Volume 52 Issue 3
Mar.  2023
Turn off MathJax
Article Contents

Zhang Xinyi, Chen Zhenlin. Research on atmospheric transmission calibration method of infrared target simulator[J]. Infrared and Laser Engineering, 2023, 52(3): 20220378. doi: 10.3788/IRLA20220378
Citation: Zhang Xinyi, Chen Zhenlin. Research on atmospheric transmission calibration method of infrared target simulator[J]. Infrared and Laser Engineering, 2023, 52(3): 20220378. doi: 10.3788/IRLA20220378

Research on atmospheric transmission calibration method of infrared target simulator

doi: 10.3788/IRLA20220378
  • Received Date: 2022-08-10
  • Rev Recd Date: 2022-09-20
  • Publish Date: 2023-03-25
  •   Objective   The infrared radiometer is an important device for the calibration of the infrared target simulator and is used as a measurement transfer standard during inspections. The primary calibration parameter for the IR target simulator's radiant energy is irradiance, so the role of the IR radiometer is to measure and calibrate its outgoing irradiance. In infrared transmission, the process will be affected by the atmosphere, including two aspects: One is the infrared radiation by atmospheric molecules, aerosol particles scattered or absorbed by the attenuation, the general use of atmospheric transmittance to characterize the degree of atmospheric attenuation of infrared radiation; second is the atmosphere itself emitted by the atmospheric range of radiation will be superimposed on the target radiation. Atmospheric corrections must be made to improve calibration and measurement accuracy.  Methods   An atmospheric transmission calibration method for infrared target simulators is proposed based on a wide dynamic infrared radiometry method based on a constant standard source. In the calibration of the IR target simulator with a horizontal homogeneous atmospheric approach, a network model of atmospheric transmittance and atmospheric range radiation at different wavelengths, temperatures, and distances is developed using the data analysis capability of convolutional neural networks (Fig.2). Based on the encoder-decoder structure, the detector output voltage under three wavelengths is used as the input of the convolutional neural network, and the test data are normalized and input to the encoder in batches for learning and training, with the batch size set to 8 and the test distance input directly in the embedding layer, and the network is trained according to the training process (Fig.3) to obtain the test distance and atmospheric transmittance and atmospheric range, the model of radiation is given as the correspondence between them.  Results and Discussions   A wide dynamic infrared radiometry method based on a constant standard source was used for multiple experiments to obtain multiple detector output voltage values, which were trained using a network model (Fig.2) to obtain network output values of atmospheric transmittance and atmospheric range radiation at different distances (Fig.7). To verify the improvement of IR radiation measurement accuracy, radiation inversion is performed, and the results of radiation inversion under different methods (Fig.9) can be obtained, and the corresponding IR radiation measurement error graph (Fig.10) and specific values (Tab.2) are shown. The experimental results show that the convolutional neural network algorithm based on the encoder-decoder structure can better predict atmospheric transmittance and atmospheric path radiation, the infrared radiometric average error in three bands of the proposed method is 3.0783%, 3.8186%, 5.3452%, which is far lower than the traditional method, reduces the influence of atmospheric transmittance and atmospheric path radiation, reduces the measurement error of infrared radiation, and improves the calibration accuracy.  Conclusions   The atmospheric transmission correction algorithm is proposed for the problem of atmospheric transmission influence using the direct measurement method. Based on the wide dynamic infrared radiation measurement method with a constant standard source, a convolutional neural network algorithm based on an encoder-decoder structure is used to obtain the relationship between atmospheric transmittance, atmospheric range radiation and waveband and test distance, and atmospheric correction is performed for different wavebands and different test distances. Compared with the traditional method, there is no need to use MODTRAN software to calculate atmospheric transmittance, atmospheric range radiation and atmospheric parameters of the measurement experiment environment, which improves the problem of low distance resolution and accuracy of MODTRAN software under close measurement conditions and improves the accuracy of infrared radiation measurement.
