Multispectral Infrared Radiation Characteristics Analysis And Temperature Measurement Method Research

Infrared radiation temperature measurement is typical non-contact technology,which has the advantages such as fast response speed and wide temperature measurement range.The purpose of this paper is to study the adaptive multi-spectral temperature measurement method.By measuring the multispectral radiance of the target at a certain temperature,the nonlinear relationship between the change rule of radiance and temperature is fully learned by advanced neural networks,and establish a temperature measurement model to accurately estimate the true temperature of the target.Main contents include:Measurement and characteristic analysis of target spectral radiance.Fourier infrared spectrometer was used to measure the radiance of the radiant target at different temperatures.Combined with infrared theory,the characteristic related to temperature were extracted including peak,area,shape,etc.Infrared multispectral radiation temperature measurement based on RBF network.By introducing the spectral emissivity model,the nonlinear correspondence between temperature and characteristic is established,and the input characteristic vector of the temperature measurement model is obtained.The infrared temperature measurement model was established by RBF network adaptive learning,and test samples were used to verify the rationality of the method.The results show that the model can obtain high-precision temperature estimate,but it may cause a large error when the input characteristic is in serious interference band.Infrared multispectral radiation temperature measurement based on PCA-ELM.In order to reduce the dependence of temperature measurement model on radiation characteristic,the multi-spectral radiance input vector was established.In order to contain enough effective temperature measurement information,the input vector usually is high dimensional and may contain redundant information to affect the accuracy.In this paper,principal component analysis(PCA)layer is established to extract the principal components of the multi-spectral radiance to improve the independence of each input variable and achieve input dimension reduction.Combined with the extreme learning machine(ELM)network for fast training,the temperature measurement model based on PCA-ELM was finally established to realize the infrared multi-spectral temperature measurement for the target with unknown emissivity.Two targets,the standard blackbody and the steel plate coating with unknown emissivity,were used to verify the effectiveness of this method.