Lightweight Fire Detection in Tunnel Environments DOI Creative Commons
Shakhnoza Muksimova, Sabina Umirzakova,

Dilnoza Abduxalikovna Babaraximova

et al.

Fire, Journal Year: 2025, Volume and Issue: 8(4), P. 134 - 134

Published: March 30, 2025

Tunnel fires pose significant challenges to public safety due their rapid development and the confined nature of tunnel environments. Traditional fire detection systems often struggle with delayed response times high false alarm rates, particularly in complex scenarios. This study proposes a lightweight hybrid deep learning (DL) model that integrates Convolutional Neural Networks (CNNs) for spatial feature extraction Long Short-Term Memory (LSTM) networks temporal analysis, offering an efficient robust solution real-time detection. Leveraging transfer learning, adapts tunnel-specific scenarios minimal training data, significantly improving its generalization capabilities. The architecture ensures computational efficiency, making it suitable deployment resource-constrained environments such as tunnels limited processing capacity. was rigorously evaluated on datasets combining simulated real-world It achieved accuracy 92%, precision 89%, recall 90%, F1 score 89.5%, outperforming state-of-the-art (SOTA) models all key metrics. Furthermore, demonstrated resilience under varied environmental conditions, including smoke density sensor failures, maintaining reliable performance. highlights potential enhancing by providing accurate, fast, dependable Future work will extend methodology other critical infrastructures optimize broader applications.

Language: Английский

Lightweight Fire Detection in Tunnel Environments DOI Creative Commons
Shakhnoza Muksimova, Sabina Umirzakova,

Dilnoza Abduxalikovna Babaraximova

et al.

Fire, Journal Year: 2025, Volume and Issue: 8(4), P. 134 - 134

Published: March 30, 2025

Tunnel fires pose significant challenges to public safety due their rapid development and the confined nature of tunnel environments. Traditional fire detection systems often struggle with delayed response times high false alarm rates, particularly in complex scenarios. This study proposes a lightweight hybrid deep learning (DL) model that integrates Convolutional Neural Networks (CNNs) for spatial feature extraction Long Short-Term Memory (LSTM) networks temporal analysis, offering an efficient robust solution real-time detection. Leveraging transfer learning, adapts tunnel-specific scenarios minimal training data, significantly improving its generalization capabilities. The architecture ensures computational efficiency, making it suitable deployment resource-constrained environments such as tunnels limited processing capacity. was rigorously evaluated on datasets combining simulated real-world It achieved accuracy 92%, precision 89%, recall 90%, F1 score 89.5%, outperforming state-of-the-art (SOTA) models all key metrics. Furthermore, demonstrated resilience under varied environmental conditions, including smoke density sensor failures, maintaining reliable performance. highlights potential enhancing by providing accurate, fast, dependable Future work will extend methodology other critical infrastructures optimize broader applications.

Language: Английский

Citations

0