\u0160\u012b kult\u016bra tiek izmantota p\u0101rtikas p\u0101rstr\u0101des r\u016bpniec\u012bb\u0101, svaigu aug\u013cu un d\u0101rze\u0146u tirgos un m\u0101jas d\u0101rzos visos kontinentos. Saska\u0146\u0101 ar Lauksaimniec\u012bbas m\u0101rketinga p\u0113t\u012bjumu centra datiem, tikai Amerikas Savienotaj\u0101s Valst\u012bs 2023. gad\u0101 svaigu aug\u013cu un d\u0101rze\u0146u tom\u0101tu ra\u017eo\u0161anas v\u0113rt\u012bba p\u0101rsniedza vair\u0101kus miljardus dol\u0101ru.<\/p>\n
Lapu slim\u012bbas ir galvenais drauds tom\u0101tu ra\u017e\u012bbai. P\u0101rtikas un lauksaimniec\u012bbas organiz\u0101cija (FAO) l\u0113\u0161, ka augu slim\u012bbas ir atbild\u012bgas par aptuveni 401\u00a0t\u016bksto\u0161i 3\u00a0t\u016bksto\u0161u ra\u017eas zaud\u0113jumu vis\u0101 pasaul\u0113<\/strong>, kas noz\u012bm\u0113 milz\u012bgas sekas p\u0101rtikas nodro\u0161in\u0101jumam un ekonomiskaj\u0101m sek\u0101m.<\/p>\n Vien s\u0113n\u012b\u0161u slim\u012bbas rada ikgad\u0113jus zaud\u0113jumus $60 miljardi vis\u0101 pasaul\u0113<\/strong>. \u012apa\u0161i tom\u0101tiem bakteri\u0101la plankumain\u012bba labv\u0113l\u012bgos slim\u012bbas apst\u0101k\u013cos var samazin\u0101t ra\u017eu par l\u012bdz 90%<\/strong>, padarot atkl\u0101\u0161anas un rea\u0123\u0113\u0161anas laiku kritiski svar\u012bgu.<\/p>\n Agr\u012bna un prec\u012bza slim\u012bbas identific\u0113\u0161ana ir iz\u0161\u0137iro\u0161ais faktors. Lauksaimnieks, kur\u0161 atkl\u0101j agr\u012bno puvi t\u0101s s\u0101kotn\u0113j\u0101 boj\u0101juma stadij\u0101, var to ierobe\u017eot ar m\u0113r\u0137tiec\u012bgu fungic\u012bdu lieto\u0161anu. Lauksaimnieks, kur\u0161 to nepamana, l\u012bdz par\u0101d\u0101s redzama defoli\u0101cija, saskaras ar ra\u017eas zaud\u0113jumiem, kurus nek\u0101da iejauk\u0161an\u0101s nevar piln\u012bb\u0101 nov\u0113rst. \u0160eit m\u0101ksl\u012bgais intelekts, \u012bpa\u0161i intelektiska tom\u0101tu lapu slim\u012bbu noteik\u0161ana, ko nodro\u0161ina dzi\u013c\u0101 m\u0101c\u012b\u0161an\u0101s, maina vien\u0101dojumu.<\/p>\n Pla\u0161\u0101ks konteksts apstiprina \u0161\u012bs p\u0101rmai\u0146as. Glob\u0101lais m\u0101ksl\u012bg\u0101 intelekta tirgus prec\u012bz\u0101s lauksaimniec\u012bbas jom\u0101 tika nov\u0113rt\u0113ts par $3,1 miljards 2024. gad\u0101<\/strong> un tiek prognoz\u0113ts, ka tas sasniegs $12,7 miljardi l\u012bdz 2034. gadam<\/strong> pie a CAGR 15,1%<\/strong> (Market.us, 2024). Inteli\u0123enta kult\u016braugu slim\u012bbu uzraudz\u012bba ir viens no visstrauj\u0101k augo\u0161ajiem segmentiem \u0161aj\u0101 papla\u0161in\u0101\u0161an\u0101s proces\u0101.<\/p>\n Pirms jebkura noteik\u0161anas sist\u0113ma var darboties, ir nepiecie\u0161ama prec\u012bza izpratne par to, ko tai l\u016bdz atrast. Tom\u0101tu lapas ietekm\u0113 pla\u0161s patog\u0113nu kl\u0101sts, katrs no tiem atst\u0101jot at\u0161\u0137ir\u012bgas, bet da\u017ereiz p\u0101rkl\u0101jo\u0161as vizu\u0101l\u0101s paz\u012bmes.<\/p>\n Agr\u012bna puve<\/strong>, ko izraisa s\u0113n\u012bte Alternaria solani<\/em>, uz vec\u0101k\u0101m lap\u0101m veido tum\u0161i br\u016bnus koncentriskus gredzenu boj\u0101jumus. Gredzeni atg\u0101dina m\u0113r\u0137a rakstu, un, audiem atmirstot, katru boj\u0101jumu ap\u0146em dzelt\u0113\u0161ana.<\/p>\n V\u0113la puve<\/strong>, ko izraisa oomic\u0113te Phytophthora infestans<\/em> \u2014 tas pats organisms, kas izrais\u012bja kartupe\u013cu badu \u012arij\u0101, \u2014 rada ar \u016bdeni pies\u016bcin\u0101tus, pel\u0113c\u012bgi za\u013cus plankumus, kas siltos, mitros apst\u0101k\u013cos \u0101tri k\u013c\u016bst br\u016bni. Tas izplat\u0101s \u0101rk\u0101rt\u012bgi \u0101tri un da\u017eu dienu laik\u0101 var izn\u012bcin\u0101t veselu lauku.<\/p>\n Septoria lapu plankums<\/strong> izpau\u017eas k\u0101 mazi, apa\u013ci plankumi ar tum\u0161i br\u016bn\u0101m mal\u0101m un gai\u0161\u0101k br\u016bnganu centru. Tas parasti s\u0101kas uz apak\u0161\u0113j\u0101m lap\u0101m un virz\u0101s uz aug\u0161u, ko izraisa s\u0113ne. Septoria lycopersici<\/em>.<\/p>\n Bakt\u0113riju plankums<\/strong>, ko izraisa Xanthomonas vesicatoria<\/em>, veido mazus, ar \u016bdeni pies\u016bcin\u0101tus plankumus, kas k\u013c\u016bst br\u016bni un st\u016braini, bie\u017ei vien ieskaujot dzeltenus oreolus. At\u0161\u0137ir\u012bb\u0101 no s\u0113n\u012b\u0161u plankumiem, bakt\u0113riju boj\u0101jumi nerea\u0123\u0113 uz \u0101rst\u0113\u0161anu ar fungic\u012bdiem.<\/p>\n Lapu pel\u0113jums<\/strong>, ko izraisa Passalora fulva<\/em>, par\u0101d\u0101s k\u0101 gai\u0161i za\u013ci vai dzelteni plankumi uz lapu aug\u0161\u0113j\u0101s virsmas, zem kuras aug ol\u012bvza\u013c\u0161 l\u012bdz pel\u0113c\u012bgi violets pel\u0113juma augums. Tas labi aug mitr\u0101 siltumn\u012bcu vid\u0113.<\/p>\n Tom\u0101tu moza\u012bkas v\u012bruss (ToMV)<\/strong> uz lap\u0101m veidojas raibi gai\u0161i un tum\u0161i za\u013ci raksti, bie\u017ei vien ar lapu \u010dokuro\u0161anos un p\u016bsl\u012b\u0161u veido\u0161anos. Neregul\u0101rais kr\u0101su sadal\u012bjums to at\u0161\u0137ir no bar\u012bbas vielu tr\u016bkuma.<\/p>\n Tom\u0101tu dzelten\u0101s lapu \u010dokuro\u0161an\u0101s v\u012bruss (TYLCV)<\/strong>, ko p\u0101rn\u0113s\u0101 baltblusi\u0146a Bemisia tabaci<\/em>, izraisa lapu malu \u010dokuro\u0161anos uz aug\u0161u, starpdz\u012bslu dzelt\u0113\u0161anu un izteiktu aug\u0161anas aizkav\u0113\u0161anos. T\u0101 ir viena no ekonomiski viskait\u012bg\u0101kaj\u0101m v\u012brusu slim\u012bb\u0101m siltos tom\u0101tu audz\u0113\u0161anas re\u0123ionos vis\u0101 pasaul\u0113.<\/p>\n Vizu\u0101l\u0101s identifik\u0101cijas izaicin\u0101jums ir iev\u0113rojams pat apm\u0101c\u012btiem agronomiem. Da\u017e\u0101du slim\u012bbu agr\u012bn\u0101s stadijas simptomi viedt\u0101lru\u0146a fotoatt\u0113l\u0101 var izskat\u012bties gandr\u012bz identiski. Gan septorijas lapu plankums, gan bakteri\u0101l\u0101 plankumain\u012bba rada mazus, apa\u013cus boj\u0101jumus. Gan agr\u012bn\u0101 puve, gan v\u0113l\u0101 puve izraisa br\u016bno audu boj\u0101eju. Vides faktori, piem\u0113ram, sl\u0101pek\u013ca defic\u012bts, aukstuma stress un smidzin\u0101\u0161anas fitotoksicit\u0101te, var atdarin\u0101t v\u012brusu simptomus.<\/p>\n T\u0101s nav tikai akad\u0113miskas sare\u017e\u0123\u012bt\u012bbas. T\u0101s tie\u0161i ietekm\u0113 to, k\u0101 j\u0101veido noteik\u0161anas datu kopas, k\u0101 j\u0101apm\u0101ca mode\u013ci un k\u0101 noteik\u0161anas sist\u0113mas j\u0101valid\u0113 pirms ievie\u0161anas.<\/p>\n Agr\u012bna atkl\u0101\u0161ana nenoz\u012bm\u0113 tikai \u0101tr\u0101ku r\u012bc\u012bbu. T\u0101 ir r\u012bc\u012bba laik\u0101, kad darb\u012bba v\u0113l ir efekt\u012bva. Fungic\u012bdi, ko lieto pie pirmaj\u0101m agr\u012bno puves boj\u0101jumu paz\u012bm\u0113m, nov\u0113r\u0161 sporul\u0101ciju un s\u0101nu izplat\u012b\u0161anos. Tie pa\u0161i fungic\u012bdi, ko lieto p\u0113c 30% vainaga defoli\u0101cijas, ir ekonomiski nelieli.