Ukubonwa Kwethagethi Yekhamera Yangaphansi kwamanzi

Ukulungiswa Kwesimiso Somfanekiso: Isistimu yokubona yamakhamera angaphansi kwamanzi ithuthukiswa ngokusekelwe kunkomba esebenzayo yamanzi. Inkomba ye-refractive yamanzi icishe ibe ngu-1.33, ehlukile kuleyo yomoya, okuholela ekukhanyeni nasekukhanyeni. Ngakho-ke, idizayini yelensi idinga ukucabangela lezi zici ukuze kuqinisekiswe izithombe ezicacile ngokuqhathaniswa. Ngokwesibonelo, ukusebenzisa amalensi akhethekile ane-engeli ebanzi kunganciphisa ukuhlanekezela kwesithombe okubangelwa ukurareka ngezinga elithile.

Ukucubungula Ngaphambi Kwesithombe: Ngenxa yemvelo eyinkimbinkimbi engaphansi kwamanzi, izithombe ngokuvamile zidinga ukucutshungulwa kusengaphambili ukuze kulungiswe umbala futhi kuthuthukise ukugqama. Lokhu kufaka phakathi ukulungiswa kombala, njengoba kunxephezela ubude begagasi obuthile bokukhanya obumuncwa amanzi, nokuthuthukisa umehluko, njengoba izithombe ezingaphansi kwamanzi ngokuvamile zinokugqama okuphansi. Izindlela ezifana nokulinganisa kwe-histogram zingathuthukisa, kwenze kube lula ukuhlukanisa izinto eziqondiwe nengemuva.

Izici Zomumo: Umumo uyisici esibalulekile sokubonwa kwethagethi engaphansi kwamanzi. Ngoba, emivubukulweni engaphansi kwamanzi, umumo wezingcezu zasendulo zokuphahlazeka komkhumbi kungase kube amabhulokhi angajwayelekile noma abe nezimo ezithile zejiyomethri. Ama-algorithms okuthola umkhawulo, njengokutholwa kwe-Canny, angasetshenziswa ukuze kukhishwe ama-edge ezinto eziqondiwe, ezisebenza njengezinkomba ezibalulekile zokubonwa.

Izici Zokuthungwa: Okuhlosiwe okuningi ngaphansi kwamanzi kunokwakheka okuhlukile. isibonelo, izixhobo zamakhorali zinendlela eyinkimbinkimbi futhi ethambile, kuyilapho amazenga ezinhlanzi anokwakheka kwawo okuhlukile. Ngokusebenzisa izindlela zokuhlaziya ukuthungwa njengokwenzeka ngokubambisana kweleveli empunga, singakwazi ukukhipha izici zokuthungwa kwezinto eziqondiwe, okufaka ukuhholo nokuqondiswa, okungasetshenziselwa ukubonwa.

Izici Zombala: Nakuba imibala ingase ihlanekezelwe, isengasetshenziswa njengesici ngokwezinga elithile. Ngokwesibonelo, ezinye izinhlanzi zasezindaweni ezishisayo zinemibala egqamile. Ngokukhipha ama-histogram ombala noma ukubala izikhathi zombala ezithombeni ezilungiswe ngombala, singasebenzisa izici zombala ukusiza ekuboneni. Ukwengeza, izinto eziphilayo ezingaphansi kwamanzi ezihlukene noma izinto zingase zibe nezici zombala ezihlukile ngaphansi kwamabhendi athile e-spectral.

Ama-Algorithms Asekelwe Isifanekiso Matching: Uma izici zomumo wento eqondiwe zichazwe kahle njengokuhlola ipayipi elingaphansi kwamanzi lapho umumo nosayizi wepayipi waziwa kusengaphambili, isithombe sesifanekiso sento eqondiwe singafaniswa nokushuthiwe. isithombe. Ngokubala izilinganiso zokufana, njenge-coefficient yokuxhumanisa okwejwayelekile, ubukhona nendawo yento eqondiwe kunganqunywa.

Ama-algorithms wokufunda ngomshini:

Ukufunda Okugadiwe: Lokhu kuhilela ukuqeqeshwa ngedathasethi yesithombe sangaphansi kwamanzi esinelebuli. Isibonelo, uma kunezithombe ezilebulwe zezinhlobo ezahlukene zezinhlanzi, izici ezifana nokuthungwa komumo, nombala zingasetshenziswa njengokufakwayo, kanye nohlobo lwenhlanzi njengelebula lokuphumayo. Ama-algorithms afana ne-Support Vector Machines (SVM) nezihlahla zezinqumo zisetshenziselwa ukuqeqeshwa kokuhlukaniswa. Imodeli eqeqeshiwe ingase isetshenziselwe ukuhlonza izinhlobo zezinhlanzi ezithombeni ezintsha zangaphansi kwamanzi.

Ukufunda Okungagadiwe: Lokhu okokuhlosiwe ngaphandle kolwazi lwangaphambili, njengemiphakathi esanda kutholwa yebhayoloji engaziwa phansi olwandle. I-Clustering algorithms, efana ne-K-means clustering, ingasetshenziswa okuphokophelwe kweqembu ngokusekelwe kuzici zabo, bese kucutshungulwa ngokuqhubekayo okuhlosiwe ngaphakathi kweqembu ngalinye.

Ama-algorithms wokufunda okujulile:

I-Convolutional Neural Networks (CNN): Lena indlela esebenzayo yokubonwa kwethagethi engaphansi kwamanzi. Isibonelo, i-CNN enezendlalelo eziningi zokuguqula, izendlalelo zokuhlanganisa, nezindlalelo ezixhumeke ngokugcwele zingakhiwa. Ngokusebenzisa inani lezithombe ezingaphansi kwamanzi njengedatha yokuqeqeshwa, inethiwekhi ingafunda ngokuzenzakalela izici zeleveli ephezulu zezinto eziqondiwe. Isibonelo, ekuboneni okuqondiwe kwerobhothi elingaphansi kwamanzi, njengamaminerali noma izingxenye zokuphahlazeka komkhumbi, i-CNN ingafunda izici eziyinkimbinkimbi zalokhu okuhlosiwe, ngaleyo ndlela izuze ukuqashelwa okunembayo okuphezulu.

I-Fusion Nezinzwa ze-Sonar: Ezindaweni ezingaphansi kwamanzi, i-sonar inganikeza ulwazi mayelana nebanga nosayizi wento eqondiwe. Ngokuthola idatha evela kumakhamera angaphansi kwamanzi nezinzwa ze-sonar, ukuqonda okubanzi kwento eqondiwe kungafinyelelwa. Isibonelo, emisebenzini yokusesha nokuhlenga ngaphansi kwamanzi, i-ar ingathola cishe indawo kanye nobubanzi balokho okungase kuqondiswe ngumuntu, bese ikhamera yangaphansi kwamanzi ingasebenzisa lolu lwazi ukuze ikwazi ukubona okubonakalayo okunembayo ukuze inqume ukuthi ingabe iyimpokophelo.

I-Fusion Nezinzwa Zokubona: Uma ikhamera engaphansi kwamanzi iyikhamera enezici eziningi, ingahlanganiswa nezinye izinzwa ezibonakalayo, njenge-LiAR, ukuze kuthuthukiswe amakhono okuqaphela okuqondiwe. Izinzwa ezihlukile zokubona zinganikeza ulwazi oluhlukile lwesici mayelana nento eqondiwe, futhi ngokuhlanganisa lolu lwazi, ukunemba nokuqina kokubonwa kwethagethi kungathuthukiswa.