Research On Compensation Method Of Atmospheric Transmittance Based On Multispectral Thermometry

When using the multispectral radiation temperature measurement method to measure the temperature of a remote high-altitude dynamic target,the radiant energy of the target needs to be transmitted through the atmosphere before reaching the detector.Due to the existence of atmospheric molecular absorption and scattering,the radiant energy will be attenuated to a certain extent,resulting in the energy signal received by the measurement system is damaged,which greatly reduces the accuracy of temperature measurement.Therefore,the research on the compensation of atmospheric transmittance in remote radiation temperature measurement is of great significance.The main research contents of this paper are as follows:The establishment and verification of atmospheric transmittance compensation model based on one-dimensional convolutional neural network.Based on the analysis of the existing thermal radiation atmospheric transmission correction method,using the nonlinear mapping relationship between the detected voltage signal and the target true temperature,a one-dimensional convolutional neural network-based atmospheric transmittance compensation model is proposed.The evaluation indicators are the mean absolute error and the root mean square error.The results of the algorithm are compared with the BP model and the support vector regression model to verify the superiority of the algorithm.The influence of the number of temperature measurement channels on the prediction accuracy is discussed,and simulation data and calibration data are used to predict the temperature of a variety of typical samples.The results show that the temperature prediction error of the atmospheric transmittance compensation model based on the one-dimensional convolutional neural network is within 10%.Establishment and verification of atmospheric transmittance compensation model based on deep belief network.The basic principles and training process of deep confidence networks are studied,the mean absolute error and root mean square error are used as model evaluation indicators,and the method of empirical selection and experimental determination is used to obtain the number of hidden layers and hidden layers of the built deep confidence network model.Hyperparameters such as the number of layer nodes and learning rate are used to establish an optimal prediction model.The influence of the number of temperature measurement channels on the prediction error is discussed,simulation data and calibration data are used to predict the temperature of a variety of typical samples,and the test results are analyzed.From the experimental results,it can be seen that the maximum error of the temperature inversion of the deep belief network model is less than 10%.In summary,the effectiveness of the atmospheric transmittance compensation method proposed in this paper has been verified,and it provides a certain reference and method reference for solving the problem of radiant energy attenuation caused by atmospheric transmittance in the process of remote temperature measurement.