  • [1] Wei Heli, Chen Xiuhong, Rao Ruizhong, et al. A moderate-spectral-resolution transmittance model based on fitting the line-by-line calculation [J]. Optics Express, 2007, 15(13): 8360-8370.
    [2] Zhou Jianbo, Wei Heli, Chen Xiuhong, et al. Progress of calculating the atmospheric absorption by K-distribution method [J]. Journal of Atmospheric and Environmental Optics, 2008(2): 92-99. (in Chinese)
    [3] Osipov V M, Borisova N F. Absorption and emission of IR radiation by the atmosphere on extended oblique tracks [J]. Journal of Optical Technology, 2014, 81(9): 510-517.
    [4] Yang Guoqing. Key technologies research on improving the measurement precision of ground-based infrared radiometry systems[D]. Changchun: Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, 2020. (in Chinese)
    [5] Wei Heli, Chen Xiuhong, Dai Congming. Combined atmospheric radiative transfer (CART) model and its applications [J]. Infrared and Laser Engineering, 2012, 41(12): 3360-3366. (in Chinese)
    [6] Zang Hongshou, Bai Yunta, Ou Yangyi. Research on atmospheric transmittance measuring method [J]. Infrared, 2009, 30(3): 26-29. (in Chinese)
    [7] Li Yunhong, Zhang Long, Wang Yannian. Second calibration of atmospheric transmission coefficients on the temperature measurement of infrared thermal imager in field [J]. Optics and Precision Engineering, 2010, 18(10): 2143-2149. (in Chinese)
    [8] Yang Ciyin, Zhang Jianping, Cao Lihua. Infrared radiation measurement based on real-time correction [J]. Journal of Infrared and Millimeter Waves, 2011, 30(3): 284-288. (in Chinese)
    [9] Guo Lihong, Guo Hanzhou, Yang Ciyin, et al. Improvement of radiation measurement precision for target by using atmosphere-corrected coefficients [J]. Optics and Precision Engineering, 2016, 24(8): 1871-1877. (in Chinese)
    [10] Chen Shuangyuan. Atmospheric infrared transmittance measurement[D]. Kunming: Yunnan Astronomical Observatory Chinese Academy of Sciences, 2019. (in Chinese)
    [11] Yang Guoqing, Li Zhou, Zhao Chen, et al. Nonlinear atmospheric correction based on neural network for infrared target radiometry [J]. Infrared and Laser Engineering, 2020, 49(5): 20190412. (in Chinese)
    [12] Pan Xin, Zhao Jian, Xu Jun. An object-based and heterogeneous segment filter convolutional neural network for high-resolution remote sensing image classification [J]. International Journal of Remote Sensing, 2019, 40(15): 5892-5916. doi:  10.1080/01431161.2019.1584687
    [13] Tardy B, Rivalland V, Huc M, et al. A software tool for atmospheric correction and surface temperature estimation of Landsat infrared thermal data [J]. Remote Sensing, 2016, 8(9): 696.
    [14] Dumoulin J, Boucher V. Infrared thermography system for transport infrastructures survey with inline local atmospheric parameter measurements and offline model for radiation attenuation evaluations [J]. Journal of Applied Remote Sensing, 2014, 8(1): 084978. doi:  10.1117/1.JRS.8.084978
    [15] Huang Zhiguo, Yin Limei, Wang Jianli, et al. Atmospheric attenuation correction based on a constant reference for high-precision infrared radiometry [J]. Applied Sciences, 2017, 7(11): 1165.
    [16] Chen Fangfang, Geng Rui, Lv Yong. Research on the transmittance model of laser infrared atmospheric transmission [J]. Infrared Technology, 2015, 37(6): 496-501. (in Chinese)
    [17] Wei Heli, Dai Congming, Tang Chaoli, et al. Atmospheric parameter model and its application in the calculation of atmospheric raditaive transfer [J]. Infrared and Laser Engineering, 2019, 48(12): 1203001. (in Chinese)
  • 加载中
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Figures(10)  / Tables(2)

Article Metrics

Article views(189) PDF downloads(47) Cited by()

Related
Proportional views

Research on atmospheric transmission calibration method of infrared target simulator

doi: 10.3788/IRLA20220378
  • Naval Aviation University, Yantai 264001, China

Abstract:   Objective   The infrared radiometer is an important device for the calibration of the infrared target simulator and is used as a measurement transfer standard during inspections. The primary calibration parameter for the IR target simulator's radiant energy is irradiance, so the role of the IR radiometer is to measure and calibrate its outgoing irradiance. In infrared transmission, the process will be affected by the atmosphere, including two aspects: One is the infrared radiation by atmospheric molecules, aerosol particles scattered or absorbed by the attenuation, the general use of atmospheric transmittance to characterize the degree of atmospheric attenuation of infrared radiation; second is the atmosphere itself emitted by the atmospheric range of radiation will be superimposed on the target radiation. Atmospheric corrections must be made to improve calibration and measurement accuracy.  Methods   An atmospheric transmission calibration method for infrared target simulators is proposed based on a wide dynamic infrared radiometry method based on a constant standard source. In the calibration of the IR target simulator with a horizontal homogeneous atmospheric approach, a network model of atmospheric transmittance and atmospheric range radiation at different wavelengths, temperatures, and distances is developed using the data analysis capability of convolutional neural networks (Fig.2). Based on the encoder-decoder structure, the detector output voltage under three wavelengths is used as the input of the convolutional neural network, and the test data are normalized and input to the encoder in batches for learning and training, with the batch size set to 8 and the test distance input directly in the embedding layer, and the network is trained according to the training process (Fig.3) to obtain the test distance and atmospheric transmittance and atmospheric range, the model of radiation is given as the correspondence between them.  Results and Discussions   A wide dynamic infrared radiometry method based on a constant standard source was used for multiple experiments to obtain multiple detector output voltage values, which were trained using a network model (Fig.2) to obtain network output values of atmospheric transmittance and atmospheric range radiation at different distances (Fig.7). To verify the improvement of IR radiation measurement accuracy, radiation inversion is performed, and the results of radiation inversion under different methods (Fig.9) can be obtained, and the corresponding IR radiation measurement error graph (Fig.10) and specific values (Tab.2) are shown. The experimental results show that the convolutional neural network algorithm based on the encoder-decoder structure can better predict atmospheric transmittance and atmospheric path radiation, the infrared radiometric average error in three bands of the proposed method is 3.0783%, 3.8186%, 5.3452%, which is far lower than the traditional method, reduces the influence of atmospheric transmittance and atmospheric path radiation, reduces the measurement error of infrared radiation, and improves the calibration accuracy.  Conclusions   The atmospheric transmission correction algorithm is proposed for the problem of atmospheric transmission influence using the direct measurement method. Based on the wide dynamic infrared radiation measurement method with a constant standard source, a convolutional neural network algorithm based on an encoder-decoder structure is used to obtain the relationship between atmospheric transmittance, atmospheric range radiation and waveband and test distance, and atmospheric correction is performed for different wavebands and different test distances. Compared with the traditional method, there is no need to use MODTRAN software to calculate atmospheric transmittance, atmospheric range radiation and atmospheric parameters of the measurement experiment environment, which improves the problem of low distance resolution and accuracy of MODTRAN software under close measurement conditions and improves the accuracy of infrared radiation measurement.

