```html
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Abalobi: 
 
 
 
 
 
 
 
 
 
 Tao Ye Hongyu Shan Jidong Wu Qiang Zhou Mingfu Ma Wenzhi Zhao Ru Ya Yuan Gao Lizheng Wu Isishwankathelo Idatha yendawo enkulu, enqanaba eliphezulu yezakhiwo ibalulekile ekucwangcisweni kommandla kunye nokwabiwa kwezibonelelo kunye nophando lweengozi zeziphepho. IPlato yeQinghai-Tibetan yindawo yesithathu yehlabathi. Nangona ukuqonda indlela abantu kunye nendalo ezisebenzisana ngayo kwiPlato yeQinghai-Tibetan kubaluleke kakhulu, oku kuye kwathintelwa ngokunqongophala kwedatha yobude besakhiwo esingafaniyo ngenxa yobubanzi kunye nokude kwendawo. Kulo nto sifundayo, senze igrafu yokuqala yezakhiwo ezikwiPlato yeQinghai-Tibetan kunye neendawo ezikuyo ngokusebenzisa imifanekiso yeGoogle enqanaba eliphezulu kunye ne-algorithm yokukhupha umda wesakhiwo yesiqonga se-AI Earth. Iziphumo zethu zibandakanya izakhiwo ezili-13.09 ezigidi ezibandakanya i-6092.7 km2, ziqinisekiswe ngamasampuli ovavanyo angama-250 x 1 km2. Idatha yaba nobugqibelele obupheleleyo obuyi-87%, isikhumbuzo se-91.9%, kunye nomyinge we-F1 ongu-64.8%, ngaloo ndlela ibonelela ngengcaciso ephambili yokuhanjiswa kwezakhiwo kwindawo efundwayo xa kuthelekiswa ne-CBRA. Umsebenzi wethu unamandla amakhulu ekukhuthazeni ulinganiso lwesityhelegu kwiifundo malunga neengozi zeziphepho kule ndawo. Isishwankathelo & Isishwankathelo Ukufika kwexesha ledijithali kuye kwandisa imfuno yedatha ethembekileyo malunga nokuhanjiswa kwezakhiwo kunye neempawu zazo , , . Idatha yokuhanjiswa kwezakhiwo ibonelela ngolwazi olubalulekileyo lwesithuba, kungelezakhiwo kuphela kodwa nangabantu kunye nezixhobo zempahla , isebenza njengomfuziselo omkhulu wokusebenza komntu. Kule minyaka ingamashumi ambalwa idlulileyo, idatha yokuhanjiswa kwezakhiwo iye yasetshenziswa kakhulu ekubekeni iliso kuphuhliso lweedolophu nezasezilalini , , ukuqonda iziphumo zokuba ngamadolophu kwintsimi yokutya, ubutyebi bemihlaba, utshintsho lwemozulu, kunye nempilo yabantu kunye nempilo , , ukuyila iindlela zophuhliso lweengingqi, kunye nokukhusela izindlu zendalo ezikwidolophu nezasezilalini , , . 1 2 3 4 5 6 7 8 9 10 11 Ukuqhubela phambili komfanekiso we-satellite kunye nomoya, kunye nophuhliso lwamva nje kwi-machine learning kunye ne-deep learning algorithms, kuye kwandisa ukufumaneka kwedatha yokuhanjiswa kwezakhiwo , , . Idatha yokuhanjiswa kwezakhiwo ibonelelwe kwifomati yeraster njengenxalenye yedatha yomhlaba/ukugubungela. Ukukhutshwa kwamva nje kubandakanya iimveliso ezintathu zobubanzi obuyi-10m ze-global land use/cover ezisekelwe kwi-Sentinel satellites, nto leyo ethi Google’s Dynamic World (DW) , Esri’s 2020 Land Cover , kunye ne-World Cover 2020 (WC) ye-European Space Agency (ESA) . Ngaphandle kolwazi lolwazi lwesithuba, abaphandi bayazama ukongeza