腦電信號特征生成優(yōu)化算法的研究.pdf

1、汕頭大學(xué)碩士學(xué)位論文腦電信號特征生成優(yōu)化算法的研究姓名：趙巖申請學(xué)位級別：碩士專業(yè)：信號與信息處理指導(dǎo)教師：柳平20090501AbstractAbstractEpilepsyresultedfromthesuperexcitabilitythesynchronousfiringofneuronsisaseriouschronicdiseasewhichwillresultinthetransientabnmalityofthecent

2、ralnervesystem.Electroencephalogram(EEG)isawellestablishedclinicaldiagnosiswhichcanprovidesomeimptantinfmationtothediagnosisofepilepsy.Howeveralongtermepilepsydetectionwithvisualinspectionisnotonlyatimeconsumingtaskfthec

3、linicaldoctsbutalsoadetectedmethodbasedonthedocts’clinicalexperience.Theautomaticdetectionoftheepilepticwavescanbehelpfultosolvetheproblemabove.ThemethodsoffeatureextractionfeatureionfromEEGsignalsaremainlydiscussedinthi

4、spaperthemainlyinnovativeachievementsareasfollows:(1)Anonlinearfeatureextractionbasedonwaveletpacketdecompositionapproximateentropy(ApEn)isproposedtosolvetheproblemthattraditionalsignalprocessingmethodsintimedomainfreque

5、ncydomaincannotsufficientlyextractthenonlinearfeatureofEEGsignals.FirstlyEEGsignalsaredecomposedintovariousfrequencybsthroughfourthlevelwaveletpacketdecomposition.Secondlyapproximatingentropy(ApEn)valuesofthewaveletcoeff

6、icientsareusedasfeaturevects.FinallytheSVMisusedasaclassifiertoverifytheeffectivenessofthismethod.Experimentalresultsshowthattheproposedmethodcangetmuchhigherclassificationaccuracy.(2)AfeatureionalgithmwithParticleSwarmO

7、ptimizationbasedonedimensionsearchingisproposedtoavoidtheproblemthatromsearchalgithmbasedfeatureionmethodsarealwaysmeeasilytrappedintolocaloptimawhensearchinginthehighdimensionalspaces.Intheproposedalgithmtheoptimization

8、probleminNdimensionaldiscretespaceisconvertedintoanoptimizationprobleminonedimensionalcontinuousspacewhichthePSOalgithmwillbeusedtosolve.Experimentalresultsshowthatthenumberoffeaturesedisreducednoticeablytheclassificatio