The Measurement Of Spectral Emissivity Based On Neural Network And DSP Technology

Spectral emissivity is one of the important parameters in object radiation. We can be acquainted ourselves with the radiation property of kinds of materials by the measurement of spectral emissivity. It is of significance to the development of the advancement of aviation, national defence.Multi-spectral radiation measurement is a better way developed in recent years on measuring spectral emissivity and the true temperature. This method can isolate spectral emissivity and the true temperature by radiation information of target, however, the flaws is the need of assuming a fixed mathematical model between emissivity and wavelength, when model is inconsistent with the tested materials, this method will cause a big error.In this paper, the author established a multi-spectral neural network model, by which got the spectral emissivity of object without assuming a fixed mathematical model. In view of the advantages of BP neural network on disposing nonlinear relationship, builded a function of emissivity and wavelength based on BP neural network, however, the sampling data of spectral emissivity can not be fired. Combining multi-spectral radiation measurement theory, transformed the function model of emissivity and wavelength, established a multi-spectral neural network model. The brightness temperature and corresponding wavelength were the sampling data of multi-spectral neural network. The brightness temperature can be got by various spectral radiation energy of target.To prove feasibility of multi-spectral neural network model, simulated four assumption emissivity with different changing rates. The results of simulation were in good agreement with the four assumption emissivity models, confirmed the feasibility of the model, then measuring the oxide steel and tungsten by the multi-spectral neural network, the results is correct. To enhance the practicality and the real-time of this model, the measurement of emissivity by use of multi-spectral neural network was realized with DSP operating system. The results were similar with MATLAB simulation ones. This model algorithm can be used as the basis for engineering practice.