Study of algorithms of temperature measurement based on digital image processing

A technology for measuring temperature based on digital image processing is presented. In order to diminish the error of neural-network calculation during testing the temperature inside a coal-fired furnace, two methods are developed in the paper. One is least squares method and the other is improved neural-network method whose inputs is different from the old neural network method. According to colorimetric temperature-measurement algorithm, the ratios among the three basic colors of CCD camera are used as the inputs of both methods, which could be helpful in reducing the influence of emissivity, soot and flame frequency. The results of experiments on a coal-fired furnace show that both algorithms are more precise than the old neural network method. The improved neural-network method is more accurate than the least squares method, but the latter is more simple and practical.