Classification of automatic skin lesions from dermoscopic images utilizing deep learning

UNIVERSIDADE ESTADUAL DE CAMPINAS SISTEMA DE BIBLIOTECAS DA UNICAMP

REPOSITÓRIO DA PRODUÇÃO CIENTIFICA E INTELECTUAL DA UNICAMP

Versão do arquivo anexado / Version of attached file:

Versão do Editor / Published Version

Mais informações no site da editora / Further information on publisher's website:

https://set.org.br/ijbe-v-5

DOI: 10.18580/setijbe.2019.14

Direitos autorais / Publisher's copyright statement:

©2019

by Sociedade Brasileira de Engenharia de Televisão. All rights reserved.



T

Abstract—Skin cancers are the most incidental in Brazil.

Thousands of Brazilians are diagnosed annually with the disease, which occurs due to the abnormal growth of the cells that make up the skin and, therefore, can give rise to several types of skin cancer, being divided into two types melanoma and non-melanoma. Skin cancer, which represents respectively 25% of the malignant tumors and about 30% of the cases of cancer in Brazil, being the majority due to the excessive exposure to the sun’s ultraviolet rays. Skin tumors are usually perceived more efficiently, and when diagnosed early, they are more likely to heal. The present study is relies on the development of an algorithm based on Deep Learning for the recognition of tumors in skin images. The AlexNet, which is a Deep Learning architecture is modified to attending our classification problem. The experiments are conducted through 1400 and 2400 images, after twice training with different optimizer, SGD is the better optimizer with99.79% of accuracy and 0.0120% of loss in training, for the scenery of 2400 images.

Index Terms—Deep Learning, Python, Biomedical Image

Processing, Skin, Melanoma.

, ,1752'8&7,21

+( PHODQRPD VNLQ FDQFHU RULJLQDWHV LQ PHODQRF\WHV PHODQLQSURGXFLQJFHOOVDVXEVWDQFHWKDWGHWHUPLQHV VNLQFRORUZKLFKLVPRUHFRPPRQLQZKLWHDGXOWVDSSHDULQJ DQ\ZKHUHRQWKHERG\RQWKHVNLQRUPXFRXVPHPEUDQHV LQ WKH IRUP RIVSRWVRU VLJQV2Q WKHRWKHU KDQGEODFNVNLQQHG LQGLYLGXDOVLWLVPRUHFRPPRQLQFOHDUDUHDVVXFKDVWKHSDOPV RI WKH KDQGV DQG VROHV RI WKH IHHW 7KLV W\SH RI GLVHDVH LV HVSHFLDOO\GDQJHURXVEHFDXVHLWFDQVSUHDGWRRWKHUWLVVXHVRI WKHERG\LHLIQRWGHWHFWHGHDUO\LWJHQHUDWHVPHWDVWDVHVWKDW PDNH WKH patient’s FOLQLFDO SLFWXUH PRUH VHULRXV >@>@

$OWKRXJK VNLQ FDQFHU LV WKH PRVW IUHTXHQW LQ %UD]LO DQG DFFRXQWV IRU DERXW  RI DOO UHJLVWHUHG PDOLJQDQFLHV 0HODQRPDUHSUHVHQWVRQO\RIPDOLJQDQWQHRSODVPVRIWKH RUJDQEHLQJWKHPRVWVHULRXVW\SHGXHWRLWVKLJKSRVVLELOLW\ RI SURYRNLQJ PHWDVWDVLV VSUHDG RI FDQFHU WR RWKHU RUJDQV

$PRQJWKHV\PSWRPVWKHUHDUHVSRWVRQWKHVNLQ ZRXQGV WKDWGRQRWKHDODQG VSRWVWKDWFKDQJHWKHLUDSSHDUDQFHDQG PHODQRPDVFDQDSSHDURQKHDOWK\VNLQRUH[LVWLQJVNLQQHYL,W PD\ DOVR IRUP EURZQ LUUHJXODU IODW RU SURWUXGLQJ VNLQ SODTXHV ZLWK GLIIHUHQW FRORU VSRWV RU KDUG EODFN RU JUD\

30LQDQJR LV SXUVXLQJ DQ 06F GHJUHH LQ (OHFWULFDO (QJ  (( DW  WKH /DE RI 9LV &RPP /&9 DW WKH 6WDWH 8QLY RI &DPSLQDV 81,&$03 SDEORGDYLG#JPDLOFRP 'U<,DQRLVWKH/&981,&$03KHDG\X]R#GHFRPIHHXQLFDPSEU $&%0RQWHLURLVFXUUHQWO\SXUVXLQJDQ3K'GHJUHHLQ((DW/&9 81,&$03PRQWHLUR#GHFRPIHHXQLFDPSEU 53DGLOKD LVFXUUHQWO\SXUVXLQJDQ3K'GHJUHHLQ((DW/&981,&$03 SDGLOKD#GHFRPIHHXQLFDPSEU **2OLYHLUDLVFXUUHQWO\SXUVXLQJDQ06FGHJUHHLQ((DW/&9 81,&$03ROLYHLUDJRPHVJDEULHO#LHHHRUJ

OXPSV 0HODQRPDV PD\ YDU\ LQ UHODWLRQ WR WKHLU DSSHDUDQFH ZKHUH VRPH DUH EURZQLVK IODW LUUHJXODU SODTXHV DQG FRQWDLQ VPDOOEODFNVSRWVDQGRWKHUVDUHWDOOEURZQLVKEURZQFRORUHG SODTXHVZLWKUHGZKLWHEODFN RUEOXHGRWVZKHUHIRUVRPH WLPHV PHODQRPD DSSHDUV DV D KDUG WXPRU RI D UHG EODFN RU JUD\FRORU>@>@>@

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

%UD]LOKDVDQHVWLPDWHGQHZFDVHVHDFK\HDURIZKLFK DUHPHQDQGDUHZRPHQ– ,1&$1DWLRQDO &DQFHU ,QVWLWXWH DQG XQIRUWXQDWHO\ WKHUH DUH  GHDWKV ZKHUH  PHQ DQG  ZRPHQ  – 6,0 0RUWDOLW\ ,QIRUPDWLRQ6\VWHP(DUO\GHWHFWLRQRIFDQFHULVDVWUDWHJ\WR ILQGDWXPRUDWDQHDUO\VWDJHDQGWKXVHQDEOHDJUHDWHUFKDQFH RI WUHDWPHQW 7KLV GHWHFWLRQ FDQ EH GRQH WKURXJK WKH LQYHVWL JDWLRQ ZLWK FOLQLFDO ODERUDWRU\ RU UDGLRORJLFDO H[DPLQDWLRQV RISHRSOHZLWKVLJQVDQGV\PSWRPVVXJJHVWLYHRIWKHGLVHDVH HDUO\GLDJQRVLVRUZLWKWKHXVHRISHULRGLFH[DPLQDWLRQV LQ SHRSOH ZLWKRXW VLJQV RU V\PSWRPV WUDFLQJ EXW EHORQJLQJ WR JURXSV ZLWK D JUHDWHU SRVVLELOLW\ RI KDYLQJ VXFK D GLVHDVH

