Wakambofadzwa here nekugona kwekamera ye smartphone yako kuona zviso muboka foto?
Zvimwe wakashamiswa nemafambisirwo anoitwa mota dzinozvityaira zvisina mutsetse, dzichiziva vanofamba netsoka nedzimwe mota nemazvo.
Izvi zvinoita semashura akaitwa zvinogoneka nekuonekwa kwechinhu, chidzidzo chinonakidza chekutsvagisa. Zvichingotaura, kutariswa kwechinhu ndiko kuzivikanwa uye kuiswa kwezvinhu mukati memifananidzo kana mavhidhiyo.
Ndiyo teknolojia inobvumira makombiyuta "kuona" uye kunzwisisa nyika yakavapoteredza.
Asi iyi nzira inoshamisa inoshanda sei? Tiri kuzviona izvozvo kudzidza kwakadzama kune yakashandura nzvimbo yechiratidzo chechinhu. Iri kuvhura nzira yehuwandu hwemaapplication ane pesvedzero yakananga pahupenyu hwedu hwezuva nezuva.
Mune ino positi, tichapfuura nemunzvimbo inonakidza yezvakadzama-yakavakirwa chidzidzo chekuzivikanwa kwechinhu, kudzidza kuti zvine mukana wekugadzirisa patsva nzira yatinodyidzana nayo tekinoroji.
Chii Chaizvo Chinonzi Object Detection?
Imwe yeiyo yakakosha kuona komputa Tasks (tasks) kuona zvinhu, izvo zvinosanganisira kutsvaga nekutsvaga zvinhu zvakasiyana mumufananidzo kana vhidhiyo.
Kana tichienzanisa nemhando yemifananidzo, panotemerwa kirasi yechinhu chimwe nechimwe, kuona kwechinhu kunoenda nhanho imwe chete nekusaziva huvepo hwechinhu chimwe nechimwe asiwo kudhirowa mabhokisi ekusungirira kutenderedza chimwe nechimwe.
Nekuda kweizvozvo, isu tinokwanisa panguva imwe chete kuziva mhando dzezvinhu zvinofarirwa uye kunyatsozviwana.
Iko kugona kuona zvinhu kwakakosha kune akawanda maapplication, kusanganisira kuzvidzivirira kutyaira, kutariswa, kuzivikanwa kwechiso, uye mifananidzo yezvokurapa.
Kubata dambudziko iri rakaoma nekukasira uye kuita chaiko-nguva, nzira dzakadzama dzekudzidza dzakashandura kuona chinhu.
Kudzidza kwakadzama kwave kubuda senzira ine simba yekukunda matambudziko aya, kushandura indasitiri yekuzivikanwa kwechinhu.
Mhuri yeR-CNN uye YOLO mhuri mbiri dzinozivikanwa dzemuenzaniso mhuri mukuzivikanwa kwechinhu chinozoongororwa muchinyorwa chino.
R-CNN Mhuri: Kuona Chinhu Chekupayona
Tsvagiridzo yekutanga yekucherechedza chinhu yakapupurira kufambira mberi kwakakura nekuda kweR-CNN mhuri, iyo inosanganisira R-CNN, Fast R-CNN, uye Inokurumidza R-CNN.
Iine matatu-module yekuvaka, R-CNN matunhu akakurudzira akashandisa CNN kuburitsa maficha, uye akaiswa zvinhu achishandisa mutsara maSVM.
R-CNN yaive yechokwadi, kunyangwe zvakatora nguva nekuti dunhu remumiriri raidiwa. Izvi zvakabatwa neFast R-CNN, iyo yakawedzera kugona nekubatanidza mamodule ese kuita modhi imwechete.
Nekuwedzera iyo Region Proposal Network (RPN) iyo yakagadzira uye yakavandudza matunhu zvikumbiro panguva yekudzidziswa, nekukurumidza R-CNN yakawedzera kuita basa uye yakawana ingangoita chaiyo-nguva yekuzivikanwa kwechinhu.
