Traffic Sign Recognition Raws li YOLOv3 Algorithm Part 3

3.3. Tsim Priori Frames Raws li K-Means Clustering Algorithm

Cov txheej txheem thauj tog rau nkoj tau siv nyob rau hauv YOLOv2, thiab tus naj npawb ntawm cov thauj tog rau nkoj tau nce mus rau cuaj hauv YOLOv3 los ua kom cov neeg sib tw tsim cov cheeb tsam zoo ib yam li cov ntawv sau npe tiag tiag thiab txhawb kev tshawb nrhiav lub network rov qab.

Muaj kev sib raug zoo ntawm cov cim cim thiab nco. Kev kos cim tuaj yeem pab peb tsim kom muaj kev ruaj khov, tsis tu ncua, thiab ua kom lub hauv paus nco, ua kom yooj yim rau kev nco txog cov ntaub ntawv ntau. Piv txwv li, thaum kawm ib hom lus, peb tuaj yeem siv cov cim cim los cim cov lus tshiab thiab cov cai sau ntawv. Thaum tshuaj xyuas keeb kwm, peb tuaj yeem siv cov cim cim los cim cov xwm txheej keeb kwm thiab sijhawm sijhawm. Ua li no, peb tuaj yeem ua kom paub paub daws teeb meem ntau dua thiab nkag siab.

Nyob rau tib lub sijhawm, kev kos cov thav duab kuj tuaj yeem txhawb nqa peb lub hlwb kev sib koom tes, yog li txhim kho peb lub cim xeeb. Vim tias peb lub cim xeeb yog raws li kev koom tes thiab kev sib txuas, los ntawm kev tsim cov cim cim, peb tuaj yeem txuas cov kev paub tshiab nrog cov kev paub uas twb muaj lawm, ua kom nco thiab nkag siab ntau dua.

Tib neeg lub peev xwm nco tau tuaj yeem cob qhia thiab txhim kho. Los ntawm kev xyaum tas li thiab kev siv cov txheej txheem nco xws li cov cim cim, peb tuaj yeem txhim kho peb lub cim xeeb thiab zoo dua nrog cov ntaub ntawv nyuaj thiab kev ua haujlwm hauv lub neej thiab kev ua haujlwm.

Nyob rau hauv luv luv, kos thav duab yog ib txoj kev nco zoo heev. Nws tuaj yeem pab peb nco qab cov ntaub ntawv tseem ceeb sai dua thiab raug. Nws tseem tuaj yeem txhawb peb lub peev xwm koom nrog thiab txhim kho peb lub cim xeeb. Cia peb nquag siv cov cim cim kom txuas ntxiv txhim kho peb cov kev txawj nco! Nws tuaj yeem pom tau tias peb yuav tsum txhim kho kev nco, thiab Cistanche deserticola tuaj yeem txhim kho kev nco zoo vim Cistanche deserticola yog cov khoom siv tshuaj suav tshuaj uas muaj ntau yam teebmeem, ib qho ntawm kev txhim kho kev nco. Kev ua tau zoo ntawm cov nqaij minced los ntawm ntau yam khoom xyaw uas nws muaj, nrog rau cov kua qaub, polysaccharides, flavonoids, thiab lwm yam. Cov khoom xyaw no tuaj yeem txhawb lub hlwb kev noj qab haus huv ntau txoj hauv kev.

Nyem Paub txhawm rau txhim kho lub cim xeeb luv luv

Nws tsis tsim nyog siv tus thawj thauj tog rau nkoj, txij li cov cim tsheb yog feem ntau me me thiab nruab nrab lub hom phiaj, nrog tsawg dua lub hom phiaj hauv TT100K dataset. Rau cov ntaub ntawv tshwj xeeb, xaiv qhov tsim nyog pib thauj tog rau nkoj txhim kho cov txiaj ntsig tshawb pom, ua kom lub network yooj yim rau kev kawm, thiab nce qhov kev tshawb pom ntawm lub thawv khi.

