{"id":11429,"date":"2025-04-02T23:23:55","date_gmt":"2025-04-02T21:23:55","guid":{"rendered":"https:\/\/geopard.tech\/?p=11429"},"modified":"2025-04-02T23:23:55","modified_gmt":"2025-04-02T21:23:55","slug":"dzilas-macisanas-datorredzes-lietojumprogrammu-loma-augu-slimibu-agrina-noteiksana","status":"publish","type":"post","link":"https:\/\/geopard.tech\/lv\/blog\/role-of-deep-learning-computer-vision-applications-for-early-plant-disease-detection\/","title":{"rendered":"Dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s datorredzes lietojumprogrammu loma augu slim\u012bbu agr\u012bn\u0101 noteik\u0161an\u0101"},"content":{"rendered":"

Augu slim\u012bbas klus\u012bb\u0101 apdraud glob\u0101lo p\u0101rtikas nodro\u0161in\u0101jumu, ik gadu izn\u012bcinot 10\u2013161 TP3 t kult\u016braugu un radot lauksaimniec\u012bbas nozarei 1 TP4 220 miljardu zaud\u0113jumus. Tradicion\u0101l\u0101s metodes, piem\u0113ram, manu\u0101las p\u0101rbaudes un laboratorijas testi, ir l\u0113nas, d\u0101rgas un bie\u017ei vien neuzticamas.<\/p>\n

Novatorisks 2025. gada p\u0113t\u012bjums,\u00a0\u201cDzi\u013c\u0101 m\u0101c\u012b\u0161an\u0101s un datorredze augu slim\u012bbu noteik\u0161an\u0101\u201d<\/em>\u00a0(Upadhyay et al.) atkl\u0101j, k\u0101 m\u0101ksl\u012bg\u0101 intelekta augu slim\u012bbu noteik\u0161ana un datorredzes lauksaimniec\u012bba p\u0101rveido lauksaimniec\u012bbu.<\/p>\n

K\u0101p\u0113c agr\u012bna augu slim\u012bbu atkl\u0101\u0161ana ir svar\u012bga glob\u0101lajai p\u0101rtikas dro\u0161\u012bbai<\/strong><\/h2>\n

Lauksaimniec\u012bb\u0101 str\u0101d\u0101 281\u00a0000\u00a0000\u00a0000 no pasaules darbasp\u0113ka, un t\u0101das valstis k\u0101 Indija, \u0136\u012bna un ASV ir vado\u0161\u0101s kult\u016braugu ra\u017eo\u0161anas jom\u0101. Neskatoties uz to, augu slim\u012bbas, ko izraisa s\u0113n\u012btes, bakt\u0113rijas un v\u012brusi, samazina ra\u017eu un rada spriedzi ekonomik\u0101.<\/p>\n

Piem\u0113ram, r\u012bsu spr\u0101dzienb\u012bstam\u0101 slim\u012bba skartajos re\u0123ionos samazina ra\u017eu par 30\u201350%, savuk\u0101rt citrusaug\u013cu za\u013co\u0161ana kop\u0161 2005. gada ir izn\u012bcin\u0101jusi 70% Floridas apels\u012bnu birz\u012bs. Agr\u012bna atkl\u0101\u0161ana ir kritiski svar\u012bga, ta\u010du daudziem lauksaimniekiem tr\u016bkst piek\u013cuves moderniem instrumentiem vai zin\u0101\u0161an\u0101m.<\/p>\n

\u0160eit noder m\u0101ksl\u012bg\u0101 intelekta vad\u012bta slim\u012bbu noteik\u0161ana, pied\u0101v\u0101jot \u0101trus, pieejamus un prec\u012bzus risin\u0101jumus, kas p\u0101rsp\u0113j tradicion\u0101l\u0101s metodes.<\/p>\n

K\u0101 m\u0101ksl\u012bgais intelekts un datorredze atkl\u0101j kult\u016braugu slim\u012bbas<\/strong><\/h2>\n

