Graphical multispectral radiation temperature inversion algorithm based on deep learning.

Area of Science:

• Metrology
• Radiometry
• Deep Learning

Background:

• Accurate multispectral radiometric temperature measurement requires an emissivity model, posing a significant challenge.
• Current neural network algorithms for this task suffer from unsatisfactory inversion accuracy and adaptability.
• Limitations exist in network selection, porting, and parameter optimization for existing methods.

Purpose of the Study:

• To improve the accuracy and adaptability of multispectral radiometric temperature measurement using deep learning.
• To address the limitations of existing neural network algorithms in this field.
• To introduce a novel data processing technique for enhanced temperature measurement.

Main Methods:

• Proposed converting one-dimensional multispectral radiometric temperature data into two-dimensional image data.
• Applied deep learning algorithms for data processing on the transformed image data.
• Conducted simulation and experimental validation to assess the method's performance.

Main Results:

• Simulated error was <0.71% (no noise) and <1.80% (5% noise), outperforming BP and GIM-LSTM algorithms.
• Experimental error was <0.83%.
• Demonstrated significant improvements in accuracy and adaptability compared to classical and advanced algorithms.

Conclusions:

• The proposed method of converting 1D data to 2D images for deep learning offers superior accuracy and adaptability in multispectral radiometric temperature measurement.
• This approach shows high research value and potential to advance the field.
• The findings suggest a new direction for developing next-generation temperature measurement technologies.