{"id":11549,"date":"2025-05-04T21:00:25","date_gmt":"2025-05-04T19:00:25","guid":{"rendered":"https:\/\/geopard.tech\/?p=11549"},"modified":"2025-05-04T21:00:25","modified_gmt":"2025-05-04T19:00:25","slug":"cmtnet-is-naujo-apibrezia-tiksluji-ukininkavima-pranokdama-tradicine-paseliu-klasifikacija","status":"publish","type":"post","link":"https:\/\/geopard.tech\/lt\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/","title":{"rendered":"CMTNet i\u0161 naujo apibr\u0117\u017eia tiksli\u0105j\u0105 \u017eemdirbyst"},"content":{"rendered":"

Tikslus pas\u0117li\u0173 klasifikavimas yra b\u016btinas \u0161iuolaikinei tiksliajai \u017eemdirbystei, nes tai leid\u017eia \u016bkininkams steb\u0117ti pas\u0117li\u0173 sveikat\u0105, prognozuoti derli\u0173 ir efektyviai paskirstyti i\u0161teklius. Ta\u010diau tradiciniai metodai da\u017enai susiduria su sud\u0117tingomis \u017eem\u0117s \u016bkio s\u0105lygomis, kai pas\u0117liai labai skiriasi r\u016b\u0161imi, augimo stadijomis ir spektriniais po\u017eymiais.<\/p>\n

Kas yra hiperspektrinis vaizdavimas ir CMTNet sistema?<\/h2>\n

Hiperspektrinis vaizdavimas (HSI) \u2013 technologija, fiksuojanti duomenis \u0161imtuose siaur\u0173, gretim\u0173 bangos ilgi\u0173 juost\u0173 \u2013 tapo revoliucine \u0161ios srities technologija. Skirtingai nuo standartini\u0173 RGB kamer\u0173 ar multispektrini\u0173 jutikli\u0173, kurie renka duomenis keliose pla\u010diose juostose, HSI pateikia i\u0161sam\u0173 kiekvieno pikselio \u201cspektrin\u012f pir\u0161t\u0173 atspaud\u0105\u201d.<\/p>\n

Pavyzd\u017eiui, sveika augmenija stipriai atspindi artimojo infraraudonojo spektro \u0161vies\u0105 d\u0117l chlorofilo aktyvumo, o streso paveikti pas\u0117liai pasi\u017eymi skirtingais sugerties modeliais. U\u017efiksuodamas \u0161iuos subtilius poky\u010dius (nuo 400 iki 1000 nanometr\u0173) didele erdvine skiriam\u0105ja geba (net 0,043 metro), HSI leid\u017eia tiksliai diferencijuoti pas\u0117li\u0173 r\u016b\u0161is, aptikti ligas ir atlikti dirvo\u017eemio analiz\u0119.<\/p>\n

Nepaisant \u0161i\u0173 privalum\u0173, esami metodai susiduria su i\u0161\u0161\u016bkiais subalansuojant vietines detales, tokias kaip lap\u0173 tekst\u016bra ar dirvo\u017eemio modeliai, su pasauliniais modeliais, tokiais kaip didelio masto pas\u0117li\u0173 pasiskirstymas. \u0160is apribojimas ypa\u010d i\u0161ry\u0161k\u0117ja triuk\u0161minguose arba nesubalansuotuose duomen\u0173 rinkiniuose, kur nedideli spektriniai skirtumai tarp pas\u0117li\u0173 gali lemti klaiding\u0105 klasifikavim\u0105.<\/p>\n

Siekdami i\u0161spr\u0119sti \u0161iuos i\u0161\u0161\u016bkius, mokslininkai suk\u016br\u0117\u00a0CMTNet<\/strong>\u00a0(Konvoliucinis ir transformatorinis tinklas) \u2013 nauja gilaus mokymosi sistema, apjungianti konvoliucini\u0173 neuronini\u0173 tinkl\u0173 (CNN) ir transformatori\u0173 stipri\u0105sias puses. CNN yra neuronini\u0173 tinkl\u0173 klas\u0117, skirta apdoroti tinklelio tipo duomenis, pvz., vaizdus, naudojant filtr\u0173 sluoksnius, kurie aptinka erdvines hierarchijas (pvz., kra\u0161tus, tekst\u016bras).<\/p>\n

\"CMTNet<\/p>\n

Transformatoriai, i\u0161 prad\u017ei\u0173 sukurti nat\u016bralios kalbos apdorojimui, naudoja sav\u0119s d\u0117mesio mechanizmus, kad modeliuot\u0173 ilgalaikes duomen\u0173 priklausomybes, tod\u0117l jie puikiai geba fiksuoti globalius modelius. Skirtingai nuo ankstesni\u0173 modeli\u0173, kurie nuosekliai apdorojo vietines ir globalias ypatybes, CMTNet naudoja lygiagre\u010di\u0105 architekt\u016br\u0105, kad vienu metu i\u0161gaut\u0173 abiej\u0173 tip\u0173 informacij\u0105.<\/p>\n

\u0160is metodas pasirod\u0117 es\u0105s labai efektyvus, pasiekdamas modern\u0173 tikslum\u0105 trijuose pagrindiniuose bepilo\u010di\u0173 orlaivi\u0173 (UAV) pagrindu sukurtuose HSI duomen\u0173 rinkiniuose. Pavyzd\u017eiui, WHU-Hi-LongKou duomen\u0173 rinkinyje CMTNet pasiek\u0117 99,58% bendr\u0105 tikslum\u0105 (OA), 0,19% pranokdamas ankstesn\u012f geriausi\u0105 model\u012f.<\/p>\n

Tradicinio hiperspektrinio vaizdavimo i\u0161\u0161\u016bkiai \u017eem\u0117s \u016bkio klasifikacijoje<\/h2>\n

