A general computing formula has been formulated for the measurement of surface temperature and the corresponding
relation between the thermal value and the true temperature of infrared images according to the principles of thermal
radiation and temperature measurement with infrared thermal imager. A least squares method and an improved neuralnetwork
method have been developed to calculate the temperature to diminish the deviation of neural-network method.
The above two methods use the ratios among the three basic colors output from the thermal infrared imager as the
independent variable or input variable, and can personalize the colorimetric temperature-measurement algorithm, so that
the influence of emissivity, soot and combustion flame on the temperature result can be reduced. Simulation results show
that the precision of these two methods are higher than that of the traditional neural network method. In addition, the
precision of the proposed neural-network method is higher than that of the least squares method.