<\/p>\n Ekonomisk\u0101 un agronomisk\u0101 lo\u0123ika ir p\u0101rliecino\u0161a: ieguldiet l\u012bdzek\u013cus agr\u012bnas atkl\u0101\u0161anas tehnolo\u0123ij\u0101s, un slim\u012bbu apkaro\u0161anas izmaksas strauji samazin\u0101sies.<\/p>\n S\u00e1nchez-S\u00e1nchez et al. (2024) l\u0113sa, ka tikai v\u012brusu slim\u012bbas samazina glob\u0101lo tom\u0101tu ra\u017eo\u0161anas v\u0113rt\u012bbu par 2 l\u012bdz 5% gad\u0101<\/strong>, \u2014 skaitlis, kas, \u0146emot v\u0113r\u0101 \u0161\u012bs ra\u017eas pasaules tirgus m\u0113rogu, noz\u012bm\u0113 zaud\u0113jumus miljardos ASV dol\u0101ru.<\/p>\n Pat neliels slim\u012bbu izrais\u012bto zaud\u0113jumu samazin\u0101jums, pateicoties agr\u012bnai m\u0101ksl\u012bg\u0101 intelekta atkl\u0101\u0161anai, var nodro\u0161in\u0101t atdevi no ieguld\u012bjumiem tehnolo\u0123ij\u0101s vienas aug\u0161anas sezonas laik\u0101 lieliem tom\u0101tu ra\u017eot\u0101jiem.<\/p>\n Prec\u012bz\u0101 lauksaimniec\u012bba ir prakse, kur\u0101 saimniec\u012bba tiek uzskat\u012bta par atsevi\u0161\u0137u zonu moza\u012bku, nevis par vienm\u0113r\u012bgu lauku. T\u0101 viet\u0101, lai katram kvadr\u0101tmetram piem\u0113rotu vien\u0101du s\u0113klas daudzumu, \u016bdens daudzumu vai fungic\u012bda devu, prec\u012bz\u0101s sist\u0113mas izmanto re\u0101llaika datus, lai main\u012btu ievades datus, pamatojoties uz faktiskajiem apst\u0101k\u013ciem katr\u0101 viet\u0101.<\/p>\n Uz datiem balst\u012bta lauksaimniec\u012bba balst\u0101s uz nep\u0101rtrauktu ciklu: sensori un att\u0113lveido\u0161anas sist\u0113mas v\u0101c lauka datus, programmat\u016bra apstr\u0101d\u0101 un interpret\u0113 \u0161os datus, un l\u0113mumu atbalsta r\u012bki interpret\u0101ciju p\u0101rv\u0113r\u0161 r\u012bc\u012bbas ieteikumos. Lai sist\u0113ma sniegtu v\u0113rt\u012bbu, katram \u0161\u012bs \u0137\u0113des posmam ir j\u0101b\u016bt prec\u012bzam.<\/p>\n Vieda kult\u016braugu apsaimnieko\u0161ana papla\u0161ina \u0161o lo\u0123iku ar\u012b uz biolo\u0123iskiem draudiem. T\u0101 viet\u0101, lai gaid\u012btu, kam\u0113r slim\u012bbas simptomi k\u013c\u016bs ac\u012bmredzami, vai pa\u013cautos uz ikned\u0113\u013cas izl\u016bko\u0161anas p\u0101rbaud\u0113m, vieda uzraudz\u012bbas sist\u0113ma atkl\u0101j slim\u012bbu pie pirm\u0101s redzam\u0101s vai spektr\u0101l\u0101s paz\u012bmes un aktiviz\u0113 br\u012bdin\u0101jumu, kas nor\u0101da infekcijas atra\u0161an\u0101s vietu, identit\u0101ti un iesp\u0113jamo smaguma pak\u0101pi.<\/p>\n Digit\u0101l\u0101 att\u0113lveido\u0161ana veido pamatdatu avotu liel\u0101kajai da\u013cai viedo slim\u012bbu noteik\u0161anas sist\u0113mu. RGB kameras uztver redzamo kr\u0101su inform\u0101ciju, ko cilv\u0113ki redz. Multispektr\u0101l\u0101s kameras uztver vi\u013c\u0146u garumus, kas p\u0101rsniedz redzamo gaismu, tostarp tuvo infrasarkano staru spektru, kas atkl\u0101j hlorofila stresu, pirms tas ir redzams ar neapbru\u0146otu aci.<\/p>\n Hiperspektr\u0101l\u0101s kameras uztver simtiem \u0161auru vi\u013c\u0146u garuma joslu un var noteikt bio\u0137\u012bmiskas izmai\u0146as molekul\u0101r\u0101 l\u012bmen\u012b, lai gan liel\u0101kajai da\u013cai saimniec\u012bbu izvieto\u0161anas t\u0101s joproj\u0101m ir d\u0101rgas.<\/p>\n Uz zemes b\u0101z\u0113ti sensori un lietu interneta (IoT) t\u012bkli papildina att\u0113lveido\u0161anu. Temperat\u016bras un mitruma sensori, kas novietoti kult\u016braugu lapotn\u0113s, sniedz mikroklimata datus, kas nor\u0101da, kad veidojas slim\u012bb\u0101m labv\u0113l\u012bgi apst\u0101k\u013ci.<\/p>\n Ilgsto\u0161s lapu mitrums, kas p\u0101rsniedz 10 stundas, apvienojum\u0101 ar temperat\u016bru no 18 l\u012bdz 22 \u00b0C, ir zin\u0101ms lakstu puves ierosin\u0101t\u0101ja sign\u0101ls \u2014 lietu interneta sist\u0113ma var izdot slim\u012bbas riska br\u012bdin\u0101jumu, pirms par\u0101d\u0101s jebk\u0101di boj\u0101jumi.<\/p>\n Droni un bezpilota lidapar\u0101ti (UAV) pie\u0161\u0137ir slim\u012bbu uzraudz\u012bbai telpisko dimensiju, ko nevar nodro\u0161in\u0101t ar zemes l\u012bme\u0146a att\u0113lveido\u0161anu. Ar multispektr\u0101lu kameru apr\u012bkots UAV var apsekot 10 hekt\u0101ru lielu tom\u0101tu lauku maz\u0101k nek\u0101 30 min\u016bt\u0113s, \u0123ener\u0113jot \u0123eoreferenc\u0113tu slim\u012bbu riska karti, kas prec\u012bzi par\u0101da, kur\u0101s rind\u0101s un zon\u0101s ir agr\u012bnas stresa paz\u012bmes.<\/p>\n Tas \u013cauj m\u0113r\u0137tiec\u012bgi apstr\u0101d\u0101t augsta riska zonas, nevis visu lauku. Ar lietu internetu (IoT) nodro\u0161in\u0101tas lauksaimniec\u012bbas sist\u0113mas integr\u0113 visas \u0161\u012bs datu pl\u016bsmas vienot\u0101 platform\u0101, lauka l\u012bme\u0146a nov\u0113rojumus ievadot m\u0101kon\u012b balst\u012bt\u0101 anal\u012btik\u0101 vai perif\u0113rijas skait\u013co\u0161anas mezglos, kur slim\u012bbu klasifik\u0101cijas algoritmi darbojas gandr\u012bz re\u0101llaik\u0101.<\/p>\n Dzi\u013c\u0101 m\u0101c\u012b\u0161an\u0101s ir ma\u0161\u012bnm\u0101c\u012b\u0161an\u0101s nozare, kur\u0101 algoritmi m\u0101c\u0101s ieg\u016bt mode\u013cus tie\u0161i no neapstr\u0101d\u0101tiem datiem \u2014 \u0161aj\u0101 gad\u012bjum\u0101 att\u0113liem \u2014, izmantojot matem\u0101tisko transform\u0101ciju hierarhiskus sl\u0101\u0146us.<\/p>\n Galven\u0101 priek\u0161roc\u012bba sal\u012bdzin\u0101jum\u0101 ar klasisko ma\u0161\u012bnm\u0101c\u012b\u0161anos ir t\u0101, ka dzi\u013cajai m\u0101c\u012b\u0161an\u0101s metodei nav nepiecie\u0161ams, lai cilv\u0113ks-eksperts manu\u0101li defin\u0113tu, k\u0101das paz\u012bmes (formas, tekst\u016bras, kr\u0101su gradientus) mekl\u0113t slim\u012bbas skartas lapas att\u0113l\u0101. Algoritms \u0161\u012bs paz\u012bmes apg\u016bst no apm\u0101c\u012bbas piem\u0113riem.<\/p>\n Un M\u0101ksl\u012bgais neironu t\u012bkls (ANN)<\/strong> (skait\u013co\u0161anas sist\u0113ma, ko br\u012bvi iedvesmoju\u0161i biolo\u0123iskie neironi) apstr\u0101d\u0101 ievades datus, izmantojot savstarp\u0113ji savienotu mezglu sl\u0101\u0146us.<\/p>\n Katram savienojumam ir skaitlisks svars, kas nosaka, cik sp\u0113c\u012bgi viena mezgla aktiv\u0101cija ietekm\u0113 n\u0101kamo. T\u012bkla apm\u0101c\u012bba noz\u012bm\u0113 \u0161o svaru piel\u0101go\u0161anu, izmantojot mar\u0137\u0113tus piem\u0113rus, l\u012bdz t\u012bkla izejas prognozes atbilst pareizaj\u0101m eti\u0137et\u0113m ar minim\u0101lu k\u013c\u016bdu.<\/p>\n A Konvolucion\u0101lais neironu t\u012bkls (CNN)<\/strong> (specializ\u0113ta neironu t\u012bkla arhitekt\u016bra, kas paredz\u0113ta att\u0113lu datiem) att\u0113l\u0101 veic matem\u0101tiskas oper\u0101cijas, ko sauc par konvol\u016bcij\u0101m. Konvol\u016bcijas laik\u0101 att\u0113ls tiek p\u0101rvilkts pa nelielu filtra logu \u2014 parasti 3 \u00d7 3 vai 5 \u00d7 5 pikse\u013ci \u2014 un katr\u0101 poz\u012bcij\u0101 tiek apr\u0113\u0137in\u0101ta sv\u0113rt\u0101 summa, izveidojot iez\u012bmju karti, kas atspogu\u013co lok\u0101los mode\u013cus, piem\u0113ram, malas, tekst\u016bras un kr\u0101su gradientus.