    • 红外辐射计是红外目标模拟器校准的重要装置,在巡检时作为量值传递标准。红外目标模拟器辐射能量的主要校准参数是辐照度,因此红外辐射计的作用就是对其出射辐照度进行测量和校准。在红外传输过程中必然会受到大气影响,若要提高校准和测量准确度必须进行大气修正。现有的大气透过率测量方法包括理论数值计算、软件计算和利用标准参考源测量等。理论数值计算常采用的方法是利用逐线积分法[1]、K 分布法[2]、MT-CDK算法和经验公式[3],但计算过程非常复杂,难以在实际工程中应用。目前广泛使用的方式是利用仿真软件进行模拟计算,较为常用的有LOWTRAN、MODTRAN、FASCODE、CART等[4-5],这种方法方便简单,但在进行实时测量时稳定性较差。除此之外,国内众多学者也提出了一系列辐射测量及大气修正的有效方法。例如:臧寿洪[6]提出直接测量法计算大气透过率;李云红[7]提出了二次修正系数;魏合理致力于大气校正软件的开发;杨词银[8]提出了一种基于实时校标的红外辐射测量方法;郭立红[9]利用线性大气修正因子对大气透过率进行修正;陈双远[10]提出了一种三元定标方法测量大气透过率;杨国庆[11]提出一种非线性大气修正方法。

      深度学习是机器学习和人工智能的一个分支,在各个领域都已经取得了一些研究成果。卷积神经网络(Convolutional Neural Networks, CNN)是深度神经网络一个代表性算法,它的特点是稀疏连接和权值共享[12],只需要用相对较少的参数就可以完成训练。

      文中基于恒定标准源的宽动态红外辐射测量方法,提出一种编码器-解码器结构的CNN,通过其强大的特征提取能力和数据分析能力建立实验环境中大气透过率和大气程辐射随着测试距离变化的模型,再利用红外辐射计进行实时测量,因其精度较高,实现了对大气透过率和大气程辐射的同时测量。

    • 红外辐射计直接测量目标辐射过程如图1所示。这个过程主要依靠探测器,但实际测量时,受环境等因素的影响,红外系统接收到的能量与目标发出的辐射能量并不完全一致。

      Figure 1.  Measuring target radiation of infrared radiometer direct measurement method

      探测器接收到的能量可分为三个部分,分别为经过大气衰减后的目标辐射、目标对周围环境的反射辐射以及探测器和被测目标之间的大气自身辐射[13-14]。所以探测器接收到的辐照度可表示为:

      式中:$ {\tau _{a\lambda }} $为大气透过率;$ {\varepsilon _\lambda } $为目标表面发射率; $ {T_0} $为目标发出温度;$ {T_u} $为环境温度;$ {T_a} $为大气温度;$ {E_{b\lambda }}({T_0}) $为被测目标发出的辐照度;$ {E_{b\lambda }}({T_u}) $为来自周围环境的辐射贡献;$ {E_{a\lambda }}({T_a}) $为红外系统和目标之间的大气程辐射。

      由基于辐亮度测量的标定参数模型可知,探测器输出电压可表示为:

      式中:$ {V_t} $为探测器输出的电压值;$ R(\lambda ) $为光学系统响应度;$ {L_{b\lambda }}({T_0}) $为标准黑体辐射源发出的辐亮度;$ {L_{b\lambda }}({T_u}) $为周围环境的辐亮度;$ {L_{a\lambda }}({T_a}) $为探测器和黑体之间路径大气程辐射的辐亮度值;$ {V_{in}} $为系统内部因素所引起的电压偏移量。当探测器在被测红外带宽范围内的响应度恒定,则公式(2)可以写作:

      由于黑体为朗伯辐射源,其辐亮度$ {L_{b\lambda }}({T_0}) $可以写作:

      也可用辐照度表示为:

      式中:$ {A_s} $为探测器接收面面积;$ f $为焦距。则有:

      对于同一个系统来说,$ k = \dfrac{{{f^2}}}{{{A_s}}} \cdot {\varepsilon _\lambda } $可以看作一个常数。所以包含测试距离的校准模型可以描述为:

      红外辐射计直接测量法具有较高机动性和高效率,往往在进行外场巡检时采用这种方法。采用红外辐射计进行直接校准时,探测器将接收到的辐照度转化为电压信号,在一定的辐射范围内,探测器的输入输出特性可以近似为线性的。进而可以得到目标发出的辐照度为:

    • 从测量原理可知,大气透过率和大气程辐射会对红外辐射计校准精度造成较大影响。在传统红外辐射测量方法中,采用的方法一般是利用仪器测量重要大气分子、大气温湿度、气溶胶及环境参数的高度分布廓线等参数,输入到大气辐射传输软件,如MODTRAN软件,进而计算出大气透射率和大气程辐射。但一般情况下,大气参数的测量过程较为繁琐复杂,在很多情况下无法得到所有的参数信息,而软件的计算精度是由输入的大气参数所决定的,所以传统方法在实际应用中具有一定的局限性。

    • 基于恒定标准源的宽动态红外辐射测量方法通过对已知辐射量的标准参考源黑体进行测量,进而可以计算出辐射传输路径上的大气透过率和大气程辐射[15]。这种方法是根据当前所处的实际大气环境进行测量的,精度明显高于利用软件计算的传统大气修正方法,可以明显改善红外辐射计测量准确度。

      在距离红外辐射计$ {d_0} $处设置一个标准黑体作为恒定标准源。将黑体温度分别设置为$ {T_1} $$ {T_2} $,利用红外辐射计分别测量这两种温度下的恒定标准源,对公式(7)进行简化,得到探测器输出电压分别为$ {V_1} $$ {V_2} $

      式中:$ {V_1} $$ {V_2} $分别为理想黑体在温度为$ {T_1} $${T_2}$时探测器的输出电压;$ R $为红外辐射计系统响应度;$ {V_{in}} $为红外辐射计系统内部的电压偏移量;$ {\tau _{a\lambda }} $为大气透射率;$ {L_{a\lambda }} $表示被测黑体和红外系统之间的大气程辐射;$ {E_{b\lambda }}({T_1}) $$ {E_{b\lambda }}({T_2}) $分别是理想情况下标准黑体辐射源设置温度$ {T_1} $$ {T_2} $时所发出的辐照度。通过对公式(9)和公式(10)解方程可以得到实际的大气透射率和大气程辐射,分别可表示为:

      式中:系统响应度$ R $和电压偏移量$ {V_{in}} $可以提前通过对红外测量系统进行辐射定标获得。

      由参考文献[16-17]可知,对辐射大气传输影响较大的参数包括水蒸气含量、二氧化碳含量、臭氧含量、气溶胶浓度、相对高度、方位夹角、能见度。由于目标模拟器校准工作通常在实验室环境下进行,且红外辐射计入瞳处与被测目标方位夹角近似为0°,所以认为红外辐射在测量过程中是水平均匀传输的。采用基于恒定标准源的宽动态红外辐射测量方法,既不需要预先设置大气模式也不需要提前建立大气辐射传输模型,很大程度上简化了大气修正的过程。除此之外,当红外辐射计和被测目标距离很近时,可以认为大气透过率为1,但随着二者距离变远,大气参数的影响也就越大,大气透过率越小。

    • 利用大气观测设备可以得到当前环境下的大气参数,将所得到的大气参数输入到软件中进而得到测试距离为$ {d_0} $时的理论大气透过率$ {\tau _{{d_0}}} $和大气程辐射值$ {L_{{d_0}}} $,则大气透过率和大气程辐射线性修正因子表示为:

      值得注意的是,大气透过率描述的是大气传输对红外辐射的衰减作用,而大气程辐射则会与目标红外辐射相叠加使测量值偏大,二者性质不同,如果同时对大气透过率和路径辐射进行线性修正,会对目标红外辐射测量结果产生相反的影响,进行大气修正时要根据实时测量条件进行考虑。在任意测量距离$ d $下,通过线性修正因子可以计算校正后的大气透过率和大气程辐射分别表示为:

