```html
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 Ababh:** 
 
 
 
 
 
 
 
 
 
 Tao Ye Hongyu Shan Jidong Wu Qiang Zhou Mingfu Ma Wenzhi Zhao Ru Ya Yuan Gao Lizheng Wu Isishwankathelo Idatha enkulu, enchanekileyo yokwabiwa kwezakhiwo yeyona ndlela ibalulekileyo yokucwangciswa kwezixeko kunye nokwabiwa kwezixhobo kunye nophando lweengozi zeziphawo. Indawo yeQinghai-Tibetan Plateau yiyona ndawo yesithathu yehlabathi. Nangona ukuqonda ulwalamano lwabantu-nokusingqongileyo kwiQinghai-Tibetan Plateau kubaluleke kakhulu, oku kuye kwathintelwa kukuphosakaliswa kwedatha ephakamileyo yezakhiwo ngenxa yobubanzi nobude bendawo. Kolu phando, siye sakha imifanekiso yezakhiwo zokuqala ezifotweyo kwiQinghai-Tibetan Plateau kunye neendawo ezijikelezileyo ngokusebenzisa umfanekiso ophezulu weGoogle kunye ne-algorithm yokukhupha imilo yesakhiwo ye-AI Earth platform. Iziphumo zethu zibandakanya izakhiwo ezili-13.09 ezigidi ezibandakanya i-6092.7 km2, eziqinisekiswe ngesampuli yokuvavanya eyi-250 x 1 km2. Idatha ibenongcaciso jikelele ye-87%, isikhumbuzo esingu-91.9%, kunye ne-F1 score ye-64.8%, ngaloo ndlela ibonelela ngenkcazo ephucukileyo yokwabiwa kwezakhiwo kwingingqi yophando xa ithelekiswa neCBRA. Umsebenzi wethu unamandla amakhulu okubonelela ngocwangciso lokuvavanywa kwengozi yeziphawo kulo mmandla. Imvelaphi & Isishwankathelo Ukuvela kweminyaka yedijithali konyusele imfuno yedatha ethembekileyo malunga nokwabiwa kwezakhiwo kunye neempawu zazo , , . Idatha yokwabiwa kwezakhiwo ibonelela ngolwazi olubalulekileyo lwesithuba, hayi nje kwezibhakabhaka kodwa kunye nabemi kunye neendawo ezibambekayo , isebenza njengomlinganisi olungileyo kwimisebenzi yabantu. Kwiiminyaka yakutshanje, idatha yokwabiwa kwezakhiwo iye yasetyenziswa kabanzi ekubekeni iliso kuphuhliso lwezixeko nemilwande , , ukuqonda iimpembelelo zokwanda kwezixeko kwisondlo, i-biodiversity, utshintsho lwemozulu, kunye nempilo yoluntu kunye nempilo , , ukwakhiwa kwezicwangciso zophuhliso lwemimandla, kunye nokhuselo lweendalo zezixeko nemilwande , , . 1 2 3 4 5 6 7 8 9 10 11 Ukuphucuka komfanekiso we-satellite kunye nomoya, kunye nokuphucuka kwamuva nje kwi-algorithms yokuphanga kobuchwepheshe kunye ne-deep learning, kuye kwandisa ukufumaneka kwedatha yokwabiwa kwezakhiwo , , . Idatha yokwabiwa kwezakhiwo iye yabonelelwa ngendlela ye-raster njengenxalenye yedatha yomhlaba/yomhlaba. Ukukhutshwa kwamva nje kubandakanya iimveliso ezintathu ze-10-m zehlabathi ezisekelwe kwi-Sentinel satellites, nto leyo i-Dynamic World (DW) kaGoogle , i-Esri ka-2020 Land Cover , kunye ne-World Cover 2020 (WC) ye-European Space Agency (ESA) . Ngaphandle kolwazi lwesithuba sokwabiwa, abaphandi baye bazama ukunamathela kulwazi lwesimo kwii-pixels zesakhiwo, njengobude besakhiwo . Ngokomzekelo, uHe  .  usebenzise idatha yokukhupha imifanekiso emininzi ukuze kwakhiwe idatha yokuqala yehlabathi eyi-30-m yesithuba kunye nedatha yedatha yedatha yedatha yeedatha yesithuba esuka ku-1990 ukuya ku-2010. Nangona ukuphucuka kokucaciswa kwesithuba, idatha yokwabiwa kwezakhiwo esekelwe kwi-raster isengakwazi ukuchaza izinto zesithuba , kwaye ukwanda kokucaciswa kwesithuba kwandisa iindleko zokugcina kunye neezokubala . 