Kev khiav ntawm K-txhais tau tias pawg algorithm kom tau txais cov ntawv sib tw yog qhia hauv daim duab 7.

Nyob rau hauv TT100K dataset, qhov txhim kho YOLOv3 network qauv suav nrog qhov ntsuas qhov ntsuas qhov ntsuas, ua rau plaub qhov ntsuas thiab kaum ob anchors: (4, 5), (5, 6), (7, 7), (7, 13), (8, 8), (9, 10), (11, 12), (13, 14), (16, 17), (20, 22), (27, 29), thiab (41, 44).

4. Kev sim thiab tshuaj xyuas cov txiaj ntsig

4.1. Dataset thiab ntsuas ntsuas

Muaj ob peb qhov loj, nthuav tawm cov ntaub ntawv qhia kev tsheb khiav tsheb, feem ntau uas siv GTSDB, tab sis GTSDB tsis zoo ib yam li Suav cov cim tsheb. CTSDB, CCTSDB, thiab TT100K, thiab lwm yam, yog cov ntaub ntawv teev npe tsheb hauv Suav teb.

CCTSDB tau nthuav dav raws li CTSDB, thiab nws cov pawg tau muab faib ua cov cim ceeb toom, cov cim qhia, thiab cov cim txwv tsis pub, tsis muaj cov ncauj lus kom ntxaws ntawm cov paib tsheb.

Lub TT100K trafficsign collection tau tsim los ntawm kev sib koom tes ntawm Tencent thiab Tsinghua University. Nws muab cov kev faib ua kom tiav thiab kev txheeb xyuas cov paib tsheb, suav nrog ntau qhov xwm txheej huab cua thiab teeb pom kev zoo, thiab yog qhov tseeb dua rau cov xwm txheej tsav tsheb tiag tiag.

Yog li ntawd, TT100K cov ntaub ntawv teev npe tsheb tau siv nyob rau hauv daim ntawv no, thiab qee qhov ntawm cov paib tsheb thiab cov ntaub ntawv qhia tau qhia hauv daim duab 8.

TT100K dataset muaj 100,000 cov duab nrog kev daws teeb meem ntawm 2048 x 2048 pixels, txawm hais tias muaj cov duab kos npe tsis muaj tsheb thauj mus los, thiab qee pawg tsuas muaj ob peb cov duab lossis cov duab du, txo cov txiaj ntsig tshawb pom.

Yog li ntawd, daim ntawv no tau tshem tawm cov duab tsis muaj npe thiab sib npaug ntawm cov duab kos npe los ntawm cov ntaub ntawv thiab xaiv 45 pawg nrog cov cim tsheb loj, qhov twg 45 pawg kos npe yog: pn, pne, i5, pll, pl40, po, pl50, pl80 , io, pl60, p26, i4, pl00, pl30, il60, l5, i2, w57, p5, p10, ip, pl120, il80, p23, pr40.ph4. 5, w59, 12, p3, w55. pm 20, pl20, pg, pl70, pm55, il100, p27, w13, p19, ph4, ph5, wo, p6.pm30, thiab w32, thiab tus naj npawb ntawm txhua hom kev kos npe yog qhia hauv daim duab 9.

Daim duab 9 qhia tau hais tias txawm tias 45 pawg nrog ntau cov cim tsheb raug xaiv, tseem muaj qhov tsis txaus ntseeg ntawm cov ntaub ntawv ntawm txhua pawg, ua rau cov qauv kev twv ua ntej tsis zoo. Raws li qhov tshwm sim, raws li tau piav qhia hauv daim duab 10, qhov kev ua haujlwm no sib npaug thiab nthuav dav cov ntaub ntawv los ntawm kev siv tactics xws li xim dithering, Gaussian suab nrov, thiab cov duab tig los xyuas kom meej tias tus nqi ntawm txhua pawg yog sib npaug li qhov ua tau.