P\u0113t\u012bjum\u0101 tika analiz\u0113ti 278 p\u0113t\u012bjumi, lai izskaidrotu, k\u0101 darbojas m\u0101ksl\u012bg\u0101 intelekta augu slim\u012bbu noteik\u0161anas sist\u0113mas. Vispirms kameras vai sensori uztver kult\u016braugu att\u0113lus. P\u0113c tam \u0161ie att\u0113li tiek apstr\u0101d\u0101ti, izmantojot algoritmus, lai identific\u0113tu slim\u012bbas paz\u012bmes.<\/p>\n

Piem\u0113ram,\u00a0RGB kameras<\/strong>\u00a0uz\u0146emiet kr\u0101su fotoatt\u0113lus, lai paman\u012btu redzamus simptomus, piem\u0113ram, lapu plankumus, savuk\u0101rt hiperspektr\u0101l\u0101s kameras atkl\u0101j sl\u0113ptus stresa sign\u0101lus, analiz\u0113jot simtiem gaismas vi\u013c\u0146u garumu.<\/p>\n

Kad att\u0113li ir uz\u0146emti, tie tiek pak\u013cauti pirmapstr\u0101dei, lai uzlabotu kvalit\u0101ti. T\u0101das metodes k\u0101 sliek\u0161\u0146u noteik\u0161ana izol\u0113 slim\u012bbas zonas p\u0113c kr\u0101sas, un malu noteik\u0161ana kart\u0113 boj\u0101jumu vai kr\u0101sas izmai\u0146u robe\u017eas.<\/p>\n

\"K\u0101<\/p>\n

P\u0113c tam dzi\u013c\u0101s m\u0101c\u012b\u0161an\u0101s mode\u013ci analiz\u0113 iepriek\u0161 apstr\u0101d\u0101tos datus.\u00a0Konvolucion\u0101lie neironu t\u012bkli (CNN)<\/strong>, visizplat\u012bt\u0101kie m\u0101ksl\u012bg\u0101 intelekta r\u012bki lauksaimniec\u012bb\u0101, sken\u0113 att\u0113lus sl\u0101ni pa sl\u0101nim, lai identific\u0113tu t\u0101dus mode\u013cus k\u0101 neparastas tekst\u016bras vai kr\u0101sas.<\/p>\n

2022. gada tiesas proces\u0101,\u00a0ResNet50<\/strong>\u2014 popul\u0101rs CNN modelis \u2014 tom\u0101tu slim\u012bbu diagnostic\u0113\u0161an\u0101 sasniedza 99,07% precizit\u0101ti.<\/p>\n

Tikm\u0113r,\u00a0Redzes transformatori (ViT)<\/strong>\u00a0sadal\u012bt att\u0113lus apgabalos un p\u0113t\u012bt to attiec\u012bbas, atdarinot to, k\u0101 cilv\u0113ki analiz\u0113 kontekstu. \u0160\u012b pieeja pal\u012bdz\u0113ja noteikt v\u012bnogul\u0101ju v\u0113nu att\u012br\u012b\u0161anas v\u012brusu ar 71% precizit\u0101ti 2020.\u00a0gada p\u0113t\u012bjum\u0101.<\/p>\n

\u201cLauksaimniec\u012bbas n\u0101kotne nav cilv\u0113ku aizst\u0101\u0161ana, bet gan vi\u0146u apr\u012bko\u0161ana ar inteli\u0123entiem instrumentiem.\u201d<\/em><\/p><\/blockquote>\n

Modernu sensoru loma m\u016bsdienu lauksaimniec\u012bb\u0101<\/strong><\/h2>\n

Da\u017e\u0101di sensori pied\u0101v\u0101 unik\u0101las priek\u0161roc\u012bbas prec\u012bzaj\u0101 lauksaimniec\u012bb\u0101.\u00a0RGB kameras<\/strong>, lai gan ir pieejamas un viegli lietojamas, ierobe\u017eot\u0101s spektr\u0101l\u0101s deta\u013cas d\u0113\u013c t\u0101s c\u012bn\u0101s ar agr\u012bn\u0101s stadijas slim\u012bb\u0101m. Turpret\u012b,\u00a0hiperspektr\u0101l\u0101s kameras<\/strong>\u00a0uztvert datus simtiem gaismas vi\u013c\u0146u garumu, atkl\u0101jot stresa sign\u0101lus, kas nav redzami ar neapbru\u0146otu aci.<\/p>\n

