Modeling And Calculation Of Infrared Radiation Atmospheric Transmittance Based On Neural Network

The infrared radiation which transfers in the atmosphere is a complex problem. Thecalculation of the atmospheric transmittance is an important part in the overall design of theinfrared measurement system. This paper researches the modeling method of the atmospherictransmittance based on neural network through the fixed-point temperature measurement ofthe standard blackbody, in order to improve the accuracy of temperature measurement. Themain research work includes three aspects as followed.1. The main impact factors analysis of the atmospheric transmittance. The absorption ofthe gas molecules in the atmosphere and the scatter of atmospheric particulates (aerosol) arethe main factors which cause the attenuation of the infrared radiation. In different weatherconditions, with the increase of distance between the measuring equipment and the target, theimpact of these factors becomes more complex, the attenuation of infrared radiation isstronger.2. Building the atmospheric transmittance static model based on BP network. Thetemperature of the standard blackbody is measured at different distances. With the increase ofdistance, the influence of atmosphere and environment and other factors are growing. So themeasuring temperature is gradually reduced, increasing the measurement error. The BPnetwork is introduced to adaptively learn the infrared radiation transmission attenuation lawthat underlies the measuring data and estimate the atmospheric transmittance at differentdistances. Then set up the correction model of the target temperature, improving the precisionof the infrared temperature measurement.3. Building the atmospheric transmittance dynamic model based on the neural network.The dynamic relationship between the atmospheric transmittance and the distance is not wellreflected, if the current measured distances and test temperature are only chosen as the inputof the network. In this paper, the distance delay unit is introduced into the input layer of thenetwork, so the current measured distance and the distance variation are selected as the input of the network and the measured temperature and the corresponding temperature variation areselected as the output of the network. Then the atmospheric transmittance is calculated. Themodel built can reflect the dynamic changes of atmospheric transmittance. The RBF networkwhich has better convergence is adopted to learn the expanded samples. Simulation resultsshow that the built dynamic model can further reduced the temperature measurement error,improving the precision of the infrared temperature measurement.