FireLite: Leveraging Transfer Learning for Efficient Fire Detection in Resource-Constrained Environments (2409.20384v1)

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.

• The paper presents a novel low-parameter CNN, FireLite, that uses transfer learning on MobileNet for efficient fire detection.
• It achieves a 99.18% validation accuracy with only 34,978 parameters, outperforming traditional models with higher complexity.
• The model integrates intelligent fire detection using IP cameras, enabling proactive fire prevention in resource-limited settings.

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

FireLite addresses the challenge of implementing efficient fire detection systems in environments with constrained resources, particularly in transportation domains where political instability increases fire risks. The necessity for such technology arises from the limitations of conventional fire detection methods, which depend heavily on manual intervention and basic alarm systems. The integration of advanced intelligent systems, such as those using IP cameras, provides a more robust approach to fire detection and prevention.

Introduction

The paper presents FireLite, a low-parameter CNN developed for fast fire detection in settings where computational resources are limited. Traditional fire detection systems often rely on human involvement which leads to delays in identifying fire threats. The research highlights how embedding intelligent fire detection systems within existing infrastructure offers proactive advantages.

Conventional strategies fall short in environments like transportation, where embedded systems provide limited memory and processing power. FireLite counteracts these constraints by leveraging transfer learning on the MobileNet architecture. This approach capitalizes on existing feature representations while significantly reducing computational overhead. With a total of 34,978 parameters, FireLite demonstrates strong performance metrics, attaining a validation accuracy of 99.18%.

Methodology

Dataset

The FireNet dataset serves as the foundation for training and testing FireLite. The dataset encompasses frames from 16 non-fire videos and 46 fire videos, supplemented by images sourced from online sites and augmented subsets of established datasets (Figure 1). The diversity within this dataset enhances the model's ability to generalize fire detection across varied contexts.

Figure 1: Number of Fire and Non-Fire Images.

Model Architecture

FireLite employs a transfer learning strategy using the MobileNet backbone pre-trained on ImageNet, optimizing only the top layers for the specific task (Figure 2). This fine-tuning process maintains the rich feature representations acquired from ImageNet, adapting them to the nuances of fire detection.

Figure 2: Architecture of the Proposed Model.

The model architecture encompasses a GlobalAveragePooling2D layer, refining the features extracted by MobileNet, followed by dense layers integrated with regularization techniques like batch normalization and dropout. This configuration mitigates overfitting while enhancing model robustness. The architecture culminates in a dense layer with softmax activation for binary classification accuracy.

Results and Discussion

FireLite achieved an accuracy of 99.18% on the FireNet Dataset, proving its efficacy in identifying fire instances. A comparison of performance metrics and confusion matrices demonstrates the model's precision, recall, and F1-score effectiveness (Figures 6). The minimal presence of false negatives and positives reinforces the model's reliability in real-world applications.

Figure 3: Training vs validation accuracy and loss.

When assessed against existing models, FireLite stands out with substantially fewer parameters while maintaining robust accuracy. Competing models such as FireNet, FireNet-v2, and FireNet-Tiny demonstrate higher parameter counts, surpassing FireLite's efficiency and accuracy balance. This head-to-head comparison (Table 1) solidifies FireLite's place as a promising solution for resource-notch environments.

Model	Accuracy	Parameters
FireLite	99.18%	34,978
FireNet	93.91%	646,818
FireNet v2	94.95%	318,460
FireNet Micro	96.78%	171,234
FireNet-Tiny	95.75%	261,922

The training and validation accuracy demonstrate FireLite's consistent learning trajectory (Figure 3), reaffirming its resistance to overfitting across numerous epochs. Iterative testing confirms the model's responsiveness and adaptability to complex input data.

Conclusion

FireLite represents a significant advancement in lightweight fire detection models for environments where computational resources are limited. By successfully leveraging transfer learning, FireLite offers a high-performance, efficient solution to proactive fire risk mitigation in transportation sectors. The model's scalability and adaptability can be further enhanced with broader dataset applications and continuous architectural refinement efforts.

Future research aims to expand FireLite's capabilities by enriching the dataset and reinforcing its robustness against diverse operational conditions. These improvements will solidify FireLite's role in addressing immediate fire safety challenges, paving the way for effective prevention strategies in environments particularly susceptible to fire hazards.