Iyunithi yokucubungula isithombe yekhamera yokuthwebula eshisayo

Esimisweni esiyinkimbinkimbi sesithombe esishisayo ikhamera, iyunithi yokucubungula isebenza njengengcweti enekhono, iqopha ngokucophelela okukhishwayo kwamasignali edijithali ngeyunithi yokucubungula isignali, iwaguqulele ezithombeni ezishisayo ezinembile, ezicacile, nezinothe ngezinga lokushisa, usekelo olubalulekile ekusetshenzisweni ngempumelelo kobuchwepheshe bezithombe ezishisayo emikhakheni ehlukahlukene.

Nakuba amasiginali edijithali asuka kuyunithi yokucubungula isignali ekhishwe umsindo kuqala, akhuliswa, futhi aqedwa ukuguqulwa kwe-analog kuya kudijithali, izithombe zisengaba nezinkinga njengokungafani okuphansi nemininingwane efiphaziwe. Iyunithi yokucubungula izithombe isebenzisa amasu okuthuthukisa isithombe ukuthuthukisa ikhwalithi yesithombe. Ngokulingana kwe-histogram, ububanzi bezinga elimpunga lesithombe liyanwetshwa, okuvumela imininingwane yokugqama nokumnyama esithombeni ukuthi yethulwe ngokucace kakhudlwana. Isibonelo, ekuhlolweni kwesithombe esishisayo semishini yezimboni, umehluko ocashile ekushiseni okungaphezulu kwendawo yokusebenza, okunzima ukuwenza ekuqaleni, ungavezwa ngokucacile ezindaweni ezihlukene zokushisa ngemuva kokucutshungulwa kokulinganisa kwe-histogram, okwenza kube lula ukuthi abasebenzi bathole ngokushesha amaphuzu angaba khona amaphutha. Kukhona futhi ama-algorithms okuhlunga okutholwa konqenqema okufana no-opharetha we-Laplacian nesihlungi se-Gaussian, esingaphinda sigqamise amakhontosi nemininingwane yezinto, kuthuthukise ukufundeka kwesithombe.

Ukusabela kwephikseli ngayinye yenzwa ye-infrared emisebeni ye-infrared ayihambisani ngokuphelele, okungaholela ezinkingeni ezifana nokukhanya okungalingani kanye nomsindo owenziwe ezithombeni. Iyunithi yokucubungula isithombe izokwenza ukulungisa okungafanayo ukuze kuqedwe lawa maphutha. Ngokusekelwe endleleni yokulungisa enamaphuzu amabili, esimweni semithombo yereferensi yokushisa okuphansi eyaziwayo, impendulo yenzwa kumthombo wereferensi ikalwa ukuze kusungulwe imodeli yokulungisa, futhi isignali yokukhiphayo yephikseli ngayinye iyalungiswa ukuze kuqinisekiswe ukuthi indawo yokushisa efanayo esithombeni inikeza ukukhanya nombala okungaguquki, okuthuthukisa ukunemba kokulinganisa izinga lokushisa.

Ukuze ubonise ngokunembile ukusatshalaliswa kwezinto, iyunithi yokucubungula isithombe yenza imephu amanani okushisa ahambisana namasignali edijithali ephikseli ngayinye futhi iwaguqulele umbala obonwayo noma amanani eleveli empunga. Ukucutshungulwa okungaphakathi kombala, ngokuya ngethebula lemephu yombala wezinga lokushisa elimisiwe, izigaba zezinga lokushisa ezihlukene zihambisana nemibala ehlukene, njengokusebenzisa okuluhlaza okwesibhakabhaka ukumela izindawo ezinethempelesha eliphansi nokubomvu kuya ezindaweni ezishisa kakhulu, okwenza ukusabalalisa kwezinga lokushisa kucace lapho uthi nhla. Ekuxilongweni kwe-imaging yezokwelapha ezishisayo, odokotela bangahlulela ngokushesha izindawo zokushisa ezingavamile zomzimba womuntu ngokusebenzisa izithombe ezishisayo ezinombala mbumbulu, zisiza ekuxilongweni kwezifo.

Kwezinye izinhlelo zokusebenza zokucabanga ezishisayo ezisezingeni eliphezulu, iyunithi yokucubungula isithombe inokuhlanganiswa kwesithombe nemisebenzi yokuqaphela into. Ukuhlanganiswa kwesithombe ukuhlanganiswa kwezithombe zesithombe ezishisayo ezinezithombe ezikhanyayo ezibonakalayo, okuhlanganisa izinzuzo zakho kokubili ukuthola kokubili izinto zolwazi lwezinga lokushisa kanye nombono ocacile wokubukeka kwezinto nendawo ezungezile. Ekuqaphelweni kwezokuphepha, lokhu kuvumela abasebenzi bezokuphepha ukuthi babe nokuqonda okuphelele kwendawo yokuqapha. ukuqaphela kusebenzisa ama-algorithms okufunda komshini, afana namanethiwekhi e-convolutional neural (CNN), ukuhlonza nokuhlukanisa izinto eziqondiwe ezithombeni ezishisayo. Ekuqapheni umlilo wehlathi, ingadubula ngokuzenzakalelayo imithombo nentuthu, futhi ikhiphe ama-alamu ngesikhathi.