Influence of Feature Extraction Duration and Step Size on ANN based Multisensor Fire Detection Performance

Abstract ANN has displayed great advantage in multisensor based fire detection. One of the major application steps in application of ANN in multisensor fire detection is feature extraction of time series. The objective of this research is to investigate feature extraction window duration and step size on fire detection performance. Some experimental results are adopted as benchmark tests, and detected fire time and failed alarm rate are the important indicators of performances. Three ANN types, namely BP, RBF and PNN are analyzed. Results indicate that both observation window duration and step size can determine ANN fire detection performance to a large extent, and a duration period of 90s with time step varies from 25s to 200s is recommended. Meanwhile, PNN might be the favorable ANN types related to the two performance parameters.