Lightweight forest smoke and fire detection algorithm based on improved YOLOv5.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Forestry Technology

Background:

• Forest fires pose significant threats, necessitating advanced automated monitoring and warning systems.
• Existing object detection algorithms like YOLOv5 face challenges with high computational demands and performance limitations in real-world applications.
• Effective smoke and fire detection is crucial for timely intervention and minimizing environmental damage.

Purpose of the Study:

• To develop a high-performance, lightweight network model for forest smoke and fire detection based on YOLOv5.
• To address the computational load and detection performance limitations of standard YOLOv5.
• To enhance the efficiency and accuracy of forest fire detection systems.

Main Methods:

• Integration of C3Ghost and Ghost modules into the Backbone and Neck networks to reduce parameters and improve feature representation.
• Incorporation of Coordinate Attention (CA) module in the Backbone to focus on critical fire/smoke features and suppress background noise.
• Implementation of a weighted feature fusion mechanism (PAN-weight) in the Neck network for improved feature integration.

Main Results:

• The proposed model achieved a 44.75% reduction in model size and a 47.46% decrease in FLOPs compared to YOLOv5s.
• Precision was increased by 2.53%, and mean average precision (mAP)@0.5 improved by 1.16%.
• Experimental results confirmed the model's superior performance and usefulness for forest fire detection.

Conclusions:

• The developed lightweight YOLOv5-based model effectively enhances forest fire detection accuracy and efficiency.
• The integration of Ghost modules, CA, and PAN-weight mechanisms significantly optimizes the network's performance.
• This research provides a valuable tool for automated forest monitoring and early warning systems, contributing to better fire management.