Lightweight Detection System with Global Attention Network (GloAN) for Rice Lodging

Area of Science:

• Agricultural Science
• Computer Vision
• Remote Sensing

Background:

• Rice lodging significantly impacts crop yield and quality, necessitating efficient monitoring.
• Traditional manual detection methods are labor-intensive and time-consuming, leading to delayed interventions.
• Internet of Things (IoT) and unmanned aerial vehicles (UAVs) offer advanced solutions for crop stress monitoring.

Purpose of the Study:

• To develop a novel, lightweight detection system for identifying rice lodging areas using UAVs.
• To improve the efficiency and accuracy of rice lodging diagnosis.
• To reduce production losses associated with rice lodging.

Main Methods:

• Leveraging UAVs for acquiring rice growth distribution data.
• Proposing a global attention network (GloAN) for efficient and accurate detection of lodging areas.
• Utilizing knowledge distillation to test the generalization ability of GloAN with other models (Xception, VGG, ResNet, MobileNetV2).

Main Results:

• The proposed GloAN system achieved a significant increase in detection accuracy with minimal computational overhead.
• GloAN demonstrated strong generalization capabilities when integrated with other deep learning models.
• An optimal mean intersection over union (mIoU) of 92.85% was achieved, highlighting the system's effectiveness.

Conclusions:

• The developed UAV-based system with GloAN offers a flexible and efficient solution for rice lodging detection.
• This technology can accelerate diagnostic processing and mitigate economic losses in rice farming.
• The findings support the adoption of advanced remote sensing techniques for precision agriculture.