Ankstyvieji hiperspektrini\u0173 duomen\u0173 analiz\u0117s metodai da\u017enai buvo sutelkti \u012f spektrinius arba erdvinius po\u017eymius, tod\u0117l rezultatai buvo nei\u0161sam\u016bs. Spektriniai metodai, tokie kaip pagrindini\u0173 komponen\u010di\u0173 analiz\u0117 (PCA), suma\u017eino duomen\u0173 sud\u0117tingum\u0105, sutelkdami d\u0117mes\u012f \u012f bangos ilgio informacij\u0105, ta\u010diau ignoruodami erdvinius ry\u0161ius tarp pikseli\u0173.<\/p>\n

Pavyzd\u017eiui, PCA transformuoja daugiama\u010dius spektrinius duomenis \u012f ma\u017eiau komponent\u0173, kurie paai\u0161kina did\u017eiausi\u0105 dispersij\u0105, taip supaprastindami analiz\u0119. Ta\u010diau \u0161is metodas neatsi\u017evelgia \u012f erdvin\u012f kontekst\u0105, pavyzd\u017eiui, pas\u0117li\u0173 i\u0161d\u0117stym\u0105 lauke. Prie\u0161ingai, erdviniai metodai, pavyzd\u017eiui, matematiniai morfologijos operatoriai, i\u0161ry\u0161kino fizinio pas\u0117li\u0173 i\u0161d\u0117stymo modelius, bet ignoravo svarbiausias spektrines detales.<\/p>\n

Matematin\u0117 morfologija naudoja tokias operacijas kaip i\u0161pl\u0117timas ir erozija, kad i\u0161 vaizd\u0173 i\u0161skirt\u0173 formas ir strukt\u016bras, pavyzd\u017eiui, ribas tarp lauk\u0173. Laikui b\u0117gant, konvoliuciniai neuroniniai tinklai (CNN) pagerino klasifikavim\u0105 apdorodami abiej\u0173 tip\u0173 duomenis.<\/p>\n

Ta\u010diau j\u0173 fiksuoti receptyv\u016bs laukai \u2013 vaizdo plotas, kur\u012f tinklas gali \u201cmatyti\u201d vienu metu \u2013 ribojo j\u0173 geb\u0117jim\u0105 u\u017efiksuoti tolimojo nuotolio priklausomybes. Pavyzd\u017eiui, 3D CNN gali b\u016bti sunku atskirti dvi soj\u0173 pupeli\u0173 veisles, turin\u010dias pana\u0161ius spektrinius profilius, bet skirtingus augimo modelius dideliame lauke.<\/p>\n

\u201eTransformers\u201c \u2013 neuroninio tinklo tipas, i\u0161 prad\u017ei\u0173 sukurtas nat\u016bralios kalbos apdorojimui, pasi\u016bl\u0117 \u0161ios problemos sprendim\u0105. Naudodami sav\u0119s d\u0117mesio mechanizmus, \u201eTransformers\u201c puikiai modeliuoja globalius duomen\u0173 ry\u0161ius. Sav\u0119s d\u0117mesys leid\u017eia modeliui \u012fvertinti skirting\u0173 \u012fvesties sekos dali\u0173 svarb\u0105, tod\u0117l jis gali sutelkti d\u0117mes\u012f \u012f atitinkamus regionus (pvz., sergan\u010di\u0173 augal\u0173 grup\u0119), ignoruodamas triuk\u0161m\u0105 (pvz., debes\u0173 \u0161e\u0161\u0117lius).<\/p>\n

Vis d\u0117lto jie da\u017enai nepastebi smulki\u0173 vietini\u0173 detali\u0173, toki\u0173 kaip lap\u0173 kra\u0161tai ar dirvo\u017eemio \u012ftr\u016bkimai. Hibridiniai modeliai, tokie kaip CTMixer, band\u0117 sujungti CNN ir Transformers, ta\u010diau tai dar\u0117 nuosekliai, pirmiausia apdorodami vietinius po\u017eymius, o v\u0117liau \u2013 globalius. Toks metodas l\u0117m\u0117 neefektyv\u0173 informacijos suliejim\u0105 ir neoptimal\u0173 na\u0161um\u0105 sud\u0117tingoje \u017eem\u0117s \u016bkio aplinkoje.<\/p>\n

Kaip veikia CMTNet: vietini\u0173 ir pasaulini\u0173 funkcij\u0173 sujungimas<\/h2>\n

\u201eCMTNet\u201c \u012fveikia \u0161iuos apribojimus naudodama unikali\u0105 trij\u0173 dali\u0173 architekt\u016br\u0105, skirt\u0105 efektyviai i\u0161gauti ir sujungti spektrinius-erdvinius, vietinius ir globalius elementus.<\/p>\n

1.<\/strong> Pirmasis komponentas, Spektrinio-erdvinio po\u017eymi\u0173 i\u0161skyrimo modulis<\/strong>, apdoroja neapdorotus HSI duomenis naudodamas 3D ir 2D konvoliucinius sluoksnius.<\/p>\n

Trima\u010diai konvoliuciniai sluoksniai vienu metu analizuoja tiek erdvinius (auk\u0161tis \u00d7 plotis), tiek spektrinius (bangos ilgis) matmenis, fiksuodami tokius modelius kaip konkre\u010di\u0173 bangos ilgi\u0173 atspindys per pas\u0117li\u0173 laj\u0105. Pavyzd\u017eiui, trimatis gr\u016bdas gali aptikti, kad sveiki kukur\u016bzai vir\u0161utiniuose lapuose atspindi daugiau artimojo infraraudonojo spinduliavimo \u0161viesos, palyginti su apatiniais.<\/p>\n

