A Deep Learning-Based Semantic Segmentation Model for Discriminating Cultivated Plants from Weeds in Agricultural Scenes

Eshonqulova, Feruza

Al-Farg'oniy avlodlari · 2026-yil

Annotatsiya

This article analyzes the problem of pixel-level separation of crop plants from weeds in agricultural images. In this study, the semantic segmentation models DeepLabV3+, U-Net, and U-Net++ were evaluated under identical conditions using the publicly available CWFID dataset. To reduce severe class imbalance, a combination of Dice and Focal loss functions was applied. According to the obtained results, the U-Net++ model achieved the highest performance on the test set, reaching 77.87% mIoU and 97.57% pixel accuracy, confirming its high effectiveness. The research findings can serve as a methodological basis for developing specialized local models for agricultural crops in future studies.

Maqola ma’lumotlari
MualliflarEshonqulova, Feruza
JurnalAl-Farg'oniy avlodlari
Nashr sanasi2026-05-30
Son2
Betlar265-272
TilRus

Kalit so‘zlar

deep learning, semantic segmentation, U-Net++, DeepLabV3+, U-Net, convolutional neural network, weed detection, precision agriculture, CWFID, глубoкoе oбучение, семaнтическaя сегментaция, U-Net++, DeepLabV3+, U-Net, свертoчнaя нейрoннaя сеть, oбнaружение сoрнякoв, тoчнoе земледелие, CWFID, chuqur o'rganish, semantik segmentatsiya, U-Net++, DeepLabV3+, U-Net, konvolyutsion neyron tarmoq, begona o'tlarni aniqlash, aniq dehqonchilik, CWFID

Ilmiy soha

Al-Farg'oniy avlodlari jurnalidan boshqa maqolalar

Al-Farg'oniy avlodlari — barcha maqolalar