ulwazi lweempawu kwiipikseli zezakhiwo, njengobude besakhiwo . Ngokomzekelo, uHe et al.  usebenzise idatha ye-remote sensing edityanisiweyo ukuze kwenziwe idatha ye-3D yobubanzi beshubhu kunye nedatha yexesha yehlabathi kwiminyaka yowe-1990 ukuya kowe-2010. Ngaphandle kokuphucuka rhoqo kobubanzi besithuba, idatha yokuhanjiswa kwezakhiwo enezigqubuthelo iraster isalungatshanga ukuchaza izinto zesithuba , kwaye ukwanda kobubanzi kwandisa kakhulu iindleko zokugcina kunye neziphakamileyo . 12 13 14 15 16 17 14 7 18 12 Idatha yokuhanjiswa kwezakhiwo kwifom yevetor, ekwayaziwa ngokuba yi-vectorized building rooftop okanye i-building footprints, yidatha yomgca ongaphandle wesakhiwo eboniswe emhlabeni kwimbono engaphezulu , , ebonelela ngolwazi olufana nendawo yendawo, ububanzi besithuba bemida, kunye nomzantsi wesakhiwo ngasinye. Ababoneleli ngeenkonzo zikawonke-wonke (umzekelo, i-Google Earth kunye ne-OpenStreetMap) babonelela ngedatha yezakhiwo ezivulekileyo ezivulekileyo ezinemida ebanzi, iinguqulelo ezikhawulezayo, kunye neendleko eziphantsi , , . Ngo-2022, i-Microsoft Corporation yasebenzisa i-deep neural network-based semantic segmentation ukukhupha imigca yezakhiwo ezingama-777 ezigidi, ezinye zazo ziqulathe iimpawu zobude besakhiwo, ngokusekelwe kwi-Bing Maps (kubandakanya imifanekiso yeMaxar kunye neAirbus) ukusuka kowe-2014 ukuya kowe-2021 kumazwekazi ngamnye ngaphandle kweTshayina. Ngomhla we-30 Meyi 2023, i-Google yakhulula idatha yezakhiwo ezili-1.8 ezigidi ezikhuphwe kwimifanekiso ye-satellite enqanaba eliphezulu le-0.5m, eyayibandakanya indawo eyi-58 yezigidi km2. Ukungabikho kwe-redundancy kunye nesakhiwo esinqabileyo sedatha yevetor emelwe ngama-vertices kunye neendlela ibonelela ngakumbi ngqo kwindawo ngokubhekiselele kubukhulu bemethi, kwaye ukwaziswa kwemithetho yetopological kuphucula ubuqudeledane bedatha yevetor . 19 20 21 22 23 13 Ngenxa yokwanda kobude bemifanekiso kunye nokuhla kweendleko zokufumana idatha, kunye nokuqhubela phambili kwakhona kwi-algorithms zokukhupha, ububanzi kunye nemida yomda yokuhanjiswa kwezakhiwo ziyaphucuka rhoqo . Ukwenzeka kwemifanekiso enqanaba eliphezulu kakhulu (VHR) kuye kwavumela ukukhupha imizobo yesakhiwo ngokusebenzisa iindlela zendabuko ezenziwe ngesandla ezisekelwe kwiimpawu  okanye iindlela ezisekelwe kwi-deep learning . Njengoko indlela yokuqala ijongene nomngeni wokungafani kwezinto zesakhiwo kunye nobukhulu, kunye nemithetho enzima yokubala kunye nokuseta kwimida, inemida xa isetyenziswa kwimifanekiso ye-remote sensing enqanaba eliphezulu . Iindlela ezisekelwe kwi-deep learning (umzekelo, i-convolutional neural networks) zibonise ukusebenza okusebenzayo kunye nokugqwesileyo ekufundeni okuzenzekelayo kweempawu eziphezulu kunye nezigqibayo kwindawo yokuhlela izakhiwo. USun et al.  bacele isicwangciso