COS-DeformDeep: Adaptive 2T2D spectral feature extraction method for improving the component identification...

Area of Science:

• Analytical Chemistry
• Spectroscopy

Background:

• Raman spectroscopy is vital for analyzing mixture components.
• Handheld Raman spectrometers offer portability and rapid on-site data acquisition.
• Quantifying components in complex mixtures using Raman spectroscopy remains challenging.

Purpose of the Study:

• To introduce COS-DeformDeep, a novel method for enhancing spectral features in handheld Raman mixture analysis.
• To improve the identification and quantification of individual components within complex mixtures.

Main Methods:

• Utilized synchronous two-trace two-dimensional correlation spectroscopy (2T2D-COS) to enhance weak signals in overlapped peaks.
• Employed deformable convolutions (DCNs) to improve deep learning model adaptability for spectral feature extraction.
• Validated the COS-DeformDeep model on three mixture datasets containing Ethanol, Diacetone alcohol, and Histidine.

Main Results:

• The COS-DeformDeep model demonstrated superior performance in mixture component identification.
• Achieved an average accuracy of 94.97%, precision of 98.45%, recall of 92.44%, and F1 score of 95.06%.
• Effectively identified components with volume-weight ratios as low as 2%.

Conclusions:

• COS-DeformDeep efficiently extracts features from weak spectral signals, enhancing recognition accuracy for low-concentration components.
• The method simplifies spectral analysis and is suitable for handheld devices, increasing accessibility.
• Significantly improves the capability of handheld Raman spectroscopy for complex mixture analysis.