<\/p>\n Vair\u0101ku konvolucion\u0101lo sl\u0101\u0146u sakrau\u0161ana \u013cauj t\u012bklam pak\u0101peniski apg\u016bt abstrakt\u0101kas iez\u012bmes: malas agr\u012bnajos sl\u0101\u0146os, formas vid\u0113jos sl\u0101\u0146os un slim\u012bb\u0101m rakstur\u012bgus mode\u013cus dzi\u013c\u0101kajos sl\u0101\u0146os.<\/p>\n \u0160\u012b hierarhisk\u0101 paz\u012bmju apguve ir tie\u0161i tas, kas padara CNN sp\u0113c\u012bgus tom\u0101tu slim\u012bbu noteik\u0161an\u0101. Boj\u0101juma ap\u013cveida robe\u017ea, t\u0101 iek\u0161\u0113j\u0101 tekst\u016bras gradients, dzelten\u012bgais oreols ap to \u2014 tas viss k\u013c\u016bst par apg\u016bstam\u0101m paz\u012bmju kombin\u0101cij\u0101m visos t\u012bkla sl\u0101\u0146os.<\/p>\n Tradicion\u0101l\u0101 ma\u0161\u012bnm\u0101c\u012b\u0161an\u0101s augu slim\u012bbu noteik\u0161anai pras\u012bja, lai jomas eksperti manu\u0101li izstr\u0101d\u0101tu funkcijas: ieg\u016btu kr\u0101su histogrammas, apr\u0113\u0137in\u0101tu tekst\u016bras deskriptorus vai m\u0113r\u012btu formas parametrus no lapu att\u0113liem, p\u0113c tam \u0161os skait\u013cus ievadot klasifikatoros, piem\u0113ram, atbalsta vektoru ma\u0161\u012bn\u0101s (SVM). \u0160is process bija darbietilp\u012bgs, atkar\u012bgs no zin\u0101\u0161an\u0101m un nestabils, ja slim\u012bbas izskats at\u0161\u0137\u012br\u0101s no apm\u0101c\u012bbas apst\u0101k\u013ciem.<\/p>\n Lobna un l\u012bdzautori (2024) apm\u0101c\u012bja kapsulas neironu t\u012bklu liela m\u0113roga datu kop\u0101 70\u00a0834 tom\u0101tu lapu att\u0113li<\/strong> un sasniedza klasifik\u0101cijas precizit\u0101ti 96.39%<\/strong> vair\u0101k\u0101s slim\u012bbu kategorij\u0101s, p\u0101rsp\u0113jot standarta CNN b\u0101zes l\u012bnijas, izmantojot tos pa\u0161us datus.<\/p>\n Lieli, daudzveid\u012bgi datu kopumi apvienojum\u0101 ar optimiz\u0113t\u0101m arhitekt\u016br\u0101m past\u0101v\u012bgi nodro\u0161ina precizit\u0101tes l\u012bmeni, kas p\u0101rsniedz to, ko var sasniegt ar tradicion\u0101lajiem datorredzes kan\u0101liem.<\/p>\n \u201cGeopard Agriculture\u201d veido tie\u0161i \u0161o sl\u0101ni. T\u0101s prec\u012bz\u0101s izp\u0113tes platforma apvieno lauka nov\u0113ro\u0161anu, slim\u012bbu identific\u0113\u0161anu un l\u0113mumu atbalstu vien\u0101 darbpl\u016bsm\u0101, ko jebkur\u0161 agronoms vai audz\u0113t\u0101js var vad\u012bt no viedt\u0101lru\u0146a.<\/p>\n Geopard vied\u0101 izl\u016bko\u0161anas sist\u0113ma identific\u0113 j\u016bsu lauka v\u0113rt\u012bg\u0101k\u0101s zonas m\u0113r\u0137tiec\u012bgai p\u0101rbaudei, nevis pieprasa vienm\u0113r\u012bgu katras rindas p\u0101rkl\u0101jumu. T\u0101 autom\u0101tiski atz\u012bm\u0113 kult\u016braugu d\u012bg\u0161anas anom\u0101lijas un novirza izl\u016bko\u0161anas centienus uz zon\u0101m, kur slim\u012bbu risks vai stress, visticam\u0101k, att\u012bst\u012bsies.<\/p>\n Tas tie\u0161i risina lauka p\u0101rkl\u0101juma probl\u0113mu, kas ierobe\u017eo manu\u0101las izl\u016bko\u0161anas programmas liel\u0101s saimniec\u012bb\u0101s. Platforma atbalsta visu galveno lauka apdraud\u0113jumu kategoriju noteik\u0161anu un re\u0123istr\u0113\u0161anu, kas attiecas uz tom\u0101tu ra\u017eo\u0161anu:<\/p>\n Zonu pl\u0101no\u0161ana ir ieb\u016bv\u0113ta izl\u016bko\u0161anas sagatavo\u0161an\u0101s posm\u0101. Geopard p\u0101rveido neapstr\u0101d\u0101tus lauka sensoru un satel\u012btu datus nep\u0101rtraukt\u0101s gradienta virsmas kart\u0113s, kas vizualiz\u0113 lauka neviendab\u012bgumu, \u013caujot agronomiem defin\u0113t apsaimnieko\u0161anas zonas pirms izl\u016bko\u0161anas s\u0101kuma. Bezsaistes zonu kartes un augsnes dati ir pieejami bez savienojuma, kas ir svar\u012bgi saimniec\u012bb\u0101s, kur mobilo sakaru p\u0101rkl\u0101jums ir nekonsekvents.<\/p>\n Visi lauka nov\u0113rojumi tiek re\u0123istr\u0113ti, izmantojot Geopard mobilo lietotni. Izl\u016bki re\u0101llaik\u0101 veic piez\u012bmes, fotogr\u0101fijas un \u0123eoreferenc\u0113tus nov\u0113rojumus, un pabeigtie izl\u016bku ieraksti tiek tie\u0161i ievad\u012bti platformas atskai\u0161u panel\u012b.<\/p>\n Lauka vad\u012bt\u0101ji var redz\u0113t, kuri draudi tika identific\u0113ti, kur tie tika atrasti, k\u0101das darb\u012bbas tika veiktas un kuras zonas joproj\u0101m tiek uzraudz\u012btas, \u2014 neapkopojot datus no pap\u012bra veidlap\u0101m vai atsevi\u0161\u0137\u0101m lietotn\u0113m.<\/p>\n \u0100rk\u0101rtas trauksmes sist\u0113ma uzrauga slim\u012bbu izplat\u012bbas mode\u013cus vis\u0101 platformas t\u012bkl\u0101 un nos\u016bta pazi\u0146ojumus, kad j\u016bsu re\u0123ion\u0101 pieaug slim\u012bbas risks. \u0160\u012b agr\u012bn\u0101s br\u012bdin\u0101\u0161anas funkcija papla\u0161ina efekt\u012bv\u0101s noteik\u0161anas logu, kas p\u0101rsniedz jebkuras atsevi\u0161\u0137as saimniec\u012bbas iek\u0161\u0113j\u0101s izl\u016bko\u0161anas iesp\u0113jas, dodot audz\u0113t\u0101jiem laiku sagatavot prevent\u012bvus pas\u0101kumus, pirms slim\u012bba sasniedz vi\u0146u laukus.<\/p>\n Geopard pieeja demonstr\u0113 praktisko integr\u0101cijas ce\u013cu, ko prec\u012bz\u0101s lauksaimniec\u012bbas p\u0113tnieki apraksta teor\u0113tiski: satel\u012btu un sensoru dati, kas inform\u0113 par izl\u016bko\u0161anas priorit\u0101\u0161u noteik\u0161anu, mobilie r\u012bki, kas \u013cauj veikt nov\u0113rojumus re\u0101llaik\u0101, un m\u0101ksl\u012bg\u0101 intelekta atbalst\u012bta draudu identific\u0113\u0161ana, kas atbalsta \u0101tr\u0101ku un m\u0113r\u0137tiec\u012bg\u0101ku l\u0113mumu pie\u0146em\u0161anu saimniec\u012bbas l\u012bmen\u012b.<\/p>\n Dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s modelis ir tikpat uzticams, cik uzticami ir dati, uz kuriem tas tika apm\u0101c\u012bts. Datu kopas sagatavo\u0161ana tom\u0101tu lapu slim\u012bbu noteik\u0161anai ir daudzpak\u0101pju process, kas nosaka jebkura mode\u013ca re\u0101l\u0101s veiktsp\u0113jas griestus.<\/p>\n Lauka att\u0113li, kas uz\u0146emti re\u0101los lauksaimniec\u012bbas apst\u0101k\u013cos \u2014 ar main\u012bgu apgaismojumu, da\u013c\u0113ju aizsegumu, \u016bdens pilieniem un augsnes fonu \u2014 ir datu kopu daudzveid\u012bbas zelta standarts, lai gan tos ir gr\u016bt\u0101k un d\u0101rg\u0101k apkopot nek\u0101 kontrol\u0113tas vides att\u0113lus.<\/p>\n Lauksaimnieku ikdienas izl\u016bko\u0161anas laik\u0101 uz\u0146emtie viedt\u0101lru\u0146u att\u0113li arvien vair\u0101k veido praktisku datu avotu, kas savieno laboratorijas apst\u0101k\u013cus ar re\u0101liem izvieto\u0161anas scen\u0101rijiem.<\/p>\n Publiski pieejamie datu kopumi ir iev\u0113rojami pa\u0101trin\u0101ju\u0161i p\u0113tniec\u012bbu. PlantVillage datu kopa<\/strong>, ko izstr\u0101d\u0101jusi Pensilv\u0101nijas \u0161tata universit\u0101te, satur vair\u0101k nek\u0101 54\u00a0000 att\u0113lu<\/strong> vesel\u012bgu un slimu augu lapu skaits 26 sug\u0101s, tostarp 10 tom\u0101tu slim\u012bbu kategorij\u0101s.