亮度、对比度敏感的缺陷,而且具有一定的泛化能力,有效克服了亮度、对比度对图像识别精度的影响。关键词:自适应;脉冲耦合神经网络;学习准则;时间序列中图分类号:TP391文献标识码:A文章编号:1004-731X(2008)11-2897-04StudyonSelf-adaptivePulseCoupledNeuralNetworkandItsApplicationinFieldsofImageProcessing111,2MAYi-de,ZHANKun,QIChun-liang(1.S
5、choolofInformationScienceandEngineering,LanzhouUniversity,Lanzhou730000,China;2.ChinaJiuquanSatelliteLaunchCenter,Lanzhou732750,China)Abstract:ThestandardPulseCoupledNeuralNetworks(PCNN)hasbeenwidelyusedintheimageprocessing,however,itishardtosetplentyparametersofPCNNefficientlywhichlimiteditscap
6、abilityforimageprocessing.Basedonthelearningrules,PCNNwasoptimizedthroughrunningitsparametersadaptively.Agradientdescentalgorithmwasadoptedtosearchparameterswhichcouldreducetheerrorbetweenthedesiredoutputandtheactualoutputgraduallyaccordingtotheleastmeansquareprinciple.ThetraditionalPCNNmodelisu
7、sedtoimagefeatureextraction,itsoutputfeaturesarerotation,scaleandshiftinvariant,butitissensitivetoillumination,thereforetheadaptiveparametersPCNNisusedforimagefeatureextractionwhenthestimuli'sillumination(intensityorcontrast