Small Object Detection in Agriculture: A Case Study on Durian Orchards Using EN-YOLO and Thermal Fusion

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Durian production in Southeast Asia faces significant yield and quality losses due to pests and diseases.
• Current manual inspection methods for durian pests and diseases are labor-intensive, inaccurate, and difficult to scale.

Purpose of the Study:

• To develop an enhanced deep learning model (EN-YOLO) for precise and automated detection of durian pests and diseases.
• To improve the accuracy, robustness, and scalability of pest and disease detection in durian cultivation.

Main Methods:

• Proposed EN-YOLO model integrating EfficientNet backbone and multimodal attention mechanisms.
• Utilized multimodal input: RGB, near-infrared, and thermal imaging for enhanced robustness.
• Optimized model architecture by removing redundant layers and adding a large-span residual edge.

Main Results:

• EN-YOLO achieved superior detection accuracy, generalization, and small-object recognition compared to YOLOv8, YOLOv5-EB, and Fieldsentinel-YOLO.
• Demonstrated 95.3% counting accuracy and strong performance in ablation and cross-scene tests.
• The system supports real-time drone deployment and integrates an expert knowledge base for decision support.

Conclusions:

• EN-YOLO offers an efficient, interpretable, and scalable solution for automated pest and disease management in smart agriculture.
• The multimodal approach enhances detection reliability under challenging environmental conditions.
• This technology facilitates intelligent decision-making for sustainable durian farming.