$PRQJ WKH PDLQ IDFWRUV VXQ H[SRVXUH HQGV XS EHLQJ WKH PDLQPHDQVRIREWDLQLQJWKHGLVHDVHWKURXJKXOWUDYLROHW89 UD\V LV D VLJQLILFDQW ULVN IDFWRU IRU PRVW PHODQRPDV ZKHUH WKH\GDPDJHWKH'1$RIVNLQFHOOVDQGWKXVDIIHFWWKHFRQWURO RIWKHLUFHOOJURZWK >@

3HRSOHZKRKDYHWDNHQWRRPXFKVXQWKURXJKWKHLUOLIHWLPH ZLWKRXWDGHTXDWHSURWHFWLRQVXQVFUHHQDQGRUVXQEORFNDUHDW LQFUHDVHG ULVN IRU PHODQRPD ZKHUH VXQEXUQV  HVSHFLDOO\ LQ FKLOGKRRGRUDGROHVFHQFHLQFUHDVHWKHULVNRIGHYHORSLQJVNLQ FDQFHU 6WLOO WDNLQJ LQWR FRQVLGHUDWLRQ WKDW JHRJUDSKLFDOO\ OLYLQJQHDUWKHHTXDWRURUDWKLJKHUDOWLWXGHDOVRLQFUHDVHVWKH ULVNRQFHWKHsun’s UD\VDUHPRUHGLUHFW7KLVSHRSOHOLYLQJDW KLJK DOWLWXGHV DUH PRUH H[SRVHG WR 89 UDGLDWLRQ >@>@

,QUHFHQW\HDUVZHDUHH[SHULHQFLQJDUHYROXWLRQLQWKHDUHD RI$UWLILFLDO,QWHOOLJHQFH$,>@ – >@DQGWKHPDLQGULYHURI WKLVUHYROXWLRQDUHWHFKQRORJLHVEDVHGRQ'HHS/HDUQLQJ'/

This open access article is distributed under a Creative Commons Attribution (CC-BY) license. http://www.set.org.br/ijbe/ doi: 10.18580/setijbe.2019.14 Web Link: http://dx.doi.org/10.18580/setijbe.2019.14

Automatic Skin Lesions Classification from

Dermoscopic Images Employing Deep Learning

Pablo Minango, Yuzo Iano, Ana Carolina Borges Monteiro, Reinaldo Padilha Fran¸ca and Gabriel Gomes de Oliveira



DQGWDNLQJLQWRDFFRXQWWKHUHFHQWDGYDQFHVLQ FRPSXWDWLRQDO SRZHU KDYH HQDEOHG WKH SRVVLELOLW\ RI GHYHORSLQJ DOJRULWKPV EDVHG RQ &RPSXWHU 9LVLRQ DQG UHODWHG WR '/ >@ – >@ 7KLV PHWKRGLQVLPSOHWHUPVLVWKHSDUWRI0DFKLQH/HDUQLQJ 0/ WKDWWKURXJKKLJKOHYHODOJRULWKPVPLPLFVWKHQHXUDOQHWZRUN RIWKHKXPDQEUDLQJHQHUDOO\XVLQJGHHSQHXUDOQHWZRUNVDQG GHSHQGLQJ RQ PDQ\ GDWD IRU WUDLQLQJ >@ – >@

,QRUGHUWRUHDFKWKHPRUHDGYDQFHG'/OHYHOWKHDUWLILFLDO QHXUDO QHWZRUNV SULQFLSOH KDV EHHQ GHYHORSHG WR VXSSRUW GLVFUHWH OD\HUV FRQQHFWLRQV DQG GDWD SURSDJDWLRQ GLUHFWLRQV DQGVXFKGDWDLVVXEMHFWHGWRVHYHUDOQRQOLQHDUSURFHVVLQJ OD\HUV WKDW VLPXODWH WKH ZD\ RI WKLQNLQJ RI QHXURQV >@>@>@>@>@

,QDVLPSOLILHGDQGJHQHULFZD\'/LVFRPSOH[DOJRULWKPV built from a stack of layers of digital ”neurons”, fed by an LPPHQVHYROXPHRIGDWDFDSDEOHRIUHFRJQL]LQJ LPDJHVDQG VSHHFK SURFHVVLQJ WKHP DQG OHDUQLQJ WR SHUIRUP H[WUHPHO\ WDVNV ZLWKRXW KXPDQ LQWHUIHUHQFH LQ WKLV VFHQDULR KDYLQJ FODVVLILFDWLRQ DQG UHFRJQLWLRQ WDVNV HVSHFLDOO\ IRU LPDJH UHFRJQLWLRQ>@–>@

3\WKRQ LV D ODQJXDJH WKDW ZDV FUHDWHG IRU SURGXFWLRQ DQG GHYHORSPHQWLQDQHDV\SUDFWLFDODQGIDVWZD\EHLQJ DQDJLOH HDV\ DQG REMHFWLYH ODQJXDJH GHPRFUDWL]LQJ LWV WHDFKLQJ PDNLQJ LW LQFUHDVLQJO\ VRXJKW DIWHU %HLQJ DQ REMHFWRULHQWHG ODQJXDJH LWV VLPSOLFLW\ LV LWV PDLQ FKDUDFWHULVWLF ,W LV DOVR RSHQDQGQRQSURILW ODQJXDJH

3\KWRQ LV LGHDO IRU VFLHQWLILF DSSOLFDWLRQV DV '/ ZKHUH LW LV DQ H[SUHVVLYH ODQJXDJH ZKHUH LW LV HDV\ WR WUDQVODWH WKH UHDVRQLQJLQWRDQDOJRULWKPDQGVWLOOWDNLQJLQWRDFFRXQWWKDW LWDOORZVWKHZRUNDQGGHYHORSPHQWZLWKGDWDVFLHQFHDQG0/ >@ >@–>@

7KXV LQ WKLV SDSHU DQ DOJRULWKP EDVHG RQ '/ ZDV GHYHORSHGZKLFKDLPVWRWKHUHFRJQLWLRQRIVNLQPHODQRPDV SDWWHUQV ZLWK WKH XVH RI  DQG  LPDJHV IRU WUDLQLQJ ZKHUHZLWKVKRZWKHDFFXUDF\DQGWKHORRVIRUHDFKWUDLQLQJ

