A multi-wavelength multi-task learning framework for risk-aware fire source classification and smoke density...

Area of Science:

• Optical Sensing
• Artificial Intelligence
• Fire Safety Engineering

Background:

• Traditional smoke detectors have limitations, including high false alarm rates and inability to differentiate fire types or quantify smoke. Existing single-task models for fire classification or smoke estimation increase computational load.
• Diverse indoor environments require advanced fire detection systems capable of distinguishing between genuine fires and nuisance sources.

Purpose of the Study:

• To develop a novel multi-task learning (MTL) framework for simultaneous fire source classification and smoke density prediction.
• To enhance the efficiency and reduce the computational overhead of fire detection systems.
• To improve the reliability of early fire detection and minimize false alarms in indoor settings.

Main Methods:

• Utilized multi-wavelength optical sensing at 460, 530, 660, and 940 nm.
• Implemented a multi-task learning framework, with CNN-LSTM architecture showing optimal performance.
• Validated the system under UL 268 standard fire scenarios, including smoldering fires, flaming fires, and cooking nuisance sources.

Main Results:

• The CNN-LSTM based MTL model achieved 97% classification accuracy and 0.668 Mean Squared Error (MSE) for smoke density prediction.
• The MTL framework demonstrated a 45% reduction in memory usage and 82% faster inference times compared to single-task approaches.
• Ablation studies confirmed the discriminative capability of the 940 nm near-infrared wavelength for differentiating fire smoke from nuisance aerosols.

Conclusions:

• The proposed MTL framework offers a computationally efficient and accurate solution for real-time, edge-based fire detection.
• This approach significantly reduces false alarms and enhances the reliability of early fire detection in various indoor environments.
• The integration of multi-wavelength optical sensing and MTL provides a robust method for advanced fire safety systems.