YOLO-ACE: Enhancing YOLO with Augmented Contextual Efficiency for Precision Cotton Weed Detection

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Effective weed management is crucial for cotton yield.
• Conventional deep learning models struggle with small/occluded weeds and high parameter counts.

Purpose of the Study:

• To introduce YOLO-ACE, an enhanced YOLOv5s model for precise and efficient weed detection in cotton.
• To improve detection accuracy and reduce computational load for agricultural applications.

Main Methods:

• YOLO-ACE integrates a Context Augmentation Module (CAM) and Selective Kernel Attention (SKAttention).
• A decoupled detection head separates classification and bounding box regression.
• The model was evaluated on the CottonWeedDet12 (CWD12) and CropWeed datasets.

Main Results:

• YOLO-ACE achieved mAP@0.5 of 95.3% and mAP@0.5:0.95 of 89.5% on the CWD12 dataset.
• The model demonstrated strong transferability, achieving 84.3% mAP@0.5 on the CropWeed dataset.
• YOLO-ACE surpasses previous benchmarks in accuracy and efficiency.

Conclusions:

• YOLO-ACE offers a robust solution for precise and parameter-efficient weed detection in cotton.
• The model's adaptability makes it suitable for diverse agricultural environments and crops.
• This advancement supports modern, demanding weed management strategies.