sokudibanisa esisekelwe kwi-parallel support vector machines ukusebenzisa ngokupheleleyo iimpawu eziphakamileyo ezikhuphwe kwi-multi-scale convolutional neural network structures kwiimilinganiselo ezahlukeneyo, ngokusebenza okugqwesileyo ekukhuphweni kwezakhiwo ezintsonyiweyo kwiindawo zedolophu. Nangona kunjalo, xa uhlela izakhiwo, ubugqibelele bemodyeli bunokukhawulelwa ngumgangatho weesampuli zokuqeqesha, nto leyo eyenza ukuba kube nzima ukwandisa. Ukusetyenziswa okunganeleyo kwemixholo ephezulu kunye nokushiya iinkcukacha ezisezantsi kwiimodeli eziphakamileyo, nto leyo ebangela ukuba imida ibe ngamafu kunye nokushiya izakhiwo ezincinci, nayo iyayithintela ukusetyenziswa kwe-deep learning ekukhuphweni kwemizobo yesakhiwo. 24 25 13 26 27 Indawo yePlato yeQinghai-Tibetan yeyona ndawo iphakame kakhulu ehlabathini, kunye nomyinge wobude ongu->4000 yeemitha ngaphezu komgangatho wolwandle, kwaye ibandakanya indawo eyi-2.5 yezigidi km2  . Bangaphezu kwe-10 yezigidi zabantu bahlala kule ndawo nangona imozulu yayo enobungozi, ubusika obubandayo nobude, ubushushu obukhulu bonyaka kunye nobushushu bemini, kunye neendawo zokuhlala zangaphakathi ezimbi. Nangona le ndawo iyeyona bhanti enkulu yendalo eTshayina , ukusebenza komntu kuchaphazele kakhulu indawo yayo enobungozi . Le ndawo ikwayindawo enengozi enkulu, enezikhalimela, ukuhlehlisa, ukudlala udaka, ukugqabhuka kwezilwandle zezilwandle, kunye neengozi zeqhwa ezibangela ukulahleka kwezinto kunye nezixhobo , , . Uphendula, idatha yokuhanjiswa kwezakhiwo enqanaba eliphezulu ibalulekile ekwenzeni imodeli yokuhanjiswa kwabantu ukuze kufundwe ngemeko-bume yomntu , , kunye nohlalutyo lokutsaleka kunye nengozi yeziphepho zendalo , . Ngaphezu koko, le ndawo inamalahle eshushu ixesha elide, izibonelelo zamandla elanga ezininzi, kunye neendawo ezaneleyo zokuqoqa amandla elanga njengamakhaya kunye neendawo ezivulelekileyo , . Idatha yesakhiwo sasemakhaya enqanaba eliphezulu ingabonelela ngesiseko sedatha esithembekileyo sokulinganisa amandla e-photovoltaic kunye nokuphucula ngokufanelekileyo imigaqo-nkqubo yokuphila yabantu abahlala kwiindawo zasemakhaya , . 28 29 30 31 32 33 34 35 36 37 38 39 40 41 Njengendawo engaphuhlanga, iPlato yeQinghai-Tibetan kunye neendawo ezikuyo zisenaso isethi epheleleyo yedatha yezakhiwo ezikwi-vectorized enqanaba eliphezulu, ngenxa yobubanzi bayo obukhulu, ukuhanjiswa kwezakhiwo okuphantsi, indawo ekude, kunye nemida yezibonelelo. Eyona datha ilungele ukusetyenziswa yileyo ibonelelwa kwinqanaba likazwelonke laseTshayina, ebandakanya i-China Building Rooftop Area data (CBRA, Liu et al., 2023b) enobubanzi obuzi-2.5m kunye neyona mapa yokuqala yomhlaba waseTshayina enobubanzi obuyi-1m (SinoLC) equka iindidi zezakhiwo . Le datha yeraster ayikwazi ukuchaza izinto zesithuba kwaye ifuna izixhobo ezinkulu zokugcina. I-data yommandla ophezulu kwiidolophu ezinkulu ezingama-90 eTshayina ezikhutshwe nguZhang et al.  izaliswe kancinci eli cala; nangona kunjalo, ziidolophu ezili-14 kuphela kwiPlato yeQinghai-Tibetan ezabandakanywa kule datha, kwaye idatha yezakhiwo ezikwi-vectorized isalungatshanga kwindawo eyi-2 yezigidi km2. 