Multispectral radiation thermometry method based on a multi-branch convolutional model

Multispectral temperature measurement is affected by unknown emissivity, and there is no algorithm that can ignore the influence of emissivity and be applicable to all materials. To solve this problem, this paper proposes a multispectral radiation thermometry method based on a multi-branch convolutional model. The core of this method is an improved multi-branch convolutional network model, which includes branches of inversion temperature, wavelength, voltage ratio, emissivity, and reference temperature. Through feature extraction and interaction, prediction results are obtained. For newly generated data sets, the maximum absolute error in simulation experiments is controlled at a level not higher than 7 K. In the prediction of actual rocket experiments, the maximum error is 9.13 K. This result indicates that the model has good generalization ability. More importantly, the model has broad applicability and can adapt to various materials and different emissivity models, providing what we believe to be new research ideas and directions for the field of multispectral radiation thermometry and is expected to promote further breakthroughs in both theory and practice in this field.