Treść książki
Przejdź do opcji czytnikaPrzejdź do nawigacjiPrzejdź do informacjiPrzejdź do stopki
Red.S.Sędziwy,SchedaeInformaticae,Vol.17/18December2009
Kraków2009,ISSN0860-0295,©byUJ
13
Fig.3.
usedtorecognizetheroad.Thisisatwo-layerPerceptron,inwhichthefirstlayer
(input)has400neurons,whilethesecondonecontains1536neurons.Thenumber
ofneuronsinthesecondlayerisalsothenumberofexaminedpixels.Totrainthe
networkthefollowingfunctionshavebeenused:
•inputlayer–radialbasistransferfunction,y=e−n
2
,
•outputlayer–logsigmoidtransferfunction,y=1
1+e−n.
Thenetworkhasbeentrainedwiththeback-propagationalgorithmusing500exam-
ples.Thenetwork’slearningcoefficientwas0.9.
Duetothefactthattheactivationfunctionofthesecondlayeriscontinuous,the
networkmayreturnvaluesfrom0to1.Inordertospecifywhichvalues,returned
bythenetwork,shouldbeclassifiedasaroad,aspecialparameterhasbeenusedin
thesystem.Itoperatesontheprinciplethatallvaluessmallerthanthisparameter
aretreatedasaroad,whileallvaluesgreaterthanthataretreatedasafieldthat
isnottheroad.Thetestsprovedthatthebestparameterforthissystemequals
0.4.Therefore,allvaluesreturnedbythenetwork,whicharelowerthan0.4are
regardedasaroadandgreatervaluesarenot.Theestimationerrorisaresultof
incorrectlydetectedpixelsdividedbyalltestedpixels,whereincorrectlydetected