Tada 2D sluoksniai patikslina \u0161ias savybes, sutelkdami d\u0117mes\u012f \u012f erdvines detales, tokias kaip augal\u0173 i\u0161sid\u0117stymas lauke. \u0160is dviej\u0173 pakop\u0173 procesas u\u017etikrina, kad b\u016bt\u0173 i\u0161saugota ir spektrin\u0117 \u012fvairov\u0117 (pvz., chlorofilo kiekis), ir erdvinis kontekstas (pvz., tarpai tarp eili\u0173).<\/p>\n

2.<\/strong> Antrasis komponentas, vietinis-globalus po\u017eymi\u0173 i\u0161skyrimo modulis<\/strong>, veikia lygiagre\u010diai. Viena \u0161aka naudoja CNN, kad sutelkt\u0173 d\u0117mes\u012f \u012f vietines detales, tokias kaip atskir\u0173 lap\u0173 tekst\u016bra ar dirvo\u017eemio plot\u0173 forma. \u0160ios savyb\u0117s yra labai svarbios norint nustatyti r\u016b\u0161is, turin\u010dias pana\u0161ius spektrinius profilius, pavyzd\u017eiui, skirtingas soj\u0173 pupeli\u0173 veisles.<\/p>\n

Kita \u0161aka naudoja transformatorius, kad modeliuot\u0173 pasaulinius santykius, pavyzd\u017eiui, kaip pas\u0117liai pasiskirsto dideliuose plotuose arba kaip netoliese esan\u010di\u0173 med\u017ei\u0173 \u0161e\u0161\u0117liai veikia spektrinius rodmenis. Apdorodama \u0161iuos elementus vienu metu, o ne nuosekliai, CMTNet i\u0161vengia informacijos praradimo, kuris kamavo ankstesnius hibridinius modelius.<\/p>\n

Pavyzd\u017eiui, nors CNN \u0161aka identifikuoja nelygius medviln\u0117s lap\u0173 kra\u0161tus, \u201eTransformer\u201c \u0161aka atpa\u017e\u012fsta, kad \u0161ie lapai yra didesnio medviln\u0117s lauko, apsupto sezam\u0173 augal\u0173, dalis.<\/p>\n

3.<\/strong> Tre\u010diasis komponentas, keli\u0173 i\u0161\u0117jim\u0173 apribojim\u0173 modulis<\/strong>, u\u017etikrina subalansuot\u0105 mokym\u0105si tarp vietini\u0173, globali\u0173 ir sujungt\u0173 po\u017eymi\u0173. Mokymo metu kiekvienam po\u017eymi\u0173 tipui taikomos atskiros nuostoli\u0173 funkcijos, tod\u0117l tinklas yra priverstas tobulinti visus savo supratimo aspektus.<\/p>\n

Nuostoli\u0173 funkcija kiekybi\u0161kai \u012fvertina skirtum\u0105 tarp prognozuot\u0173 ir faktini\u0173 ver\u010di\u0173, vadovaudamasi modelio koregavimais. Pavyzd\u017eiui, vietini\u0173 po\u017eymi\u0173 praradimas gali nubausti model\u012f u\u017e neteising\u0105 lap\u0173 kra\u0161t\u0173 klasifikavim\u0105, o bendras praradimas i\u0161taiso didelio masto pas\u0117li\u0173 pasiskirstymo klaidas.<\/p>\n

\u0160ie nuostoliai sujungiami naudojant svorius, optimizuotus atsitiktin\u0117s paie\u0161kos b\u016bdu \u2013 tai technika, kuri i\u0161bando \u012fvairius svori\u0173 derinius, siekiant maksimaliai padidinti tikslum\u0105. \u0160is procesas sukuria patikim\u0105 ir pritaikom\u0105 model\u012f, galint\u012f apdoroti \u012fvairius \u017eem\u0117s \u016bkio scenarijus.<\/p>\n

CMTNet na\u0161umo vertinimas bepilo\u010di\u0173 orlaivi\u0173 hiperspektriniuose duomen\u0173 rinkiniuose<\/h2>\n

Nor\u0117dami \u012fvertinti CMTNet, tyr\u0117jai j\u012f i\u0161band\u0117 su trimis bepilo\u010diais orlaiviais (UAV) gautais hiperspektriniais duomen\u0173 rinkiniais i\u0161 Uhano universiteto. \u0160ie duomen\u0173 rinkiniai yra pla\u010diai naudojami nuotolinio steb\u0117jimo etalonas d\u0117l savo auk\u0161tos kokyb\u0117s ir \u012fvairov\u0117s:<\/p>\n

