Related Experiment Video
Updated: Jan 7, 2026
08:16
Collecting and Processing Drone-based Remotely Sensed Data for Use in Forest Recovery Monitoring
Published on: October 24, 2025
416
BDM-YOLOv8n: A high-performance model for accurate fire detection in aerial imagery
Laohu Yuan1, Peng Zhou1, Zhiyuan Wang1
1College of Aerospace Engineering, Shenyang Aerospace University, Shenyang, 110136, China.
Summary
This study introduces the BMD-YOLOv8n model for enhanced Unmanned Aerial Vehicle (UAV) based fire detection. The model significantly improves accuracy in challenging conditions, offering a more reliable solution for early wildfire detection.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Remote Sensing
Background:
- Unmanned Aerial Vehicle (UAV) aerial imagery is crucial for fire detection.
- Long-distance imaging faces challenges like limited angles, complex backgrounds, and environmental interference, reducing accuracy.
- Existing methods struggle with reliability in diverse and challenging fire detection scenarios.
Purpose of the Study:
- To propose the BMD-YOLOv8n model to overcome limitations in UAV-based fire detection.
- To enhance the accuracy and reliability of fire detection from aerial imagery.
- To improve the performance of object detection models for early wildfire identification.
Main Methods:
- A new dataset, AeroFlame, was created for UAV fire detection.
- Developed a Bidirectional Feature Pyramid Network with P2 layer (P2-BiFPN) for multi-scale feature fusion and small target retention.
- Integrated a Diverse Branch Block, Cross Stage Partial, and Efficient Layer Aggregation Network (DBBCSPELAN) module for superior feature extraction.
- Incorporated a Simple Attention Module (SimAM) to improve computational efficiency and reduce background noise.
- Replaced the standard CIoU loss with Multi-Perspective Distance-IoU (MPDIoU) loss for precise bounding box regression.
Main Results:
- BMD-YOLOv8n demonstrated superior performance over the baseline YOLOv8n, with improvements of 3.5% in precision, 3.7% in recall, and 3.5% in mAP50.
- Achieved high detection accuracies of 92.4% on the FLAM dataset and 91.8% on the FASDD dataset.
- The model effectively handles challenges posed by long-distance imaging and complex environmental factors.
Conclusions:
- The proposed BMD-YOLOv8n model significantly enhances fire detection capabilities using UAV imagery.
- The integration of novel modules and loss functions leads to improved accuracy and reliability.
- This research offers a robust solution for early wildfire detection, crucial for environmental protection.