<\/p>\n Tas ir kalpojis par apm\u0101c\u012bbas pamatu simtiem public\u0113tu tom\u0101tu slim\u012bbu noteik\u0161anas mode\u013cu un joproj\u0101m ir vispla\u0161\u0101k izmantotais etalonu datu kopums \u0161aj\u0101 jom\u0101.<\/p>\n No da\u017e\u0101diem avotiem apkopotie neapstr\u0101d\u0101tie att\u0113li satur troksni, nekonsekventus izm\u0113rus un kr\u0101su kalibr\u0113\u0161anas at\u0161\u0137ir\u012bbas, kas mode\u013ca apm\u0101c\u012bb\u0101 var rad\u012bt ne\u012bstus mode\u013cus. Priek\u0161apstr\u0101de standartiz\u0113 datus, pirms tie non\u0101k model\u012b.<\/p>\n Katram att\u0113lam uzraudz\u012btas m\u0101c\u012b\u0161an\u0101s datu kop\u0101 ir j\u0101b\u016bt pamatotai eti\u0137etei: kurai slim\u012bbas kategorijai tas pieder vai vai lapa ir vesel\u012bga. \u0160\u012b mar\u0137\u0113\u0161ana j\u0101veic vai j\u0101apstiprina augu patologiem, ne tikai lauksaimniec\u012bbas \u0123eneralistiem, jo vizu\u0101l\u0101 slim\u012bbu p\u0101rkl\u0101\u0161an\u0101s padara amatieru anot\u0101cijas neuzticamas.<\/p>\n Slim\u012bbu klasifik\u0101cijas anot\u0101cija klases l\u012bmen\u012b ir sam\u0113r\u0101 vienk\u0101r\u0161a, bet objektu noteik\u0161anas mode\u013cu ierobe\u017eojo\u0161\u0101 lodzi\u0146a anot\u0101cija \u2014 prec\u012bza boj\u0101juma atra\u0161an\u0101s vietas atz\u012bm\u0113\u0161ana att\u0113l\u0101 \u2014 prasa iev\u0113rojami vair\u0101k laika un zin\u0101\u0161anu par katru att\u0113lu.<\/p>\n P\u0113tnieku aprindas ir izv\u0113rt\u0113ju\u0161as desmitiem tom\u0101tu lapu slim\u012bbu klasifik\u0101cijas arhitekt\u016bru. Izpratne par to, kuras arhitekt\u016bras domin\u0113 un k\u0101p\u0113c, pal\u012bdz prakti\u0137iem pie\u0146emt p\u0101rdom\u0101tus l\u0113mumus, ievie\u0161ot \u0161\u012bs sist\u0113mas.<\/p>\n CNN pamatmode\u013ci slim\u012bbu klasifik\u0101cijai atbilst standarta modelim: konvolucion\u0101lie sl\u0101\u0146i paz\u012bmju ieguvei, apvienojo\u0161ie sl\u0101\u0146i, kas samazina telpiskos izm\u0113rus, vienlaikus saglab\u0101jot domin\u0113jo\u0161\u0101s paz\u012bmes, un piln\u012bb\u0101 savienoti sl\u0101\u0146i beig\u0101s, kas kart\u0113 ieg\u016bt\u0101s paz\u012bmes slim\u012bbu klases varb\u016bt\u012bb\u0101m.<\/p>\n Agr\u012bnais darbs ar PlantVillage datu kopu par\u0101d\u012bja, ka pat pietic\u012bgi CNN ar 5\u20137 sl\u0101\u0146iem var sasniegt precizit\u0101ti, kas p\u0101rsniedz 90%, t\u012bros, laboratorij\u0101 ieg\u016btos att\u0113los.<\/p>\n M\u0101c\u012bbu p\u0101rne\u0161ana<\/strong> (prakse, kad tiek s\u0101kts ar modeli, kas iepriek\u0161 apm\u0101c\u012bts liel\u0101 visp\u0101r\u012bg\u0101 datu kop\u0101, un tiek preciz\u0113ts ar konkr\u0113tai jomai paredz\u0113tu datu kopu) p\u0101rveidoja tom\u0101tu slim\u012bbu noteik\u0161anas p\u0113t\u012bjumus, \u013caujot apm\u0101c\u012bt augstas precizit\u0101tes mode\u013cus ar relat\u012bvi nelieliem lauksaimniec\u012bbas datu kopumiem.<\/p>\n 1. VGG16 un VGG19<\/strong>, ko izstr\u0101d\u0101jusi Vizu\u0101l\u0101s \u0123eometrijas grupa Oksford\u0101, izmanto 16 vai 19 svara sl\u0101\u0146us ar vienm\u0113r\u012bg\u0101m 3\u00d73 konvol\u016bcij\u0101m. Tie joproj\u0101m ir uzticami b\u0101zes avoti tom\u0101tu slim\u012bbu klasifik\u0101cijai, parasti sasniedzot 94\u201397% precizit\u0101ti p\u0113c slim\u012bbu datu kopu preciz\u0113\u0161anas.<\/p>\n 2. ResNet<\/strong> (Residual Network) ieviesa izlai\u0161anas savienojumus, kas \u013cauj gradientiem pl\u016bst tie\u0161i p\u0101ri sl\u0101\u0146iem, atrisinot izz\u016bdo\u0161\u0101 gradienta probl\u0113mu, kas iepriek\u0161 ierobe\u017eoja apm\u0101c\u012bbas dzi\u013cumu. ResNet50, kas prec\u012bzi noregul\u0113ts uz tom\u0101tu slim\u012bbu datiem, jaun\u0101kajos p\u0113t\u012bjumos konsekventi sasniedz 96\u201398% precizit\u0101ti.<\/p>\n 3. Bl\u012bvs t\u012bkls<\/strong> papla\u0161ina izlai\u0161anas savienojuma koncepciju, savienojot katru sl\u0101ni ar katru n\u0101kamo sl\u0101ni bl\u012bv\u0101 blok\u0101, maksim\u0101li palielinot funkciju atk\u0101rtotu izmanto\u0161anu un radot kompaktus mode\u013cus ar sp\u0113c\u012bgu klasifik\u0101cijas veiktsp\u0113ju.<\/p>\n 4. Efekt\u012bvs t\u012bkls<\/strong>, ko izstr\u0101d\u0101jusi Google Brain, vienlaikus m\u0113rogo t\u012bkla platumu, dzi\u013cumu un iz\u0161\u0137irtsp\u0113ju, izmantojot salikto koeficientu. EfficientNetB0 ar uzman\u012bbas moduli tika sasniegts 99,39% precizit\u0101te<\/strong> par augu slim\u012bbu klasifik\u0101ciju p\u0113t\u012bjum\u0101, ko public\u0113ja Gonz\u00e1lez-Briones et al. (2025), ar atbilsto\u0161u veiktsp\u0113ju perif\u0113rijas ier\u012b\u010du izvieto\u0161anai.<\/p>\n 5. MobileNet<\/strong>, kas \u012bpa\u0161i izstr\u0101d\u0101ta resursiem ierobe\u017eot\u0101m ier\u012bc\u0113m, izmanto dzi\u013cum\u0101 atdal\u0101mas konvol\u016bcijas, lai iev\u0113rojami samazin\u0101tu skait\u013co\u0161anas apjomu, vienlaikus saglab\u0101jot augstu precizit\u0101ti, padarot to par v\u0113lamo arhitekt\u016bru viedt\u0101lru\u0146u un perif\u0113rijas m\u0101ksl\u012bg\u0101 intelekta izvieto\u0161anai prec\u012bzaj\u0101 lauksaimniec\u012bb\u0101.<\/p>\n Jaun\u0101kie p\u0113t\u012bjumi ir p\u0101rsniegu\u0161i standarta CNN robe\u017eas un virz\u012bju\u0161ies uz arhitekt\u016br\u0101m, kas var uztvert liel\u0101ka att\u0101luma telpisk\u0101s attiec\u012bbas lapu att\u0113los.<\/p>\n Redzes transformatori (ViT)<\/strong> (neironu t\u012bkli, kas att\u0113lu iel\u0101piem piem\u0113ro Transformer uzman\u012bbas meh\u0101nismu, kas s\u0101kotn\u0113ji tika izstr\u0101d\u0101ts dabisk\u0101s valodas apstr\u0101dei) ir uzr\u0101d\u012bju\u0161i labus rezult\u0101tus augu slim\u012bbu noteik\u0161an\u0101, ja ir pieejami pietiekami daudz apm\u0101c\u012bbas datu.<\/p>\n At\u0161\u0137ir\u012bb\u0101 no CNN, kas apstr\u0101d\u0101 lok\u0101los att\u0113lu re\u0123ionus, izmantojot konvol\u016bcijas, ViT vienlaic\u012bgi apg\u016bst attiec\u012bbas starp visiem att\u0113lu iel\u0101piem, \u013caujot tiem noteikt telpiski sadal\u012btus mode\u013cus vis\u0101 lap\u0101.<\/p>\n Uzman\u012bbu balst\u012bti CNN hibr\u012bdi<\/strong> Apvienojiet konvol\u016bciju lok\u0101lo paz\u012bmju ieguves stiprumu ar uzman\u012bbas meh\u0101nismiem, kas \u013cauj modelim koncentr\u0113t apstr\u0101des resursus uz slim\u012bbai visatbilsto\u0161\u0101kajiem att\u0113lu re\u0123ioniem.<\/p>\n Sasniegts uz Si\u0101mas t\u012bklu balst\u012bts viegls ietvars tom\u0101tu slim\u012bbu noteik\u0161anai 96,97% precizit\u0101te Plant Village tom\u0101tu apak\u0161kop\u0101<\/strong> ar tikai aptuveni 2,96 miljoniem parametru (Frontiers in Plant Science, 2025), kas pier\u0101da, ka augsta precizit\u0101te un aparat\u016bras efektivit\u0101te nav savstarp\u0113ji izsl\u0113dzo\u0161i m\u0113r\u0137i.