Iterative algorithm computational spectrometer based on a single-hidden-layer neural network.

Area of Science:

• Spectroscopy
• Computational Imaging
• Machine Learning

Background:

• Computational spectrometers offer promising applications in hyperspectral detection.
• Iterative algorithms enable hardware integration and in situ measurements but struggle with reconstruction accuracy due to ill-posed problems.
• Neural networks provide high-precision spectral reconstruction but demand significant computational resources, hindering integration into embedded systems.

Purpose of the Study:

• To improve the reconstruction accuracy of iterative algorithm-based computational spectrometers.
• To leverage neural networks to mitigate the ill-posed nature of spectral reconstruction problems.
• To enable fast and high-precision in situ measurements using computational spectrometers.

Main Methods:

• Spectral reconstruction was performed using iterative algorithms on a public dataset.
• A single-hidden-layer neural network was trained to map iterative reconstruction results to original spectra.
• The hybrid approach was validated through simulations and experimental results.

Main Results:

• The proposed method effectively alleviates the ill-posed problem in iterative spectral reconstruction.
• Neural network integration significantly improved the reconstruction accuracy of computational spectrometers.
• The approach demonstrated low computational resource requirements, suitable for embedded systems.

Conclusions:

• Combining neural networks with iterative algorithms enhances computational spectrometer performance.
• This hybrid model offers a viable solution for achieving high-precision, in situ spectral measurements.
• The research holds potential for advancing hyperspectral detection technologies.