General-purpose mid-infrared micro-spectrometer based on hierarchical residual CNN and data augmentation

Area of Science:

• Optoelectronics
• Spectroscopy
• Artificial Intelligence

Background:

• Reconstructive spectrometers offer broad response and snapshot capabilities, but face challenges with sparse sampling and generalization.
• Computational techniques and microstructures are key to advancing spectrometer design.

Purpose of the Study:

• To develop and demonstrate a mid-infrared micro-spectrometer with enhanced spectral resolution and reliability.
• To address limitations in sparse sampling and data-driven reconstruction for micro-spectrometers.

Main Methods:

• Utilized a grating-integrated lead selenide detector array for spectral sampling.
• Employed a hierarchal residual convolutional neural network (HRCNN) for spectral reconstruction.
• Implemented data augmentation to improve the generalization ability of the HRCNN model.

Main Results:

• Achieved a spectral resolution of 15 nm across the 2.5-5 µm mid-infrared range.
• Demonstrated excellent reliability with an average reconstruction error of ~1E-4 for over one hundred chemical species, including untrained ones.
• Validated the effectiveness of HRCNN and data augmentation in overcoming sparse sampling challenges.

Conclusions:

• The developed mid-infrared micro-spectrometer shows high performance and reliability, promoting reconstructive spectrometer strategies.
• This approach offers a promising solution for compact, high-resolution spectral analysis in various chemical applications.
• The integration of advanced computational methods with micro-optic hardware is crucial for future spectrometer development.