<\/p>\n Ansamb\u013ca m\u0101c\u012bbu mode\u013ci<\/strong> apvienot prognozes no vair\u0101k\u0101m neatkar\u012bgi apm\u0101c\u012bt\u0101m arhitekt\u016br\u0101m, apr\u0113\u0137inot vid\u0113jo v\u0113rt\u012bbu vai balsojot starp t\u0101m, lai ieg\u016btu gal\u012bgo prognozi, kas ir robust\u0101ka nek\u0101 jebkur\u0161 atsevi\u0161\u0137s modelis. Vu et al. (2024) izmantoja ResNet50 ar iez\u012bmju palielin\u0101\u0161anas metod\u0113m, lai ar \u0161o pieeju uzlabotu klasifik\u0101cijas veiktsp\u0113ju.<\/p>\n Abdullah et al. (Agronomy, 2024) sal\u012bdzin\u0101ja YOLOv8s, YOLOv5 un Faster R-CNN slimu tom\u0101tu lapu noteik\u0161an\u0101 un atkl\u0101ja, ka YOLOv8s sasniedza vid\u0113jo precizit\u0101ti (mAP) 92.5%<\/strong>, p\u0101rsp\u0113jot YOLOv5 ar 89,1% un \u0101tr\u0101ku R-CNN ar 77,5%, vienlaikus demonstr\u0113jot ar\u012b \u0101tr\u0101ku secin\u0101jumu \u0101trumu un maz\u0101ku mode\u013ca nospiedumu.<\/p>\n Re\u0101llaika noteik\u0161anas lietojumprogramm\u0101m lauk\u0101 YOLOv8 klases mode\u013ci pied\u0101v\u0101 vislab\u0101ko precizit\u0101tes un apstr\u0101des \u0101truma l\u012bdzsvaru, padarot tos labi piem\u0113rotus dronu uzst\u0101d\u012b\u0161anai vai izvieto\u0161anai malu ier\u012bc\u0113s.<\/p>\n Izvietojama intelektu\u0101la slim\u012bbu noteik\u0161anas sist\u0113ma ir vair\u0101k nek\u0101 tikai apm\u0101c\u012bts modelis. T\u0101 ir pilna cikla process, kas p\u0101riet no neapstr\u0101d\u0101tu att\u0113lu uz\u0146em\u0161anas uz r\u012bc\u012bbas pl\u0101niem slim\u012bbu p\u0101rvald\u012bbas jom\u0101.<\/p>\n Galvenais cauru\u013cvads sast\u0101v no pieciem sec\u012bgiem posmiem, katrs no kuriem p\u0101rveido datus pirms to nodo\u0161anas n\u0101kamajam.<\/p>\n 1. Att\u0113lu ievade pie\u0146em neapstr\u0101d\u0101tus lapu att\u0113lus no jebkura uztver\u0161anas avota \u2014 lauka viedt\u0101lru\u0146a, bezpilota lidapar\u0101tam (UAV) piestiprin\u0101tas kameras vai fiks\u0113tas lapotnes kameras siltumn\u012bc\u0101. Ievades apstr\u0101des modu\u013ci p\u0101rbauda att\u0113la iz\u0161\u0137irtsp\u0113ju un atz\u012bm\u0113 izpl\u016bdu\u0161us vai nelietojamus att\u0113lus, pirms tie non\u0101k apstr\u0101des \u0137\u0113d\u0113.<\/p>\n 2. Priek\u0161apstr\u0101des posm\u0101 tiek piem\u0113rotas 6.2. sada\u013c\u0101 aprakst\u012bt\u0101s normaliz\u0101cijas, izm\u0113ru mai\u0146as un kvalit\u0101tes uzlabo\u0161anas darb\u012bbas, nodro\u0161inot, ka ievades dati atbilst apm\u0101c\u012bt\u0101 mode\u013ca paredz\u0113tajam form\u0101tam.<\/p>\n 3. Iez\u012bmju ekstrakcijas proces\u0101 priek\u0161apstr\u0101d\u0101tais att\u0113ls tiek palaists caur apm\u0101c\u012bta dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s mode\u013ca konvolucion\u0101lajiem sl\u0101\u0146iem. \u0160aj\u0101 posm\u0101 modelis p\u0101rveido neapstr\u0101d\u0101tus pikse\u013cu datus kompakt\u0101 skaitlisk\u0101 att\u0113lojum\u0101 (iez\u012bmju vektor\u0101), kas kod\u0113 lapas slim\u012bbas izrais\u012bt\u0101jas vizu\u0101l\u0101s \u012bpa\u0161\u012bbas.<\/p>\n 4. Slim\u012bbu klasifik\u0101cija piln\u012bb\u0101 savienotos sl\u0101\u0146us un softmax izejas funkciju piem\u0113ro paz\u012bmju vektoram, apr\u0113\u0137inot varb\u016bt\u012bbas r\u0101d\u012bt\u0101ju katrai slim\u012bbas kategorijai. Kategorija ar visaugst\u0101ko varb\u016bt\u012bbu k\u013c\u016bst par prognoz\u0113to diagnozi.<\/p>\n 5. L\u0113mumu atbalsta izvade p\u0101rveido klasifik\u0101cijas rezult\u0101tu praktisk\u0101 ieteikum\u0101: identific\u0113t\u0101s slim\u012bbas nosaukums, ticam\u012bbas l\u012bmenis, ieteikt\u0101 p\u0101rvald\u012bbas darb\u012bba (m\u0113r\u0137tiec\u012bgs fungic\u012bds, biokontroles l\u012bdzeklis, skarto augu no\u0146em\u0161ana) un \u0123eoreferenc\u0113ti atra\u0161an\u0101s vietas dati, ja att\u0113ls ir uz\u0146emts ar GPS apr\u012bkotu ier\u012bci.<\/p>\n Re\u0101l\u0101 ievie\u0161anas situ\u0101cij\u0101 lauksaimnieks atver mobilo lietotni un nofotograf\u0113 skarto lapu. Att\u0113ls tiek vai nu nos\u016bt\u012bts uz m\u0101ko\u0146serveri, kur\u0101 darbojas noteik\u0161anas modelis, vai ar\u012b apstr\u0101d\u0101ts lok\u0101li ier\u012bc\u0113, izmantojot saspiestu malas modeli.<\/p>\n Da\u017eu sekun\u017eu laik\u0101 lietotne atgrie\u017e diagnozi: \u201cAgr\u012bna puve \u2014 94% p\u0101rliec\u012bba. Ieteicam\u0101 darb\u012bba: skartaj\u0101 zon\u0101 lietot mankozeba b\u0101zes fungic\u012bdu 1,5 kg\/ha dev\u0101.\u201d.<\/p>\n "N\u0101kamo 5 dienu laik\u0101 nov\u0113rojiet blakus eso\u0161os augus." \u0122eoreferenc\u0113tais rezult\u0101ts tiek re\u0123istr\u0113ts saimniec\u012bbas digit\u0101laj\u0101 vesel\u012bbas kart\u0113, un, ja viena un t\u0101 pati slim\u012bba par\u0101d\u0101s vair\u0101k\u0101s zon\u0101s, sist\u0113ma pie\u0161\u0137ir paaugstin\u0101ta riska br\u012bdin\u0101jumu visam lauka blokam.<\/p>\n Demilie (2024), apskats 161 publik\u0101cija<\/strong> par dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s balst\u012btu augu slim\u012bbu noteik\u0161anu atkl\u0101ja, ka tom\u0101ti bija visvair\u0101k p\u0113t\u012bt\u0101 kult\u016bra visos p\u0113t\u012bjumos, kas veido 39% no vis\u0101m publik\u0101cij\u0101m<\/strong> \u2014 vair\u0101k nek\u0101 divas reizes liel\u0101ks p\u0101rkl\u0101jums nek\u0101 otrai visvair\u0101k p\u0113t\u012btajai kult\u016brai (r\u012bsiem 16% l\u012bmen\u012b), apstiprinot \u0161\u012bs kult\u016bras unik\u0101lo ekonomisk\u0101s noz\u012bmes un slim\u012bbu ievainojam\u012bbas krustpunktu.<\/p>\n Tom\u0101tu slim\u012bbu noteik\u0161anas p\u0113t\u012bjumu briedums noz\u012bm\u0113, ka prakti\u0137i var piek\u013c\u016bt pla\u0161am valid\u0113tu arhitekt\u016bru un iepriek\u0161 apm\u0101c\u012btu mode\u013cu kl\u0101stam, nevis veidot visu no nulles.<\/p>\n Pareizas nov\u0113rt\u0113\u0161anas metrikas izv\u0113le ir tikpat svar\u012bga k\u0101 pareizas arhitekt\u016bras izv\u0113le, jo \u012bpa\u0161i slim\u012bbu noteik\u0161an\u0101, kur viltus negat\u012bviem rezult\u0101tiem (re\u0101las slim\u012bbas neesam\u012bba) ir at\u0161\u0137ir\u012bgas izmaksas nek\u0101 viltus pozit\u012bviem rezult\u0101tiem (vesel\u012bgas lapas k\u013c\u016bdaina identific\u0113\u0161ana k\u0101 slima).<\/p>\n Precizit\u0101te<\/strong> m\u0113ra visu pareizo progno\u017eu \u012bpatsvaru. Tas ir visbie\u017e\u0101k zi\u0146otais r\u0101d\u012bt\u0101js, ta\u010du tas var b\u016bt maldino\u0161s, ja slim\u012bbu klases ir nel\u012bdzsvarotas \u2014 modelis, kas vienm\u0113r prognoz\u0113 \u201cvesel\u012bgu\u201d st\u0101vokli datu kop\u0101 ar 90% veseliem att\u0113liem, sasniedz 90% precizit\u0101ti, vienlaikus b\u016bdams piln\u012bgi neder\u012bgs slim\u012bbu noteik\u0161anai.