Piem\u0113ram, p\u0113tnieki izmantoja hiperspektr\u0101lo att\u0113lveido\u0161anu, lai 2022. gad\u0101 diagnostic\u0113tu \u0101bolu Valsa v\u0113zi ar precizit\u0101ti 98%. Tom\u0113r \u0161\u012bs kameras maks\u0101\u00a010\u00a0000\u2013<\/span>10<\/span>,<\/span>000\u2013<\/span><\/span><\/span><\/span>50 000, padarot tos p\u0101r\u0101k d\u0101rgus mazajiem lauksaimniekiem.<\/p>\n

Termisk\u0101s kameras<\/strong>\u00a0sniegt citu skat\u012bjumu, m\u0113rot infekciju izrais\u012bt\u0101s temperat\u016bras izmai\u0146as. 2019.\u00a0gada p\u0113t\u012bjum\u0101 tika atkl\u0101ts, ka ar citrusaug\u013cu za\u013cumu infic\u0113t\u0101m lap\u0101m ir at\u0161\u0137ir\u012bgi siltuma mode\u013ci, kas \u013cauj to agr\u012bni atkl\u0101t.<\/p>\n

Tikm\u0113r,\u00a0multispektr\u0101l\u0101s kameras<\/strong>\u2014vidusce\u013ca risin\u0101jums \u2014 izsekot hlorofila l\u012bmenim, lai nov\u0113rt\u0113tu augu vesel\u012bbu.<\/p>\n

\u0160ie sensori 2014. gad\u0101 kart\u0113ja kvie\u0161u sv\u012btraino r\u016bsu, pal\u012bdzot lauksaimniekiem efekt\u012bv\u0101k m\u0113r\u0137\u0113t apstr\u0101des metodes. Neskatoties uz to priek\u0161roc\u012bb\u0101m, sensoru izmaksas un vides faktori, piem\u0113ram, v\u0113j\u0161 vai nevienm\u0113r\u012bgs apgaismojums, joproj\u0101m rada probl\u0113mas.<\/p>\n

Publisk\u0101s datu kopas: m\u0101ksl\u012bg\u0101 intelekta lauksaimniec\u012bbas mugurkauls<\/strong><\/h2>\n

Uzticamu m\u0101ksl\u012bg\u0101 intelekta mode\u013cu apm\u0101c\u012bbai ir nepiecie\u0161ams milz\u012bgs apjoms mar\u0137\u0113tu datu.\u00a0PlantVillage datu kopa<\/strong>, bezmaksas resurss ar 87\u00a0000 att\u0113liem, kuros att\u0113loti 14 kult\u016braugi un 26 slim\u012bbas, ir k\u013cuvis par zelta standartu p\u0113tniekiem.<\/p>\n

Vair\u0101k nek\u0101 901\u00a0TP3T p\u0113t\u012bjum\u0101, kas cit\u0113ti rakst\u0101, \u0161is datu kopums tika izmantots mode\u013cu apm\u0101c\u012bbai un test\u0113\u0161anai. V\u0113l viens svar\u012bgs resurss ir\u00a0Maniokas slim\u012bbas datu kopa<\/strong>, ietver 11\u00a0670 maniokas moza\u012bkas slim\u012bbas att\u0113lus un ar CNN mode\u013ciem sasniedza 96% precizit\u0101ti.<\/p>\n

Tom\u0113r joproj\u0101m past\u0101v nepiln\u012bbas. Ret\u0101m slim\u012bb\u0101m, piem\u0113ram, prie\u017eu nematodei, ir maz\u0101k nek\u0101 100 mar\u0137\u0113tu att\u0113lu, kas ierobe\u017eo m\u0101ksl\u012bg\u0101 intelekta sp\u0113ju t\u0101s noteikt. Turkl\u0101t liel\u0101kaj\u0101 da\u013c\u0101 datu kopu ir laboratorij\u0101 uz\u0146emti att\u0113li, kuros nav \u0146emti v\u0113r\u0101 re\u0101l\u0101s pasaules main\u012bgie, piem\u0113ram, laika apst\u0101k\u013ci vai apgaismojums.<\/p>\n