42 13 Ngoko ke, le sifundo sijonge ukwenza imizobo yezakhiwo ezikwiPlato yeQinghai-Tibetan kunye neengingqi ezikuyo ngokudibanisa imifanekiso ye-satellite enqanaba eliphezulu kunye ne-deep learning algorithm. Idatha yethu yaqinisekiswa ngokusebenzisa iisampuli zovavanyo ezibandakanya iigridi ezingama-250 x 1 km2 kwiingingqi ezahlukeneyo, ziphumela kubugqibelele obupheleleyo obuyi-91.92% kunye nomyinge we-F1 ongu-64.81%. Iindlela Isakhiwo Kulo sifundo, sasebenzisa i-algorithms zokukhupha izakhiwo kwi-platform ye-AI Earth ukwenza idatha yezakhiwo ezikwi-vectorized kwiPlato yeQinghai-Tibetan kunye nommandla wayo okufutshane eTshayina (Jongi. 1). Izingongoma eziphambili zesakhiwo sethu zibandakanya: (1) ukulungiswa kunye nokulungiswa kwakhona kwedatha ye-satellite kunye nedatha yenkxaso; (2) ukukhupha izakhiwo ezikwi-vectorized usebenzisa i-platform ye-AI Earth; (3) ukuqinisekiswa ngokusebenzisa idatha yommandla wekhaya eyenziwe ngesandla.    Isakhiwo sokukhupha izakhiwo kunye nokuqinisekiswa. Indawo yokuFunda Esi sifundo sigxile kakhulu kwiPlato yeQinghai-Tibetan kunye neengingqi zayo ezikufutshane kumazantsi-ntshona eTshayina phantsi kwesakhiwo esingundoqo soBumodeli yesibini yeNzululwazi yePlato yeTibetan kunye neNkqubo yoPhando ; ngokweendawo, ibandakanya iSithili saseTibet, iProvinsi yaseQinghai, intshona yeProvinsi yaseYunnan, intshona yeProvinsi yaseSichuan, intshona-ntshona yeProvinsi yaseGansu, kunye nesithili saseXinjiang esingaphantsi kweYanga. Umndilili wobude bendawo yokuFunda ngu-4000m ngaphezu komgangatho wolwandle. Ukuhanjiswa kwabantu kunye nezakhiwo kwindawo yokuFunda kuchaphazeleka kakhulu bubude kunye nemozulu (Jongi. 2), ubukhulu becala ibandakanya umntla womgca ukusuka kwiSithili saseJilong ukuya kwiSithili saseQilian eQinghai ; empuma, zihanjiswe kwiindawo ezithe tyaba kumda ongase mpuma wommandla, kubandakanywa iingxowa zomlambo eProvinsi yaseYunnan, intshona yeProvinsi yaseSichuan, kunye neNdawo yoMlambo iYellow River Basin. Kwiindawo ezikumazantsi-ntshona empuma, abantu kunye nezakhiwo zihanjiswe ikakhulu kwii-agro-pastoral areas ezincinci kwiingxowa eziphambili zomlambo, kunye nemigca yendlela yendlela. 