<\/p>\n Precizit\u0101te<\/strong> m\u0113ra, cik liela da\u013ca no atkl\u0101tajiem slim\u012bbas gad\u012bjumiem ir patiesi pozit\u012bvi gad\u012bjumi, fiks\u0113jot viltus trauksmju l\u012bmeni. Augsta precizit\u0101te noz\u012bm\u0113, ka modelis reti izraisa nevajadz\u012bgus \u0101rst\u0113\u0161anas ieteikumus.<\/p>\n Atsauk\u0161ana (jut\u012bgums)<\/strong> m\u0113ra, cik liela da\u013ca no faktiski slimajiem augiem ir pareizi atz\u012bm\u0113ti. Augsta atpaz\u012bstam\u012bba noz\u012bm\u0113, ka modelis reti kad nepamana re\u0101las infekcijas, kas ir svar\u012bg\u0101ks r\u0101d\u012bt\u0101js slim\u012bbu p\u0101rvald\u012bb\u0101.<\/p>\n F1 rezult\u0101ts<\/strong> ir precizit\u0101tes un atcer\u0113\u0161an\u0101s sp\u0113jas harmoniskais vid\u0113jais, kas nodro\u0161ina vienotu l\u012bdzsvarotu r\u0101d\u012bt\u0101ju, kas soda mode\u013cus, kuri upur\u0113 vienu otra lab\u0101. Tas ir v\u0113lamais r\u0101d\u012bt\u0101js, ja gan viltus pozit\u012bviem, gan viltus negat\u012bviem rezult\u0101tiem ir iev\u0113rojamas izmaksas.<\/p>\n Specifiskums<\/strong> m\u0113ra, cik prec\u012bzi modelis identific\u0113 patiesi vesel\u012bgas lapas k\u0101 vesel\u012bgas, kas ir svar\u012bgi, lai nov\u0113rstu nevajadz\u012bgu pestic\u012bdu lieto\u0161anu slim\u012bb\u0101m neskart\u0101s kult\u016br\u0101s.<\/p>\n Port\u0101ls Apjukuma matrica<\/strong> par\u0101da piln\u012bgu progno\u017eu sadal\u012bjumu vis\u0101s klas\u0113s, atkl\u0101jot, kuri slim\u012bbu p\u0101ri tiek visbie\u017e\u0101k jaukti \u2014 kritiska inform\u0101cija apm\u0101c\u012bbas datu vai mode\u013ca arhitekt\u016bras pilnveido\u0161anai.<\/p>\n Port\u0101ls ROC-AUC (uztv\u0113r\u0113ja darb\u012bbas raksturlielums \u2014 laukums zem l\u012bknes)<\/strong> m\u0113ra kop\u0113jo mode\u013ca at\u0161\u0137iram\u012bbu visos klasifik\u0101cijas sliek\u0161\u0146os, un v\u0113rt\u012bba 1,0 apz\u012bm\u0113 perfektu diskrimin\u0101ciju, bet 0,5 \u2013 nejau\u0161\u012bbas l\u012bme\u0146a veiktsp\u0113ju.<\/p>\n Augstas precizit\u0101tes mode\u013ca p\u0101rvieto\u0161ana no p\u0113tniec\u012bbas piez\u012bmju gr\u0101mati\u0146as uz funkcion\u0113jo\u0161u saimniec\u012bbas sist\u0113mu prasa atrisin\u0101t at\u0161\u0137ir\u012bgu probl\u0113mu kopumu nek\u0101 mode\u013cu apm\u0101c\u012bba. Izvieto\u0161anas vid\u0113m ir aparat\u016bras ierobe\u017eojumi, savienojam\u012bbas ierobe\u017eojumi un latentuma pras\u012bbas, kas ietekm\u0113 arhitekt\u016bras un infrastrukt\u016bras izv\u0113li.<\/p>\n Viedt\u0101lru\u0146u lietotnes ir vispieejam\u0101kais ievie\u0161anas veids mazajiem un vid\u0113ja m\u0113roga lauksaimniekiem. Lietotnes, kas veidotas uz MobileNet vai EfficientNet-Lite mode\u013ciem, piln\u012bb\u0101 veic secin\u0101jumus ier\u012bc\u0113, un datu ieg\u016b\u0161anas br\u012bd\u012b nav nepiecie\u0161ams interneta savienojums.<\/p>\n Prec\u012bz\u0101s lauksaimniec\u012bbas jom\u0101 visaugst\u0101k\u0101s precizit\u0101tes modelis ir bezv\u0113rt\u012bgs, ja vien tas nevar darboties ier\u012bc\u0113, kas jau ir lauksaimnieka kabat\u0101.<\/p><\/blockquote>\n Tas ir \u0101rk\u0101rt\u012bgi svar\u012bgi saimniec\u012bb\u0101m lauku vai re\u0123ionos ar zemu interneta piesl\u0113gumu. Lauksaimnieks nofotograf\u0113 aizdom\u012bgu lapu, 1\u20133 sekun\u017eu laik\u0101 sa\u0146em slim\u012bbas prognozi un rezult\u0101tu re\u0123istr\u0113 datub\u0101z\u0113, kas apkopo vesel\u012bbas datus vis\u0101 ve\u0123et\u0101cijas period\u0101.<\/p>\n Edge AI<\/strong> (m\u0101ksl\u012bg\u0101 intelekta secin\u0101jumu palai\u0161ana tie\u0161i uz aparat\u016bras, kas atrodas datu v\u0101k\u0161anas punkt\u0101, nevis datu s\u016bt\u012b\u0161ana uz att\u0101lo serveri) atrisina m\u0101kon\u012b balst\u012btas noteik\u0161anas latentuma un savienojam\u012bbas probl\u0113mas. Specializ\u0113tas perif\u0113rijas ier\u012bces, piem\u0113ram, NVIDIA Jetson s\u0113rija vai Google Coral TPU pa\u0101trin\u0101t\u0101ji, var palaist saspiestus CNN mode\u013cus ar \u0101trumu vair\u0101k nek\u0101 30 kadri sekund\u0113, nodro\u0161inot nep\u0101rtrauktu lapu uzraudz\u012bbu re\u0101llaik\u0101 no fiks\u0113t\u0101m kamer\u0101m, kas uzst\u0101d\u012btas uz siltumn\u012bcu slied\u0113m vai lauka ap\u016bde\u0146o\u0161anas konstrukcij\u0101m.<\/p>\n Mode\u013ca saspie\u0161anas metodes \u2014 kvant\u0113\u0161ana (mode\u013ca svaru skaitlisk\u0101s precizit\u0101tes samazin\u0101\u0161ana), atzaro\u0161ana (maznoz\u012bm\u012bgu t\u012bkla savienojumu no\u0146em\u0161ana) un zin\u0101\u0161anu destil\u0101cija (maza mode\u013ca apm\u0101c\u012bba, lai atdarin\u0101tu lielu modeli) \u2014 padara to iesp\u0113jamu, neupur\u0113jot pie\u0146emamu precizit\u0101ti.<\/p>\n Saimniec\u012bb\u0101m, kas p\u0101rsniedz 20\u201330 hekt\u0101rus, p\u0101rbaude zemes l\u012bmen\u012b nevar nodro\u0161in\u0101t nepiecie\u0161amo telpisko p\u0101rkl\u0101jumu, lai atkl\u0101tu slim\u012bbu uzliesmojumus pirms to izplat\u012b\u0161an\u0101s. Bezpilota lidapar\u0101tu sist\u0113mas, kas apr\u012bkotas ar multispektr\u0101l\u0101m kamer\u0101m, uztver slim\u012bbas stresa sign\u0101lus visa lauka l\u012bmen\u012b.<\/p>\n Uz\u0146emtie att\u0113li tiek padoti uz ieb\u016bv\u0113tu perif\u0113rijas procesoru vai p\u0101rs\u016bt\u012bti uz zemes staciju, kur noteik\u0161anas modelis identific\u0113 infic\u0113t\u0101s zonas un \u0123ener\u0113 \u0123eoreferenc\u0113tu karti, kas p\u0101rkl\u0101j saimniec\u012bbas digit\u0101los zemes gabala ierakstus.<\/p>\n Tas slim\u012bbas p\u0101rvald\u012bbu maina no reakt\u012bvas (rea\u0123\u0113jot p\u0113c simptomu par\u0101d\u012b\u0161an\u0101s) uz telpiski proakt\u012bvu (rea\u0123\u0113jot uz konkr\u0113t\u0101m lauka koordin\u0101t\u0101m, kur tiek atkl\u0101ts agr\u012bns stress).<\/p>\n \u0160aj\u0101 jom\u0101 ir pan\u0101kts iev\u0113rojams progress, ta\u010du god\u012bgs atliku\u0161o izaicin\u0101jumu nov\u0113rt\u0113jums nov\u0113r\u0161 p\u0101rm\u0113r\u012bgu p\u0101rliec\u012bbu par izvieto\u0161anas l\u0113mumiem.<\/p>\n Ierobe\u017eoti lauka datu kopumi<\/strong> joproj\u0101m ir visizplat\u012bt\u0101k\u0101 probl\u0113ma. Liel\u0101k\u0101 da\u013ca augstas precizit\u0101tes mode\u013cu tiek apm\u0101c\u012bti un nov\u0113rt\u0113ti, izmantojot PlantVillage, kas izmanto kontrol\u0113ta apgaismojuma vienas lapas att\u0113lus uz t\u012bra fona. Re\u0101l\u0101 lauka veiktsp\u0113ja iev\u0113rojami samazin\u0101s, ja tie pa\u0161i mode\u013ci saskaras ar sare\u017e\u0123\u012btiem, daudzelementu att\u0113liem, kas uz\u0146emti main\u012bgos apgaismojuma apst\u0101k\u013cos.<\/p>\n Apgaismojuma apst\u0101k\u013cu vari\u0101cijas<\/strong> \u2014 tie\u0161i saules stari, kas izraisa atsp\u012bdumus uz lapu virsm\u0101m, apm\u0101cies laiks, kas saplacina tekst\u016bras nianses, vai \u0113na no kult\u016braugu lapotnes, \u2014 maina boj\u0101jumu redzamo kr\u0101su un tekst\u016bru t\u0101d\u0101 veid\u0101, kas var pasliktin\u0101t klasifik\u0101ciju par 5\u201315 procentpunktiem sal\u012bdzin\u0101jum\u0101 ar kontrol\u0113tiem apst\u0101k\u013ciem.