Super resolution convolutional neural network for feature extraction in spectroscopic data.

Area of Science:

• Physics
• Data Analysis
• Machine Learning

Background:

• Two-dimensional (2D) peak finding is crucial in physics experiments.
• Traditional derivative-based methods are unstable with complex data or low signal-to-noise ratios.

Purpose of the Study:

• To propose a novel, stable, and accurate 2D peak finding method.
• To leverage convolutional neural networks (CNNs) for peak tracking in experimental data.

Main Methods:

• Formalizing peak tracking as an inverse problem solvable by CNNs.
• Utilizing underlying physics principles to generate synthetic training data for the CNN.
• Generalizing the trained CNN to real experimental data.

Main Results:

• The CNN method achieves comparable or superior results to traditional derivative-based methods.
• Demonstrated robustness in complex data landscapes and low signal-to-noise conditions.
• Successful generalization of the trained network on real experimental data.

Conclusions:

• The proposed CNN-based peak finding method offers a more stable and accurate alternative.
• This approach is adaptable to various physics experiments with known physical processes.
• Highlights the potential of machine learning in scientific data analysis.