DINOv3 Meets YOLO26 for Weed Detection in Vegetable Crops

Researchers developed a foundational crop-weed detection model combining DINOv3 vision transformer with YOLO26 architecture, achieving significant improvements in precision agriculture applications. The model showed up to 14% better performance on cross-domain datasets while maintaining real-time processing at 28.5 fps despite increased computational requirements.

• →DINOv3-YOLO26 model achieved up to 5.4% mAP50 improvement on current season data and 14% on historical datasets compared to standard YOLO26.
• →The research addresses the critical limitation of scarce annotated weed-crop datasets by curating 199,388 filtered images from an initial collection of 618,642.
• →Despite 45.6% more parameters and 2.9x increase in inference latency, the model maintains real-time performance at approximately 28.5 frames per second.
• →The integration uses a dual-backbone architecture with feature alignment loss to enhance feature fusion with minimal computational overhead.
• →The curated dataset and software programs will be made publicly available, potentially advancing precision agriculture research.