I-Convolutional neural network (CNN): Ifundisa amakhompyutha ukuthi angabona kanjani

• About the Author:
• Igama lomlobi

  I-Quantumrun Foresight
• December 1, 2023

Isifinyezo sokuqonda

I-Convolutional Neural Networks (CNNs) ibalulekile ekuhlukaniseni izithombe nasekubonweni kwekhompyutha, aguqula indlela imishini ehlonza futhi iqonde ngayo idatha ebonakalayo. Balingisa umbono womuntu, bacubungule izithombe ngokusebenzisa izendlalelo zokuhlanganisa, ukuhlanganisa, nezixhunywe ngokugcwele ukuze kukhishwe isici nokuhlaziya. Ama-CNN anezinhlelo zokusebenza ezihlukene, okuhlanganisa ukuthengiswa kwezincomo zomkhiqizo, izimoto zokuthuthukiswa kokuphepha, ukunakekelwa kwezempilo kokutholwa kwesimila, nobuchwepheshe bokubonwa kobuso. Ukusetshenziswa kwazo kudlulela ekuhlaziyweni kwamadokhumenti, izakhi zofuzo, nokuhlaziya izithombe zesathelayithi. Ngokuhlanganiswa kwawo okukhulayo emikhakheni eyahlukene, ama-CNN aphakamisa ukukhathazeka kokuziphatha, ikakhulukazi mayelana nobuchwepheshe bokubona ubuso nobumfihlo bedatha, egqamisa isidingo sokucatshangelwa ngokucophelela kokuthunyelwa kwawo.

I-Convolutional neural network (CNN) umongo

Ama-CNN ayimodeli yokufunda ejulile egqugquzelwe indlela abantu nezilwane abasebenzisa ngayo amehlo abo ukuhlonza izinto. Amakhompyutha abanalo leli khono; uma “bebuka” isithombe, sihunyushwa sibe amadijithi. Ngakho, ama-CNN ahlukaniswa kwamanye amanethiwekhi e-neural ngamakhono awo athuthukile okuhlaziya idatha yesithombe nesignali yomsindo. Aklanyelwe ukufunda ngokuzenzakalelayo nangokuguquguqukayo ukuhlelwa kwendawo yezici, kusukela kumaphethini asezingeni eliphansi ukuya ezingeni eliphezulu. Ama-CNN angasiza ikhompuyutha ekutholeni amehlo “omuntu” futhi ayinikeze ukubona kwekhompuyutha, eyivumela ukuthi imunce wonke amaphikseli nezinombolo ezibonayo futhi isize ekuqashelweni kwesithombe nasekuhlukaniseni ngezigaba. 

I-ConvNets isebenzisa imisebenzi yokuvula kumephu yesici ukuze isize umshini ekunqumeni lokho okubonayo. Le nqubo inikwe amandla izendlalelo ezintathu eziyinhloko: i-convolutional, i-pooling, kanye nezendlalelo ezixhunywe ngokugcwele. Okubili kokuqala (i-convolutional kanye nokuhlanganisa) kwenza ukukhishwa kwedatha, kuyilapho isendlalelo esixhumeke ngokugcwele sikhiqiza okukhiphayo, njengokuhlukanisa. Imephu yesici idluliswa isuka kungqimba iye kungqimba kuze kube yilapho ikhompuyutha isibona sonke isithombe. Ama-CNN anikezwa ulwazi oluningi ngangokunokwenzeka ukuze kutholwe izici ezihlukahlukene. Ngokutshela amakhompyutha ukuthi abheke imiphetho nemigqa, le mishini ifunda ukuhlonza izithombe ngokushesha nangokunembile ngamanani angenakwenzeka kubantu.