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– >@

,QWKLVZD\'/OHDUQVDORQHDQGLVIHGE\WKHPXOWLSOHGDWD JHQHUDWHGDWDOOWLPHVWKURXJKPXOWLSOHDQDO\]HVLQWKHVHDOHG GDWDVHW EHLQJ DEOH WR GHFLSKHU IRU H[DPSOH QDWXUDO ODQJXDJH DQG WR UHODWH WHUPV JHQHUDWLQJ D FRUUHVSRQGLQJ PHDQLQJ DV ZHOO DV XVHG LQ LPDJH SURFHVVLQJ DQG FRPSXWHU YLVLRQ 6XFK DSSURDFKHVDOORZDFRPSOH[DQGDEVWUDFWUHSUHVHQWDWLRQRIWKH GDWD IRUPLQJ DQ RUGHUO\ FODVVLILFDWLRQ WKURXJK WKH XVH RI DOJRULWKPVWKDWGRQRWUHTXLUHDSUHSURFHVVLQJRIWKHGDWD DQG

DXWRPDWLFDOO\JHQHUDWHRSWLPDOUHVXOWVWKURXJKWKHKLHUDUFKLFDO OD\HUV RI UHSUHVHQWDWLRQ FRQVLVWLQJ RI WKH WHFKQLTXH ZKHUH VXFK OD\HUV DUH QRQOLQHDU GDWD >@ >@ – >@

1RZDGD\V WKH LPSRUWDQFH RI '/ FDQ EH VHHQ E\ D GLUHFW FRQQHFWLRQ WR QHXUDO QHWZRUNV ZKLFK SURYLGHG IRU H[DPSOH WKH DXWRPDWLF OHDUQLQJ XVHG IRU WKH FUHDWLRQ RI DXWRQRPRXV FDUVDVZHOODVWKH VLJQLILFDQWLPSURYHPHQWRIWKHVHDUFKHV RQ WKH ZHE )XUWKHU 'HHS /HDUQLQJ WHFKQLTXHV KDYH EHHQ enhancing computers’ ability to classify, recognize, detect, and GHVFULEHWKHPRYHUWLPHLQRUGHUWRXQGHUVWDQG>@>@–>@ '/ DQG 0/ DUH FRQFHSWV WKDW JR KDQG LQ KDQG DUH FORVHO\ OLQNHG ZLWK $UWLILFLDO ,QWHOOLJHQFH $, 7KH\ DUH QRW V\QRQ\PVDQG'/HYROYHGIURPPDFKLQHOHDUQLQJ6LQFH$,LV D UDWKHU EURDG WHUP D EUDQFK RI FRPSXWHU VFLHQFH LQ ZKLFK UHVHDUFKLVGHYHORSHGWRILQGVROXWLRQVWKDWVLPXODWHWKHKXPDQ FDSDFLW\ IRU UHDVRQLQJ 0/ LV DOUHDG\ XVLQJ FRQFHSWV RI $, WKURXJKDOJRULWKPVPDNLQJWKHPDFKLQHVDEOHWRRUJDQL]HGDWD DQG LGHQWLI\ SDWWHUQV ZLWKRXW QHFHVVDULO\ D SUHSURJUDPPLQJ PDNLQJLQVLPSOHWHUPV$,LVVLPLODUWRWKHSURFHVVRIKXPDQ UHDVRQLQJ UHODWHG WR WKLV SURFHVV RI OHDUQLQJ E\ LQSXWV DQG WKXVILQGLQJSDWWHUQV0RUHRYHUILQDOO\WKHUHLV'/ZKHUH LW FDQ EH FRQVLGHUHG WR EH DW D GHHSHU OHYHO ZLWK KLJKOHYHO DOJRULWKPVZKHUHPDFKLQHVEHJLQWRDFWVLPLODUO\WRWKHQHXUDO QHWZRUN RI WKH KXPDQ EUDLQ 7KXV WKH GDWD DUH VXEPLWWHG WR VHYHUDO OD\HUV LQ WKH VDPH ZD\ DV D QHWZRUN RI QHXURQV LQ RUGHUWRUHFRJQL]HLPDJHVDQGRUVSHHFK)RUH[DPSOHZLWKRXW

KXPDQLQWHUYHQWLRQLQWKHVHTXHQFH>@ – >@

6RHVVHQWLDOO\'/LVDQHPHUJLQJWHFKQRORJLFDOZLWKLQWKH ILHOG RI $, ZKLFK LV FRQVLGHUHG D VXEFDWHJRU\ RI 0/ '/ XVHVQHXUDOQHWZRUNVWRLPSURYHVSHHFKUHFRJQLWLRQFRPSXWHU YLVLRQ LPDJH SURFHVVLQJ DQG QDWXUDO ODQJXDJH SURFHVVLQJ >@ – >@

'XH WR WKH LWHUDWLYH QDWXUH RI '/ PXFK FRPSXWDWLRQDO SRZHULVQHHGHGWRVROYHSUREOHPVZKHUHLWVFRPSOH[LW\JURZV DV WKH QXPEHU RI OD\HUV HPSOR\HG DQG FRQVHTXHQWO\ WKH QHFHVVDU\ GDWD YROXPHV DUH JHWWLQJ ODUJH HQRXJK WR LPSURYH WKHQHWZRUNWUDLQLQJ7KHG\QDPLFQDWXUHRI'/SUHVHQWV WKH RSSRUWXQLW\ WR LQWURGXFH G\QDPLF EHKDYLRUV WR DQDO\WLFV UHODWHG WR WKH GDWDVHW WKDW LV GXH WR WKH FDSDFLW\ IRU FRQWLQXRXV LPSURYHPHQW DQG DGDSWDWLRQ WR FKDQJHV LQ WKH SDWWHUQ RI XQGHUO\LQJLQIRUPDWLRQ >@ – >@

Technologically speaking, DL performs the ”training” of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users’ LQWHUHVW DIWHU WDNLQJ DFWLRQ RQ WKHLU SODWIRUP MXVW DV IURP SUHYLRXV EHKDYLRUV V\VWHPV VXFK DV 6LUL DQG &RUWDQD DUH SDUWLDOO\IHGE\'/LVDOVRWKHEDVHWHFKQRORJ\IRUWRROVOLNH *RRJOH7UDQVODWHDPRQJRWKHUV >@ – >@

,Q VKRUW ZLWK DQ HQRUPRXV DPRXQW RI FRPSXWLQJ SRZHU today’s machines have the possibility and ability to recognize ‹



REMHFWV DQG WUDQVODWH YRLFH LQ UHDOWLPH DPRQJ PDQ\ RWKHU FRPSOH[DFWLYLWLHVRIKXPDQOLIHILQDOO\PDNLQJWKH$,DSSOL FDEOH>@ – >@

