A YOLOv7 incorporating the Adan optimizer based corn pests identification method.

Area of Science:

• Agricultural Entomology
• Computer Vision
• Machine Learning

Background:

• Accurate identification of major corn pests (e.g., corn borer, armyworm, bollworm) is vital for effective pest management.
• Existing machine learning and neural network methods face challenges with high training costs and suboptimal recognition accuracy.

Purpose of the Study:

• To develop a more efficient and accurate maize pest identification system.
• To improve upon existing object detection models for agricultural applications.

Main Methods:

• A YOLOv7-based object detection model was employed for maize pest identification.
• The Adan optimizer was integrated into the YOLOv7 architecture to enhance computational efficiency and model robustness.
• Data augmentation techniques were utilized to create a comprehensive dataset for training.

Main Results:

• The proposed YOLOv7 model with the Adan optimizer achieved a mean Average Precision (mAP) of 96.69% and a precision of 99.95%.
• The improved model required only 1/2-2/3 of the computing power of the original YOLOv7 network.
• Performance metrics showed significant improvements compared to the original YOLOv7 and other common object detection models.

Conclusions:

• The YOLOv7 model incorporating the Adan optimizer offers a highly accurate and computationally efficient solution for corn pest detection.
• This approach represents a State-of-the-Art (SOTA) method for real-time, accurate pest identification in complex agricultural environments.