Fast and Accurate Residential Fire Detection Using Wireless Sensor Networks

Abstract

Prompt and accurate residential fire detection is important for on-time fire extinguishing and consequently reducing damages and life losses. To detect fire sensors are needed to measure the environmental parameters and algorithms are required to decide about occurrence of fire. Recently, wireless sensor networks (WSNs) have been used for environmental monitoring and real-time event detection because of their low implementation costs and their capability of distributed sensing and processing. Although there are several works on fire detection using WSNs, they have rarely paid sufficient attention to investigate the optimal sensor sets and usage of suitable artificial intelligence (AI) methods. Therefore, by aiming at residential fire detection, this paper investigates proper sensor sets and proposes AI-based techniques for fire detection in WSNs. The proposed methods are evaluated in terms of detection accuracy rate and computational complexity.

Fire Data Analysis and Feature Reduction Using Computational Intelligence Methods

“…However, investigation of gas sensors such as carbon monoxide (CO) demonstrated that gas sensors contribute the most to fire detection techniques [1,13,14]. Recent studies show that temperature, ionization, photoelectric, and CO sensors together make an optimal sensor set for residential fire detection [1,2,3]. Therefore, we use these four sensors to analyze residential fire data and further to detect residential fire.…”

“…By looking back to the basic notions on residential fire detection, it is seen that earlier works were conducted by only one sensor type [1]. Temperature, flame detector and air-obscuration sensors are mostly used as an individual sensor for residential fire detection [1,2]. However, investigation of gas sensors such as carbon monoxide (CO) demonstrated that gas sensors contribute the most to fire detection techniques [1,13,14].…”

“…The exact contribution of each sensor to the fire detection varies based on the classifier or detection algorithm. Nevertheless, it has been reported that CO is the most contributing feature [1,2,3]. In Table 2, we report detection accuracies of the 16 possible combinations of sensors using decision trees, neural networks, and naive Bayes classifiers.…”

“…As a result, having temperature, ionization, photo and CO sensors are proposed as the minimum set of sensors for fire detection [1,2,3]. Although the fire detection can be conducted with only one sensor, fire detection using all four sensors is more robust to environmental noise and provides extra support for sensor failures.…”

“…A careful analysis of existing research in the field of fire detection reveals that while much attention has been paid to developing complex detection algorithms, not equal attention has been paid to analysis of fire data to identify the most effective features and the optimal set of features (sensors). In this paper, by applying data analysis and feature extraction techniques originated from computational intelligent field on two fire data sets, i.e., wildfire and residential fire, we identify the most contributing features (sensors) to the process of fire detection [1,2]. We then apply our distributed naive Bayes and neural network outlier detection techniques designed for WSNs on the extracted features and present the performance evaluation.…”