Kubva kuR-CNN kuenda kuFaster R-CNN
Mhuri yeR-CNN, inomiririra "Region-Based Convolutional Neural Networks," yakatanga kufambira mberi mukuonekwa kwechinhu.
Mhuri iyi inosanganisira R-CNN, Fast R-CNN, uye Faster R-CNN, ayo ese akagadzirirwa kubata chinhu chenzvimbo uye mabasa ekuzivikanwa.
Iyo yekutanga R-CNN, yakaunzwa muna 2014, yakaratidza kushandiswa kwakabudirira kweiyo convolutional neural network yekuonekwa kwechinhu uye kuiswa munzvimbo.
Zvakatora zano rematanho matatu raisanganisira zano redunhu, kudhirowa neCNN, uye kupatsanurwa kwechinhu chine mutsara Wekutsigira Vector Machine (SVM) classifiers.
Kutevera kutangwa kweFast R-CNN muna 2015, matambudziko ekumhanya akagadziriswa nekubatanidza chikumbiro chedunhu uye kuiswa mumhando imwe chete, kudzikisa zvakanyanya kudzidziswa uye nguva yekufungidzira.
Nekukurumidza R-CNN, yakaburitswa muna 2016, yakavandudza kukurumidza uye kurongeka nekubatanidza Region Proposal Network (RPN) panguva yekudzidziswa kukurumidza kukurudzira nekudzokorora nzvimbo.
Nekuda kweizvozvo, Kurumidza R-CNN yakazvimisikidza seimwe yeanotungamira maalgorithms emabasa ekuona chinhu.
Kubatanidzwa kwevagadziri veSVM kwakakosha pakubudirira kwemhuri yeR-CNN, kushandura nzvimbo yekuona komputa uye kuisa nzira yezvichaitwa mune ramangwana mukudzidza kwakadzama-kwakavakirwa chinhu kuona.
Strengths:
- High localization chinhu kuona chokwadi.
- Kurongeka uye kushanda nesimba kwakaenzana nedhizaini yakabatana yekukurumidza R-CNN.
Kushaya simba:
- Inference neR-CNN uye Fast R-CNN inogona kuve yakaoma.
- Kuti ikurumidze R-CNN kuti ishande nepamusoro payo, zvakawanda zvedunhu zvikumbiro zvingangove zvichidikanwa.
YOLO Mhuri: Chinhu Kuonekwa mune Chaiyo-Nguva
Mhuri yeYOLO, yakavakirwa pachirevo che "Iwe Chete Chete" inosimbisa kucherechedzwa kwechinhu-chaiyo-nguva uchibayira chaiko.
Yekutanga YOLO modhi yaive neneural network imwechete yaifanotaura zvakananga mabhokisi ekusungirira uye mavara ekirasi.
Kunyangwe paine kushomeka kwekufanotaura, YOLO inogona kushanda nekumhanya kunosvika 155 mafuremu pasekondi. YOLOv2, inozivikanwawo seYOLO9000, yakagadzirisa kumwe kukanganisa kwemodhiyo yekutanga nekufanotaura zviuru zvipfumbamwe zvezvinhu uye kusanganisira mabhokisi eanchor ekufanotaura kwakasimba.
YOLOv3 yakatowedzera kunatsiridza, iine yakawedzera maficha detector network.
Kushanda Kwemukati kweYOLO Mhuri
Mamodheru ekuzivikanwa kwechinhu mumhuri yeYOLO (Iwe Chete Chete) abuda sebudiriro inocherechedzwa muchiratidzo chekombuta.
YOLO, iyo yakaunzwa muna 2015, inoisa pamberi nekukurumidza uye-chaiyo-nguva-chaiyo chinhu kuzivikanwa nekufungidzira zvakananga mabhokisi ekusungirira uye makirasi mavara.
Kunyangwe kumwe kurongeka kwakabairwa, inoongorora mafoto munguva-chaiyo, ichiita kuti ive yakakosha kune-yakakosha nguva yekushandisa.
