Adaptive compressed sensing of Raman spectroscopic profiling data for discriminative tasks

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Data Science

Background:

• Raman spectroscopy offers rapid, non-invasive physio-chemical fingerprinting for discriminative tasks.
• Compressed sensing (CS) can enhance Raman spectrometry by reducing memory, acquisition time, and sensor costs.
• Traditional CS methods focus solely on signal processing, neglecting categorical information crucial for classification.

Purpose of the Study:

• To propose a novel method optimizing CS hyper-parameters for Raman spectroscopy using both spectral and categorical data.
• To improve the efficiency and effectiveness of CS in discriminative tasks involving spectroscopic data.
• To demonstrate the application of optimized CS in real-world identification scenarios.

Main Methods:

• Developed a method to optimize CS hyper-parameters (sampling ratio, basis matrix, regularization rate) by integrating spectral and categorical information.
• Applied the proposed method to a case study involving formula milk brand identification using Raman spectrometry.
• Evaluated the discriminative power and efficiency of the reconstructed signals under optimized CS parameters.

Main Results:

• The proposed method effectively optimizes CS for Raman spectroscopy, preserving discriminative information.
• A formula milk brand identification task achieved 100% classification accuracy.
• This accuracy was maintained while reducing the original signal sampling to just 20%.

Conclusions:

• Integrating signal and categorical information significantly enhances CS performance in discriminative spectroscopic tasks.
• Optimized CS offers a cost-effective and efficient solution for Raman spectrometry, reducing data requirements without compromising accuracy.
• The proposed approach demonstrates the potential for broader applications in chemical analysis and identification.