43 44    Indawo yokuFunda (idatha yabantu ivela kwi-LandScan; i-DEM ivela kwiZiko leZibonelelo kunye neDatha yeLwazi). Imifanekiso yeSatellite Idatha yemifanekiso ye-satellite enqanaba eliphezulu evulekileyo yafunyanwa kwi-Rivermap Co. ( ), eyayifunyenwe kumanyano lwe-Google Earth lwemifanekiso ye-satellite kunye nedatha yomoya. Phakathi kwazo, imifanekiso ye-satellite ibandakanya i-DigitalGlobe’s QuickBird kunye ne-WorldView commercial satellites, kunye ne-aerial photography ivela kwi-BlueSky e-UK kunye ne-Sanborn e-US . Kwindawo nganye, kwakukho iiqoqo zemifanekiso emininzi enobubanzi obufika kwi-0.15m kwiindawo ezithile. Ukudityaniswa kwedatha okunjalo kuye kwasetshenziswa kabanzi kumqolo wezinto kwiziganeko eziyinkimbinkimbi , , ,  kwaye inamandla okwenza iimephu ezinkulu zobungakanani bolwazi lweentlobo zezinto . http://www.rivermap.cn/index.html 45 46 47 48 49 50 Ububanzi bonke bendawo yethu yokuFunda yi-3.06 yezigidi km2, kwaye ubungakanani obulindelekileyo bedatha yemifanekiso ye-satellite yi-29.4 TB, kunye nobubanzi besithuba obuyi-0.6 m. Ngokubheka le milinganiselo, ii-fishnets ezingama-0.175° x 0.175° zenziwa kwindawo yethu yokuFunda ukuze kuvumele ukuhlanganiswa kobungakanani obuncinci bokukhuphela, kunye ne-fishnets ezili-10,033 ezisetyenzisiweyo ukugubungela yonke indawo yokuFunda. Inani langempela lee-fishnets ezikhutshiweyo belingaphantsi kwi-5921 ngezizathu ezimbini. Okokuqala, njengoko inxalenye enkulu yendawo yokuFunda ibandakanya iindawo zokuhlala ezingezizo zabantu apho izakhiwo zingekho, sakhuphela ngabom ii-fishnets ezingenawo nawaphi na ama-pixels okwakhiwa kwiimveliso ze-ESA World Cover. Okwesibini, iidolophu/izithili ezenziweyo ngamacandelo ezakhiwo (nto leyo ethi, i-Lhasa, i-Shannan, i-Kunming, i-Xining, i-Haidong, i-Zhangye, i-Baiyin, i-Lanzhou, i-Chengdu, i-Dali, i-Lijiang, i-Kunming, i-Zhaotong, kunye ne-Yuxi) sele zifunyenwe kwidatha yobubanzi besakhiwo esikwi-vectorized yezi dolophu zingama-90 eTshayina  zaphuma. Imifanekiso yakhutshwa phakathi kukaSeptemba 2022 noJanuwari 2023, enobubanzi obuyi-0.6 m, ubungakanani bomsonto omnye bungama-3 GB, kunye nobubanzi bonke obuyi-2.72 TB. Imifanekiso ekhutshiweyo yayithathwe ikakhulu ukusuka kowe-2019 ukuya kowe-2021, kodwa imifanekiso yeendawo ezikude ezithile ingathathwa njengoko sele ikwii-2001 (Jongi. 3). 13    Ii-fishnets zokukhuphela imifanekiso kunye nexesha lokudutshulwa kwemifanekiso. Idatha encedisayo Idatha yethu encedisayo ibandakanya iimephu zomhlaba eziphakamileyo, imodeli yedijithali yobude (DEM), i-normalized difference vegetation index (NDVI), kunye nedatha yokuhanjiswa kwabantu ukuze kuqhubeke uhlalutyo lwe-cluster ngeenjongo zokuthatha isampuli (Itafile 1). Kuhlobo lomhlaba kunye nendawo eyakhiweyo, imveliso ye-ESA’s World Cover evela kuZanaga et al.

Part of HackerNoon's growing list of open-source research papers, promoting free access to academic material.

Le audio iveliswe ngolwimi lokuqala lwebali!

I-iMephu Ikhombisa Izakhiwo Eziyi-13 lezigidi Kwingingqi enye Yenqanaba Eliphezulu Eliyiyo Emhlabeni

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps