Intelligent spectral algorithm for pigments visualization, classification and identification based on Raman spectra.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Chemometrics

Background:

• Raman spectroscopy is a powerful molecular vibrational technique for chemical analysis.
• Rapid field detection applications are expanding, requiring robust data processing methods.
• Accurate identification and classification of materials like pigments are crucial in various fields.

Purpose of the Study:

• To investigate Raman spectral pretreatment and classification algorithms for pigment identification.
• To evaluate the performance of multiple algorithms, including SVM, KNN, and ANN.
• To assess the impact of signal-to-noise ratio (SNR) on algorithm accuracy and explore denoising solutions.

Main Methods:

• Utilized spectral data from nearly 300 pigments.
• Applied and compared over five classification algorithms (SVM, KNN, ANN, etc.).
• Introduced SNR as a metric for algorithm stability and evaluated denoising algorithms (flattopwin, hanning, blackman).

Main Results:

• Most classification algorithms achieved approximately 90% accuracy for pigment identification.
• Algorithm accuracy sharply decreased at low SNR, with a peak accuracy of 39.46% at SNR=1.
• Denoising techniques improved accuracy to 80% even at SNR=1, demonstrating robustness.

Conclusions:

• Intelligent algorithms, combined with effective denoising, can reliably solve challenges in Raman spectral detection.
• The developed methods offer a pathway for accurate and efficient pigment visualization, classification, and identification.
• This research validates the use of advanced algorithms for robust Raman spectral analysis in demanding conditions.