      式中:$ \tau {'_d} $$ L{'_d} $分别为由MODTRAN计算出距离为$ d $时的大气透过率和大气程辐射理论值。

      在公式(13)的基础上进行改进,有增强大气透过率线性修正因子,可表示为:

      式中:$ d $为红外辐射计和被测目标之间的测试距离;[ ]表示取整运算。则通过增强线性修正因子计算校正后的大气透过率可以表示为:

    • 利用神经网络的非线性拟合技术,将Modtran4.0在不同距离下的理论大气透过率和大气程辐射作为输入,将基于恒定标准源的宽动态红外辐射测量方法得到的大气透过率和大气程辐射测量值作为输出,使用BP神经网络进行训练,找到二者的非线性关系,进而计算出大气透过率和大气程辐射。

    • 各种神经网络架构已被开发用于不同类型的任务。在图像处理相关任务领域,如图像分类、分割、超分辨率和人脸识别,CNN因其优越性和较少的可训练参数而被人们所熟知。与一般的图像处理类似,采用编码器-解码器结构的CNN算法作为大气校正问题的主要框架。

      文中的网络模型如图2所示,CNN的架构分为三个主要模块,包括卷积编码器、Embedding层和卷积解码器。网络由3层编码器和4层解码器组成,其中每个编码器由一个卷积层(Conv)、一个归一化层(Batchnorm)以及一个带泄露修正线性单元函数(LeakyReLU)作为激活层构成,每个解码器由转置卷积层(Convtrans)、归一化层、改进激活层以及激活层(ReLU)构成。网络的参数设置如表1所示,包括编码器、解码器及Embedding层的卷积核、步幅和填充。输入为探测器输出电压和测试距离,其中编码器所需的电压值是在实验室环境下探测器接收到标准黑体辐射源的测试数据。图2中,输入的特征向量为探测器输出电压$ V $$ (c,T,\lambda ) $为描述$ V $的特征参数,其中,$ c $是限定维度的常量,这里取1,$ T $$ \lambda $分别代表红外辐射的温度和波段。测试距离$ d $是直接输入隐藏层中的全连接层中。

      Figure 2.  Convolutional neural network model of encoder-decoder architecture

      Operation layerNumber of filtersSize of each filterStridePaddingOutput size
      1×9×3
      EncoderConv_1644×32×211×2×64
      Conv_21285×32×211×1×128
      Conv_36418×11×111×1×64
      Embedding layersEmbedding_distance----64
      Embedding_1----64
      Embedding_2----64
      DecoderConvtrans_125618×11×112×2×128
      Convtrans _21285×42×214×4×64
      Convtrans _3645×32×218×8×2
      Convtrans_424×32×212×3×9

      Table 1.  Network parameter settings

      为了提高模型的训练效率,减少输入的空间维度,同时降低对样本训练集大小和计算资源的需求,对编码器进行改进,不采用传统卷积法中的最大值池化,而是采用步进卷积,将输入压缩成一个潜在空间的向量。

      在保证模型精度的前提下,降低模型复杂度,利用编码器将输入的电压值降维成一个一维向量。然而,为了纳入红外辐射计和被测目标的位置关系,增加了一个独立的Embedding层,在其中处理测试距离。编码器和独立的Embedding层输出一个向量,将这个向量作为潜在空间处理的输入。潜在空间包含两个Embedding层,每层包含64个神经元,在每一个Embedding层之后都要进行LeakyReLU运算。

      理论上,潜在空间需要将混合代表向量计算为需要的大气透过率和大气程辐射,所以需要解码器将大气透过率和大气程辐射扩展为可读维度作为输入,所以采用转置卷积层进行上采样,对向量升高维度,四个转置卷积层的参数见表1,在转置卷积层后放置一个批量归一化层和LeakyReLU的组合,防止梯度爆炸。将编码器最后一层的LeakyReLU修改为ReLU,用于消除输出结果中的负参数。

