RESUMO
A multi-wavelength analysis method is introduced to measure the temperature of basic oxygen furnace flame. In this study, USB4000 spectrometer was applied to obtain radiation spectrum of flame within wavelength range 200-1 100 nm, from which the flame temperature and monochromatic emissivity was derived by Levenberg-Marquart modeling method. Wavelet neural network was applied to process the spectral measurement data, which could cancel the assumption model of emissivity and wavelengths. It is a kind of valid method to acquire the true temperature and spectral emissivity. Each neuron in the hidden layer of a feed-forward network is a combination of the sigmoidal activation function (SAF) and morlet wavelet activation function (WAF). The output of the hidden neuron is the product of the output from these two activation functions.