The Fire-ViT Model for Tunnel Fire Detection with Vision Transformer Improvement

Abstract

Smoke detection in tunnels presents unique challenges, including low light conditions, visual obstructions like smoke and headlights, and the need to analyze ultrahigh-resolution images. To address these challenges, this study introduces an innovative model named Fire-ViT, leveraging the vision transformer (ViT) architecture. Unlike traditional convolutional neural networks that often struggle with false positives under these complex conditions, Fire-ViT incorporates an attention mechanism and multiperceptron layer, significantly enhancing its capability to discern details in high-resolution images specific to tunnel environments. The model’s performance is outstanding, achieving an accuracy rate of 99.87% on a high-resolution tunnel image dataset, markedly surpassing that of conventional models. Notably, Fire-ViT not only elevates detection accuracy and robustness but also cuts training time in half. This efficiency, coupled with its adaptability to the intricate tunnel environment, makes Fire-ViT an ideal solution for early warning systems against fires in tunnels, fulfilling the demand for high-standard, fine-grained fire detection.

Received: 15 June 2024 | Revised: 13 August 2024 | Accepted: 14 September 2024

Conflicts of Interest
The authors declare that they have no conflicts of interest in this work.

Data Availability Statement
The data that support the findings of this study are openly available in Kaggle at https://www.kaggle.com/datasets/xiaobingliu323/ tunnelfire2024/data.

Author Contribution Statement
Xiaobing Liu：Conceptualization, Methodology, Software, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration, Funding acquisition. Vladimir Y. Mariano: Methodology, Validation, Resources, Writing – review & editing, Supervision, Project administration.