A Knowledge-Guided refinement framework for Late-Season Rice–Weed segmentation using UAV imagery under canopy closure

Key Points

• The aim is to develop a framework for accurately mapping rice and weeds using UAV imagery during late-season.
• Proposed a knowledge-guided refinement framework for segmentation.
• Utilized WeedGraphNet for improved performance compared to U-Net baselines.
• Evaluated performance using held-out tiles of UAV imagery.
• Achieved overall accuracy of 0.974 and mean IoU of 0.883.
• Class-specific IoUs were 0.970 for rice and 0.820 for weed.
• Weed F1-score reached 0.901, indicating strong classification ability.