    • 在将实验所得到的电压值输入设计的网络中,其架构见图2。测试数据包含三个波段的测试数据,且涉及到的温度点较多,笔者观察到,随着温度升高,探测器输出电压也随之增大,数据分布跨度大,量纲不统一,直接放入网络中进行拟合的话,模型精度会有所降低。对于这种情况有两种处理方式:一是增加卷积层数量,数量越多准确率越高,但这种方法无疑会增加计算负担,使其更难训练;另一种是对数据进行预处理,降低输入特征之间的差异,消除量纲对模型的影响,减少运算量提高性能。为此,对实验数据进行归一化处理。

      采用min-max归一化方法,对原始数据线性变换,将数据缩放到[0,1]范围内,则归一化后的样本数据可以表示为:

      式中:$ x $为样本数据;$ {x_{\max }} $$ {x_{\min }} $分别为样本数据中的最大值和最小值。

    • 建立编码器-解码器结构的CNN算法模型,实验过程中三个波段下的测试数据经归一化后分批输入编码器进行学习训练,批尺寸大小设为8,测试距离在Embedding层直接输入,当前距离下的大气透过率和对应的大气程辐射作为解码器输出,利用卷积神经网络进行学习训练,得到测试距离和大气透过率、大气程辐射的模型,给出两者的对应关系。

      文中提出的基于编码器-解码器结构的CNN算法的大气修正训练过程如图3所示。在训练阶段利用Adam优化器进行网络性能优化,学习率设为0.001。

      Figure 3.  Encoder-decoder CNN model training flowchart

    • 为了验证分析结果和提出的校准算法,用实验室搭建的红外辐射计和标准黑体辐射源进行了辐射测量校准实验。该设备的红外探测器工作在1~16 μm的波段,光敏面尺寸为1 mm×1 mm,通过计算机控制红外辐射计内的滤光片轮转动,可实现对1~3、3~5、8~14 μm三个波段的红外辐射测量。 采用的标准黑体辐射源的温度范围为50~1 000 ℃,温度分辨力为0.01 ℃,有效辐射面大小为Φ50 mm,工作波长范围为1.4~14 μm,在此范围内具有较高的有效发射率,不小于0.995。

    • 在一定积分时间内,红外辐射计内探测器输出的电压值与辐照度之间呈线性关系。为了得到系统响应度和偏置电压值,采用平行光管定标法对红外辐射计进行辐射定标,流程如图4所示,实验装置如图5所示。标准黑体辐射源采用高温腔型黑体,平行光管可以对黑体发出的辐射进行均匀扩束,从而得到平行辐射。调节标准黑体辐射源的温度并得到探测器输出电压,便可以计算出红外辐射计的定标响应参数。

      Figure 4.  Infrared radiometer calibration flowchart

      Figure 5.  Experiment device of infrared radiometer calibration

      设置标准黑体辐射源的温度范围为100~900 ℃,根据普朗克公式可以计算出对应的黑体辐射照度,对1~3、3~5、8~14 μm三个波段下测量的实验结果进行拟合,结果如图6所示。拟合方程可写作:

      由公式(20)可知,红外辐射计的系统响应为1.206×104,偏置电压为2.582×10−3 V。

      Figure 6.  Infrared radiometer calibration results

    • 为了验证模型性能,在测试距离范围0.5~3 m,间隔0.5 m,采用基于恒定标准源的宽动态红外辐射测量方法进行多组实验,得到多组探测器输出电压值,使用图2中所构建的结构对网络进行训练。训练后各距离的大气透过率和大气程辐射的网络输出值如图7所示。

      图7中,随着测试距离增加,大气透过率减小且大气程辐射增加,距离越远,大气传输对红外辐射的影响越明显;黑体温度几乎不会对大气透过率和大气程辐射产生影响; 在不同波长下的大气传输也不同,可能导致这种现象的原因是在空气中气体成分对某些特定波段影响较大,例如水蒸气分子和气溶胶三个波段都有较为明显的影响,二氧化碳及其他微量气体成分(N2O和CH4)对3~5 μm波段影响较大,而臭氧对长波8~14 μm波段影响较大。