    \n
  1. WHU-Hi-LongKou<\/strong>\u0160is duomen\u0173 rinkinys apima 550 \u00d7 400 pikseli\u0173 su 270 spektrini\u0173 juost\u0173 ir 0,463 metro erdvine skiriam\u0105ja geba. 0,463 metro erdvin\u0117 skiriamoji geba rei\u0161kia, kad kiekvienas pikselis atitinka 0,463 m \u00d7 0,463 m plot\u0105 ant \u017eem\u0117s, leid\u017eiant\u012f identifikuoti atskirus augalus. Jame yra devyni\u0173 r\u016b\u0161i\u0173 pas\u0117liai, tokie kaip kukur\u016bzai, medviln\u0117 ir ry\u017eiai, su 1 019 mokomaisiais pavyzd\u017eiais ir 203 523 bandomaisiais pavyzd\u017eiais.<\/li>\n
  2. WHU-Hi-Hanchuan<\/strong>\u0160iame 1 217 \u00d7 303 pikseli\u0173 rai\u0161kos ir 0,109 metro skiriamosios gebos duomen\u0173 rinkinyje yra 16 \u017eem\u0117s dangos tip\u0173, \u012fskaitant bra\u0161kes, soj\u0173 pupeles ir plastiko lak\u0161tus. Didesn\u0117 skiriamoji geba (0,109 m) leid\u017eia gauti smulkesnes detales, pavyzd\u017eiui, atskirti jaunus ir subrendusius soj\u0173 pupeli\u0173 augalus. Mokymo ir bandym\u0173 imtys i\u0161 viso sudar\u0117 atitinkamai 1 289 ir 256 241.<\/li>\n
  3. WHU-Hi-HongHu<\/strong>\u0160is didel\u0117s skiriamosios gebos (0,043 metro) duomen\u0173 rinkinys, turintis 940 \u00d7 475 pikseli\u0173 ir 270 juost\u0173, apima 22 klases, tokias kaip medviln\u0117, rapsai ir \u010desnak\u0173 daigai. Esant 0,043 m skiriamajai gebai, matomi atskiri lapai ir dirvo\u017eemio \u012ftr\u016bkimai, tod\u0117l jis idealiai tinka smulkiagr\u016bd\u017eiui klasifikavimui. Jame yra 1 925 mokymo pavyzd\u017eiai ir 384 678 bandomieji pavyzd\u017eiai.<\/li>\n<\/ol>\n

    \"Didel\u0117s<\/p>\n

    Modelis buvo apmokytas NVIDIA TITAN Xp GPU naudojant \u201ePyTorch\u201c, mokymosi sparta buvo 0,001, o paketo dydis \u2013 100. Mokymosi sparta lemia, kiek modelis koreguoja savo parametrus mokymo metu \u2013 per didel\u0117 sparta gali vir\u0161yti optimalias vertes; per ma\u017ea sparta gali pad\u0117ti apmokyti dar sunkiau.<\/p>\n

    Kiekvienas eksperimentas buvo pakartotas de\u0161imt kart\u0173, siekiant u\u017etikrinti patikimum\u0105, o \u012fvesties fragmentai \u2013 ma\u017ei viso vaizdo segmentai \u2013 buvo optimizuoti iki 13 \u00d7 13 pikseli\u0173 naudojant tinklelio paie\u0161k\u0105 \u2013 metod\u0105, kuris tikrina skirtingus fragment\u0173 dyd\u017eius, kad b\u016bt\u0173 rastas efektyviausias.<\/p>\n

    CMTNet pasiekia modern\u0173 tikslum\u0105 pas\u0117li\u0173 klasifikavime<\/h2>\n

    \u201eCMTNet\u201c pasiek\u0117 puiki\u0173 rezultat\u0173 visuose duomen\u0173 rinkiniuose, pranokdama esamus metodus tiek bendru tikslumu (OA), tiek konkre\u010dioms klas\u0117ms b\u016bdingu na\u0161umu. OA matuoja teisingai klasifikuot\u0173 pikseli\u0173 procentin\u0119 dal\u012f visose klas\u0117se, o vidutinis tikslumas (AA) apskai\u010diuoja vidutin\u012f tikslum\u0105 kiekvienoje klas\u0117je, pa\u0161alindamas disbalans\u0105.<\/p>\n

    \u201eWHU-Hi-LongKou\u201c duomen\u0173 rinkinyje \u201eCMTNet\u201c pasiek\u0117 99,58% OA, 0,19% vir\u0161ydamas \u201eCTMixer\u201c. Sud\u0117tingose klas\u0117se su ribotais mokymo duomenimis, pavyzd\u017eiui, medviln\u0117s (41 pavyzdys), \u201eCMTNet\u201c vis tiek pasiek\u0117 99,53% tikslum\u0105. Pana\u0161iai ir \u201eWHU-Hi-HanChuan\u201c duomen\u0173 rinkinyje jis pagerino arb\u016bz\u0173 (22 pavyzd\u017ei\u0173) tikslum\u0105 nuo 82,42% iki 96,11%, parodydamas geb\u0117jim\u0105 apdoroti nesubalansuotus duomenis naudojant efektyv\u0173 po\u017eymi\u0173 suliejim\u0105.<\/p>\n

    Vizualiai palyginus klasifikavimo \u017eem\u0117lapius, pasteb\u0117ta ma\u017eiau suskaidyt\u0173 plot\u0173 ir lygesn\u0117s ribos tarp lauk\u0173, palyginti su tokiais modeliais kaip 3D-CNN ir \u201eVision Transformer\u201c (ViT). Pavyzd\u017eiui, \u0161e\u0161\u0117li\u0173 paveiktame WHU-Hi-HanChuan duomen\u0173 rinkinyje CMTNet suma\u017eino d\u0117l \u017eemo saul\u0117s kampo atsirandan\u010dias klaidas, o \u201eResNet\u201c neteisingai klasifikavo soj\u0173 pupeles kaip pilkus stogus.<\/p>\n

    \"CMTNet<\/p>\n

    \u0160e\u0161\u0117liai kelia unikal\u0173 i\u0161\u0161\u016bk\u012f, nes jie kei\u010dia spektrinius para\u0161us \u2013 \u0161e\u0161\u0117lyje esantis soj\u0173 pupeli\u0173 augalas gali atspind\u0117ti ma\u017eiau artimojo infraraudonojo spinduliavimo \u0161viesos, primindamas neaugalij\u0105. Pasitelkdama pasaulin\u012f kontekst\u0105, CMTNet atpa\u017eino, kad \u0161ie \u0161e\u0161\u0117lyje esantys augalai yra didesnio soj\u0173 pupeli\u0173 lauko dalis, taip suma\u017eindama paklaidas.<\/p>\n