Mosaic mus kom ze nyeem plaub daim duab ntawm ib lub sij hawm, teev thiab hloov cov xim gamutof txhua daim duab, npaj lawv nyob rau hauv plaub cov lus qhia thiab tom qab ntawd stitches cov duab ua ke los tsim lub hom phiaj qhov tseeb thav duab.

Txoj kev txhim kho stitches plaub cov duab, sib npaug rau kev suav cov tsis muaj plaub daim duab nrog ib qho kev nkag. Qhov no tuaj yeem txo tus naj npawb ntawm cov duab rau batch input, txo qhov nyuaj ntawm kev cob qhia thiab kev cob qhia tus nqi, txhim kho kev cob qhia ceev, thiab ua kom muaj txiaj ntsig ntau ntawm cov qauv hauv cov ntaub ntawv, uas yog qhov zoo rau kev kawm.

Hauv daim ntawv no, cov kev ntsuas ntsuas ntawm COCO dataset, suav nrog mAPou - 050APs, APM, AP, thiab ob peb lwm yam kev ntsuas, tau siv los ntsuas qhov ua tau zoo ntawm tus qauv. Tshwj xeeb tshaj yog, feem ntau ntawm cov paib tsheb khiav hauv TT100K cov ntaub ntawv teev npe tsheb tau koom nrog cov hom phiaj me me, yog li kev saib xyuas tshwj xeeb yuav tsum tau them rau kev tshawb pom qhov tseeb ntawm cov hom phiaj me. Cov ntsiab lus tshwj xeeb ntawm cov ntsuas ntsuas ntsuas yog raws li hauv qab no:

AP: Thaj chaw hauv qab PR nkhaus, qhov twg PR yog qhov tseeb thiab rov qab los, raws li:

API {{0}}.50: Thaum IoU qhov pib raug teeb tsa rau 0.50, nws yog qhov nruab nrab ntawm txhua pawg ntawm AP hauv cov ntaub ntawv, uas yog qhov ntsuas ntsuas ntawm PASCAL VOC cov ntaub ntawv thiab sib raug. rau APIoU=0.50 hauv COCO kev ntsuam xyuas indexmAPloU= 0.50: Thaum loU qhov pib raug teeb tsa rau 0.50, nws yog qhov nruab nrab ntawm txhua pawg ntawm AP hauv dataset, uas yog qhov ntsuas ntsuas ntawm PASCAL VOC dataset thiab sib raug rau APloU=0.5 hauv COCO qhov ntsuas ntsuas.

APs: tus nqi nruab nrab ntawm mAP rau cov khoom me me: thaj tsam < 322, thiab loU=ntau yam (0.5, 1.00, 0.05) rau tag nrho o10 ua.

APm: cov khoom nruab nrab: 322 < thaj tsam < 962, thiab loU=ntau yam (0.5, 1.00, 0.05) txhais tau tias tus nqi ntawm mAP rau ib tag nrho ntawm 10 IoUs.

AP: tus nqi nruab nrab ntawm mAP rau cov khoom loj: cheeb tsam > 962, thiab loU=ntau yam (0.5, 1.00, 0.05 rau tag nrho ntawm 10 IoUs.

4.2. Kev Tshawb Fawb Kev Tshawb Fawb thiab Kev Tshawb Fawb

4.2.1. Txhim kho YOLOv3 Kev sim sib piv

Peb YOLOv3 tes hauj lwm nrog cov txheej txheem txhim kho tau muab piv thiab sim hauv qhov kev tshawb fawb no, siv TT100K tsheb kos npe cov ntaub ntawv thiab cov duab nkag uas yog 608 × 608 pixelsin loj. Daim duab 11 qhia txog daim ntawv qhia thiab AR ntawm M-YOLOv3 tau kawm ntawm TT100 dataset.