YOLOv2 yakabatanidza mabhokisi eanchor ekubata nezvikero zvezvinhu zvakasiyana uye akadzidziswa pane akawanda dataset kutarisira pamusoro pe9,000 chinhu makirasi.
Muna 2018, YOLOv3 yakasimbisa mhuri zvakatowedzera neyakadzama detector network, ichiwedzera huchokwadi pasina kupira kuita.
Mhuri yeYOLO inofanotaura mabhokisi anosungirirwa, zvingangoitika zvekirasi, uye zvibodzwa zvekupokana nekupatsanura mufananidzo kuita grid. Inonyatso kusanganisa kumhanya uye kurongeka, ichiita kuti igone kushandiswa mukati autonomous cars, kutariswa, hutano, nezvimwe.
Iyo YOLO yakatevedzana yakashandura chiziviso chechinhu nekupa chaiyo-nguva mhinduro pasina kupa kwakakosha chokwadi.
Kubva paYOLO kuenda kuYOLOv2 neYOLOv3, mhuri iyi yaita budiriro huru mukuvandudza kuzivikanwa kwechinhu mumaindasitiri, kumisa chiyero chemazuva ano kudzidza kwakadzama-kwakavakirwa masisitimu ekuona zvinhu.
Strengths:
- Kuongorora zvinhu munguva-chaiyo pamitengo yakakwira furemu.
- Kugadzikana mukusungirira kufungidzira kwebhokisi kunounzwa muYOLOv2 uye YOLOv3.
Kushaya simba:
- YOLO modhi dzinogona kusiya huchokwadi mukutsinhana nekumhanya.
Muenzaniso Kuenzanisa Kwemhuri: Kururama vs
Kana mhuri dzeR-CNN neYOLO dzichifananidzwa, zviri pachena kuti kunyatsoita uye kushanda zvakanaka kwakakosha kutengeserana. Mhando dzemhuri dzeR-CNN dzinokunda mukurongeka asi dzinononoka panguva yekufungidzira nekuda kwekuvaka kwavo kwematatu-module.
Mhuri yeYOLO, kune rumwe rutivi, inoisa pamberi-chaiyo-nguva kuita, ichipa yakatanhamara kumhanya uku ichirasika imwe chaiyo. Sarudzo pakati peidzi mhuri dzemhando inotarwa nezvinodiwa chaizvo zvekushandisa.
Mhando dzemhuri dzeR-CNN dzinogona kuve dzakanakira mabasa anoda kunyatsojeka, nepo YOLO mhando dzemhuri dzakakodzera kushandiswa-chaiyo-nguva.
Beyond Object Recognition: Real-World Applications
Kupfuura mabasa akajairwa ekucherechedzwa kwechinhu, kudzidza kwakadzama-kwakavakirwa chinhu kuona kwakawana kuwanda kwekushandisa.
Kuchinjika kwayo uye kurongeka kwakagadzira mikana mitsva muzvikamu zvakasiyana, kugadzirisa matambudziko akaomarara uye kushandura mabhizinesi.
Autonomous Vehicles: Kuisa Mwero weKutyaira Kwakachengeteka
Kuonekwa kwechinhu kwakakosha mumotokari dzinozvimiririra pakuvimbisa kufamba kwakachengeteka uye kwakavimbika.
Mienzaniso yekudzidza kwakadzama kupa ruzivo rwakakosha kune anozvimirira ekutyaira masisitimu nekuziva nekuisa munharaunda vanofamba netsoka, vanochovha mabhasikoro, dzimwe mota, uye njodzi dzinogona kuitika mumigwagwa.
Aya mamodheru anoita kuti mota dzitore sarudzo dzenguva-chaiyo uye kudzivirira kudhumhana, zvichitiswededza pedyo neramangwana umo motokari dzinozvityaira dzinogara pamwe chete nevatyairi vevanhu.
