Research and application of deep learning object detection methods for forest fire smoke recognition

Area of Science:

• Computer Science
• Environmental Science
• Artificial Intelligence

Background:

• Forest fires pose significant ecological and economic threats, exacerbated by climate change.
• Effective monitoring and early warning systems are crucial for mitigating wildfire damage.
• Deep learning offers promising solutions for automated fire detection.

Purpose of the Study:

• To evaluate the effectiveness of the YOLOv11x algorithm for deep learning-based forest fire smoke recognition.
• To develop an efficient fire detection model to enhance early detection capabilities.
• To improve the generalizability of fire detection models using diverse datasets.

Main Methods:

• Utilized the YOLOv11x algorithm for object detection in forest fire scenarios.
• Trained the model on two public datasets: WD (Wildfire Dataset) and FFS (Forest Fire Smoke).
• Evaluated model performance using metrics such as precision, recall, and mean average precision (mAP50, mAP50-95).

Main Results:

• YOLOv11x demonstrated strong performance with precision of 0.949, recall of 0.850, mAP50 of 0.901, and mAP50-95 of 0.786.
• The model showed superior performance in smoke detection (mAP@0.5 = 0.962) compared to flame detection (mAP@0.5 = 0.841).
• 86.89% of test samples achieved confidence scores above 0.85, indicating high reliability.

Conclusions:

• The YOLOv11x algorithm is highly effective for forest fire smoke recognition.
• The model provides robust technical support for early fire warning systems.
• Findings offer valuable insights for designing intelligent monitoring systems for wildfire prevention.