Cov txiaj ntsig tshawb pom rau ntau qhov ntau thiab tsawg ntawm cov hom phiaj tau pom nyob rau hauv daim duab 12 thiab Table 1. Ntawm lawv, YOLOv3-DK tau txais lub tswv yim ntawm kev txhim kho qhov kev poob haujlwm DioU poob thiab qhov rov ua haujlwm thauj tog rau nkoj; YOLOv3-SPP tau txais kev pom zoo ntawm qhov chaw fusion ntawm cov qauv pyramidpooling; YOLOv3-4l tau txais lub tswv yim ntawm kev ntxiv plaub qhov kev kwv yees featurelayer nrog 152 × 152 nplai; thiab M-YOLOv3 yog YOLOv3 network qauv siv tag nrho cov tswv yim txhim kho.

Daim duab 1 thiab daim duab 12 qhia tias qhov nruab nrab qhov tseeb ntawm qhov qub YOLOv3 yam tsis siv cov tswv yim yog 68.9%. Hauv qhov sib piv, daim duab qhia kev hloov kho YOLOv3 nrog txhua txoj hauv kev yog 77.3%, kev txhim kho ntawm 8.4% hauv kev tshawb pom.

Lub DioU poob functionand re-clustering thauj tog rau nkoj txheej txheem txhim kho kom paub tseeb tseeb los ntawm 1.3%; Txawm li cas los xij, kev txhim kho yog vim muaj kev poob haujlwm sai dua thaum kev cob qhia, uas ua rau lub hom phiaj lub thawv regression ruaj khov thiab txhim kho tus nqi rov qab. Ntau qhov kev txhim kho hauv mAP tau pom hauv YOLOv3, uas suav nrog SPP qauv thiab ua tiav 73.2%.

Cov qauv SPP ua ke nrog cov yam ntxwv hauv zos thiab thoob ntiaj teb, txhim kho daim ntawv qhia tshwj xeeb lub peev xwm los nthuav qhia nws tus kheej thiab ua kom pom tseeb qhov tseeb.Siv cov txheej txheem ntawm kev ntxiv plaub txheej txheej txheej txheej nrog 152 × 152 nplai, lawvAP kuj tseem muaj txiaj ntsig zoo dua.

Qhov tseeb ntawm cov phiaj xwm me me tau txhim kho los ntawm 10.5% thaum piv rau YOLOv3, uas ua rau siv tag nrho cov yam ntxwv ntiav hauv lub network rau kev kwv yees me me, ua rau muaj txiaj ntsig zoo nrhiav pom, tab sis ntawm tus nqi ntawm kev sib txuas ntxiv thiab kev ua haujlwm. . Qhov kev txhim kho zoo tshaj plaws yogM-YOLOv3, uas ua ke nrog peb txoj kev txhim kho thiab ua tiav ib qho mAP ntawm 77.3%, uas yog 8.4% siab dua li qhov qub YOLOv30 qhov nruab nrab qhov tseeb. Daim duab 13 qhia txog cov txiaj ntsig ntawm M-YOLOv3 ntawm TT100K.

4.2.2. Kev sib piv ntawm Kev Txhim Kho YOLOv3 Algorithm nrog Lwm Cov algorithms

M-YOLOv3 tau muab piv nrog rau ob peb lwm lub hom phiaj kev tshawb nrhiav algorithmsto ntxiv validate qhov kev paub txog kev txhim kho network, thiab cov txiaj ntsig tau tshwm sim hauv Table 2.

