Related Experiment Video
Updated: Jul 10, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.8K
Early Fire Detection Using Long Short-Term Memory-Based Instance Segmentation and Internet of Things for Disaster
1Department of Information Technology, Faculty of Computing and Information Technology, King Abdulaziz University, P.O. Box 344, Rabigh 21911, Saudi Arabia.
Sensors (Basel, Switzerland)
|November 25, 2023
Summary
A new hybrid deep learning model, IS-CNN-LSTM, effectively detects fires with high accuracy and low false alarms. This advanced fire detection system utilizes instance segmentation and Internet of Things (IoT) devices for timely alerts.
Area of Science:
- Computer Science
- Artificial Intelligence
- Environmental Science
Background:
- Fire outbreaks pose significant global risks, exacerbated by population growth and climate change.
- Existing fire detection methods struggle with accuracy, complexity, and detecting small or distant fires.
Purpose of the Study:
- To develop a novel hybrid model (IS-CNN-LSTM) for accurate fire detection and intensity analysis.
- To address challenges in detecting small fires and fires from long distances.
- To create an optimized fire detection system with low false alarm rates.
Main Methods:
- Implementation of a 57-layer Convolutional Neural Network (CNN) model integrated with Long Short-Term Memory (LSTM) networks.
- Utilizing instance segmentation for distinguishing fire from non-fire events.
- Employing a key-frame extraction algorithm to reduce model complexity and integrating Internet of Things (IoT) devices for real-time alerts.
Main Results:
- Achieved 95.25% classification accuracy on a public dataset of fire and normal videos.
- Demonstrated a low false positive rate (FPR) of 0.09% and a false negative rate (FNR) of 0.65%.
- The model achieved a rapid prediction time of 0.08 seconds.
Conclusions:
- The proposed IS-CNN-LSTM model offers a highly accurate and efficient solution for fire detection and intensity analysis.
- The integration of CNN, LSTM, instance segmentation, and IoT devices provides a robust system for real-time fire monitoring and alerting.
- This advanced approach effectively addresses limitations of previous fire detection techniques, offering significant potential for mitigating fire-related damages.
