Related Experiment Video
Updated: Dec 11, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
876
An Efficient Fire Detection Method Based on Multiscale Feature Extraction, Implicit Deep Supervision and Channel
Summary
This study introduces an improved vision-based fire detection system using convolutional neural networks. The new method achieves higher accuracy and a better balance of model size and speed for real-time fire detection.
Area of Science:
- Computer Vision
- Artificial Intelligence
- Fire Safety Engineering
Background:
- Vision-based fire detection systems are crucial for early warning.
- Current convolutional neural network (CNN) methods struggle to balance accuracy, model size, and detection speed.
- A need exists for more efficient and accurate fire detection algorithms.
Purpose of the Study:
- To develop an accurate and efficient vision-based fire detection method.
- To improve the tradeoff between accuracy, model size, and speed in fire detection.
- To enhance the discriminative ability of fire-like objects in complex scenes.
Main Methods:
- Utilized a multiscale feature extraction mechanism for richer spatial details.
- Employed an implicit deep supervision mechanism with dense skip connections to improve information flow.
- Integrated a channel attention mechanism to enhance feature map contributions.
Main Results:
- Achieved 95.3% accuracy, outperforming the suboptimal method by 2.5%.
- Demonstrated a 3.76% increase in speed on GPU compared to the suboptimal method.
- Reduced model size by 63.64% compared to the suboptimal method.
Conclusions:
- The proposed method offers a superior balance of accuracy, speed, and model size for vision-based fire detection.
- Enhanced feature extraction and attention mechanisms contribute to improved fire detection performance.
- This approach represents a significant advancement for real-time fire detection applications.