A peanut and weed detection model used in fields based on BEM-YOLOv7-tiny

Area of Science:

• Agricultural Engineering
• Computer Vision
• Machine Learning

Background:

• Weed characteristics in peanut fields vary significantly with growth stages, necessitating adaptable detection models.
• Current mechanical weeding systems require improved real-time identification of peanuts and weeds for different weeding periods.

Purpose of the Study:

• To develop a generalized peanut and weed detection model for intelligent mechanical weeding in peanut fields.
• To enhance the accuracy and efficiency of weed identification and localization across various weeding stages.

Main Methods:

• Proposed the BEM-YOLOv7-tiny target detection model incorporating ECA, MHSA, and BiFPN modules.
• Utilized the SIoU loss function to optimize model convergence and field detection performance.
• Evaluated model performance based on precision, recall, mAP, F1-score, positioning error, and detection speed.

Main Results:

• The BEM-YOLOv7-tiny model demonstrated improved performance metrics compared to the original YOLOv7-tiny for both weed and all targets.
• Achieved precision, recall, mAP, and F1 improvements of 1.6%, 4.9%, 4.4%, and 3.2% for weed targets, respectively.
• Exhibited a peanut positioning offset error below 16 pixels and a detection speed of 33.8 f/s, meeting real-time requirements.

Conclusions:

• The BEM-YOLOv7-tiny model offers a robust solution for real-time peanut and weed detection and localization.
• This technology provides essential technical support for the advancement of intelligent mechanical weeding in peanut cultivation.
• The model's adaptability to different weeding periods addresses a critical need in precision agriculture.