,,, 0(7+2'2/2*<

1RZDGD\V SDWWHUQ UHFRJQLWLRQ DQG PHGLFDO LPDJHV FODVVLILFDWLRQ XVLQJ 'HHS /HDUQLQJ DUH SRVVLEOH WR FRPSXWDWLRQDODGYDQFHVDQGODUJHGDWDVHWV>@

&RQYROXWLRQDO 1HXUDO 1HWZRUNV &11 LV DQ DUFKLWHFWXUH RI'HHS/HDUQLQJLQVSLUHGE\WKHKXPDQYLVXDOFRUWH[&11 LV XWLOL]HG LQ VHYHUDO PHGLFDO DSSOLFDWLRQ DV VNLQV OHVLRQ FODVVLILFDWLRQ SUREOHPV >@ SDQFUHDV VHJPHQWDWLRQ >@ H[WUDFWLRQ RI YHVVHOV LQ IXQGXV LPDJHV >@ DQG EUDLQ WXPRU VHJPHQWDWLRQ>@ >@

,Q WKLVZRUNZH HPSOR\ WKHDUFKLWHFWXUHRI $OH[1HW >@ ZKHUHKDVILYHFRQYROXWLRQDOOD\HUVDQGWKUHHIXOO\FRQQHFWHG OD\HUV 7KLV DUFKLWHFWXUH LV XVHG WR REWDLQ WKH ELQDU\ FODVVLILFDWLRQEHWZHHQEHQLJQDQGPDOLJQFDQFHU

A. Dataset Description

7KH ,QWHUQDWLRQDO 6NLQ ,PDJLQJ &ROODERUDWLRQ ,6,& >@ GDWDVHW RI RSHQ DFFHVV ZDV XWLOL]HG IRU WUDLQLQJ $OH[1HW DUFKLWHFWXUH,6,&LVDQRUJDQL]DWLRQWKDWLPSXOVHV0HODQRPD 3URMHFWE\SURYLGLQJWKHDFFHVVRIWKHLULPDJHVIRUDFDGHPLFV DQG LQGXVWULHV ZLWK WKH SXUSRVH RI GHFUHDVH PRUWDOLW\ UDWH RFFDVLRQHGIRUWKHVNLQFDQFHUNQRZQDV0HODQRPDEHFDXVH LI VNLQ FDQFHU LV HDUO\ UHFRJQL]HG LQ LWV LQLWLDO VWDJHV WKH SUREDELOLW\RIFXUHLV EHWWHU

,6,& LV D FROOHFWRU GDWDVHW ZLWK DOPRVW  LPDJHV RI PDQ\ GLIIHUHQW VNLQ OHVLRQV ZLWK WKHLU RZQ KLVWRSDWKRORJLFDO GLDJQRVWLF2XUSULQFLSDOIRFXVLVWKHFODVVLILFDWLRQRILPDJHV ZLWK EHQLJQ DQG PDOLJQDQW FDQFHU ZKLFK UHSUHVHQWV ELQDU\ FODVVLILFDWLRQ

,Q WKLV ZRUN $OH[1HW DUFKLWHFWXUH LV WUDLQHG WZLFH ZLWK GLIIHUHQW TXDQWLWLHV RI LPDJHV 7KLV SURFHGXUH LV GHVFULEHG EHORZ

B. Architecture Description

:H XVH WKH LGHD RI $OH[1HW >@ ZKLFK ZDV XVHG DQG DSSOLHG WR UHFRJQL]H WKH ,PDJH1HW /DUJH 6FDOH 9LVXDO 5HFRJQLWLRQ &KDOOHQJH )LJ  GHSLFWV WKH GHWDLOHG FRQILJXUDWLRQ 7KH VWUXFWXUH RI $OH[1HW FRQVLVWV RI ILYH FRQYROXWLRQDO OD\HUV ZKLFK LQFOXGH WKH FRQFHSW RI 0D[3RROLQJ>@DQG%DWFK1RUPDOL]DWLRQ >@ LQ HDFK RQH RI WKHLU OD\HUV

7KH RULJLQDO $OH[1HW DUFKLWHFWXUH ZDV PRGLILHG IRU DGDSWLQJRXUSUREOHP7KHILUVWSDUWRIWKLVDUFKLWHFWXUHLVWKH EORFN RI LQSXW LPDJHV GHWHUPLQHG E\ WKH EDWFK VL]H K\SHUSDUDPHWHU 7KH RXWSXW RI WKH ILUVW EORFN ,1 SDVVHV WKURXJKILYHFRQYROXWLRQDOEORFNVWRREWDLQWKHEHWWHUDFFXUDF\ GXULQJ WKH WUDLQLQJ FODVVLILFDWLRQ ZKHUH HDFK RQH RI WKHVH FRQYROXWLRQDO EORFNV DUH FRPSRVHG IRU VPDOO FRQYROXWLRQDO ILOWHUV DOVR QDPHG NHUQHOV 7KH VL]H RI WKHVH NHUQHOV DUH PDWULFHVRI î IRU&î IRU&DQGî IRU& &DQG&VHH)LJ(DFKRQHRIWKHVHEORFNVLVHTXLSSHG ZLWK5HFWLILHG/LQHDU8QLW5H/8DVWKHDFWLYDWLRQIXQFWLRQ RIHDFKOD\HU

)LJ$OH[1HWDUFKLWHFWXUHDGDSWHG WRRXUSUREOHP

,Q )LJ  WKH ODVW WKUHH OD\HUV DUH NQRZQ DV GHQVH OD\HUV 'XULQJWKHWUDLQLQJVWHSLQHDFKRQHRIWKHGHQVHOD\HUVHYHU\ QHXURQKDVDSUREDELOLW\SRIEHLQJWHPSRUDOO\RII,WPHDQVWKDW

LWLVGHDFWLYDWHGEXWSUREDEO\LWZLOOEHDFWLYDWHGGXULQJWKHQH[W VWHSV7KLVSURFHVVLVNQRZQDV'URSRXW>@

$IWHUHDFKEDWFKVL]HRILPDJHVDUHILOWHUHGWKHUHVXOWV SDVVWR WKH GHQVH OD\HU VWDJH ZKLFK LV FRPSRVHG RI WKUHHGHQVH OD\HUV 7KLVZRUNPRGLILHVWKHODVWRQHOD\HU RXWSXWLQRUGHUWRSHUIRUPD ELQDU\ FODVVLILFDWLRQ EHWZHHQ EHQLJQ DQG PDOLJQDQW DV ZHOO DV WKHLU DFWLYDWLRQ IXQFWLRQ ZDV PRGLILHG IURP D 6RIWPD[ WR D 6LJPRLG_{ZKLFKLVYHU\XVHGIRUELQDU\FODVVLILFDWLRQV SUREOHPV}

