Leaf Classification Utilizing a Convolutional1

《Leaf Classification Utilizing a Convolutional1》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、ProceedingsoftheIEEESoutheastCon2015,April9-12,2015-FortLauderdale,FloridaLeafClassificationUtilizingaConvolutionalNeuralNetworkTimothyJ.Jassmann,RahmanTashakkori,andR.MitchellParryDepartmentofComputerScienceAppalachianStateUniversityBoone,NC28608,USA{jassmanntj,tashakkorir,parryrm}@appstate.ed

2、uAbstract—Treeidentificationisanimportanttaskinrecognitionwhichobtained98.19%accuracy.Simardetal.biologicalresearch.Sincenotallbiologistsarequalifiedforthis[5]trainedaclassifierfortheMNISThandwritingtaskandexpertsarenotaccessibleallthetime,anapplicationrecognitiontrainingsetthatperformedwith99.

3、6%accuracy.toassistinplantidentificationwouldbeextremelyuseful.ThispaperdescribesanapplicationclassifiesthetypeofatreeII.PREPROCESSINGbasedonapictureofoneofitsleaves.ThesystemdevelopedTheoriginalleafimageisresizedto60x80pixelsinorderforthisresearchutilizesaconvolutionalneuralnetworkinantoreduce

4、computingtime.Next,theZeroComponentsandroidmobileapplicationtoclassifynaturalimagesofleaves.Analysis(ZCA)whiteningtechnique[6]isappliedtotheimagetomakethefeatureslessredundant.Keywords—leafclassification,convolutionalneuralnetwork,imageprocessingPixeldatainimageshasseveralredundanciessinceI.INT

5、RODUCTIONpixelsneareachotherareusuallyhighlycorrelated.TheZCAwhiteningtechniqueisapreprocessingstepthatObjectclassificationisanimportantareaofresearchinreducestheseredundanciesbymakingthefeatureslesscomputervision.Historically,objectclassificationhascorrelatedwitheachotherandensuringthatthefeat

6、ureshaverequiredsegmentationoftheobjectfromtheimagefollowedthesamevariance[6].TheformulafortheZCAwhiteningbyclassification,howeverdevelopmentofconvolutionalmatrixWisshownin(1)[6],whereUcontainstheneuralnetworkshasrecentlyallowedforobjectclassificationeigenvectorsofthecovariancematrixofthematrix

7、tobeofnaturalimages.ThispaperdescribesanapplicationforwhitenedandDisthediagonalmatrixcontainingtheleafclassificationthatutilizesaconvolutionalneuralcorrespondingeigenvalues.TheZCAwhitenedimageisnetwork.Anapplicationofthistype,ifpu