Umthelela ophazamisayo

Nakuba ama-CNN evamise ukusetshenziselwa ukubonwa kwesithombe nemisebenzi yokuhlukanisa, angaphinde asetshenziselwe ukutholwa nokuhlukaniswa. Isibonelo, ekuthengiseni, ama-CNN angasesha ngokubonakalayo ukuze abone futhi ancome izinto ezihambisana nekhabethe elikhona. Kwezezimoto, lawa manethiwekhi angaqaphela izinguquko ezimweni zomgwaqo njengokutholwa komugqa womzila ukuze kuthuthukiswe ukuphepha. Kwezokunakekelwa kwempilo, ama-CNN asetshenziselwa ukuhlonza kangcono izimila ezinomdlavuza ngokuhlukanisa la maseli alimele ezithweni ezinempilo eziwazungezile. Ngaleso sikhathi, ama-CNN athuthukise ubuchwepheshe bokubona ubuso, okuvumela izinkundla zokuxhumana ukuthi zibone abantu ezithombeni futhi zinikeze izincomo zokumaka. (Kodwa-ke, i-Facebook inqume ukumisa lesi sici ngo-2021, icaphuna ukukhathazeka okukhulayo kwezimiso zokuziphatha kanye nezinqubomgomo zokulawula ezingacacile ekusebenziseni lobu buchwepheshe). 

Ukuhlaziywa kwedokhumenti nakho kungathuthuka ngama-CNN. Bangakwazi ukuqinisekisa umsebenzi obhalwe ngesandla, bawuqhathanise nesizindalwazi sokuqukethwe okubhalwe ngesandla, bahumushe amagama, nokuningi. Bangase baskene amaphepha abhalwe ngesandla abalulekile ebhange kanye nezezimali noma ukuhlukaniswa kwemibhalo eminyuziyamu. Kuzakhi zofuzo, lawa manethiwekhi angahlola amasiko amaseli ocwaningo lwezifo ngokuhlola izithombe nokwenza imephu nokuhlaziya okubikezelayo ukuze kusizwe ochwepheshe bezokwelapha ekuthuthukiseni izindlela zokwelapha ezingase zibe khona. Okokugcina, izendlalelo ze-convolutional zingasiza ekuhlukaniseni izithombe zesathelayithi futhi zihlonze ngokushesha ukuthi ziyini, okungasiza ekuhloleni indawo.

Izicelo ze-convolutional neural network (CNN)

Ezinye izinhlelo zokusebenza ze-convolutional neural network (CNN) zingafaka: 

• Ukwanda kokusetshenziswa ekuxilongeni kokunakekelwa kwezempilo, okuhlanganisa i-radiology, ama-x-ray, nezifo zofuzo.
• Ukusetshenziswa kwama-CNN ukuze kuhlukaniswe izithombe ezisakazwayo ezisuka emkhathini naseziteshini, namarover enyanga. Ama-ejensi wezokuvikela angasebenzisa ama-CNN kumasathelayithi agadayo nama-drones ukuze kuhlonzwe ukuzimela kanye nokuhlolwa kokuphepha noma izinsongo zezempi.
• Ubuchwepheshe obuthuthukisiwe bokubona uhlamvu lwemibhalo ebhalwe ngesandla nokubonwa kwesithombe.
• Izinhlelo zokusebenza ezithuthukisiwe zokuhlunga ngerobhothi ezindaweni zokugcina izimpahla nasezindaweni zokugaywa kabusha.
• Ukusetshenziswa kwazo ekuhlukaniseni izigebengu kanye nabantu abanentshisekelo kubo kumakhamera asemadolobheni noma angaphakathi okugada. Nokho, le ndlela kungaba ngaphansi kokuchema.
• Izinkampani eziningi ezibuzwayo mayelana nokusebenzisa kwazo ubuchwepheshe bokubona ubuso, okuhlanganisa ukuthi ziqoqa kanjani futhi zisebenzisa kanjani idatha.

Imibuzo ongaphawula ngayo

• Ucabanga ukuthi iyiphi enye indlela ama-CNN angathuthukisa ngayo umbono wekhompyutha nendlela esiyisebenzisa ngayo nsuku zonke?
• Yiziphi ezinye izinzuzo ezingaba khona zokuqashelwa okungcono kwesithombe nokuhlukaniswa?