    WHU-Hi-HongHu duomen\u0173 rinkinyje modelis puikiai atskyr\u0117 spektriniu po\u017ei\u016briu pana\u0161ius augalus, pvz., skirtingas bastini\u0173 dar\u017eovi\u0173 veisles, pasiekdamas 96,54% tikslum\u0105.\u00a0Brassica parachinensis<\/em>.<\/p>\n

    Abliacijos tyrimai \u2013 eksperimentai, kuri\u0173 metu pa\u0161alinami komponentai, siekiant \u012fvertinti j\u0173 poveik\u012f \u2013 patvirtino kiekvieno modulio svarb\u0105. Vien tik keli\u0173 i\u0161\u0117jim\u0173 apribojim\u0173 modulio prid\u0117jimas padidino OA 1,52% WHU-Hi-HongHu sistemoje, pabr\u0117\u017edamas jo vaidmen\u012f tobulinant po\u017eymi\u0173 suliejim\u0105. Be \u0161io modulio vietiniai ir global\u016bs po\u017eymiai buvo derinami atsitiktinai, tod\u0117l klasifikacijos buvo nenuoseklios.<\/p>\n

    Skai\u010diavimo kompromisai ir praktiniai aspektai<\/h2>\n

    Nors CMTNet tikslumas yra neprilygstamas, jo skai\u010diavimo s\u0105naudos yra didesn\u0117s nei tradicini\u0173 metod\u0173. Mokymasis WHU-Hi-HongHu duomen\u0173 rinkinyje u\u017etruko 1 885 sekundes, palyginti su 74 sekund\u0117mis, kai buvo naudojamas \u201eRandom Forest\u201c (RF) \u2013 ma\u0161ininio mokymosi algoritmas, kuris mokymo metu kuria sprendim\u0173 med\u017eius.<\/p>\n

    Ta\u010diau \u0161is kompromisas yra pateisinamas tiksliojoje \u017eemdirbyst\u0117je, kur tikslumas tiesiogiai veikia derliaus prognozes ir i\u0161tekli\u0173 paskirstym\u0105. Pavyzd\u017eiui, neteisingai priskyrus sergant\u012f pas\u0117l\u012f sveikam, gali kilti nekontroliuojami kenk\u0117j\u0173 protr\u016bkiai, nuniokojantys i\u0161tisus laukus.<\/p>\n

    Realaus laiko taikymams ateityje b\u016bt\u0173 galima i\u0161tirti modeli\u0173 glaudinimo metodus, tokius kaip perteklini\u0173 neuron\u0173 gen\u0117jimas arba svori\u0173 kvantavimas (suma\u017einant skaitmenin\u012f tikslum\u0105), siekiant sutrumpinti vykdymo laik\u0105 neprarandant na\u0161umo. Gen\u0117jimas pa\u0161alina ma\u017eiau svarbias jungtis i\u0161 neuroninio tinklo, pana\u0161iai kaip med\u017eio \u0161ak\u0173 gen\u0117jimas siekiant pagerinti jo form\u0105, o kvantavimas supaprastina skaitmeninius skai\u010diavimus, pagreitindamas apdorojim\u0105.<\/p>\n

    Hiperspektrinio pas\u0117li\u0173 klasifikavimo ateitis naudojant CMTNet<\/h2>\n

    Nepaisant s\u0117km\u0117s, CMTNet susiduria su apribojimais. Na\u0161umas \u0161iek tiek suprast\u0117ja stipriai \u0161e\u0161\u0117liuotose srityse, kaip matyti WHU-Hi-HanChuan duomen\u0173 rinkinyje (97.29% OA ir 99.58% gerai ap\u0161viestoje LongKou planetoje). \u0160e\u0161\u0117liai apsunkina klasifikavim\u0105, nes suma\u017eina atspind\u0117tos \u0161viesos intensyvum\u0105, pakeisdami spektrinius profilius.<\/p>\n

    Be to, klas\u0117s su itin ma\u017eais mokomaisiais pavyzd\u017eiais, pavyzd\u017eiui, siauralap\u0117s sojos pupel\u0117s (20 pavyzd\u017ei\u0173), atsilieka nuo t\u0173, kuri\u0173 duomen\u0173 gausu. Ma\u017eas im\u010di\u0173 dydis riboja modelio geb\u0117jim\u0105 i\u0161mokti \u012fvairius variantus, pavyzd\u017eiui, lap\u0173 formos skirtumus d\u0117l dirvo\u017eemio kokyb\u0117s.<\/p>\n

    B\u016bsimuose tyrimuose b\u016bt\u0173 galima integruoti multimodalinius duomenis, tokius kaip LiDAR auk\u0161\u010dio \u017eem\u0117lapiai arba terminis vaizdavimas, siekiant pagerinti atsparum\u0105 \u0161e\u0161\u0117liams ir u\u017etemimams. LiDAR (\u0161viesos aptikimas ir diapazono nustatymas) naudoja lazerio impulsus 3D reljefo modeliams kurti, kurie, analizuojant auk\u0161\u010dio skirtumus, gal\u0117t\u0173 pad\u0117ti atskirti pas\u0117lius nuo \u0161e\u0161\u0117li\u0173.<\/p>\n

    Be to, terminis vaizdavimas fiksuoja \u0161ilumos para\u0161us, suteikdamas papildom\u0173 u\u017euomin\u0173 apie augal\u0173 sveikat\u0105 \u2013 stres\u0105 patiriantys augalai da\u017enai turi auk\u0161tesn\u0119 laj\u0173 temperat\u016br\u0105 d\u0117l suma\u017e\u0117jusios transpiracijos. Pusiau pri\u017ei\u016brimo mokymosi metodai, kurie naudoja nepa\u017eym\u0117tus duomenis (pvz., bepilo\u010di\u0173 orlaivi\u0173 vaizdus be rankini\u0173 anotacij\u0173), taip pat gali pagerinti ret\u0173 augal\u0173 r\u016b\u0161i\u0173 na\u0161um\u0105.<\/p>\n