Kuwedzera Kubudirira uye Chengetedzo muRetail Indasitiri
Bhizinesi rekutengesa rakambundira zvakadzika kudzidza-kwakavakirwa chinhu kuona kuti ivandudze zvakanyanya mashandiro ayo.
Kuona zvinhu zvinobatsira mukuzivikanwa uye kutsvaga zvigadzirwa pamasherufu ezvitoro, zvichibvumira kuti pave nekuwedzera kushanda uye kudzikiswa kwemamiriro ekunze.
Zvakare, masisitimu ekutarisa akashongedzerwa nealgorithms yekuona chinhu anobatsira mukudzivirira kubiwa uye kuchengetedza kwekuchengetedza chitoro.
Medical Imaging Advancement in Healthcare
Kuona kwakadzama-kwakavakirwa chinhu chave chishandiso chakakosha mukufungidzira kwekurapa muchikamu chehutano.
Inobatsira varapi vehutano mukuona zvisirizvo muX-rays, MRI scans, uye mimwe mifananidzo yekurapa, yakaita sekenza kana kukanganisa.
Kuzivikanwa kwechinhu kunobatsira mukutanga kuongororwa uye kuronga kurapwa nekuona uye kuratidza nzvimbo dzakanangana dzekunetsekana.
Kuvandudza Kuchengetedzeka Kuburikidza Nekuchengetedza uye Kuongorora
Kuonekwa kwechinhu kunogona kubatsira zvakanyanya mukuchengetedza uye kuongorora maapplication.
Kudzika kudzidza algorithms batsira mhomho yevarindi, kuona maitiro ekufungira, uye kuona njodzi dzinogona kuitika munzvimbo dzeveruzhinji, nhandare dzendege, nenzvimbo dzekufambisa.
Aya masisitimu anogona kunyevera vashandi vekuchengetedza munguva-chaiyo nekuramba vachiongorora mafidhiyo evhidhiyo, kudzivirira kutyorwa kwekuchengetedza, uye nekuona kuchengetedzwa kweveruzhinji.
Zvipingamupinyi Zvazvino uye Tarisiro Yeramangwana
Pasinei nekufambira mberi kwakakosha mukudzidza kwakadzama-kwakavakirwa kuona chinhu, matambudziko anoramba aripo. Kuvanzika kwedata chinhu chakakomba, sezvo kuwonekwa kwechinhu kazhinji kunosanganisira kutonga ruzivo rwakadzama.
Rimwe dambudziko guru nderekuona kusimba pakurwisa vavengi.
Vatsvagiri vachiri kutsvaga nzira dzekuwedzera modhi generalization uye kududzira.
Nekutsvagisa kunoenderera mberi kwakanangana nekuzivikanwa kwezvinhu zvakawanda, vhidhiyo yekutevera chinhu, uye chaiyo-nguva 3D kuzivikanwa kwechinhu, ramangwana rinoita serakajeka.
Tinofanira kutarisira mhinduro dzakanyatsojeka uye dzinoshanda munguva pfupi sezvo mhando dzekudzidza dzakadzama dzichiramba dzichikura.
mhedziso
Kudzidza kwakadzama kwakasandura kuona kwechinhu, kuchiunza nguva yekunyatsojeka uye kugona. Mhuri dzeR-CNN neYOLO dzakatamba mabasa akakosha, rimwe nerimwe riine hunyanzvi hwakasiyana kune mamwe maapplication.
Kuziva kwakadzama-kwakavakirwa chinhu kushandura zvikamu nekuvandudza chengetedzo nekubudirira, kubva kumotokari dzinozvimiririra kuenda kune hutano.
Ramangwana rekuonekwa kwechinhu rinoratidzika kujeka kupfuura nakare kose sezvo tsvakiridzo inofambira mberi, kugadzirisa matambudziko nekuongorora nzvimbo itsva.
Tiri kuona kuzvarwa kwezera idzva mukuona kwekombuta sezvatinombundikira simba rekudzidza kwakadzama, nekuonekwa kwechinhu kunotungamirira nzira.
Leave a Reply