红外辐射亮度的RBF网络建模及其光谱发射率估计  

Modeling infrared radiance and calculating spectral emissivity based on RBF network

作　　者：席剑辉[1] 徐振方[1] 傅莉[1] 王琦[1] 

机构地区：[1]沈阳航空航天大学自动化学院,辽宁沈阳110136

出　　处：《红外与激光工程》2016年第B05期17-22,共6页Infrared and Laser Engineering

基　　金：辽宁省自然科学基金联合封闭基金项目(2015020069);中航创新基金项目(cxy2012SH18);沈阳市科技创新团队项目(src201204)

摘　　要：建立一种基于RBF神经网络的目标红外辐射亮度建模方法,进而实现对目标光谱发射率的估计。通过FTIR光谱仪测量目标表面3~14μm波段的红外辐射特性,亮度光谱会受到二氧化碳、水蒸气等的吸收及大气辐射的干扰。文中首先结合红外传输理论选择有效学习样本;然后基于RBF网络对样本进行充分学习,建立目标红外辐射亮度模型;利用所建模型估计大气吸收和杂散干扰波段的亮度,最终计算出较完整的目标光谱发射率。黑体测试结果与理论发射率比较,最大相对误差为1.5%。测温验证的结果也表明文中所建的RBF神经网络可以有效地对目标光谱发射率进行估计。A method of modeling infrared radiance based on RBF neural network was built, then the target spectral emissivity was estimated. When measuring the infrared radiation characteristics of the target surface in the 3-14 μm band by FTIR spectrometer, the infrared radiance will be absorbed by carbon dioxide, water vapor, etc, and affected by some stray radiation. In this paper, the effective learning samples were firstly selected combined with the theory of infrared transmission. Then the samples based on the RBF network were fully learned, and a target infrared radiance model was built. And this model was used to estimate the radiance in the band of atmospheric absorption and stray radiation. A more complete target spectral emissivity curve was finally calculated. Compared the calculating results of blackbody with theoretical emissivity, the maximum relative error is 1.5%. The verification of temperature measurement also shows that the RBF neural networks can be built efficiently to estimate target spectral emissivity.

关 键 词：光谱发射率 辐射亮度 RBF网络 

分 类 号：TN219[电子电信—物理电子学]