    Naudodami nuoseklumo reguliavim\u0105 \u2013 apmokydami model\u012f gauti stabilias prognozes \u0161iek tiek pakeistose to paties vaizdo versijose \u2013 tyr\u0117jai gali panaudoti nepa\u017eym\u0117tus duomenis, kad pagerint\u0173 apibendrinim\u0105.<\/p>\n

    Galiausiai, CMTNet diegimas periferiniuose \u012frenginiuose, tokiuose kaip dronai su integruotais GPU, gal\u0117t\u0173 sudaryti s\u0105lygas steb\u0117ti realiuoju laiku nuotoliniuose laukuose. Diegimas periferiniuose \u012frenginiuose suma\u017eina priklausomyb\u0119 nuo debes\u0173 kompiuterijos, suma\u017eindamas dels\u0105 ir duomen\u0173 perdavimo i\u0161laidas. Ta\u010diau tam reikia optimizuoti model\u012f, atsi\u017evelgiant \u012f ribot\u0105 atmint\u012f ir apdorojimo gali\u0105, galb\u016bt naudojant lengvas architekt\u016bras, tokias kaip \u201cMobileNet\u201d arba \u017eini\u0173 distiliavim\u0105, kai ma\u017eesnis \u201cstudento\u201d modelis imituoja didesn\u012f \u201emokytojo\u201c model\u012f.<\/p>\n

    I\u0161vada<\/h2>\n

    CMTNet yra reik\u0161mingas \u017eingsnis \u012f priek\u012f hiperspektrinio augal\u0173 klasifikavimo srityje. Suderinus CNN ir Transformers, jis sprend\u017eia ilgalaikius objekt\u0173 i\u0161skyrimo ir suliejimo i\u0161\u0161\u016bkius, suteikdamas \u016bkininkams ir agronomams galing\u0105 tiksliosios \u017eemdirbyst\u0117s \u012frank\u012f.<\/p>\n

    Taikymo sritys apima nuo lig\u0173 aptikimo realiuoju laiku iki dr\u0117kinimo grafik\u0173 optimizavimo \u2013 visa tai yra labai svarbu tvariam \u016bkininkavimui klimato kaitos ir gyventoj\u0173 skai\u010diaus augimo metu. Kadangi bepilo\u010di\u0173 orlaivi\u0173 technologijos tampa vis labiau prieinamos, tokie modeliai kaip CMTNet atliks lemiam\u0105 vaidmen\u012f u\u017etikrinant pasaulin\u012f apr\u016bpinim\u0105 maistu.<\/p>\n

    B\u016bsimi patobulinimai, tokie kaip lengvesn\u0117s architekt\u016bros ir multimodalin\u0117 duomen\u0173 sintez\u0117, gal\u0117t\u0173 dar labiau padidinti j\u0173 prakti\u0161kum\u0105. Nuolat diegiant inovacijas, CMTNet gal\u0117t\u0173 tapti i\u0161mani\u0173j\u0173 \u016bkininkavimo sistem\u0173 visame pasaulyje kertiniu akmeniu, u\u017etikrinan\u010diu efektyv\u0173 \u017eem\u0117s naudojim\u0105 ir atspari\u0105 maisto gamyb\u0105 ateities kartoms.<\/p>\n

    Nuoroda: <\/strong>Guo, X., Feng, Q. ir Guo, F. CMTNet: hibridinis CNN transformatori\u0173 tinklas, skirtas bepilo\u010di\u0173 orlaivi\u0173 (UAV) pagr\u012fstam hiperspektriniam pas\u0117li\u0173 klasifikavimui tiksliojoje \u017eemdirbyst\u0117je. Sci Rep 15, 12383 (2025). https:\/\/doi.org\/10.1038\/s41598-025-97052-w<\/a><\/p>","protected":false},"excerpt":{"rendered":"

