Foxtail Millet Ear Detection Method Based on Attention Mechanism and Improved YOLOv5

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate identification of foxtail millet ears is challenging due to dense planting, morphological variations, shading, and complex backgrounds.
• Existing detection methods struggle with occlusion and small object detection in agricultural settings.

Purpose of the Study:

• To develop a lightweight and efficient object detection model for foxtail millet ears.
• To improve the accuracy and speed of foxtail millet ear detection in complex field environments.

Main Methods:

• Proposed a lightweight foxtail millet ear detection method based on an improved YOLOv5 architecture.
• Integrated the GhostNet module to reduce model parameters and computation.
• Incorporated the Coordinate Attention (CA) mechanism and utilized the Efficient Intersection over Union (EIOU) loss function.

Main Results:

• The improved YOLOv5 model achieved high performance metrics: 97.70% recall, 93.80% precision, 95.81% F1 score, and 96.60% mean Average Precision (mAP).
• Demonstrated significantly better detection performance compared to YOLOv3_tiny, YOLOv5-Mobilenetv3small, and YOLOv5-Shufflenetv2.
• Achieved a fast average detection time of 0.0181 s per image with a compact model size of 8.12 MB.

Conclusions:

• The proposed method effectively addresses challenges in detecting occluded and small foxtail millet ears.
• Provides technical support for rapid and accurate identification of foxtail millet ears, crucial for yield improvement and intelligent agriculture.