Lai risin\u0101tu \u0161o probl\u0113mu, t\u0101di projekti k\u0101 AI4Ag izmanto lauku att\u0113lu p\u016b\u013ca resursus, lai apkopotu tos no lauksaimniekiem vis\u0101 pasaul\u0113, cen\u0161oties izveidot stabil\u0101kas un re\u0101listisk\u0101kas datu kopas.<\/p>\n

M\u0101ksl\u012bg\u0101 intelekta veiktsp\u0113jas m\u0113r\u012b\u0161ana: precizit\u0101te, precizit\u0101te un kas vair\u0101k<\/strong><\/h2>\n

\"M\u0101ksl\u012bg\u0101<\/p>\n

P\u0113tnieki izmanto vair\u0101kus r\u0101d\u012bt\u0101jus, lai nov\u0113rt\u0113tu m\u0101ksl\u012bg\u0101 intelekta augu slim\u012bbu noteik\u0161anas sist\u0113mas.\u00a0Precizit\u0101te<\/strong>\u2014pareizo diagno\u017eu procentu\u0101l\u0101 da\u013ca \u2014 sv\u0101rst\u0101s no\u00a076.9% agr\u012bnajos mode\u013cos<\/strong>\u00a0uz\u00a099.97% modern\u0101s sist\u0113m\u0101s, piem\u0113ram, EfficientNet-B5<\/strong>.<\/p>\n

Tom\u0113r precizit\u0101te vien var b\u016bt maldino\u0161a. Precizit\u0101te m\u0113ra, cik daudz atz\u012bm\u0113to slim\u012bbu ir re\u0101las (izvairoties no viltus trauksm\u0113m), savuk\u0101rt atsauk\u0161ana izseko, cik daudz faktisku infekciju tiek atkl\u0101tas.<\/p>\n

Piem\u0113ram,\u00a0Maska R-CNN<\/strong>, objektu noteik\u0161anas modelis, sasniedza 93,5% precizit\u0101ti zeme\u0146u antraknozes noteik\u0161an\u0101, bet tikai 45% precizit\u0101ti kokvilnas sak\u0146u puves noteik\u0161an\u0101.<\/p>\n

Port\u0101ls\u00a0F1 rezult\u0101ts<\/strong>\u00a0l\u012bdzsvaro precizit\u0101ti un atcer\u0113\u0161anos, pied\u0101v\u0101jot holistisku snieguma skat\u012bjumu. 2023.\u00a0gada izm\u0113\u0123in\u0101jum\u0101,\u00a0PlantViT<\/strong>\u2014hibr\u012bds m\u0101ksl\u012bg\u0101 intelekta modelis \u2014 ieguva 98,61% F1 v\u0113rt\u0113jumu PlantVillage datu kop\u0101.<\/p>\n

Objektu noteik\u0161anai,\u00a0vid\u0113j\u0101 precizit\u0101te (mAP)<\/strong>\u00a0ir kritiski svar\u012bgs.\u00a0\u0100tr\u0101ks R-CNN<\/strong>, popul\u0101rs modelis, \u0101bolu slim\u012bbu izm\u0113\u0123in\u0101jumos sasniedza 73,07% mAP, kas noz\u012bm\u0113, ka vairum\u0101 gad\u012bjumu tas pareizi atrada un klasific\u0113ja infekcijas.<\/p>\n

Probl\u0113mas, kas kav\u0113 m\u0101ksl\u012bg\u0101 intelekta att\u012bst\u012bbu lauksaimniec\u012bb\u0101<\/strong><\/h2>\n

Neskatoties uz t\u0101 potenci\u0101lu, m\u0101ksl\u012bg\u0101 intelekta vad\u012bta slim\u012bbu atkl\u0101\u0161ana saskaras ar \u0161\u0137\u0113r\u0161\u013ciem. Pirmk\u0101rt, datu tr\u016bkums nomoka retas vai jaunas slim\u012bbas.<\/p>\n