Multispectral radiometric thermometry inversion algorithm using multisource information fusion.

Abstract

Listen

Translate article

This paper proposes a multispectral temperature measurement method integrating multiple information sources to address inaccuracies in target temperature measurements in multispectral thermometry when emissivity is unknown. This algorithm leverages convolutional neural networks for image recognition to convert one-dimensional voltage data into two-dimensional voltage spectral images. To enhance accuracy, it combines emissivity trend inversion with emissivity constraint algorithms. Additionally, a temperature-constraint function algorithm is incorporated to improve the computational speed. Through simulation experiments and comparative tests, the algorithm achieved a relative error below 0.08% and an absolute error below 2 K within a temperature range of 1000 to 4000 K, with an average runtime under 5 ms. The experimental results demonstrate that the method not only addresses efficiency issues in multispectral thermometry but also enhances temperature measurement accuracy, providing crucial technical support for industrial applications of multispectral thermometry.