Abstract:Fire hazards are extremely dangerous, particularly in sectors such as the transportation industry, where political unrest increases the likelihood of their occurrence. By employing IP cameras to facilitate the setup of fire detection systems on transport vehicles, losses from fire events may be prevented proactively. However, the development of lightweight fire detection models is required due to the computational constraints of the embedded systems within these cameras. We introduce FireLite, a low-parameter convolutional neural network (CNN) designed for quick fire detection in contexts with limited resources, in response to this difficulty. With an accuracy of 98.77\%, our model -- which has just 34,978 trainable parameters achieves remarkable performance numbers. It also shows a validation loss of 8.74 and peaks at 98.77 for precision, recall, and F1-score measures. Because of its precision and efficiency, FireLite is a promising solution for fire detection in resource-constrained environments.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to achieve efficient fire detection in resource - constrained environments (such as embedded systems and IP cameras). Specifically, fire is especially dangerous in the transportation industry, especially in politically unstable regions where the risk of fire is higher. Traditional fire detection methods rely on human intervention or simple alarm systems and usually cannot provide timely responses, resulting in serious damage or even loss of life. To meet this challenge, the author proposes a lightweight convolutional neural network (CNN) named "FireLite", which aims to detect fire quickly and accurately while adapting to the computational limitations of resource - constrained devices. The following are the specific problems and solutions mentioned in the paper: ### 1. **Problem Background** - **Fire Risk**: Especially in the transportation industry, fire can lead to serious accidents, affecting everything from passenger trains to cargo ships. - **Limitations of Traditional Methods**: Traditional fire detection methods rely on human intervention or simple alarm systems and cannot provide timely responses. - **Resource - Constrained Environments**: Devices such as embedded systems and IP cameras usually have limited memory and processing power, so lightweight models need to be developed to adapt to these limitations. ### 2. **Proposed Method** - **FireLite Model**: The author designed a lightweight CNN model with only 34,978 parameters, which can operate efficiently in resource - constrained environments. - **Transfer Learning**: By using the pre - trained MobileNet architecture and combining transfer learning techniques, FireLite can utilize high - level features in the ImageNet dataset, thereby improving detection performance. - **Model Optimization**: By fine - tuning some layers of MobileNet and introducing regularization techniques (such as batch normalization and dropout), FireLite reduces the risk of overfitting while maintaining high accuracy. ### 3. **Experimental Results** - **Performance Metrics**: FireLite has an accuracy rate of 99.18% on the FireNet dataset, a validation loss of 8.74, and performs well in terms of precision, recall, and F1 - score. - **Comparison with Existing Models**: Compared with other existing lightweight fire detection models (such as FireNet, FireNet - v2, FireNet - Tiny, etc.), FireLite can still maintain high performance with a significantly reduced number of parameters. ### 4. **Conclusion** - FireLite achieves fast and accurate fire detection in resource - constrained environments through its efficient architecture and transfer learning techniques. - The lightweight design of this model makes it suitable for devices such as embedded systems and IP cameras, providing a new solution for fire prevention in the transportation industry. In summary, this paper aims to solve the deficiencies of existing methods in resource - constrained environments by developing a lightweight and efficient fire detection model, thereby improving the timeliness and accuracy of fire detection.

FireLite: Leveraging Transfer Learning for Efficient Fire Detection in Resource-Constrained Environments

Deep Learning Based Fire Detection System For Surveillance Videos

An efficient deep learning architecture for effective fire detection in smart surveillance

KutralNet: A Portable Deep Learning Model for Fire Recognition

Research on Fire Detection in Laboratories Based on CNN and Transfer Learning

FireNet: A Specialized Lightweight Fire & Smoke Detection Model for Real-Time IoT Applications

Deep Learning-Based Fire Detection for Enhanced Safety Systems

Early Fire Detection Using Long Short-Term Memory-Based Instance Segmentation and Internet of Things for Disaster Management

FireNet: A Lightweight and Efficient Multi-Scenario Fire Object Detector

Developing an Integrated Computer Vision Image Processing Technique to raise an Early Alarm for Fire Detection

Accelerated Fire Detection and Localization at Edge

Detection of forest fire using deep convolutional neural networks with transfer learning approach

Image-Based Fire Detection Using Dynamic Threshold Grayscale Segmentation and Residual Network Transfer Learning

Hybrid CNN-ViT architecture to exploit spatio-temporal feature for fire recognition trained through transfer learning

A visualized fire detection method based on convolutional neural network beyond anchor

Implementation of Lightweight Convolutional Neural Networks with an Early Exit Mechanism Utilizing 40 nm CMOS Process for Fire Detection in Unmanned Aerial Vehicles

Real-Time Early Indoor Fire Detection and Localization on Embedded Platforms with Fully Convolutional One-Stage Object Detection

A modified vision transformer architecture with scratch learning capabilities for effective fire detection

An Indoor Fire Detection Method Based on Multi-Sensor Fusion and a Lightweight Convolutional Neural Network

A lightweight fire detection algorithm for small targets based on YOLOv5s

Fire Detection From Image and Video Using YOLOv5