A Method for On-Board Fire Detection Based on the Integration of Expert Systems and Neural Networks

Abstract

Fire accidents (especially large-scale fires) pose significant threats to human society, such as forest fires and chemical plant explosions, which can cause substantial loss of life, health, and economic damage. However, current fire detection using remote sensing satellites is mostly for post-disaster confirmation rather than pre-disaster warning, lacking a high-timeliness, high-accuracy onboard fire detection and warning scheme. On the other hand, the significant improvement in satellite payload technology and the increasing richness of satellite remote sensing data products have made the processing of remote sensing data products increasingly difficult. In fire detection, the satellite detection scheme determines the satellite's application capability and further determines whether the satellite can maximize its effectiveness. Based on the intelligent detection application requirements for onboard fire targets, this paper focuses on solving the problems of the existing fire detection models, such as the difficulty in eliminating high-reflective objects and other false fire targets, the large data volume when using multi-band combinations that cannot ensure onboard processing timeliness, and the poor environmental adaptability of existing fire detection schemes. A high-timeliness, high-confidence, and highly adaptable high-orbit satellite multi-spectral onboard fire intelligent detection scheme is proposed. By integrating expert system feature maps for fire confirmation, the scheme meets the high-frequency inspection and rapid warning needs for fires, supporting the integrated application of satellite and ground systems, and will significantly enhance the early warning detection efficiency of satellite fire detection.