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

Area of Science:

• Thermophysics
• Optical Engineering
• Machine Learning

Background:

• Multispectral temperature measurement is hindered by unknown material emissivity.
• Existing algorithms struggle with emissivity variations and broad material applicability.
• Accurate temperature sensing is critical in diverse industrial and scientific applications.

Purpose of the Study:

• To develop a robust multispectral radiation thermometry method independent of emissivity.
• To create a versatile algorithm applicable to a wide range of materials.
• To enhance the accuracy and reliability of non-contact temperature measurements.

Main Methods:

• Implementation of an improved multi-branch convolutional neural network (CNN).
• The CNN incorporates branches for temperature inversion, wavelength, voltage ratio, emissivity, and reference temperature.
• Utilized feature extraction and interaction for accurate temperature prediction.

Main Results:

• Simulation experiments demonstrated a maximum absolute error below 7 K.
• Real-world rocket experiments showed a maximum error of 9.13 K.
• The model exhibited strong generalization capabilities across different datasets and conditions.

Conclusions:

• The proposed method effectively addresses the challenge of unknown emissivity in multispectral thermometry.
• The multi-branch CNN model demonstrates broad applicability to various materials and emissivity models.
• This research offers new avenues for advancements in radiation thermometry theory and practice.