:LWKWKHXVHRI$OH[1HWDUFKLWHFWXUHPRGLILHGDQGSUHYLRXVO\ GHVFULEHG WKH ZHLJKWV RI WKH RXWSXW DUH VHQW WR DQ RSWLPL]HU GXULQJHDFKRQHHSRFKZLWKWKHSXUSRVHRIWR PHDVXUHDFFXUDF\ DQG ORVV 7KH UHVXOWV RI WKHVH RSWLPL]HUV DUH WHVWHG LQ HDFK RQH HSRFK XQWLO DWWDLQLQJ WKHLU FRQYHUJHQFH 7R DWWDLQ WKLV SUHPLVH ZH XVH WKUHH GLIIHUHQW W\SHV RI RSWLPL]HU ZKLFK DUH 6WRFKDVWLF *UDGLHQW'HVFHQW6*' 506SURSDQG$GDP>@:HFRQVLGHU DQ LQLWLDO OHDUQLQJ UDWH RI  IRU WKH WKUHH RSWLPL]HUV ZLWK D %LQDU\ FURVVHQWURS\ ORVV IXQFWLRQ >@ )XUWKHUPRUH DQ HDUO\ VWRSSLQJ FULWHULRQ ZDV LPSOHPHQWHG WR LQWHUUXSW WKH WUDLQLQJ SURFHVV DIWHU WKH WUDLQLQJ DFFXUDF\ “AUC” does not improve on HSRFKV

_{:KHQ DSSOLHG D 6LJPRLG DFWLYDWLRQ IXQFWLRQ LQ WKH ODVW OD\HU 7KH}

DUFKLWHFWXUHRQO\OHDUQVWRSUHGLFWYDOXHVEHWZHHQDQG)RUWKLVUHDVRQLV RQHRI WKH PRVW XVHG LQ ELQDU\ FODVVLILFDWLRQ

2019 SET - Brazilian Society of Television Engineering / ISSN (Print): 2446-9246/ISSN (Online): 2446-9432

This open access article is distributed under a Creative Commons Attribution (CC-BY) license. http://www.set.org.br/ijbe/ doi: 10.18580/setijbe.2019.14 Web Link: http://dx.doi.org/10.18580/setijbe.2019.14



,9 5(68/76$1'',6&866,21

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î  SL[HOV 5*% DQG QRUPDOL]HG VXEWUDFW IURP WKH PHDQDQGGLYLGHE\WKHVWDQGDUGGHYLDWLRQ

)RUWKHVHFRQGWUDLQLQJZHFRQVLGHUHGLPDJHVZKLFK DUH GLYLGHG LQWR  LPDJHV RI EHQLJQ DQG  LPDJHV RI FDQFHU 7KH VDPH FRQGLWLRQV GHVFULEHG DERYH ZHUH WDNLQJ LQ WKLVWUDLQLQJ

7KH HYDOXDWLRQ RI RXU PHWKRGRORJ\ ZDV SHUIRUPHG XVLQJ WKH IROORZLQJ PHWULFV LQ WKH WUDLQLQJ GDWDVHW

‡ Accuracy ZKLFK UHSUHVHQWV WKH FRUUHFW SUHGLFWLRQV LQ HDFK

RQH FODVV GLYLGHG E\ WKH WRWDO QXPEHU RI SUHGLFWLRQV

‡ LossZKLFKFRUUHVSRQGVWRWKHQXPEHURILPDJHVZURQJ

FODVVLI\FRQVLGHULQJWKHWUXH ODEHO

)LJVKRZVWKHDFFXUDF\DQGWKHORVVRIRXUPHWKRGRORJ\ IRU WKH ILUVW WUDLQLQJ 7KH 6*' RSWLPL]HU RXWSHUIRUPV WKH $GDPDQG506RSWLPL]HUGXHWR6*'DFKLHYHVDSHUFHQWDJH RIDQGLQOHDUQLQJDQGORVVUHVSHFWLYHO\

7DEOH ,, VKRZV WKH FRPSDULVRQ SHUIRUPDQFH RI WKH WKUHH RSWLPL]HUV 7$%/(,, $&&85$&<$1'/266&203$5,62162)7+(',))(5(177<3(62) 237,0,=(56 $GDP (SRFKV 506SURS (SRFKV 6*' (SRFKV $FF     97.66  /RVV     0.0567  )RUWKHVHFRQGWUDLQLQJZHFDQDSSUHFLDWHWKHSHUIRUPDQFH RIWKHPRGHOLQ)LJ$JDLQLQWKLVFDVH6*'LVEHWWHURYHU WKH RWKHUV RSWLPL]HUV DFKLHYHG D  RI OHDUQLQJ DQG RI ORVV

7DEOH ,,, VKRZV WKH SHUIRUPDQFH RI WKH WKUHH RSWLPL]HUV ZKHUH ZH DSSUHFLDWH WKDW 6*' KDV EHWWHU DFFXUDF\ DQG ORVV UHVXOWV

&RQFHUQLQJ WKH HSRFKV RI HDFK RQH RSWLPL]HU $GDP DQG 506SURS ZHUH VWRSSHG ZLWK WKH FULWHULRQ RI HDUO\ VWRSSLQJ EHFDXVH WKH\ DWWDLQ DQ DFFXUDF\ FRQYHUJHQFH DQG GR QRW LPSURYHPRUHLIZHFRQWLQXHGZLWKPRUHHSRFKVRIWUDLQLQJ

9 &21&/86,21

:HKDYHSURSRVHGDELQDU\FODVVLILFDWLRQWRGHWHFWWKH HDUO\ VWDJHV RI PHODQRPD 7KH VROXWLRQ XVHG ZDV $OH[1HW

D E )LJ3HUIRUPDQFHRIWKHPRGHOZLWKWUDLQLQJLPDJHVD$FFXUDF\ E /RVV 7$%/(,,, $&&85$&<$1'/266&203$5,62162)7+(',))(5(177<3(62) 237,0,=(56 $GDP (SRFKV 506SURS (SRFKV 6*' (SRFKV $FF     99.79  /RVV     0.0120 

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

$&.12:/('*0(17

:H WKDQN WKH 6WDWH 8QLYHUVLW\ RI &DPSLQDV 81,&$03 %UD]LO IRU SURYLGLQJ DOO WKH QHFHVVDU\ LQIUDVWUXFWXUH IRU WKH GHYHORSPHQWRIWKLVVWXG\

