Esikhathini semisebenzi yethu, nsuku zonke sibhekana nenkinga yokunquma izinto ezibalulekile zentuthuko. Uma sicabangela ukuguquguquka okuphezulu kwentuthuko yemboni ye-IT, isidingo esanda njalo esivela ebhizinisini nakuhulumeni sobuchwepheshe obusha, isikhathi ngasinye lapho sinquma i-vector yentuthuko futhi sitshale amandla ethu kanye nezimali kumandla esayensi enkampani yethu, siyaqinisekisa ukuthi lonke ucwaningo lwethu namaphrojekthi ayisisekelo futhi ahlukene ngokwemikhakha.
Ngakho-ke, ngokuthuthukisa ubuchwepheshe bethu obuyinhloko - uhlaka lwe-HIEROGLYPH lokuqaphela idatha, sikhathazekile ngakho kokubili ukuthuthukisa ikhwalithi yokuqashelwa kwedokhumenti (umugqa wethu webhizinisi oyinhloko) kanye nethuba lokusebenzisa ubuchwepheshe ukuxazulula izinkinga ezihlobene nokuqashelwa. Esihlokweni sanamuhla sizokutshela ukuthi, ngokusekelwe enjinini yethu yokuqaphela (amadokhumenti), senze kanjani ukuqaphela izinto ezinkulu, ezibaluleke kakhulu ekusakazweni kwevidiyo.
Ukwakheka kwenkinga
Usebenzisa intuthuko ekhona, yakha uhlelo lokuqaphela ithangi elenza kube lula ukuhlukanisa into, kanye nokunquma izinkomba eziyisisekelo zejiyomethri (ukuma kanye nebanga) ezimeni ezingalawulwa kahle ngaphandle kokusebenzisa imishini ekhethekile.
Isixazululo
Sikhethe indlela yokufunda yomshini yezibalo njenge-algorithm eyinhloko yokuxazulula inkinga. Kodwa enye yezinkinga eziyinhloko zokufunda komshini isidingo sokuba nenani elanele ledatha yokuqeqeshwa. Ngokusobala, izithombe zemvelo ezitholwe ezigcawini zangempela eziqukethe izinto esizidingayo azitholakali kithi. Ngakho-ke, kwanqunywa ukuba kuphendukele ekukhiqizeni idatha edingekayo yokuqeqeshwa, ngenhlanhla . Kodwa nokho, kubonakale kungekona okwemvelo kithina ukuhlanganisa ngokuphelele idatha yalo msebenzi, ngakho-ke isakhiwo esikhethekile salungiselelwa ukulingisa izigcawu zangempela. Imodeli iqukethe izinto ezahlukahlukene ezilingisa indawo yasemaphandleni: ukumbozwa kwezindawo, izihlahla, izihlahla, uthango, njll. Izithombe zithathwe kusetshenziswa ikhamera yedijithali yefomethi. Phakathi nenqubo yokuthwebula isithombe, ingemuva lesigcawu lishintshe kakhulu ukwenza ama-algorithms aqine kakhulu ekushintsheni kwengemuva.
Izinto ezihlosiwe kwakuyizinhlobo ezi-4 zamathangi empi: T-90 (Russia), M1A2 Abrams (USA), T-14 (Russia), Merkava III (Israel). Izinto bezibekwe ezindaweni ezihlukene zepholigoni, ngaleyo ndlela kwandiswa uhlu lwama-engeli abonakalayo amukelekayo ento. Izithiyo zobunjiniyela, izihlahla, izihlahla nezinye izici zezwe zidlale indima enkulu.
Ngakho, ezinsukwini ezimbalwa siqoqe isethi eyanele yokuqeqeshwa kanye nokuhlola okwalandela kwekhwalithi ye-algorithm (amashumi ambalwa ezinkulungwane zezithombe).
Banqume ukuhlukanisa ukuqashelwa kona kube izingxenye ezimbili: ukwenziwa kwasendaweni kwento nokuhlukaniswa kwento. Ukwenziwa kwendawo kwenziwa kusetshenziswa i-classifier ye-Viola no-Jones eqeqeshiwe (ngemuva kwakho konke, ithangi liyinto evamile eqinile, engebi kakhulu kunobuso, ngakho-ke indlela "yemininingwane-impumputhe" ye-Viola noJones isheshe ifake into ehlosiwe). Kodwa siphathise ukuhlukaniswa nokuzimisela kwe-engeli kunethiwekhi ye-convolutional neural - kulo msebenzi kubalulekile kithi ukuthi umtshina ukhombe ngempumelelo lezo zici, ezithi, zihlukanise i-T-90 ne-Merkava. Ngenxa yalokho, kwakungenzeka ukwakha ukwakheka okusebenzayo kwama-algorithms okuxazulula ngempumelelo inkinga yokwenziwa kwasendaweni kanye nokuhlukaniswa kwezinto zohlobo olufanayo.
Okulandelayo, sethule uhlelo oluwumphumela kuwo wonke amapulatifomu ethu akhona (i-Intel, i-ARM, i-Elbrus, i-Baikal, i-KOMDIV), senze ama-algorithms anzima ngokwekhompiyutha ukwandisa ukusebenza (sesivele sibhale ngalokhu izikhathi ezimbalwa ezihlokweni zethu, isibonelo lapha. noma ) futhi kuzuzwe ukusebenza okuzinzile kohlelo kudivayisi ngesikhathi sangempela.
Njengomphumela wazo zonke izenzo ezichaziwe, sithole umkhiqizo wesofthiwe ogcwele ngokugcwele onezici ezibalulekile zamaqhinga nezobuchwepheshe.
I-Smart Tank Reader
Ngakho-ke, sethula kuwe ukuthuthukiswa kwethu okusha - uhlelo lokubona izithombe zamathangi ekusakazweni kwevidiyo I-Smart Tank Reader, okuthi:
- Ixazulula inkinga "yomngane noma isitha" ngesethi ethile yezinto ngesikhathi sangempela;
- Inquma imingcele yejiyomethri (ibanga lento, ukuma okukhethiwe kwento);
- Isebenza ezimweni zezulu ezingalawulwa, kanye nasesimweni sokuvinjwa ingxenye yento ngezinto zangaphandle;
- Ukusebenza okuzimele ngokuphelele ocingweni oluqondiwe, okuhlanganisa nokungabikho kokuxhumana komsakazo;
- Uhlu lwezakhiwo ezisekelwayo zeprosesa: Elbrus, Baikal, KOMDIV, kanye ne-x86, x86_64, ARM;
- Uhlu lwezinhlelo zokusebenza ezisekelwayo: I-Elbrus OS, i-AstraLinux OS, i-Atlix OS, kanye ne-MS Windows, i-macOS, ukusabalalisa okuhlukahlukene kwe-Linux okusekela i-gcc 4.8, i-Android, i-iOS;
- Intuthuko yasekhaya ngokuphelele.
Ngokuvamile, esiphethweni sezindatshana zethu ezimayelana ne-Habré, sinikeza isixhumanisi esimakethe, lapho noma ubani osebenzisa umakhalekhukhwini wakhe angadawuniloda inguqulo yedemo yohlelo lokusebenza ukuze empeleni ahlole ukusebenza kobuchwepheshe. Lesi sikhathi, ngokucabangela imininingwane yesicelo esiwumphumela, sifisa ukuthi bonke abafundi bethu bangalokothi babhekane nenkinga yokunquma ngokushesha ukuthi ithangi liyingxenye ethile yini.
Source: www.habr.com