    Tikslus pas\u0117li\u0173 klasifikavimas yra b\u016btinas \u0161iuolaikinei tiksliajai \u017eemdirbystei, leid\u017eian\u010diai \u016bkininkams steb\u0117ti pas\u0117li\u0173 sveikat\u0105, prognozuoti derli\u0173 ir efektyviai paskirstyti i\u0161teklius. Ta\u010diau tradiciniai metodai da\u017enai\u2026<\/p>","protected":false},"author":210157960,"featured_media":11556,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_coblocks_attr":"","_coblocks_dimensions":"","_coblocks_responsive_height":"","_coblocks_accordion_ie_support":"","_eb_attr":"","content-type":"","_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_wpcom_ai_launchpad_first_post":false,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"{title}\n\n{excerpt}\n\n{url}","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2},"_wpas_customize_per_network":false,"jetpack_post_was_ever_published":false},"categories":[1657,1377],"tags":[],"class_list":["post-11549","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-precision-farming","category-crop-monitoring"],"acf":[],"yoast_head":"\nCMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification - GeoPard Agriculture<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/geopard.tech\/lt\/tinklarastis\/cmtnet-is-naujo-apibrezia-tiksluji-ukininkavima-pranokdama-tradicine-paseliu-klasifikacija\/\" \/>\n<meta property=\"og:locale\" content=\"lt_LT\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification - GeoPard Agriculture\" \/>\n<meta property=\"og:description\" content=\"Accurate crop classification is essential for modern precision agriculture, enabling farmers to monitor crop health, predict yields, and allocate resources efficiently. Traditional methods, however, often...\" \/>\n<meta property=\"og:url\" content=\"https:\/\/geopard.tech\/lt\/tinklarastis\/cmtnet-is-naujo-apibrezia-tiksluji-ukininkavima-pranokdama-tradicine-paseliu-klasifikacija\/\" \/>\n<meta property=\"og:site_name\" content=\"GeoPard - Precision agriculture Mapping software\" \/>\n<meta property=\"article:publisher\" content=\"https:\/\/www.facebook.com\/geopardAgriculture\/\" \/>\n<meta property=\"article:published_time\" content=\"2025-05-04T19:00:25+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification-1024x576.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1024\" \/>\n\t<meta property=\"og:image:height\" content=\"576\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"dementievgeopard\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@geopardagri\" \/>\n<meta name=\"twitter:site\" content=\"@geopardagri\" \/>\n<meta name=\"twitter:label1\" content=\"Written by\" \/>\n\t<meta name=\"twitter:data1\" content=\"dementievgeopard\" \/>\n\t<meta name=\"twitter:label2\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 minu\u010di\u0173\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/\"},\"author\":{\"name\":\"dementievgeopard\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/#\\\/schema\\\/person\\\/dd217733c742620adc57befbbcd84a8a\"},\"headline\":\"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification\",\"datePublished\":\"2025-05-04T19:00:25+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/\"},\"wordCount\":2064,\"commentCount\":0,\"publisher\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/#organization\"},\"image\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i0.wp.com\\\/geopard.tech\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1\",\"articleSection\":[\"Precision Farming\",\"Crop monitoring\"],\"inLanguage\":\"lt-LT\",\"potentialAction\":[{\"@type\":\"CommentAction\",\"name\":\"Comment\",\"target\":[\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#respond\"]}]},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/\",\"url\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/\",\"name\":\"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification - GeoPard Agriculture\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/i0.wp.com\\\/geopard.tech\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1\",\"datePublished\":\"2025-05-04T19:00:25+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#breadcrumb\"},\"inLanguage\":\"lt-LT\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"lt-LT\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#primaryimage\",\"url\":\"https:\\\/\\\/i0.wp.com\\\/geopard.tech\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1\",\"contentUrl\":\"https:\\\/\\\/i0.wp.com\\\/geopard.tech\\\/wp-content\\\/uploads\\\/2025\\\/05\\\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1\",\"width\":3600,\"height\":2025},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/blog\\\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/geopard.tech\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/#website\",\"url\":\"https:\\\/\\\/geopard.tech\\\/\",\"name\":\"GeoPard - Precision agriculture software\",\"description\":\"Precision agriculture Mapping software\",\"publisher\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/#organization\"},\"alternateName\":\"GeoPard\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/geopard.tech\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"lt-LT\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/#organization\",\"name\":\"GeoPard Agriculture\",\"alternateName\":\"GeoPard\",\"url\":\"https:\\\/\\\/geopard.tech\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"lt-LT\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"https:\\\/\\\/i0.wp.com\\\/geopard.tech\\\/wp-content\\\/uploads\\\/2026\\\/07\\\/geopard_logo_transparentbackground.png?fit=512%2C68&ssl=1\",\"contentUrl\":\"https:\\\/\\\/i0.wp.com\\\/geopard.tech\\\/wp-content\\\/uploads\\\/2026\\\/07\\\/geopard_logo_transparentbackground.png?fit=512%2C68&ssl=1\",\"width\":512,\"height\":68,\"caption\":\"GeoPard Agriculture\"},\"image\":{\"@id\":\"https:\\\/\\\/geopard.tech\\\/#\\\/schema\\\/logo\\\/image\\\/\"},\"sameAs\":[\"https:\\\/\\\/www.facebook.com\\\/geopardAgriculture\\\/\",\"https:\\\/\\\/x.com\\\/geopardagri\",\"https:\\\/\\\/www.linkedin.com\\\/company\\\/geopard-agriculture\\\/\",\"https:\\\/\\\/www.instagram.com\\\/geopardagriculture\\\/\",\"https:\\\/\\\/www.youtube.com\\\/channel\\\/UCiaPGLAhRPNh-s85dXdC-Sw\",\"https:\\\/\\\/www.g2.com\\\/products\\\/geopard-agriculture-precision-farming-software\\\/reviews\"]},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/geopard.tech\\\/#\\\/schema\\\/person\\\/dd217733c742620adc57befbbcd84a8a\",\"name\":\"dementievgeopard\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"lt-LT\",\"@id\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/7c09adac9c20b00454199df5a41ea67c812a6c97e948bf109899836ddde31354?s=96&d=identicon&r=g\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/7c09adac9c20b00454199df5a41ea67c812a6c97e948bf109899836ddde31354?s=96&d=identicon&r=g\",\"contentUrl\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/7c09adac9c20b00454199df5a41ea67c812a6c97e948bf109899836ddde31354?s=96&d=identicon&r=g\",\"caption\":\"dementievgeopard\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"CMTNet