Fire detection and feature extraction apparatus and method based on dual optical wavelength

Optical data is collected from an optical sensor of a dual wavelength, and in order to detect the fire from the collected optical data, an average value of a first wavelength, an average value of a second wavelength, and a ratio of the average values of the two wavelengths are calculated, and an amount of change of a slope of the ratio is used to detect the fire and determine the fire occurrence time. From the determined fire occurrence time, fire features are extracted from the optical data in real time according to defined rules to configure a data set. The data set may be used for learning and inference techniques to identify a fire or non-fire, a fire source, a combustion material, and the like.