A unified RT-DETR–Mask2Former framework for segmentation-guided plant leaf disease detection and classification in real-world agricultural environments
7 affiliations
- 1Department of Electronics and Communication Engineering, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology
- 2Department of General Medicine, Saveetha Institute of Medical and Technical Sciences
- 3Department of Computer Science and Engineering, Vel Tech Rangarajan Dr. Sagunthala R&D Institute of Science and Technology
- 4Division of Data Science and Artificial Intelligence, Computer Science and Engineering, Chalmers University of Technology
- 5Division of Data Science and Artificial Intelligence, Computer Science and Engineering, University of Gothenburg
- 6Department of Computers, Electronics and Automation, Stefan cel Mare University of Suceava
- 7Department of Electronic and Telecommunication, Constanta Maritime University
Abstract
Plant leaf disease image analysis under real-world field conditions remains challenging due to cluttered backgrounds, illumination variation, overlapping leaves, scale changes, and irregular lesion patterns. This study presents a unified RT-DETR–Mask2Former framework for segmentation-guided plant leaf disease detection and classification. In the proposed pipeline, RT-DETR first localizes leaf regions from complex agricultural images, after which Mask2Former generates fine-grained lesion masks within the detected regions. The refined lesion masks are used to isolate disease-affected regions, enabling EfficientNet-B4 to extract disease-focused feature representations for classification. An EMA-normalized adaptive task-weighting procedure was applied with AdamW optimization to modulate the relative gradient scales of the separately supervised modules. Model development and internal evaluation were conducted using leakage-controlled, group-aware PlantVillage (PV) partitions, whereas 4,200 field-acquired images were reserved exclusively for independent external evaluation. On the complete 38-category PV internal test, accuracy and macro-F1 were 98.34% and 97.97%. On the nine-category source-matched subset, accuracy and macro-F1 were 97.43% and 96.77%, versus 84.93% and 80.80% in the field. PV-to-field performance declined from 95.63% to 89.28% for leaf-localization [email protected] and from 92.47%/86.00% to 77.90%/63.80% for lesion-positive Dice/IoU. Lesion guidance improved field macro-F1 from 73.97% to 80.80%. The complete pipeline required approximately 285 GFLOPs and 63 ms per image (16 FPS), demonstrating a practical accuracy–efficiency trade-off while retaining a measurable field-domain gap.
Related
- An integrated instance segmentation and deep feature fusion model for robust leaf species classification in natural environments
- A Computer Vision-Based Framework for Multilevel Egg Fertility Classification Using L*a*b* Color Segmentation, Wavelet Texture Features, and Support Vector Machine
- Chili-Seg: a high-precision detection and instance segmentation model for chili flowers in complex environments
- Pollinator Conservation Strategies: The Effectiveness of Agroecological Infrastructure in Agricultural Environments.
- From segmentation to morphometry: A post-processed U-Net framework for corneal endothelial cell analysis
BibTeX
@article{kumar2026unified,
title = {{A unified RT-DETR–Mask2Former framework for segmentation-guided plant leaf disease detection and classification in real-world agricultural environments}},
author = {C H. Mohan Sai Kumar and Golthu Anusha and V. Dhilip Kumar and Oana Geman and Roxana Toderean and Mihaela Hnatiuc},
journal = {Frontiers in Plant Science},
year = {2026},
month = oct,
doi = {10.3389/fpls.2026.1931819},
url = {https://www.frontiersin.org/articles/10.3389/fpls.2026.1931819/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A unified RT-DETR–Mask2Former framework for segmentation-guided plant leaf disease detection and classification in real-world agricultural environments
AU - C H. Mohan Sai Kumar
AU - Golthu Anusha
AU - V. Dhilip Kumar
AU - Oana Geman
AU - Roxana Toderean
AU - Mihaela Hnatiuc
PY - 2026
DA - 2026/10/15
JO - Frontiers in Plant Science
DO - 10.3389/fpls.2026.1931819
UR - https://www.frontiersin.org/articles/10.3389/fpls.2026.1931819/full
AB - Plant leaf disease image analysis under real-world field conditions remains challenging due to cluttered backgrounds, illumination variation, overlapping leaves, scale changes, and irregular lesion patterns. This study presents a unified RT-DETR–Mask2Former framework for segmentation-guided plant leaf disease detection and classification. In the proposed pipeline, RT-DETR first localizes leaf regions from complex agricultural images, after which Mask2Former generates fine-grained lesion masks within the detected regions. The refined lesion masks are used to isolate disease-affected regions, enabling EfficientNet-B4 to extract disease-focused feature representations for classification. An EMA-normalized adaptive task-weighting procedure was applied with AdamW optimization to modulate the relative gradient scales of the separately supervised modules. Model development and internal evaluation were conducted using leakage-controlled, group-aware PlantVillage (PV) partitions, whereas 4,200 field-acquired images were reserved exclusively for independent external evaluation. On the complete 38-category PV internal test, accuracy and macro-F1 were 98.34% and 97.97%. On the nine-category source-matched subset, accuracy and macro-F1 were 97.43% and 96.77%, versus 84.93% and 80.80% in the field. PV-to-field performance declined from 95.63% to 89.28% for leaf-localization [email protected] and from 92.47%/86.00% to 77.90%/63.80% for lesion-positive Dice/IoU. Lesion guidance improved field macro-F1 from 73.97% to 80.80%. The complete pipeline required approximately 285 GFLOPs and 63 ms per image (16 FPS), demonstrating a practical accuracy–efficiency trade-off while retaining a measurable field-domain gap.
SN - 1664-462X
ER -