A low-power embedded system for fire monitoring and detection using a multilayer perceptron

Fire monitoring and detection systems can evaluate data from environmental or image sensors in order to predict occurrences of fire. It is a complex procedure that requires a significant amount of energy as input data is usually acquired from multiple sensors and the algorithms generally have an increased complexity. This paper introduces a low-power fire monitoring and detection system that utilizes data from two environmental sensors. As a predictive algorithm for fire occurrences, it uses a multilayer perceptron (MLP) with a combination of different optimizations, developing a model with low memory requirements and high -accuracy predictions. The accuracy of the proposed system was verified using a dataset created by the environmental sensors for fire incidents and its performance was compared to existing approaches. An evaluation of the proposed system's power consumption and memory requirements is also presented.