Table 2 qhia tau hais tias M-YOLOv3 muaj qhov siab tshaj mAP ntawm 77.3%, thiab SSD muaj qhov ua tau zoo tshaj plaws ntawm lub sijhawm, nrog rau FPS ntawm 42. Piv nrog rau tus thawj YOLOv3 algorithm, qhov nruab nrab precision txhais tau zoo heev, txawm hais tias lub sijhawm tiag tiag. kev ua tau zoo raug txo. Piv nrog rau ib-theem algorithm SSD, mAP tau txhim kho los ntawm 12%, tab sis tseem muaj qhov sib txawv hauv kev ua haujlwm tiag tiag. Piv nrog rau ob-theem lub hom phiaj detectionalgorithm Faster-RCNN, FPS tau txhim kho mus rau 22, thiab mAP kuj tau txhim kho los ntawm 1.7%, uas txhim kho qhov kev tshawb pom ceev, nrog rau kev tshawb pom qhov tseeb. Qhov kev sim tau qhia tias M-YOLOv3 ua tau zoo dua ntawm kev tshawb pom qhov tseeb thiab nrawm.

4.2.3. Txhim kho kev lees paub qhov cuam tshuam ntawm YOLOv3 ntawm cov cim tsheb hauv ib puag ncig tshwj xeeb

Vim muaj ntau yam, xws li lub teeb pom kev muaj zog, hmo ntuj, thiab cov xwm txheej tshwj xeeb ntawm cov cim kev sib tw tsheb, uas yuav cuam tshuam rau kev tshawb pom kev tsheb khiav thiab kev lees paub hauv lub ntiaj teb kev tsav tsheb tiag tiag, nws kuj tseem yuav tsum xav txog cov qauv kev lees paub ntawm cov paib tsheb hauv qhov chaw tshwj xeeb. Hauv qhov xwm txheej tshwj xeeb, tus qauv hloov khoYOLOv3 tau ua haujlwm kom paub txog cov paib tsheb, raws li tau pom hauv daim duab 13.

Hauv daim duab 14, qhov kev tshawb pom ntawm YOLOv3 yog piv nrog M-YOLOv3 hauv ib puag ncig tshwj xeeb. Raws li pom nyob rau hauv daim duab 14(b1,c1), YOLOv3 algorithm ua tsis tau tejyam mus ntes theobscured tsheb kos npe rau nyob rau hauv cov ntaub ntawv ntawm ib tug obscured tsheb ciav hlau kos npe rau, thaum lub txhim kho YOLOv3algorithm raug txheeb xyuas qhov tsis pom kev tsheb ciav hlau; raws li qhia hauv daim duab 14(b2,c2), YOLOv3 algorithm muaj teeb meem ntawm kev tshawb pom tsis tseeb thiab tsis pom kev rau kev paub txog kev tsheb khiav nyob rau hauv ib puag ncig ntawm lub teeb pom kev zoo, thaum qhov kev txhim kho YOLOv3algorithm tau lees paub tag nrho cov paib tsheb kom raug.

Kev txhim kho YOLOv3 algorithmin tau nce plaub qhov kev kwv yees qhov ntsuas rau cov hom phiaj me me, txhim kho cov txiaj ntsig ntawm cov hom phiaj me me, thaum YOLOv3 algorithm muaj teeb meem nrog kev kuaj tsis pom thiab tsis muaj kev ntseeg siab rau cov hom phiaj me, raws li pom hauv daim duab 14(b3,c3); nyob rau hauv qhov pom kev tsis pom kev, xws li thaum hmo ntuj, qhov hloov kho YOLOv3 algorithm pom cov paib tsheb, raws li tau piav qhia hauv daim duab 14(b4,c4); Txawm li cas los xij, YOLOv3 txoj kev tsis pom lub hom phiaj. Raws li qhov tshwm sim, nyob rau hauv cov xwm txheej tshwj xeeb, qhov hloov kho tshiab YOLOv3 algorithm tseem tau txais txiaj ntsig zoo dua kev tshawb pom.

5. Cov lus xaus

Kev kuaj pom kev tsheb khiav thiab kev lees paub lub network raws li kev hloov kho YOLOv3 tau tawm tswv yim hauv qhov kev tshawb fawb no, los daws cov teeb meem ntawm cov hom phiaj me me uas tsis yooj yim rau kev txheeb xyuas thiab tsis tshua pom qhov tseeb hauv kev tshawb nrhiav thiab kev txheeb xyuas cov haujlwm.