12 13 14 15 16 17 14 et al 7 18 12 Idatha yokwabiwa kwezakhiwo kwifomathi ye-vector, ekwabizwa ngokuba yimiphezulu yezakhiwo ezifotweyo okanye imifanekiso yezakhiwo, yimida yedatha yesakhiwo ebekwe emhlabeni xa ibukwe ngaphezulu , , ebonelela ngolwazi olufana nendawo yendawo, ububanzi besithuba bemida, kunye nomsebenzi wesakhiwo esinye. Ababoneleli ngeenkonzo zoluntu (umzekelo, i-Google Earth kunye ne-OpenStreetMap) babonelela ngedatha yezakhiwo ezifotweyo ezivulweyo ezinegqebelo elibanzi, ukuhlaziywa okukhawulezayo, kunye neendleko eziphantsi , , . Ngo-2022, i-Microsoft Corporation yasebenzisa ukwahlula kwe-semantic esekelwe kwi-deep neural network ukukhupha imida yezakhiwo ezili-777 ezigidi, ezinye zazo zineempawu zobude besakhiwo, esekelwe kwi-Bing Maps (kubandakanya imifanekiso ye-Maxar kunye ne-Airbus) ukusuka ku-2014 ukuya ku-2021 kuwo wonke amazwekazi ngaphandle kweChina. Ngomhla wama-30 kuMeyi 2023, i-Google yakhuphelwa idatha yezakhiwo ezili-1.8 yezigidi ezikhuphwe kumfanekiso we-satellite ophezulu we-0.5-m, ugubungela indawo eyi-58 yezigidi km2. Ukungaphazamiseki okuphantsi kunye nolwakhiwo oluqinileyo lwedatha ye-vector emelwe ngama-vertices kunye nemizila kubonelela ngqo yendawo ephezulu engazukuxhomekeka kubukhulu bemitha, kwaye ukungeniswa kwemithetho ye-topological kuphucula ubuqhetseba bedatha ye-vector . 19 20 21 22 23 13 Ngenxa yokwanda kobude bomfanekiso kunye nokuhla kweendleko zokufumana idatha, kunye nokuphucuka kwamuva nje kwi-algorithms zokukhupha, ububanzi kunye negqebelo ledatha yokwabiwa kwezakhiwo liyaphucuka rhoqo . Ukubaluleka kwemifanekiso enezinto eziphezulu kakhulu (VHR) kuye kwenza ukukhuphwa kweemifanekiso zezakhiwo ngokusebenzisa iindlela ezindala eziphathwa ngesandla okanye iindlela ezisekelwe kwi-deep learning . Njengoko indlela yangaphambili ijongene nemiceli-mikhomo yokwahluka kobume bezakhiwo kunye nobukhulu, kunye nemigaqo enzima kunye nezicwangciso zokubeka, ilinganiselwe xa isetyenziswa kwimifanekiso emikhulu, enezinto eziphezulu kakhulu yokukhupha imifanekiso . Iindlela ezisekelwe kwi-deep learning (umzekelo, i-convolutional neural networks) ziye zabonisa ukusebenza okusebenzayo kunye nokugqwesileyo ekufundeni ngokuzenzekelayo iimfashini eziphezulu nezibonisa ngokucacileyo kwi-segmentation yezakhiwo. USun  .  uye wacebisa isicwangciso sokudibanisa esekelwe kwi-parallel support vector machines ukuze kusetyenziswe ngokupheleleyo iimfashini ezikhuphwe kwi-multi-scale convolutional neural network structures kwiindawo ezahlukeneyo, ngokusebenza okugqwesileyo ekukhuphweni kwezakhiwo ezininzi kwizindawo zasemzini. Nangona kunjalo, xa uphawula izakhiwo, uchaneko lwemodeli lunokulinganiselwa bubungqino bezampulu zokufunda, nto leyo eyenza kube nzima ukwandisa. Ukusetyenziswa okunganeleyo kwezinto eziphezulu kunye nokugqithisa iinkcukacha ezisezantsi kwiimodeli ezinzulu, eziphumela ekubeni imida ingacaci kunye nokugqithisa izakhiwo ezincinci, kukwathintela ukusetyenziswa kwe-deep learning ekukhuphweni kweemifanekiso zezakhiwo. 24 13 26 et al 27 Indawo yeQinghai-Tibetan Plateau yeyona ndawo iphakamileyo ehlabathini, ngokumgangatho ophakathi ongaphaya kwe-4000 yeemitha ngaphezu komgangatho wolwandle, kwaye igubungela indawo eyi-2.5 yezigidi km2  . Abemi abangaphezu kwe-10 yezigidi bahlala kulo mmandla nangona imozulu yawo ekhohlakeleyo, ubusika obubandayo nobunzima, ubuninzi bokutshintshatshintsha komoya kunye nobushushu bemini, kunye neemeko ezibi zokufudumeza ngaphakathi. Nangona le mimandla yeyona ndawo enkulu yokhuselo lwendalo e-China , imisebenzi yabantu iye yaba nefuthe elikhulu kwindalo enobutsha bayo . Le mimandla iphinde