5()(5(1&(6

>@ “AnDOLVH GH GDGRV GDV FDPSDQKDV GH SUHYHQFÜD˜R DR FDˆQFHU GD SHOH SURPRYLGDV SHOD 6RFLHGDGH %UDVLOHLUD



D

E

)LJ3HUIRUPDQFHRIWKHPRGHOZLWKLPDJHVRIWUDLQLQJD$FFXUDF\ E/RVV

GH'HUPDWRORJLDGHD2005,” SWAnais Brasileiros

de Dermatologia YRO  SS – 'HF 

,661  >2QOLQH@ $YDLODEOH KWWS   ZZZ VFLHOREU  VFLHORSKS" VFULSW VFL DUWWH[W SLG 6   QUP LVR

>@ C. M. Balch, “Cutaneous melanoma.,” Annals of

Surgi-cal Oncology 

>@ :Clark, L. From, E. Bernardino, and M. Mihm, “The KLVWRJHQHVLV DQG ELRORJLF EHKDYLRU RI SULPDU\ KXPDQ PDOLJQDQW PHODQRPDV RI WKH skin,” Cancer Research YRO  QR  SS – 0DU  ,661   >2QOLQH@ $YDLODEOH KWWSHXURSHSPFRUJ DEVWUDFW0('

>@ - 6FRWWR DQG 7 5 )HDUV “The DVVRFLDWLRQ RI VRODU XOWUD YLROHWDQGVNLQPHODQRPDLQFLGHQFHDPRQJFDXFDVLDQV in the united states,” Cancer InvestigationYROQR SS –  30,'  '2,  HSULQW KWWSVGRLRUJ >2QOLQH@ $YDLODEOH KWWSVGRL RUJ >@ 1 5D]PMRR\ 9 9 (VWUHOD + - /RVFKL DQG : Farfan, “A comprehensive survey of new PHWDKHXULVWLF algorithms,” Recent Advances in Hybrid Metaheuristics

for Data Clustering 

>@ 1 5D]PMRR\ 0 $VKRXULDQ 9 9 (VWUHOD DQG + - /RVFKL, “EntropyEDVHG EUHDVW FDQFHU GHWHFWLRQ LQ GLJ LWDO PDPPRJUDPV XVLQJ ZRUOG FXS RSWLPL]DWLRQ DO

gorithm,” International Journal of Swarm Intelligence

Research (IJSIR),661

>@ 15D]PMRR\0$VKRXULDQ99(VWUHOD+- /RVFKL 5 3 )UDQFÜD DQG 0 9LVKQHYVNL “&RPSXWHUDLGHG GLDJQRVLV RI VNLQ FDQFHU $ review,” Current Medical

Imaging Reviews,661 3ULQW

>@ 15D]PMRR\DQG99Estrela, “Applications of LPDJH SURFHVVLQJ DQG VRIW FRPSXWLQJ V\VWHPV LQ agriculture,” '2, 

>@ ' - +HPDQWK DQG 9 9 (VWUHOD Deep Learning for

Image Processing ApplicationsVHU$GYDQFHVLQ3DUDO

OHO &RPSXWLQJ ,26 3UHVV  ,6%1  >2QOLQH@ $YDLODEOH KWWSVERRNVJRRJOHFRPEUERRNV"

LG YV)9'Z$$4%$-

>@ , *RRGIHOORZ < %HQJLR DQG $ &RXUYLOOH Deep

Learning 0,7 3UHVV  KWWSZZZ GHHSOHDUQLQJERRNRUJ

>@ / 'HQJ DQG ' <X “Deep learning: Methods DQG applications,” Found. Trends Signal Process. YRO   QR  SS – -XQ ,661  '2,>2QOLQH@$YDLODEOH KWWSG[GRLRUJ

>@ &'RQJ&&/R\.+HDQG; 7DQJ“Learning D GHHS FRQYROXWLRQDO QHWZRUN IRU LPDJH VXSHUresolution,” LQ Computer Vision – ECCV 2014 ' )OHHW 7 3DMGOD % 6FKLHOH DQG 7 7X\WHODDUV (GV &KDP 6SULQJHU ,QWHUQDWLRQDO 3XEOLVKLQJ  SS – ,6%1  

>@ 0 $ 1LHOVHQ Neural networks and deep learning PLVF  >2QOLQH@ $YDLODEOH KWWS QHXUDOQHWZRUNVDQGGHHSOHDUQLQJFRP

>@ *-6/LWMHQV7.RRL%(%HMQRUGL$$$ 6HWLR ) &LRPSL0*KDIRRULDQ-$:0YDQGHU/DDN % YDQ *LQQHNHQ DQG & , 6DQFKH] “A VXUYH\ RQ GHHS learning in medical image analysis.,” Medical Image

AnalysisYROSS–>2QOLQH@$YDLODEOH

KWWSGEOSXQL WULHUGHGEMRXUQDOVPLDPLDKWPO /LWMHQV.%6&*/*6

>@ < /Y < 'XDQ : .DQJ = /L DQG )< :DQJ “Traffic flow prediction with big data: A deep OHDUQLQJ approach,” IEEE Transactions on Intelligent

Trans-portation Systems YRO  SS – 

>@ <&KHQ=/LQ;=KDR*:DQJDQG<*X“Deep OHDUQLQJEDVHG FODVVLILFDWLRQ RI hyperspectral data,”

IEEE J. Sel. Topics in Applied Earth Observations and Remote Sensing YRO  QR  SS – -XQ

 ,661  '2,

-67$56

>@ 13DSHUQRW30F'DQLHO6-KD0)UHGULNVRQ=% Celik, and A. Swami, “The limitations of deep learning in adversarial settings,” in 2016 IEEE European Symp.

on Sec. and Priv. (EuroS P) 0DU SS –

'2, (XUR63

>@ , $UHO ' 5RVH DQG 7 .arnowski, “Deep machine OHDUQLQJ  D QHZ IURQWLHU LQ DUWLILFLDO LQWHOOLJHQFH UH search [research frontier].,” IEEE Comp. Int. Mag.YRO  SS – -DQ 

This open access article is distributed under a Creative Commons Attribution (CC-BY) license.

2019 SET - Brazilian Society of Television Engineering / ISSN (Print): 2446-9246/ISSN (Online): 2446-9432

http://www.set.org.br/ijbe/ doi: 10.18580/setijbe.2019.14 Web Link: http://dx.doi.org/10.18580/setijbe.2019.14



>@ ' 6KHQ * :X DQG +I. Suk, “Deep learning LQ medical image analysis,” Annual Review of Biomedical

Engineering YRO  0DU  '2, DQQXUHY ELRHQJ

>@ <Bengio, “Learning deep architectures for ai,” Found.