<\/p>\n Fona sare\u017e\u0123\u012bt\u012bba<\/strong> lauka att\u0113los par\u0101d\u0101s neatbilsto\u0161a vizu\u0101l\u0101 inform\u0101cija. Augsne, mul\u010da, ap\u016bde\u0146o\u0161anas pilienu l\u012bnijas, aug\u013ci un citas lapas var par\u0101d\u012bties vien\u0101 att\u0113la kadr\u0101 ar slimu lapu, un mode\u013ci, kas nav \u012bpa\u0161i apm\u0101c\u012bti sare\u017e\u0123\u012btam fonam, bie\u017ei jauc fona elementus ar slim\u012bbas paz\u012bm\u0113m.<\/p>\n Vair\u0101kas slim\u012bbas<\/strong> uz vienas lapas rada klasifik\u0101cijas izaicin\u0101jumu, ar kuru liel\u0101k\u0101 da\u013ca vienas eti\u0137etes mode\u013cu nav paredz\u0113ti. Lapai, kur\u0101 vienlaikus izpau\u017eas bakt\u0113riju plankums un agr\u012bn\u0101 puve, ir nepiecie\u0161ama vair\u0101ku eti\u0137e\u0161u klasifik\u0101cijas iesp\u0113ja, kas palielina apm\u0101c\u012bbas sare\u017e\u0123\u012bt\u012bbu un datu kopas anot\u0101cijas pras\u012bbas.<\/p>\n Klases nel\u012bdzsvarot\u012bba<\/strong> Apm\u0101c\u012bbas datu kop\u0101s \u2014 daudz vair\u0101k att\u0113lu par izplat\u012bt\u0101m slim\u012bb\u0101m, piem\u0113ram, agr\u012bnu puvi, nek\u0101 ret\u0101m slim\u012bb\u0101m, piem\u0113ram, moza\u012bkas v\u012brusu \u2014 mode\u013ci ir p\u0101r\u0101k p\u0101rliecin\u0101ti par bie\u017ei p\u0101rst\u0101v\u0113t\u0101m klas\u0113m un neuzticami maz\u0101k p\u0101rst\u0101v\u0113t\u0101m klas\u0113m.<\/p>\n Mode\u013ca interpret\u0113jam\u012bba<\/strong> ir b\u016btisks \u0161\u0137\u0113rslis uztic\u012bbai saimniec\u012bbas l\u012bmen\u012b. Kad modelis klasific\u0113 lapu k\u0101 slimu ar 97% ticam\u012bbas pak\u0101pi, liel\u0101k\u0101 da\u013ca lauksaimnieku un agronomu v\u0113las saprast, kuras vizu\u0101l\u0101s paz\u012bmes noteica \u0161o l\u0113mumu, pirms r\u012bkoties saska\u0146\u0101 ar ieteikumu.<\/p>\n P\u0113tniec\u012bbas robe\u017ea vienlaikus virz\u0101s pa vair\u0101k\u0101m slied\u0113m, risinot da\u017e\u0101das iepriek\u0161 min\u0113t\u0101s probl\u0113mas da\u013cas.<\/p>\n 1. Izskaidrojamais m\u0101ksl\u012bgais intelekts (XAI)<\/strong> r\u012bki \u2014 jo \u012bpa\u0161i Grad-CAM (Gradient-weighted Class Activation Mapping), kas \u0123ener\u0113 siltuma kartes p\u0101rkl\u0101jumus, kas par\u0101da, uz kuriem lapas att\u0113la re\u0123ioniem modelis koncentr\u0113j\u0101s, pie\u0146emot l\u0113mumu, \u2014 tie\u0161i risina interpret\u0113jam\u012bbas probl\u0113mu.<\/p>\n Grad-CAM vizualiz\u0101cija, kas par\u0101da mode\u013ca uzman\u012bbas koncentr\u0113\u0161anos uz tum\u0161u boj\u0101juma gredzenu, sniedz pamatojuma veidu, ko agronomi var nov\u0113rt\u0113t un kam uztic\u0113ties.<\/p>\n 2. Redzes transformatori (ViT)<\/strong> turpina g\u016bt popularit\u0101ti tom\u0101tu slim\u012bbu noteik\u0161an\u0101, uzlabojoties pirmsapm\u0101c\u012bbas strat\u0113\u0123ij\u0101m un samazinoties datu pras\u012bb\u0101m ViT preciz\u0113\u0161anai.<\/p>\n To sp\u0113ja uztvert glob\u0101lus lapu l\u012bme\u0146a mode\u013cus, nevis tikai lok\u0101las paz\u012bmes, padara tos \u012bpa\u0161i daudzsolo\u0161us v\u012brusu slim\u012bbu atkl\u0101\u0161an\u0101, kas ietekm\u0113 visu lapu virsmas sadal\u012bjumu, nevis rada lokaliz\u0113tus boj\u0101jumus.<\/p>\n 3. Feder\u0101l\u0101 m\u0101c\u012b\u0161an\u0101s<\/strong> risina lauka datu kopas probl\u0113mu, \u013caujot mode\u013cus apm\u0101c\u012bt sadarb\u012bb\u0101 vair\u0101k\u0101s saimniec\u012bb\u0101s, necentraliz\u0113jot sensit\u012bvus datus.<\/p>\n Katra saimniec\u012bba apm\u0101ca lok\u0101lu modeli, izmantojot savus att\u0113lus, un tiek koplietoti tikai mode\u013ca parametru atjaunin\u0101jumi (nevis pa\u0161i att\u0113li), lai uzlabotu centr\u0101lo glob\u0101lo modeli. Tas saglab\u0101 lauksaimnieku datu priv\u0101tumu, vienlaikus iev\u0113rojami papla\u0161inot apm\u0101c\u012bbas datu daudzveid\u012bbu.<\/p>\n Prec\u012bza slim\u012bbu p\u0101rvald\u012bba netiks defin\u0113ta p\u0113c viena mode\u013ca precizit\u0101tes \u2014 to noteiks sist\u0113mas intelekts, kas savieno atkl\u0101\u0161anu, l\u0113mumu pie\u0146em\u0161anu un r\u012bc\u012bbu.<\/p><\/blockquote>\n 4. Pa\u0161vad\u012bta m\u0101c\u012b\u0161an\u0101s<\/strong> veic mode\u013cu iepriek\u0161\u0113ju apm\u0101c\u012bbu liel\u0101s nemar\u0137\u0113tu augu att\u0113lu kolekcij\u0101s, lai apg\u016btu visp\u0101r\u012bgus vizu\u0101los att\u0113lojumus, un p\u0113c tam prec\u012bzi piel\u0101go mazus mar\u0137\u0113tus slim\u012bbu datu kopumus. Tas samazina anot\u0101ciju slogu un \u013cauj izstr\u0101d\u0101t augstas kvalit\u0101tes mode\u013cus slim\u012bbu kategorij\u0101m, kur\u0101s mar\u0137\u0113tu piem\u0113ru ir maz.<\/p>\n 5. Multimod\u0101la slim\u012bbu noteik\u0161ana<\/strong> integr\u0113 lapu att\u0113lu datus ar spektr\u0101lo sensoru r\u0101d\u012bjumiem, meteorolo\u0123isko staciju datiem un v\u0113sturiskiem slim\u012bbu ierakstiem vienot\u0101 mode\u013ca ievades model\u012b. Vizu\u0101lo un vides datu apvieno\u0161ana var uzlabot noteik\u0161anas veiktsp\u0113ju, kas p\u0101rsniedz to, ko atbalsta tikai att\u0113lu dati, \u012bpa\u0161i slim\u012bb\u0101m, kuru vizu\u0101lajiem simptomiem seko bio\u0137\u012bmiskas izmai\u0146as, ko var noteikt spektr\u0101lajos parakstos.<\/p>\n P\u0101reja no augstas precizit\u0101tes p\u0113tniec\u012bbas mode\u013ciem uz konsekventi uzticamu ievie\u0161anu saimniec\u012bbu l\u012bmen\u012b prasa m\u0113r\u0137tiec\u012bgu darbu vair\u0101k\u0101s front\u0113s.<\/p>\n Re\u0101l\u0101s pasaules izvieto\u0161anas valid\u0101cija<\/strong> Lai god\u012bgi raksturotu veiktsp\u0113jas at\u0161\u0137ir\u012bbu starp kontrol\u0113tu apst\u0101k\u013cu un lauka apst\u0101k\u013cu noteik\u0161anu, ir nepiecie\u0161ams ne tikai PlantVillage etalonu tests, bet ar\u012b da\u017e\u0101d\u0101s \u0123eogr\u0101fisk\u0101s viet\u0101s un lauksaimniec\u012bbas sist\u0113m\u0101s.<\/p>\n Stabila lauka l\u012bme\u0146a noteik\u0161ana<\/strong> b\u016bs nepiecie\u0161ami m\u0113r\u0137tiec\u012bgi veidoti lauka datu kopumi, kas apkopoti vair\u0101k\u0101s aug\u0161anas sezon\u0101s vair\u0101k\u0101s valst\u012bs, ar sistem\u0101tisku laika apst\u0101k\u013cu dokument\u0101ciju att\u0113lu uz\u0146em\u0161anas laik\u0101.<\/p>\n Integr\u0101cija ar lietu internetu (IoT) un vied\u0101s lauksaimniec\u012bbas infrastrukt\u016bru<\/strong> \u2014 slim\u012bbu atkl\u0101\u0161anas br\u012bdin\u0101jumu tie\u0161a savieno\u0161ana ar automatiz\u0113t\u0101m ap\u016bde\u0146o\u0161anas un m\u0113slo\u0161anas sist\u0113m\u0101m, dronu izsmidzin\u0101\u0161anas platform\u0101m un saimniec\u012bbu p\u0101rvald\u012bbas programmat\u016bru \u2014 nosl\u0113gs apli starp atkl\u0101\u0161anu un r\u012bc\u012bbu.