Multi-Stage Feature Selection Based Intelligent Classifier for Classification of Incipient Stage Fire in Building

Area of Science:

• Sensor Technology
• Artificial Intelligence
• Fire Safety Engineering

Background:

• Early fire detection is crucial for safety.
• Traditional methods may have limitations in detecting nascent fires.
• Utilizing novel sensing technologies can enhance detection capabilities.

Purpose of the Study:

• To propose an early fire detection algorithm using a low-cost sensor array.
• To analyze the "smellprint" of various fire sources and building materials.
• To develop a reliable classification model for fire detection.

Main Methods:

• Utilizing off-the-shelf gas, dust, temperature, and humidity sensors.
• Collecting and analyzing odour profile data from common fire sources and materials.
• Applying feature extraction, multi-stage feature selection, and Principal Component Analysis (PCA).
• Employing a hybrid PCA-PNN (Probabilistic Neural Network) approach for classification.

Main Results:

• PCA-based dimension reduction enhanced classification accuracy.
• The algorithm demonstrated high reliability across varying ambient conditions (temperature, humidity).
• Detection performance was robust against sensor drift and different gas concentrations/heating temperatures.

Conclusions:

• The proposed PCA-PNN algorithm offers a reliable and accurate method for early fire detection.
• Low-cost sensor arrays combined with advanced data processing can effectively identify fire "smellprints".
• This approach provides a promising solution for enhanced fire safety systems.