Trends Mach. Learn. YRO  QR  SS – -DQ 

,661  '2, 

>2QOLQH@ $YDLODEOH

KWWSG[GRLRUJ

>@ 9 0QLK . .DYXNFXRJOX ' 6LOYHU $ $ 5XVX - 9HQHVV0*%HOOHPDUH$*UDYHV0 5LHGPLOOHU $ . )LGMHODQG * 2VWURYVNL 6 3HWHUVHQ & %HDW WLH$6DGLN,$QWRQRJORX+.LQJ' .XPDUDQ D. Wierstra, S. Legg, and D. Hassabis, “HumanOHYHO control through deep reinforcement learning,” Nature YRO  QR  SS – )HE  ,661  >2QOLQH@ $YDLODEOH KWWSG[GRLRUJQDWXUH

>@ : 6DPHN 7 :LHJDQG DQG . 0XŽOOHU “Explain DEOH DUWLILFLDO LQWHOOLJHQFH 8QGHUVWDQGLQJ YLVXDOL]LQJ DQG interpreting deep learning models,” CoRR YRO DEV  DU;LY  >2Q OLQH@ $YDLODEOH KWWSDU[LYRUJDEV

>@ )3HGUHJRVD*9DURTXDX[$*UDPIRUW 90LFKHO % 7KLULRQ2*ULVHO0%ORQGHO33UHWWHQKRIHU 5 :HLVV 9 'XERXUJ - 9DQGHUSODV $ 3DVVRV ' &RXUQDSHDX 0 %UXFKHU 0 3HUURW DQG ( 'XFKHVQD\ “Scikitlearn: Machine learning in python,” J. Mach.

Learn. Res.YROSS–1RY,661

 >2QOLQH@ $YDLODEOH KWWSGODFPRUJFLWDWLRQFIP"LG  >@ 7E. Oliphant, “Python for scientific computing,”

Com-puting in Science EngineeringYROQRSS–

0D\  ,661  '2, 0&6( 

>@ 0 ) Sanner, “Python: A programming language IRU software integration and development.,” Journal of

Molecular Graphics Modelling YRO   SS –



>@ ) Sebastiani, “Machine learning in automated WH[W categorization,” ACM Comput. Surv.YROQRSS – 0DU  ,661  '2,  >2QOLQH@$YDLODEOH KWWSGRLDFPRUJ 

>@ , + :LWWHQ DQG ( )UDQN Data Mining: Practical

Machine Learning Tools and TechniquesQGVHU0RU

JDQ .DXIPDQQ 6HULHV LQ 'DWD 0DQDJHPHQW 6\VWHPV 0RUJDQ.DXIPDQQ 

>@ 0$EDGL3 %DUKDP-&KHQ=&KHQ$ 'DYLV - 'HDQ 0 'HYLQ 6 *KHPDZDW * ,UYLQJ 0 ,VDUG 0 .XGOXU - /HYHQEHUJ 5 0RQJD 6 0RRUH ' * 0XUUD\%6WHLQHU37XFNHU99DVXGHYDQ3:DUGHQ 0 :LFNH<<Xand X. Zheng, “Tensorflow: A system IRU ODUJHscale machine learning,” in 12th USENIX

Symp. on Op. Systems Des. and Implem. (OSDI 16)

 SS – >2QOLQH@ $YDLODEOH KWWSVZZZ XVHQL[RUJV\VWHPILOHVFRQIHUHQFHRVGLRVGLDEDGL SGI

>@ (  $OSD\GLQ Introduction to Machine Learning QG 7KH 0,7 3UHVV  ,6%1 ; 

>@ -'HQJ:'RQJ56RFKHU//L.DL/LDQG/L)HL Fei, “Imagenet: A largeVFDOH KLHUDUFKLFDO LPDJH database,” in 2009 IEEE Conference on Computer

Vision and Pattern Recognition -XQ  SS –

 '2, &935

>@ &19DVFRQFHORVand B. N. Vasconcelos, “IncreasLQJ GHHSOHDUQLQJPHODQRPDFODVVLILFDWLRQE\FODVVLFDODQG expert knowledge based image transforms,” CoRRYRO DEV  DU;LY  >2Q OLQH@ $YDLODEOH KWWSDU[LYRUJDEV

>@ + 5RWK $ )DUDJ / /X ( % 7XUNEH\ DQG 5  0 Summers, “Deep convolutional QHWZRUNV   IRU pancreas segmentation in CT imaging,” CoRR YRO DEV  DU;LY  >2Q OLQH@ $YDLODEOH KWWSDU[LYRUJDEV

>@ 0 0HOLQVFDN 3 Prentasic, and S. Loncaric, “Retinal vessel segmentation using deep neural networks,”

VIS-APP 2015 - 10th International Conference on Computer Vision Theory and Applications; VISIGRAPP, Proceed-ings YRO  SS – -DQ  '2, 



>@ 0 +DYDHL $ 'DY\ ' :DUGH)DUOH\ $ %LDUG $ &&RXUYLOOH<%HQJLR &3DO3-RGRLQDQG H. Larochelle, “Brain tumor segmentation with GHHS neural networks,” CoRR YRO DEV  DU;LY  >2QOLQH@ $YDLODEOH KWWSDU[LY RUJDEV

>@ 6 3HUHLUD $ 3LQWR 9 $OYHV DQG & 6LOYD “Deep FRQYROXWLRQDO QHXUDO QHWZRUNV IRU WKH VHJPHQWDWLRQ RI JOLRPDV LQ PXOWLVHTXHQFH mri,” YRO  -DQ  SS – ,6%1  '2,  

>@ $ .UL]KHYVN\ , 6XWVNHYHU DQG * ( +LQWRQ “Im DJHQHW FODVVLILFDWLRQ ZLWK GHHS FRQYROXWLRQDO QHXUDO networks,” Neural Information Processing SystemsYRO -DQ'2, 

>@ , SRUMHFW Public skin lesion image archive KWWSVZZZLVLFDUFKLYHFRPWRS:LWK+HDGHU RQO\+HDGHU7RSJDOOHU\

>@ + :X and X. Gu, “MaxSRROLQJ GURSRXW IRU UHJ XODUL]DWLRQ of convolutional neural networks,” CoRR YRO DEV  DU;LY  >2Q OLQH@$YDLODEOH KWWSDU[LYRUJDEV >@ S. Ioffe and C. Szegedy, “Batch normalization: Accel

HUDWLQJ GHHS QHWZRUN WUDLQLQJ E\ UHGXFLQJ LQWHUQDO FR variate shift,” CoRRYRODEVDU;LY  >2QOLQH@ $YDLODEOH KWWSDU[LYRUJDEV 

>@ $ *URQ Hands-On Machine Learning with

Scikit-Learn and TensorFlow: Concepts, Tools, and Tech-niques to Build Intelligent Systems, 1st. O’Reilly 0HGLD

,QF,6%1 >@ S. Ruder, “An overview of gradient descent optimiza

WLRQ algorithms,” CoRR YRO DEV 



DU;LY  >2QOLQH@ $YDLODEOH KWWSDU[LY RUJDEV

>@ .-DQRFKDDQG:M. Czarnecki, “On loss IXQFWLRQVIRU deep neural networks in classification,” CoRR YRO DEV  DU;LY  >2Q OLQH@ $YDLODEOH KWWSDU[LYRUJDEV