<\/p>\n Prognoz\u0113jo\u0161a slim\u012bbu prognoz\u0113\u0161ana,<\/strong> Apvienojot pa\u0161reiz\u0113jos slim\u012bbu atkl\u0101\u0161anas datus ar laika prognoz\u0113\u0161anas mode\u013ciem un v\u0113sturiskiem slim\u012bbu izplat\u012bbas mode\u013ciem, tiks main\u012bta paradigma no rea\u0123\u0113jo\u0161as atkl\u0101\u0161anas uz prevent\u012bvu p\u0101rvald\u012bbu: ieteikt prevent\u012bva rakstura darb\u012bbas pirms simptomu par\u0101d\u012b\u0161an\u0101s.<\/p>\n Autonom\u0101s lauksaimniec\u012bbas sist\u0113mas<\/strong> \u2014 Bezpilota lidapar\u0101tu (UAV) flotes, kas veic nep\u0101rtrauktu lauka uzraudz\u012bbu, iez\u012bm\u0113 slim\u012bbu zonas un koordin\u0113 darb\u012bbu ar automatiz\u0113t\u0101m smidzin\u0101t\u0101ju vien\u012bb\u0101m bez cilv\u0113ka iejauk\u0161an\u0101s, \u2014 atspogu\u013co horizontu, uz kuru virz\u0101s pa\u0161reiz\u0113j\u0101s prec\u012bz\u0101s lauksaimniec\u012bbas tehnolo\u0123ijas.<\/p>\n Inteli\u0123enta tom\u0101tu lapu slim\u012bbu noteik\u0161ana, izmantojot dzi\u013co m\u0101c\u012b\u0161anos, vairs nav eksperiment\u0101la tehnolo\u0123ija. T\u0101 ir nobriedusi, labi valid\u0113ta lietojumprogramma ar arvien vair\u0101k recenz\u0113tu pier\u0101d\u012bjumu, kas apstiprina t\u0101s sp\u0113ju prec\u012bzi, \u0101tri un par str\u0101d\u0101jo\u0161iem lauksaimniekiem pieejamu cenu identific\u0113t tom\u0101tu slim\u012bbas. S\u0101kot ar pamata CNN mode\u013ciem, kas apm\u0101c\u012bti PlantVillage datu kop\u0101, l\u012bdz uzman\u012bbas vad\u012bt\u0101m hibr\u012bdarhitekt\u016br\u0101m, kas sasniedz 99%+ precizit\u0101ti, tehnisk\u0101s iesp\u0113jas tagad p\u0101rsniedz izvieto\u0161anas infrastrukt\u016bru, kas pieejama liel\u0101kajai da\u013cai lauksaimniec\u012bbas uz\u0146\u0113mumu.<\/p>\n Ce\u013c\u0161 uz priek\u0161u ir skaidrs. Prec\u012bz\u0101s lauksaimniec\u012bbas sist\u0113mas, kas integr\u0113 uz att\u0113liem balst\u012btu slim\u012bbu noteik\u0161anu ar lietu interneta sensoru t\u012bkliem, bezpilota lidapar\u0101tu uzraudz\u012bbu un prognoz\u0113jo\u0161iem laikapst\u0101k\u013cu mode\u013ciem, \u0161aj\u0101 desmitgad\u0113 noteiks konkur\u0113tsp\u0113j\u012bgu tom\u0101tu ra\u017eo\u0161anu. M\u0101ksl\u012bgais intelekts prec\u012bz\u0101s lauksaimniec\u012bbas tirg\u016b strauji aug. CAGR 15,1%<\/strong> virzien\u0101 $12,7 miljardi l\u012bdz 2034. gadam<\/strong> liecina, ka \u0161\u012bs invest\u012bcijas jau tiek \u012bstenotas pla\u0161\u0101 m\u0113rog\u0101.<\/p>","protected":false},"excerpt":{"rendered":" Inteli\u0123enta tom\u0101tu lapu slim\u012bbu noteik\u0161ana atrodas divu sp\u0113c\u012bgu sp\u0113ku, kas p\u0101rveido glob\u0101lo lauksaimniec\u012bbu, krustpunkt\u0101: pa\u0161as tom\u0101tu ra\u017eas ekonomiskais svars un\u2026<\/p>","protected":false},"author":210157960,"featured_media":13712,"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_hide_featured_image":false,"_jetpack_featured_image_duplicate_dismissed":0,"_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":[1657,1370],"tags":[],"class_list":["post-13709","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-precision-farming","category-scouting"],"acf":[],"yoast_head":"\n![\"K\u0101p\u0113c](\"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/06\/Why-Tomato-Cultivation-and-Disease-Detection-Matter.png?resize=564%2C564&ssl=1\")
<\/p>\nTom\u0101tu lapu slim\u012bbas: atkl\u0101\u0161anas darba p\u0101rskats<\/h2>\n
1. Bie\u017e\u0101k sastopam\u0101s tom\u0101tu lapu slim\u012bbas un to izrais\u012bt\u0101ji<\/h3>\n
2. Slim\u012bbas simptomi un kodola noteik\u0161anas izaicin\u0101jums<\/h3>\n
\n
Agr\u012bnas slim\u012bbu atkl\u0101\u0161anas kritisk\u0101 loma kult\u016braugu apsaimnieko\u0161an\u0101<\/h2>\n
\n
Prec\u012bz\u0101 lauksaimniec\u012bba un vied\u0101s slim\u012bbu uzraudz\u012bbas sist\u0113mas<\/h2>\n
1. Prec\u012bz\u0101s lauksaimniec\u012bbas pamatj\u0113dzieni<\/h3>\n
2. Slim\u012bbu uzraudz\u012bbas tehnolo\u0123ijas m\u016bsdienu prec\u012bzaj\u0101 lauksaimniec\u012bb\u0101<\/h3>\n
![\"Slim\u012bbu](\"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/06\/Disease-Monitoring-Technologies-in-Modern-Precision-Farming.png?resize=540%2C540&ssl=1\")
<\/p>\nDzi\u013c\u0101 m\u0101c\u012b\u0161an\u0101s tom\u0101tu lapu slim\u012bbu noteik\u0161anai<\/h2>\n
1. Dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s pamati att\u0113lu klasifik\u0101cijai<\/h3>\n
![\"Dzi\u013c\u0101](\"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/06\/Deep-Learning-for-Tomato-Leaf-Disease-Detection.png?resize=563%2C563&ssl=1\")
<\/p>\n2. K\u0101p\u0113c dzi\u013c\u0101 m\u0101c\u012b\u0161an\u0101s p\u0101rsp\u0113j tradicion\u0101l\u0101s metodes<\/h3>\n
\n
Geopard Agriculture izmanto\u0161ana slim\u012bbu noteik\u0161an\u0101 re\u0101los laukos<\/h2>\n
Ko Geopard pied\u0101v\u0101 tom\u0101tu slim\u012bbu apkaro\u0161anai<\/h3>\n
\n
Mobil\u0101 izpilde, zi\u0146o\u0161ana un \u0101rk\u0101rtas br\u012bdin\u0101jumi<\/h3>\n
![\"Geopard](\"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/06\/Use-of-Geopard-Agriculture-in-Disease-Detection-for-Real-Fields.png?resize=580%2C580&ssl=1\")
<\/p>\nDatu kopas sagatavo\u0161ana noteik\u0161anas mode\u013ca izveidei<\/h2>\n
1. Att\u0113lu ieg\u016b\u0161anas avoti<\/h3>\n
2. Datu pirmapstr\u0101des so\u013ci<\/h3>\n
\n
3. Datu kopas anot\u0101cija un mar\u0137\u0113\u0161ana<\/h3>\n
Dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s arhitekt\u016bras, ko izmanto tom\u0101tu slim\u012bbu noteik\u0161anai<\/h2>\n
1. Standarta konvolucion\u0101lie neironu t\u012bkli<\/h3>\n
2. M\u0101c\u012bbu p\u0101rne\u0161ana ar iepriek\u0161 apm\u0101c\u012bt\u0101m arhitekt\u016br\u0101m<\/h3>\n
3. Hibr\u012bdie un uzlabotie mode\u013ci<\/h3>\n
![\"Dzi\u013c\u0101s](\"https:\/\/i0.wp.com\/geopard.tech\/wp-content\/uploads\/2026\/06\/Deep-Learning-Architectures-Used-for-Tomato-Disease-Detection.png?resize=605%2C605&ssl=1\")
<\/p>\nInteli\u0123enta slim\u012bbu atkl\u0101\u0161anas sist\u0113ma<\/h2>\n
1. Sist\u0113mas arhitekt\u016bra<\/h3>\n
2. Piln\u012bgas noteik\u0161anas sist\u0113mas darbpl\u016bsma praks\u0113<\/h3>\n
Veiktsp\u0113jas nov\u0113rt\u0113\u0161anas r\u0101d\u012bt\u0101ji<\/h2>\n
Tom\u0101tu slim\u012bbu noteik\u0161ana re\u0101llaik\u0101: izvieto\u0161ana<\/h2>\n
1. Viedt\u0101lru\u0146u lietojumprogrammas slim\u012bbu diagnostikai mobilaj\u0101m ier\u012bc\u0113m<\/h3>\n
2. Edge AI izvieto\u0161ana: secin\u0101jumu veik\u0161ana ier\u012bc\u0113<\/h3>\n
3. Uz droniem balst\u012btas uzraudz\u012bbas sist\u0113mas liela m\u0113roga saimniec\u012bbu uzraudz\u012bbai<\/h3>\n
Dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s slim\u012bbu atkl\u0101\u0161anas izaicin\u0101jumi<\/h2>\n
Jaun\u0101kie sasniegumi intelektu\u0101laj\u0101 slim\u012bbu noteik\u0161an\u0101<\/h2>\n
Turpm\u0101kie p\u0113t\u012bjumu virzieni: kas \u0161ai jomai joproj\u0101m ir nepiecie\u0161ams<\/h2>\n
Secin\u0101jums<\/h2>\n