Tus tshiab spatial pyramidal pooling qauv enabled fusion ntawm lub zos thiab lub ntiaj teb no cov yam ntxwv nyob rau hauv txoj kev tshawb no, nrog rau ntxiv rau lub thib plaub feature kwv yees teev rau me me lub hom phiaj los txhim kho cov txiaj ntsig ntawm cov phiaj xwm me. Txhawm rau ua kom lub hom phiaj thav duab regression ruaj khov, DioU poob tau siv, uas muaj kev sib koom ua ke sai dua thiab ua tau zoo dua nrog cov phiaj xwm rov qab.

Kev tshawb pom lub network qhov raug tau raug kho kom zoo dua los ntawm kev puas tsuaj rau lub sijhawm tiag tiag network tsawg li sai tau. MAP nce 8.4 ntsiab lus. Qhov hloov kho YOLOv3 algorithm txhim kho lub network txoj kev nyuaj thiab txo qis qhov kev tshawb nrhiav ceev. Txawm li cas los xij, kev tshawb nrhiav lub sijhawm tiag tiag tseem yog txoj hauv kev ntev; yog li ntawd, thaj chaw tshawb fawb tom ntej no yuav txhawb kev tshawb nrhiav ceev kom ua tiav cov txiaj ntsig ntawm kev tshawb nrhiav lub sijhawm.

Sau Cov Kev Pabcuam: Cov txheej txheem thiab kev sau ntawv-kev npaj ua ntej, AL thiab CG; formalanalysis thiab kev tshawb nrhiav, YS; data curation, NX; cov peev txheej, AL; validation, WH Txhua tus neeg sau ntawv tau nyeem thiab pom zoo rau cov ntawv luam tawm ntawm cov ntawv sau.

Kev Pab Nyiaj: Qhov project no tau txais kev txhawb nqa los ntawm Shandong Provincial Higher Educational Youth Innovation Science thiab Technology Program (Grant No.2019KJB019), Shandong Provincial NaturalScience Foundation ntawm Tuam Tshoj (Grant No. ZR2021MF131, ZR2015EL019, thiab ZR2020ME126), thiab lub Foundation Tuam Tshoj (Grant No. 61601265 thiab 51505258). Qhov project no tau txais nyiaj los ntawm Tuam Tshoj Postdoctoral Science Foundation (Grant No. 2021M701405), qhib Project of State Key Laboratory of Mechanical Behavior and System Safety of Traffic Engineering Structures, Tuam Tshoj (Grant No. 1903), Open Project of Hebei Traffic Safety and Control Key Laboratory, Tuam Tshoj (Grant No. JTKY2019002), thiab Major Science thiab Technology Innovation Project hauv ShandongProvince (Grant No. 2022CXGC020706).

Institutional Review Board Statement: Tsis siv tau.

Cov Lus Qhia Txog Kev Pom Zoo: Tsis siv tau.

Cov Lus Qhia Muaj Cov Ntaub Ntawv: Tsis siv tau.

Kev lees paub: Peb ua tsaug rau txhua tus kws sau ntawv rau lawv qhov kev koom tes rau kev sau tsab xov xwm no.

Kev tsis sib haum xeeb ntawm kev txaus siab: Cov neeg sau ntawv tshaj tawm tsis muaj teeb meem ntawm kev txaus siab.

