Abstract：Multispectral radiation thermometry is a method of creating multiple spectral channels in an instrument, using multiple spectral radiance information of the measured target, and processing the data to obtain the temperature. This method has no special requirements for the measured object, making it particularly suitable for simultaneous measurement of temperature and material emissivity of high-temperature targets. Due to the influence of unknown spectral emissivity, the problem of multispectral radiometric temperature inversion can be summarized as an underdetermined equation-solving problem under emissivity constraints. Therefore, multispectral radiometric temperature inversion algorithms have always been a difficult and hot research topic in this field. With the continuous development of deep learning, to fully utilize the precise feature recognition ability of deep learning algorithms in the image field to solve the problem of multispectral radiometric temperature inversion, this paper proposes a multi-channel convolutional neural network (MC-CNN) based on the fusion of Markov Transition Field (MTF) and Gramian Angular Field (GAF) for multispectral radiometric temperature inversion. Deep learning algorithms have obvious advantages in the field of image feature recognition. First, MTF and GAF methods convert one-dimensional spectral voltage data into two-dimensional images with spectral temperature features. Then, the images carrying spectral temperature features are input into an improved convolutional neural network for training, thereby achieving temperature inversion. The simulation results show that for 8 spectral channel data, the average absolute error of inversion at 1K evenly distributed temperature points between 2 355 and 2 624 K is 16.6 K, and the average relative error is 0.7%. Compared with the theoretical value, the inversion error of the measured data of the rocket tail flame is within ±16.5 K, indicating a high inversion accuracy. This method is not affected by unknown emissivity and directly uses spectral voltage data to invert temperature values, further improving the multispectral radiation temperature measurement theory.
[1] Zheng S, Cai W G, Liu B, et al. Fuel., 2023, 333: 126391.
[2] SUN Bo-jun, SUN Xiao-gang, DAI Jing-min(孙博君，孙晓刚，戴景民). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2021, 41(5): 1336.
[3] Zheng S, Yang Y, Li X Y, et al. Fuel Processing Technology, 2020, 204: 106423.
[4] Zou Z, Fu X, Zhang Y, et al. Infrared Physices & Technology, 2022, 127: 104385.
[5] Chen L W, Zhang X Q, Gao S, et al. Optics Express, 2022, 30(20): 36603.
[6] Zheng S, Cai W G, Sui R, et al. Fuel, 2022, 323: 124328.
[7] Yang Z, Wang B, Dai J, et al. Results in Physics, 2023, 53: 107014.
[8] Liang M, Wang Y, Wang C. Optics Express, 2023, 31(20): 33638.
[9] Liang M, Wang Y, Wang C, et al. Measurement Science and Technology, 2022, 33(6): 65004.
[10] Dong P K, Zhou L, Liu Z H, et al. Applied Optics, 2024, 63(30): 7900.
[11] Zhang X, Han Y. Photonics., 2024, 11(8): 767.
[12] Jiang H, Tang R, Wang C Y, et al. IEEE Sensors Journal, 2024, 24(7): 10125.
[13] Sun X G, Yuan G B, Dai J M, et al. International Journal of Thermophysics, 2005, 26(4): 1255.