i\u0161 naujo apibr\u0117\u017eia tiksli\u0105j\u0105 \u017eemdirbyst\u0119, pranokdama tradicin\u0119 pas\u0117li\u0173 klasifikacij\u0105 \u2013 \u201eGeoPard Agriculture\u201c","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/geopard.tech\/lt\/tinklarastis\/cmtnet-is-naujo-apibrezia-tiksluji-ukininkavima-pranokdama-tradicine-paseliu-klasifikacija\/","og_locale":"lt_LT","og_type":"article","og_title":"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification - GeoPard Agriculture","og_description":"Accurate crop classification is essential for modern precision agriculture, enabling farmers to monitor crop health, predict yields, and allocate resources efficiently. Traditional methods, however, often...","og_url":"https:\/\/geopard.tech\/lt\/tinklarastis\/cmtnet-is-naujo-apibrezia-tiksluji-ukininkavima-pranokdama-tradicine-paseliu-klasifikacija\/","og_site_name":"GeoPard - Precision agriculture Mapping software","article_publisher":"https:\/\/www.facebook.com\/geopardAgriculture\/","article_published_time":"2025-05-04T19:00:25+00:00","og_image":[{"width":1024,"height":576,"url":"https:\/\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification-1024x576.png","type":"image\/png"}],"author":"dementievgeopard","twitter_card":"summary_large_image","twitter_creator":"@geopardagri","twitter_site":"@geopardagri","twitter_misc":{"Written by":"dementievgeopard","Est. reading time":"10 minu\u010di\u0173"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#article","isPartOf":{"@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/"},"author":{"name":"dementievgeopard","@id":"https:\/\/geopard.tech\/#\/schema\/person\/dd217733c742620adc57befbbcd84a8a"},"headline":"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification","datePublished":"2025-05-04T19:00:25+00:00","mainEntityOfPage":{"@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/"},"wordCount":2064,"commentCount":0,"publisher":{"@id":"https:\/\/geopard.tech\/#organization"},"image":{"@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#primaryimage"},"thumbnailUrl":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1","articleSection":["Precision Farming","Crop monitoring"],"inLanguage":"lt-LT","potentialAction":[{"@type":"CommentAction","name":"Comment","target":["https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#respond"]}]},{"@type":"WebPage","@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/","url":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/","name":"CMTNet i\u0161 naujo apibr\u0117\u017eia tiksli\u0105j\u0105 \u017eemdirbyst\u0119, pranokdama tradicin\u0119 pas\u0117li\u0173 klasifikacij\u0105 \u2013 \u201eGeoPard Agriculture\u201c","isPartOf":{"@id":"https:\/\/geopard.tech\/#website"},"primaryImageOfPage":{"@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#primaryimage"},"image":{"@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#primaryimage"},"thumbnailUrl":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1","datePublished":"2025-05-04T19:00:25+00:00","breadcrumb":{"@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#breadcrumb"},"inLanguage":"lt-LT","potentialAction":[{"@type":"ReadAction","target":["https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/"]}]},{"@type":"ImageObject","inLanguage":"lt-LT","@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#primaryimage","url":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1","contentUrl":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1","width":3600,"height":2025},{"@type":"BreadcrumbList","@id":"https:\/\/geopard.tech\/blog\/cmtnet-redefines-precision-agriculture-by-outperforming-traditional-crop-classification\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/geopard.tech\/"},{"@type":"ListItem","position":2,"name":"CMTNet Redefines Precision Agriculture By Outperforming Traditional Crop Classification"}]},{"@type":"WebSite","@id":"https:\/\/geopard.tech\/#website","url":"https:\/\/geopard.tech\/","name":"GeoPard \u2013 tiksliosios \u017eemdirbyst\u0117s programin\u0117 \u012franga","description":"Tiksliosios \u017eemdirbyst\u0117s kartografavimo programin\u0117 \u012franga","publisher":{"@id":"https:\/\/geopard.tech\/#organization"},"alternateName":"GeoPard","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/geopard.tech\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"lt-LT"},{"@type":"Organization","@id":"https:\/\/geopard.tech\/#organization","name":"GeoPard \u017dem\u0117s \u016bkis","alternateName":"GeoPard","url":"https:\/\/geopard.tech\/","logo":{"@type":"ImageObject","inLanguage":"lt-LT","@id":"https:\/\/geopard.tech\/#\/schema\/logo\/image\/","url":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/07\/geopard_logo_transparentbackground.png?fit=512%2C68&ssl=1","contentUrl":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/07\/geopard_logo_transparentbackground.png?fit=512%2C68&ssl=1","width":512,"height":68,"caption":"GeoPard Agriculture"},"image":{"@id":"https:\/\/geopard.tech\/#\/schema\/logo\/image\/"},"sameAs":["https:\/\/www.facebook.com\/geopardAgriculture\/","https:\/\/x.com\/geopardagri","https:\/\/www.linkedin.com\/company\/geopard-agriculture\/","https:\/\/www.instagram.com\/geopardagriculture\/","https:\/\/www.youtube.com\/channel\/UCiaPGLAhRPNh-s85dXdC-Sw","https:\/\/www.g2.com\/products\/geopard-agriculture-precision-farming-software\/reviews"]},{"@type":"Person","@id":"https:\/\/geopard.tech\/#\/schema\/person\/dd217733c742620adc57befbbcd84a8a","name":"dementjevgeopardo","image":{"@type":"ImageObject","inLanguage":"lt-LT","@id":"https:\/\/secure.gravatar.com\/avatar\/7c09adac9c20b00454199df5a41ea67c812a6c97e948bf109899836ddde31354?s=96&d=identicon&r=g","url":"https:\/\/secure.gravatar.com\/avatar\/7c09adac9c20b00454199df5a41ea67c812a6c97e948bf109899836ddde31354?s=96&d=identicon&r=g","contentUrl":"https:\/\/secure.gravatar.com\/avatar\/7c09adac9c20b00454199df5a41ea67c812a6c97e948bf109899836ddde31354?s=96&d=identicon&r=g","caption":"dementievgeopard"}}]}},"jetpack_publicize_connections":[],"jetpack_likes_enabled":true,"jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/pdiCPa-30h","jetpack_featured_media_url":"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2025\/05\/CMTNet-Redefines-Precision-Agriculture-By-Outperforming-Traditional-Crop-Classification.png?fit=3600%2C2025&ssl=1","_links":{"self":[{"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/posts\/11549","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/users\/210157960"}],"replies":[{"embeddable":true,"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/comments?post=11549"}],"version-history":[{"count":0,"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/posts\/11549\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/media\/11556"}],"wp:attachment":[{"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/media?parent=11549"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/categories?post=11549"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/geopard.tech\/lt\/wp-json\/wp\/v2\/tags?post=11549"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}