Cov ntaub ntawv

1. De la Escalera, A.; Armingol, JM; Mata, M. Traffic kos npe thiab tsom xam rau cov tsheb ntse. Duab Vis. Tshuab xam zauv. 2003, 21, 247–258. [CrossRef]

2. Saadna, Y.; Behloul, A. Cov ntsiab lus ntawm kev kuaj pom kev tsheb khiav thiab kev faib tawm. Int. J. Multimed. Cov ntaub ntawv. Retr. 2017, 6, 193–210. [CrossRef]

3. Boumediene, M.; Kub, C.; Basset, M.; Ouamri, A. Daim duab peb sab kev tsheb ciav hlau kuaj pom raws li RSLD algorithm. Mach. Vis. 2013, 24, 1721–1732. [CrossRef]

4. Maldonado-Bascón, S.; Lafuente-Arroyo, S.; Gil-Jimenez, P.; Gomez-Moreno, H.; Lopez-Ferreras, F. Txoj kev-kos npe tshawb pom thiab paub raws li kev txhawb nqa vector tshuab. IEEE Trans. Intell. Transp. Syst. 2007, 8, 264–278. [CrossRef]

5. Bahlmann, C.; Zhu, Y.; Ramesh, V.; Pellkofer, M.; ib. Koehler, T. A system rau kev kuaj pom kev tsheb khiav, taug qab, thiab paub txog siv xim, duab, thiab cov lus qhia. Hauv Kev Ua Haujlwm ntawm IEEE Kev Ua Haujlwm. Intelligent Vehicles Symposium, 2005, Las Vegas, NV, USA, 6–8 Lub Rau Hli 2005; ib., 255–260.

6. Ren, S.; Nws, K.; Girshick, R.; Sun, J. Faster R-CNN: Mus rau Lub Sijhawm Kev Tshawb Fawb Txog Lub Sijhawm Nrog Lub Zej Zog Kev Pom Zoo Networks.Adv. Cov ntaub ntawv Neural. Txheej txheem. Syst. 2015, 28, 91–99. [CrossRef] [PubMed]

7. Li, W. Anguelov, D.; Erhan, D.; Szegedy, C.; Raub, S.; Fu, C.-Y.; Berg, AC SSD: Tib Txhaum MultiBox Detector. Hauv EuropeanConference on Computer Vision; Springer: Cham, Switzerland, 2016; ib., 21–37.

8. Redmon, J.; Divvala, S.; Girshick, R.; Farhadi, A. Koj tsuas saib ib zaug: Kev sib koom ua ke, tshawb pom cov khoom hauv lub sijhawm. Hauv Kev Ua Haujlwm ntawm IEEE Lub Rooj Sib Tham ntawm Computer Vision thiab Pattern Recognition, Las Vegas, NV, USA, 27–30 Lub Rau Hli 2016; IEEE: Piscataway, NJ, USA, 2016; Ib., 779–788.

9. Wang, Z.; Guo, H. Kev tshawb fawb txog kev kuaj pom kev tsheb khiav raws li kev sib txuas lus neural network. Hauv Kev Ua Haujlwm ntawm 12th InternationalSymposium ntawm Visual Information Communication thiab Interaction, Shanghai, Tuam Tshoj, 20-22 Cuaj hlis 2019; pp. 1–5.

10. Han, C.; Gao, G.; Zhang, Y. Real-time me me tsheb ciav hlau kos npe nrog kho sai-RCNN. Multimedia. Cov cuab yeej Appl. Xyoo 2019, 78, 13263–13278. [CrossRef]

11. Zhang, J.; Huang, M. Jin, X.; Li, X. Ib lub sij hawm ntawm lub sij hawm ntawm Suav tsheb ciav hlau nrhiav pom algorithm raws li hloov kho YOLOv2. Algorithms2017, 10, 127. [CrossRef]

12. Zhou, Z.; Liang, D.; Zhang, S.; Huang, X.; Li, B.; Hu, S. Traffic-signal detection and classification in the wild. Hauv Kev Ua Haujlwm ntawm IEEE Conference on Computer Vision and Pattern Recognition 2016, Las Vegas, NV, USA, 27–30 June 2016; Ib., 2110–2118.

For more information:1950477648nn@gmail.com