LiteMS-YOLO: a lightweight framework for small target detection in complex wheat field environments.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Wheat spike detection is crucial for accurate yield estimation in precision agriculture.
• Existing methods face challenges due to small target size, dense distribution, and complex field conditions.

Purpose of the Study:

• To develop a lightweight and efficient object detection framework for wheat spike detection.
• To improve spatial-semantic feature interaction and multi-scale feature representation for enhanced accuracy.

Main Methods:

• Proposed LiteMS-YOLO, a lightweight object detection framework based on YOLO26n.
• Integrated Feature Complementary Mapping (FCM) and Multi-Kernel Perception (MKP) modules.
• Implemented targeted redundancy reduction strategies to decrease model complexity.

Main Results:

• LiteMS-YOLO achieved mAP50 of 92.28% and mAP50-95 of 52.56% on a combined dataset.
• The model uses only 0.627 million parameters, a significant reduction compared to YOLO26n and YOLOv8n.
• Demonstrated competitive accuracy with substantially lower parameter count.

Conclusions:

• LiteMS-YOLO offers an excellent balance between detection accuracy and computational efficiency.
• The framework is well-suited for real-time deployment in resource-constrained agricultural environments.
• Facilitates improved wheat yield estimation through advanced precision agriculture techniques.