資源描述:
《滿意特征選擇及其應(yīng)用》由會(huì)員上傳分享,免費(fèi)在線閱讀,更多相關(guān)內(nèi)容在行業(yè)資料-天天文庫(kù)。
1、第23卷第1期控制理論與應(yīng)用Vo1.23No.12006年2月ControlTheory&ApplicationsFeb.2006文章編號(hào):1000-8152(2006)01-0019-06滿意特征選擇及其應(yīng)用112張葛祥,金煒東,胡來(lái)招(1.西南交通大學(xué)電氣工程學(xué)院,四川成都610031;2.電子對(duì)抗國(guó)防科技重點(diǎn)實(shí)驗(yàn)室,四川成都610036)摘要:實(shí)際應(yīng)用中的特征選擇是一個(gè)滿意優(yōu)化問(wèn)題.針對(duì)已有特征選擇方法較少考慮特征獲取代價(jià)和特征集維數(shù)的自動(dòng)確定問(wèn)題,提出一種滿意特征選擇方法(SFSM),將樣本分類
2、性能、特征集維數(shù)和特征提取復(fù)雜性等多種因素綜合考慮.給出特征滿意度和特征集滿意度定義,設(shè)計(jì)出滿意度函數(shù),導(dǎo)出滿意特征集評(píng)價(jià)準(zhǔn)則,詳細(xì)描述了特征選擇算法.雷達(dá)輻射源信號(hào)特征選擇與識(shí)別的實(shí)驗(yàn)結(jié)果顯示,SFSM在計(jì)算效率和選出特征的質(zhì)量方面明顯優(yōu)于順序前進(jìn)法、新特征選擇法和多目標(biāo)遺傳算法.證實(shí)了SFSM的有效性和實(shí)用性.關(guān)鍵詞:優(yōu)化;滿意優(yōu)化;特征選擇;識(shí)別中圖分類號(hào):TP18,O235文獻(xiàn)標(biāo)識(shí)碼:ASatisfactoryfeatureselectionanditsapplications112ZHANGGe
3、xiang,JINWeidong,HULaizhao(1.SchoolofElectricalEngineering,SouthwestJiaotongUniversity,ChengduSichuan610031,China;2.NationalEWLaboratory,ChengduSichuan610036,China)Abstract:Featureselectionisessentiallyasatisfactoryoptimizationprobleminengineeringapplicat
4、ions.Mostoftheexistingfeatureselectionmethodsdidnotconsiderthecostoffeatureextractionandautomaticdecisionofthedimensionoffeaturesubse.tInthispaper,anovelapproachcalledsatisfactoryfeatureselectionmethod(SFSM)isproposed.SFSMconsiderscompromisinglyclassificatio
5、nperformanceoffeaturesamples,thedimensionoffeaturesetandthecomplexityoffeatureextraction.Featuresatisfactoryrateandfeaturesetsatisfactoryratearedefined.Severalsatisfactoryratefunctionsaredesigned.Satisfactoryfeaturesetevaluationcriterionisgiveninamathematica
6、lway.Satisfactoryfeatureselectionalgorithmisdescribedindetai.lExperimentalresultsofradaremittersignalfeatureselectionandrecognitionshowthatSFSMissuperiortosequentialforwardselectionusingdistancecriterion,newfeatureselectionmethodandmultiobjectivegeneticalgo
7、rithmincomputingefficiencyandfeaturequalities.Hence,thevalidityandapplicabilityoftheproposedmethodareverified.Keywords:optimization;satisfactoryoptimization;featureselection;recognition[1]1引言(Introduction)難以保證得到的特征集是最優(yōu)解,所以,在實(shí)際在模式識(shí)別、機(jī)器學(xué)習(xí)和數(shù)據(jù)挖掘等領(lǐng)域中,特應(yīng)用中,人們總是尋找
8、可計(jì)算和實(shí)用的類分離判據(jù)征選擇因能降低特征向量維數(shù)、減少特征提取代價(jià)、和性能較好的搜索算法,以期獲得較好的次優(yōu)[2,3]簡(jiǎn)化分類器設(shè)計(jì)和提高識(shí)別率而成為近年來(lái)一個(gè)充解.由此可知,特征選擇實(shí)質(zhì)上是一個(gè)滿意優(yōu)化滿生機(jī)和活力的研究課題,受到人們的廣泛關(guān)問(wèn)題,得到的解均是滿意解.而且,已有的特征選擇[1~3]注.特征選擇的任務(wù)是利用模式樣本集內(nèi)部信算法較少考慮特征集的維數(shù)和特征獲取的代價(jià),造息