Deep Learning Based Real Time Fire Detection and Alert System

Abstract

Fire accidents are a major threat to human lives, property, and the environment. Traditional fire detection systems that rely on smoke and heat sensors often respond slowly and may not work effectively in large or open areas. They can also generate false alarms, which reduces their reliability.To overcome these challenges, this paper presents an intelligent fire detection system based on a hybrid deep learning model that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The CNN is used to analyse individual image frames and extract important visual features such as flame colour, texture, and smoke patterns. Meanwhile, the LSTM captures the changes that occur over time by analysing consecutive frames, helping the system understand how fire develops and spreads.By combining both spatial and temporal analysis, this hybrid model improves detection accuracy and significantly reduces false alarms caused by fire-like elements such as sunlight reflections or artificial lighting. The system is trained and tested on a dataset containing both fire and non-fire images and videos, including indoor, outdoor, and forest fire situations. The performance of the model is evaluated using standard metrics such as accuracy, precision, recall, and F1-score. The experimental results show that the proposed CNN-LSTM model performs better than traditional CNN-based methods, offering improved reliability and real-time fire detection capability.

Introduction

The text presents a fire detection system using a hybrid deep learning approach that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The motivation arises from the limitations of traditional sensor-based fire detection systems, which suffer from delayed response, limited coverage, and high false alarm rates, especially in complex environments.

The proposed system uses CNN to extract spatial features such as flame color, texture, and smoke patterns from video frames, while LSTM analyzes temporal sequences to capture motion and progression of fire over time. This combination helps distinguish real fire from fire-like objects such as lights, reflections, or headlights, thereby reducing false alarms.

The methodology includes data collection from diverse fire and non-fire scenarios, preprocessing (frame extraction, resizing, normalization, augmentation), and training a hybrid CNN–LSTM model using binary classification. The system is evaluated using metrics like accuracy, precision, recall, and F1-score.

Results show that the hybrid model outperforms standalone CNN approaches, achieving high detection accuracy with reduced false positives and false negatives. The confusion matrix confirms strong performance in correctly identifying both fire and non-fire cases.

Conclusion

This paper presented an intelligent Fire Detection System built on a hybrid deep learning architecture that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The proposed method overcomes the limitations of traditional sensor-based fire detection systems as well as standalone image-based models by integrating both spatial and temporal feature learning into a unified framework. In this architecture, the CNN effectively extracts spatial features from individual frames, including flame colour distribution, texture patterns, and smoke characteristics. Meanwhile, the LSTM network captures temporal relationships and motion patterns across consecutive frames. The integration of these two components significantly enhances detection accuracy and minimizes false positives caused by fire-like elements such as sunlight reflections, artificial lighting, and other bright surfaces.

References

This paper presented an intelligent Fire Detection System built on a hybrid deep learning architecture that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks. The proposed method overcomes the limitations of traditional sensor-based fire detection systems as well as standalone image-based models by integrating both spatial and temporal feature learning into a unified framework. In this architecture, the CNN effectively extracts spatial features from individual frames, including flame colour distribution, texture patterns, and smoke characteristics. Meanwhile, the LSTM network captures temporal relationships and motion patterns across consecutive frames. The integration of these two components significantly enhances detection accuracy and minimizes false positives caused by fire-like elements such as sunlight reflections, artificial lighting, and other bright surfaces.

Copyright

Copyright © 2026 Sk. Haseena Begum, O. Sandeep Kumar, M. Suresh Babu, M. Naga Durga Prasad, G. Sundar Rohith Kiran. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.