红外目标模拟器校准装置构建和校准技术研究

Study of calibration device construction and calibration technology for infrared target simulator

摘要:

在对传统红外目标模拟器校准装置分析的基础上，结合辐射照度校准理论、方法和误差来源三方面，构建了红外目标模拟器校准装置，开展了辐照度测量系统内部杂散辐射建模仿真和抑制的研究、红外大气传输修正方法的研究和测试结果数据拟合算法的研究，分别提出了基于复合蒙特卡洛法的内部杂散辐射数学模型、基于编码器-解码器结构的卷积神经网络红外辐射大气传输校准算法和基于粒子群优化的极限学习机算法，并在实验室条件下进行了辐照度计量校准实验。分别验证了在追迹光线数相等的情况下复合蒙特卡洛法的精度更高，复合蒙特卡洛法标准偏差为6.1940×10-8，平均误差为0.19293，且仿真结果体现了红外辐射计各部分杂散辐射对探测器入瞳面接收到的杂散辐射造成的影响；基于编码器-解码器结构的卷积神经网络算法能够较好的预测大气透过率和大气程辐射，在三个波段下的平均误差为3.0783%，3.8186%，5.3452%，低于传统方法，降低了大气透过率和大气程辐射的影响；通过与GA-ELM模型、ELM模型进行对，验证了与传统数据拟合方法相比，基于PSO-ELM的方法在1μm~3μm、3μm~5μm、8μm~14μm三个波段下的拟合精度都有所提高，决定系数分别为0.9925、0.9913、0.9814，平均相对误差分别为0.1242%、0.7157%、0.7471%有效提高了红外辐射测量准确度。

Abstract:

Based on the analysis of the traditional infrared target simulator calibration device, the infrared target simulator calibration device was constructed by combining three aspects of radiometric calibration theory, method and error sources, and the research on the simulation and suppression of internal spurious radiation modeling of the irradiance measurement system, the research on the infrared atmospheric transmission correction method and the research on the test result data fitting algorithm were carried out, and the compound Monte Carlo method based The mathematical model of internal spurious radiation, the calibration algorithm of infrared radiation atmospheric transmission based on the encoder-decoder structure of convolutional neural network and the algorithm of limit learning machine based on particle swarm optimization are proposed, and the irradiance metrology calibration experiments are carried out under laboratory conditions. It is verified that the accuracy of the compound Monte Carlo method is higher with equal number of tracer lines, and the standard deviation of the compound Monte Carlo method is 6.1940×10-8 with a mean error of 0.19293, and the simulation results reflect the influence of stray radiation from each part of the infrared radiometer on the stray radiation received at the incoming pupil surface of the detector; the convolutional neural network algorithm based on the encoder-decoder structure can The average error in the three bands is 3.0783%, 3.8186%, 5.3452%, which is lower than the traditional method and reduces the influence of atmospheric transmittance and atmospheric range radiation; by comparing with GA-ELM model and ELM model, it is verified that compared with the traditional data fitting method, the PSO-ELM-based method can predict the atmospheric transmittance and atmospheric range radiation in 1μm The accuracy of the PSO-ELM method is improved in the three bands of ~3μm, 3μm~5μm and 8μm~14μm, and the coefficients of determination are 0.9925, 0.9913 and 0.9814, respectively, and the average relative errors are 0.1242%, 0.7157% and 0.7471%, respectively, which effectively improve the accuracy of infrared radiation measurement.