yaba yindawo yentlekele ezininzi, kunye neentlekele zenyikima, ukudilika komhlaba, ukudilika komdaka, ukudilika kwezilwandle zomkhenkce, kunye neentlekele zeqhwa eziye zabangela ukulahleka kwezibambiso kunye neempahla , , . Ukuphendula, idatha echane ngqo yezakhiwo ibalulekile ekumodeli ukwabiwa kwemisebenzi yabantu kumfundweyo odibeneyo lwabantu-nokusingqongileyo , , kunye nokuvavanywa kokufumanisa kunye nomngcipheko weentlekele zendalo , . Ukongeza, le mimandla inamawonga elanga amade, imithombo enamandla elanga, kunye neendawo ezigcweleyo zokuqokelelwa kwamandla elanga ezifana nezindlu kunye neendawo ezivulekileyo , . Idatha yezakhiwo zasemilwande enezinto eziphezulu ingabonelela ngesiseko sedatha esithembekileyo sokuvavanya ubuchule be-photovoltaic kunye nokuphucula ngempumelelo imigaqo-nkqubo yokuphila yabahlali basemilwande , . 28 29 30 31 32 33 34 35 36 37 38 39 40 41 Njengendawo engaphuhlanga, iQinghai-Tibetan Plateau kunye neendawo zayo ezikufutshane isengenasethi epheleleyo yedatha yezakhiwo eziphakathi kwezinto eziphezulu, ngenxa yendawo yayo enkulu, ukwabiwa kwezakhiwo okuncinci, indawo ekude, kunye nezithintelo zezixhobo. Idatha elungele ukusetyenziswa kakhulu yileyo ibonelelwa kwinqanaba likazwelonke lase-China, ebandakanya i-China Building Rooftop Area data (CBRA, Liu  ., 2023b) ye-2.5-m kunye neyokuqala yase-China yomhlaba ezitheni ezine-1-m ezibandakanya iindidi zezakhiwo (SinoLC) . Le datha ye-raster ayikwazi ukuchaza izinto zesithuba kwaye ifuna izixhobo ezininzi zokugcina. Idatha yommandla ophahla lwesakhiwo kwizixeko ezingama-90 zase-China ekhutshwe nguZhang  .  yakhe yagqibezela eli gcuntswana; nangona kunjalo, zizixeko ezili-14 kuphela kwiQinghai-Tibetan Plateau ezibandakanyiweyo kule datha, kwaye idatha yesakhiwo ezifotweyo isasalele indawo eyi-2 yezigidi km2. et al 42 et al 13 Ngoko ke, olu phando lujolise ekubeni kwenziwe imifanekiso yezakhiwo ezifotweyo kwiQinghai-Tibetan Plateau kunye neendawo zayo ezikufutshane ngokudibanisa umfanekiso we-satellite ophezulu kunye ne-algorithm ye-deep learning. Idatha yethu iye yagqithiselwa kusetyenziswa iisampulu zovavanyo ezibandakanya iigridi ezingama-250 x 1 km2 kwiindawo ezahlukeneyo, kubangela ingcaciso jikelele ye-91.92% kunye ne-F1 score ye-64.81%. Iindlela Isakhiwo Kolum phando, sisebenzise i-algorithms zokukhupha izakhiwo ezivela kwi-AI Earth platform ukwenza idatha yesakhiwo esifotweyo kwiQinghai-Tibetan Plateau kunye nommandla wayo ojikelezileyo e-China (Umzekelo  ). Iindidi eziphambili zesakhiwo sethu zibandakanya: (1) ukulungiswa kunye nokulungiswa kwangaphambili kwedatha ye-satellite kunye nedatha eyongezelelweyo; (2) ukukhuphwa kwezakhiwo ezifotweyo kusetyenziswa i-AI Earth platform; (3) ukugqithiselwa kusetyenziswa idatha yezakhiwo ezifotwe ngesandla. 1    Isakhiwo sokukhuphwa kwezakhiwo kunye nokugqithiselwa. Indawo yophando Olu phando luye lwagxila ikakhulu kwiQinghai-Tibetan Plateau kunye neendawo zayo ezikufutshane kumzantsi-ntshona we-China phantsi kwesakhiwo esikhulu soHlola lwesayensi yeQinghai-Tibetan Plateau kunye neNkqubo yophando ; ngokwentlalo, ibandakanya iTibetan Autonomous Region, iProvinces zeQinghai, intshona yeYunnan Province, intshona yeSichuan Province, umzantsi-ntshona weGansu Province, kunye nomzantsi-mpuma weXinjiang Autonomous Region. Umgangatho ophakathi wengingqi yophando yi-4000-m ngaphezu komgangatho wolwandle. Ukwabiwa kwabemi kunye nezakhiwo