Treść książki
Przejdź do opcji czytnikaPrzejdź do nawigacjiPrzejdź do informacjiPrzejdź do stopki
Red.S.Sędziwy,SchedaeInformaticae,Vol.17/18December2009
Kraków2009,ISSN0860-0295,©byUJ
10
properlane.Thispaperpresentsasolutionfortheroadedgedetectionproblem,
whichusesneuralnetworks.Usingthiswonderfultool–i.e.neuralnetworks–
wasdictatedbythedesiretobuildasystemthatdoesnotusecomplexalgorithms
toidentifyanimagebutisequallyeffectiveorbetter.Oneofthemainobjectives
whilecreatingthissystemwasitsabilitytoeasilyadapttovariousroadconditions.
Theneuralnetworksusedhavebeentrainedon500samples.Thesamplescontain
theoriginalphotosandimagesofaselectedroad.Inthecourseoftheresearchtwo
solutionsarose.ThefirstsolutionistouseasinglePerceptrontorecognizetheroad.
ThesecondsolutionistoclassifythephotosusingaKohonennetworkandestablish
aseparatenetworkforeachclass.Thesecondchapterofthispaperdescribesa
theoreticalapproachtotheproblem.Adescriptionofasolutionispresentedin
thethird,mainchapter.Conclusionsdrawnwhilestudyingthesaidproblemand
suggestionsforfurtherdevelopmentofthesystemareincludedinthelastchapter.
2.Imagerecognition
Byanimagewewillunderstandatwo-dimensionalillustration.Thetaskofthe
imagerecognitionprocessistoassignatestobjecttoagrade.Thisassignmentis
basedonasequence,forwhichthecorrectclassificationisknownandthisimme-
diatelybringsaforementionedneuralnetworkstomind.Inordertodefineimage
recognitionproperly,wemustfirstdefinetheequivalencerelationshipK,known
alsoasaclassification.Thisrelationship(K⊂DxD)isdefinedonasetofrec-
ognizedobjects(D)andsplitsitintoacollectionofequivalenceclassesD
ithat
correspondtoindividualimages.ThenumberofclassesgeneratedbyKisequalto
LandIisacollectionofindexesoftheseclasses,thereforewecanwrite:
D=UDi,
∀µ,ν∈I,µ±νD
µ∩Dν=∅,
∀dµ,dν∈D(d
µ,dν>∈K⇒∃
i∈I(d
µ∈Di)∧(dν∈Di).
Itfollowstherepresentation:
A:D→I,
∀d∈D∃i∈IA(d)=i≡d∈D
i.
(1)
(2)
(3)
(4)
(5)
Therecognitionalgorithmshouldperformthefollowingmapping:ˆ
A:D→I∪{i0},
wherei0meanslackofresponse.Itisthesubmissionofthreeothermappings:
A=F◦C◦B.
ˆ
B:D→X–selectionoffeatures,
(6)