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

Area of Science:

• Engineering
• Computer Science
• Safety Science

Background:

• Indoor fires cause significant global casualties and economic losses.
• Accurate and early indoor fire detection is vital for mitigating these threats.
• Resource-constrained embedded platforms require efficient detection methods.

Purpose of the Study:

• To propose an accurate and efficient indoor fire detection method for resource-constrained embedded platforms.
• To leverage multi-sensor fusion and a lightweight convolutional neural network (CNN).
• To enhance the robustness and performance of embedded fire detection systems.

Main Methods:

• Applied Savitzky-Golay (SG) filter for heterogeneous sensor data cleaning.
• Utilized Gramian Angular Field (GAF) to transform time-series data into matrices.
• Integrated sensor data into a 3D matrix, preserving temporal dependencies.
• Developed a lightweight CNN by reducing network blocks, channels, and layers.
• Employed Fire Dynamic Simulator (FDS) for data simulation to enhance network robustness.

Main Results:

• Achieved an impressive accuracy of 99.1% in experimental validation.
• The proposed lightweight CNN demonstrated a small number of parameters and low computational requirements.
• The method is highly suitable for resource-constrained embedded platforms.

Conclusions:

• The proposed multi-sensor fusion and lightweight CNN method offers a highly accurate and efficient solution for early indoor fire detection.
• This approach effectively addresses the limitations of resource-constrained embedded systems.
• The method enhances safety by enabling timely and reliable fire detection.