>@ )&KROOHWet al.KerasKWWSVNHUDVLR 

>@ J. D. Hunter, “Matplotlib: A 2d graphics environment,”

Computing in Science & EngineeringYRO QRSS

–'2, 0&6(

Reinaldo Padilha Franc¸a. %6F LQ &RPSXWHU

(QJLQHHULQJ IURP 8QLYHUVLW\ 5HJLRQDO &HQWHU RI (VStULWR 6DQWR GH 3LQKDO   3UHVHQWO\ KH LV D 3K'&DQGLGDWHE\'HSDUWPHQWRI&RPPXQLFDWLRQV '(&20 )DFXOW\ RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ )((& DW 6WDWH 8QLYHUVLW\ RI &DPSLQDV 81,&$03 DQG D UHVHDUFKHU DW WKH /DERUDWRU\RI9LVXDO&RPPXQLFDWLRQV/&9+HLV DOVR WKH 3URFHHGLQJV &KDLU RI WKH %UD]LOLDQ 6\PSRVLXP RQ 7HFKQRORJ\ %76\P +LV LQWHUHVW LQFOXGHV SURJUDPPLQJ DQG GHYHORSPHQW LQ &  &  -DYD DQG 1(7

ODQJXDJHV 6LPXODWLRQ 2SHUDWLQJ 6\VWHPV 6RIWZDUH (QJLQHHULQJ :LUHOHVV DQG 1HWZRUN ,QWHUQHW RI 7KLQJV %URDGFDVWLQJ ,PDJH 3URFHVVLQJ 0XOWLPHGLDDQG7HOHFRPPXQLFDWLRQV6\VWHPV

Pablo Minango. %6F LQ (OHFWURQLF (QJLQHHULQJ

IURP WKH 8QLYHUVLGDG 3ROLWHFQLFD 6DOHVLDQD 836 4XLWR (FXDGRU LQ  &XUUHQWO\ KH LV DQ 06F FDQGLGDWH E\ 'HSDUWPHQW RI &RPPXQLFDWLRQV )DFXOW\ RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ )((& DW 6WDWH 8QLYHUVLW\ RI &DPSLQDV – 81,&$03 +LV UHVHDUFK LQWHUHVWV LQFOXGH 'HHS /HDUQLQJ 0DFKLQH /HDUQLQJ 'LJLWDO ,PDJH 3URFHVVLQJZLWK0HGLFDOVLPDJHV

Yuzo Iano. %6F  06F  DQG 3K'

GHJUHHV  LQ (OHFWULFDO (QJ DW 81,&$03 %UD]LO +H KDV EHHQ ZRUNLQJ LQ WKH WHFKQRORJLFDO SURGXFWLRQILHOGZLWKSDWHQWJUDQWHGILOHG SDWHQW DSSOLFDWLRQV DQG  SURMHFWV FRPSOHWHG ZLWK UHVHDUFK DQG GHYHORSPHQW DJHQFLHV +H KDV VXSHUYLVHG  GRFWRUDOWKHVHV, 49 master’s dissertations, 74 XQGHU JUDGXDWH DQG  VFLHQWLILF LQLWLDWLRQ ZRUNV +H KDV participated in more than 100 master’s H[DPLQDWLRQ ERDUGV  GRFWRUDO GHJUHHV DXWKRU RI  ERRNV DQG PRUHWKDQSXEOLVKHGDUWLFOHV+HLV FXUUHQWO\ 3URIHVVRU DW 81,&$03 (GLWRULQ&KLHI RI WKH 6(7 ,QWHUQDWLRQDO -RXUQDO RI %URDGFDVW(QJLQHHULQJDQG*HQHUDO&KDLURIWKH%UD]LOLDQ6\PSRVLXPRQ 7HFKQRORJ\ %76\P +H KDV H[SHULHQFH LQ (OHFWULFDO (QJLQHHULQJ ZLWK NQRZOHGJHLQ7HOHFRPPXQLFDWLRQV(OHFWURQLFVDQG,QIRUPDWLRQ7HFKQRORJ\ PDLQO\ LQ WKH ILHOG RI DXGLRYLVXDO FRPPXQLFDWLRQV DQG PXOWLPHGLD

Gabriel Gomes de Oliveira. %6F LQ &LYLO

(QJLQHHULQJIURPWKH81,38QLYHUVLW\LQ+H LVSXUVXLQJDmaster’s GHJUHHDW81,&$03)DFXOW\ RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ )((& &RPPXQLFDWLRQV 'HSDUWPHQW '(&20 /DERUDWRU\ RI 9LVXDO &RPPXQLFDWLRQ /&9 &XUUHQWO\ KHUHVHDUFKHV DQG VWXGLHV WKH DUHD 6PDUW &LWLHV ,QWHUQHW RI 7KLQJV HQYLURQPHQWDO HQJLQHHULQJDQGF\EHU SK\VLFDO V\VWHPV

Ana Carolina Borges Monteiro. *UDGXDWHG LQ

%LRPHGLFLQH IURP &HQWUR 8QLYHUVLWDULR $PSDUHQVH 81,),$  &XUUHQWO\ VKH SXUVXHV D 3K' &DQGLGDWHE\ 'HSDUWPHQWRI&RPPXQLFDWLRQV '( &20 )DFXOW\ RI (OHFWULFDO DQG &RPSXWHU (QJLQHHULQJ )((& DW 6WDWH 8QLYHUVLW\ RI &DPSLQDV 81,&$03 DQG D UHVHDUFKHU DW WKH /DERUDWRU\ RI 9LVXDO &RPPXQLFDWLRQV /&9 6KH LV DOVR WKH 5HJLVWUDWLRQ &KDLU RI WKH %UD]LOLDQ 6\PSRVLXP RQ 7HFKQRORJ\ %76\P DQG KDV H[SHUWLVHLQWKHDUHDVRI&OLQLFDO$QDO\VLVDQGGLJLWDO LPDJH SURFHVVLQJ

WKURXJK 0$7/$% VRIWZDUH 6KH KDV SHUIRUPHG ZRUN UHVHDUFK H[SHULPHQWVSURMHFWV DQG LQWHUQVKLS LQ PXQLFLSDO KRVSLWDO DQG ZRUNV DV D UHYLHZHU

This open access article is distributed under a Creative Commons Attribution (CC-BY) license.

2019 SET - Brazilian Society of Television Engineering / ISSN (Print): 2446-9246/ISSN (Online): 2446-9432

http://www.set.org.br/ijbe/ doi: 10.18580/setijbe.2019.14 Web Link: http://dx.doi.org/10.18580/setijbe.2019.14

SET INTERNATIONAL JOURNAL OF BROADCAST ENGINEERING - SET IJBE V.5, 2019, Article 14, 7p.