kwingingqi yophando kuye kwaba nefuthe elikhulu ngumgangatho kunye nemozulu (Umzekelo  ), ingakumbi emantla omlenze ukusuka eJilong County eTibet ukuya eQilian County eQinghai ; empuma, zihleli kwiindawo eziphakamileyo kumda ongaphambili wommandla, kubandakanya iintshela zemilambo yaseYunnan Province, intshona yeSichuan Province, kunye neXining-Lanzhou Yellow River Basin. Kumgangatho wephondo entshona yomlenze, abemi nezakhiwo zihlala zihleli kunye neendawo ezilima-zilwela ezincinci emacaleni emilambo emikhulu, kunye nemizila yothutho lwemigqiggo. 43 2 44    Indawo yophando (idatha yabemi ivela kwi-LandScan; i-DEM ivela kwi-Resource and Environment Science and Data Centre). Imifanekiso yeSatellite Idatha yomfanekiso we-satellite enezinto eziphezulu ezifumanekayo ngokukhululekileyo yafunyanwa kwi-Rivermap Co. ( ), eyafunyanwa kumanyano lwe-Google Earth lomfanekiso we-satellite kunye nedatha yomoya. Phakathi kwazo, umfanekiso we-satellite ubukhulu becala uvela kwi-DigitalGlobe's QuickBird kunye ne-WorldView satellites zorhwebo, kwaye ukudutshulwa komoya kuthathwa kwi-BlueSky e-UK kunye neSanborn e-US . Kwiindawo nganye, kwakukho ingqokelelo yemifanekiso emininzi enezinto eziphezulu ezifikelela kuma-0.15 m kwiindawo ezithile. Ukudityaniswa kwedatha enjalo kuye kwasetyenziswa kabanzi ekuboneni izinto kwiindawo ezinzima , , ,  kwaye inamandla okwenza imephu enkulu, enezinto eziphezulu zohlobo lwezinto . http://www.rivermap.cn/index.html 45 46 47 48 49 50 Indawo yommandla wethu yonke yi-3.06 yezigidi km2, kwaye ubungakanani obuqikelelweyo bedatha yomfanekiso we-satellite yi-29.4 TB, enobubanzi besithuba obu-0.6 m. Ngoku, oku kuyakhulu, sijonge ngaphakathi kwiigridi ezili-0.175° x 0.175° kwindawo yophando ukuvumela iipakethi ezincinci zokukhuphelwa, kunye neegridi ezili-10,033 ezisetyenziselwa ukugubungela yonke indawo yophando. Inani langempela leegridi ezikhutshelweyo lalingaphantsi, lama-5921, ngenxa yezizathu ezimbini. Okokuqala, njengoko inxalenye enkulu yendawo yophando ibandakanya iindawo zokuhlala zabantu ezingahlaliyo apho izakhiwo zingekho, sixhomekeke ngokungakhethiyo iigridi ezingenazo naziphi na iipiksels zakhiwo ezivela kwimveliso ye-ESA World Cover. Okwesibini, izixeko/izithili ezinezakhiwo zazo ezifotweyo (oko kukuthi, 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 yommandla ophahla lwesakhiwo kuzo zonke izixeko ezingama-90 zase-China  kwaye zakhutshwa. Imifanekiso yakhutshelwa phakathi kukaSeptemba 2022 noJanuwari 2023, enobubanzi obu-0.6 m, ubungakanani besiqwenga esinye esimalunga ne-3 GB, kunye nobukhulu obungu-2.72 TB. Imifanekiso ekhutshelweyo yayingenawo u-2019 ukuya ku-2021, kodwa imifanekiso yeendawo ezikude kakhulu ingathathwa ngo-2001 (Umzekelo  ). 13 3    Iigridi zokukhuphelwa komfanekiso kunye nexesha lokufota umfanekiso. Idatha eyongezelelweyo Idatha yethu eyongezelelweyo ibandakanya iimephu eziphakamileyo zomhlaba, idatha yedatha yobude bomhlaba (DEM), isalathiso sophando lwendalo (NDVI), kunye nedatha yokwabiwa kwabantu kumfundweyo wokuhlanganisa ukuze kusetyenziswe iisampulu (Itheyibhile  ). 1

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

Le audio iveliswe ngolwimi lokuqala lwebali!

I-AI iyakala izakhiwo eziyi-13 lezigidi kwenye yeeNdawo eziKude kunawo onke eHlabathini

About Author

IZIMVO

ZIJONGE IIMPAWU

ELI NQAKU LINIKEZELWE KU

Related Stories

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps