This framework demonstrates improved detection of small agricultural fire targets using UAV technology, suggesting a significant advancement in monitoring systems.
Key Points
The goal is to improve the early detection of agricultural fires by leveraging UAV-based remote sensing technologies.
Developed a hybrid dataset using a ROI-guided synthesis pipeline with latent diffusion models.
Introduced EF-YOLO, a detector optimized for small target detection with SPD-Conv and a high-resolution P2 head.
Implemented a Dual-Path Frequency–Spatial Enhancement (DP-FSE) module to reduce background noise.
EF-YOLO achieved a mAP of 40.2% on sub-pixel targets, surpassing YOLOv8s by 15.4 percentage points.
The model indicates a recall of 88.7% for fire detection.
It processes at a real-time speed of 78 FPS, suitable for immediate application in agricultural monitoring systems.