DGS-Yolov7-Tiny: a lightweight pest and disease target detection model suitable for edge computing environments

Area of Science:

• Agricultural Technology
• Computer Vision
• Artificial Intelligence

Background:

• Traditional pest detection models are computationally intensive, limiting real-time use in edge computing.
• Efficient and accurate pest detection is crucial for modern agriculture and crop health management.

Purpose of the Study:

• To develop a lightweight pest detection model optimized for edge computing environments.
• To improve the precision and efficiency of pest detection in smart agriculture applications.

Main Methods:

• Proposed DGS-YOLOv7-Tiny, a lightweight model based on YOLOv7-Tiny.
• Incorporated a Global Attention Module for enhanced context aggregation and small object detection.
• Introduced DGSConv, a novel fusion convolution, to reduce parameters while retaining feature information.
• Replaced Leaky ReLU with SiLU and CIOU with SIOU to improve gradient flow and convergence speed.

Main Results:

• DGS-YOLOv7-Tiny achieved 95.53% precision, 92.88% recall, and 96.42% mAP@0.5 on a tomato leaf pest dataset.
• The model has 4.43 million parameters and 10.2 GFLOPs computational complexity.
• Achieved a high inference speed of 168 FPS, demonstrating suitability for edge computing.

Conclusions:

• DGS-YOLOv7-Tiny offers a significant improvement in pest detection efficiency and computational requirements for edge devices.
• The model provides a practical and effective solution for real-time pest detection in smart agriculture.
• This research holds substantial theoretical and practical value for advancing agricultural technology.