Intelligent Inter- and Intra-Row Early Weed Detection in Commercial Maize Crops.

Area of Science:

• Agricultural Science
• Computer Science
• Artificial Intelligence

Background:

• Intra-row weed competition significantly reduces crop productivity due to proximity and occlusion.
• Effective weed management is crucial for optimizing crop yields and reducing operational costs.

Purpose of the Study:

• To evaluate deep learning models for accurate weed and crop identification in maize fields.
• To compare the performance of Faster R-CNN, RT-DETR, and YOLOv11 for weed detection.

Main Methods:

• A dataset of maize fields was created, including three weed species: Cyperus rotundus, Echinochloa crus-galli, and Solanum nigrum.
• Deep learning models (Faster R-CNN, RT-DETR, YOLOv11) were trained and evaluated for weed and crop identification.
• Hardware evaluations were conducted to assess field deployment viability.

Main Results:

• YOLOv11 achieved the highest performance with 97.5% mean average precision (mAP) at 34 frames per second (FPS).
• Faster R-CNN and RT-DETR showed mAP of 91.9% and 97.2% at 11 and 27 FPS, respectively.
• YOLOv11m was identified as the most suitable for field deployment due to high precision (94.4% mAP) and lower energy use.

Conclusions:

• Advanced deep learning models, particularly YOLOv11, are effective for precise inter- and intra-row weed management.
• Early-stage weed detection with minimal crop interference is feasible using these AI technologies.
• Integration into agricultural machinery can enhance weed control, lower costs, and support sustainable farming.