Real-time soybean pest detection system integrating UAV and Jetson based on improved YOLO

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Insect pests are a major threat to soybean yield and quality.
• Current pest detection models suffer from high computational complexity, excessive parameters, and low accuracy.

Purpose of the Study:

• To develop a lightweight, high-precision, and robust portable soybean pest recognition system.
• To overcome the limitations of existing pest detection models for smart agriculture.

Main Methods:

• Preprocessing a close-range dataset from an Unmanned Aerial Vehicle (UAV) perspective.
• Optimizing the backbone network with Hierarchical Weight Decoupling (HWD), C3 module with Kernel-2 decomposition and Multi-Channel Attention (C3K2-MCA), and Spatial Pyramid Pooling-Fast with Squeeze-and-Excitation Version 2 (SPPF-SEV2).
• Utilizing the Normalized Wasserstein Distance (NWD) loss function for improved accuracy and robustness.

Main Results:

• The optimized model achieved a high precision rate of 96.2%.
• The UAV-Jetson-YOLO system demonstrated rapid and cost-effective pest detection capabilities.
• The system is deployable on NVIDIA Jetson devices for practical smart agriculture applications.

Conclusions:

• The proposed lightweight and high-precision soybean pest recognition system effectively addresses the limitations of current models.
• The developed system offers a viable solution for rapid, cost-effective pest detection in smart agriculture.
• Further advancements in AI and UAV technology can significantly benefit agricultural practices.