      Figure 7.  Atmospheric transmittance and atmospheric path radiation at different distances

    • 在常规的辐射测量方法中,大气透过率和大气程辐射是用Modtran4.0计算的,而线性修正和非线性修正方法也必须要使用Modtran进行模拟计算,这就使得测量精度很大程度上取决于输入的大气参数的完整度和准确度。实验过程中,地面平均温度约为20 ℃,压力约为1 021 kPa,相对湿度约77%,海拔11 m,能见度23 km。对MODTRAN软件输入相关大气参数,计算水平距离为1~3 m,间隔为1 m。

      采用基于恒定标准源的宽动态红外辐射测量方法,将标准黑体辐射源放置于距离红外辐射计2 m处,将温度设置为200 ℃和700 ℃,此时在三个波段下Modtran4.0、线性修正方法、非线性修正方法及编码器-解码器结构的CNN算法得到的大气透过率和大气程辐射如图8所示。

      为了验证红外辐射测量精度的提高,进行辐射反演,设置黑体温度为500 ℃,此时的辐照度理论值和利用Modtran4.0、线性修正方法、非线性修正方法及编码器-解码器结构的CNN算法得到的大气透过率和大气程辐射进行辐射反演的结果如图9所示,对应的红外辐射测量误差如图10所示,具体数值如表2所示。

      Figure 8.  Calculation results of atmospheric transmittance and atmospheric path radiation with different methods

      Figure 9.  Comparison of radiative inversion results with different methods

      Figure 10.  Errors of radiative inversion results with different methods

      从误差曲线可以看出,编码器-解码器结构的CNN算法的误差明显小于其他方法,且误差相对稳定,考虑到误差大小可能主要是由算法引起的。当测试距离较小时,Modtran4.0的误差较大,且Modtran4.0的对于距离的分辨率为1 m,不适合实验室环境下的近距离测量。线性修正方法计算结果趋势相同,这种方法认为实验环境为大气分布均匀,仅进行了线性校正;而非线性修正方法需要利用Modtran4.0模拟数据和实测值进行拟合,在近距离测量时无法获得大量Modtran4.0数据,样本量小,导致拟合精度较差。

      Measurement errorMeasurement method1-3 μm3-5 μm8-14 μm
      Mean relative errorModtran4.04.4408%7.7930%9.2662%
      Encoder-decoder CNN3.0783%3.8186%5.3452%
      Linear correction method3.9120%5.9204%8.2427%
      Nonlinear correction method3.7673%5.3866%7.8982%
      Maximum relative errorModtran4.04.6787%7.9853%9.4576%
      Encoder-decoder CNN3.1332%4.1722%5.4866%
      Linear correction method4.0767%6.0062%8.4335%
      Nonlinear correction method3.7945%5.6752%8.1451%

      Table 2.  Measurement error

    • 在基于恒定标准源的宽动态红外辐射测量方法的基础上,提出了一种基于编码器-解码器结构的CNN算法的红外目标模拟器的大气传输校准方法,从而提高校准精度。利用编码器-解码器结构的CNN算法得到大气透过率、大气程辐射与波段、测试距离之间的关系,针对不同波段、不同测试距离进行大气修正。与传统方法相比,不需要使用Modtran软件计算大气透过率、大气程辐射,所以不必测量实验环境的大气参数,也改进了近距离测量条件下软件的计算结果数据少、精度不高的问题。通过辐射反演进行验证,在水平均匀大气近距离条件下,编码器-解码器结构的CNN算法的红外辐射在三个波段下的平均误差分别为3.0783%、3.8186%、5.3452%,与传统的软件计算方法相比,提高了红外辐射校准精度。

Reference (17)

Catalog

    /

    DownLoad:  Full-Size Img  PowerPoint
    Return
    Return