IV-YOLO: an information vortex-based progressive fusion method for accurate rice detection.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Remote Sensing

Background:

• UAV remote sensing in precision agriculture faces challenges with adhered rice plant features and background interference.
• Traditional models struggle with individual plant-level detection due to these image complexities.

Purpose of the Study:

• To develop an advanced deep learning model for accurate individual rice plant detection.
• To overcome limitations of existing models in handling feature adhesion and background clutter in UAV imagery.

Main Methods:

• Proposed an Information Vortex-based progressive fusion YOLO (IV-YOLO) model.
• Introduced a Multi-scale Spiral Information Vortex (MSIV) module for feature disentanglement and background decoupling.
• Constructed a Gradual Feature Fusion Neck (GFEN) for effective feature representation.

Main Results:

• The IV-YOLO model achieved a Precision of 0.8581 on the DRPD dataset.
• Outperformed YOLOv5-YOLOv11 and FRPNet across all evaluated metrics.
• Demonstrated superior performance in disentangling adhered features and reducing background interference.

Conclusions:

• IV-YOLO offers a robust technical solution for individual rice plant monitoring.
• The model facilitates the large-scale implementation of precision agriculture through accurate detection.
• The developed modules effectively address challenges in UAV-based agricultural remote sensing.