[Adaptive "3R" De-Noising Algorithm Based on Near Infrared Bi-Spectrum].

Area of Science:

• Spectroscopy
• Chemometrics
• Signal Processing

Background:

• Near-infrared (NIR) spectroscopy is widely used for chemical analysis.
• Noise in spectral data can significantly hinder accurate quantitative analysis.
• Existing de-noising methods often involve parameter tuning or can alter spectral features.

Purpose of the Study:

• To develop an adaptive de-noising algorithm for NIR spectra.
• To evaluate the performance of the proposed algorithm against established methods.
• To assess the impact of de-noising on subsequent chemometric modeling.

Main Methods:

• Synchronous collection of NIR transmission and absorption spectra.
• Decomposition of spectra into intrinsic mode functions (IMFs) using Ensemble Empirical Mode Decomposition (EEMD).
• Development of the "3R" algorithm to identify and remove noise IMFs based on correlation and self-correlation analysis.

Main Results:

• The "3R" algorithm effectively identified and removed noise components.
• Simulated and real NIR spectra de-noised by the "3R" method showed superior performance in BP neural network models for chlorophyll prediction.
• The "3R" method exhibited minimal distortion of spectral peak positions and intensities compared to EMD, EEMD, and wavelet decomposition.

Conclusions:

• The "3R" algorithm provides an adaptive and robust solution for NIR spectral de-noising.
• It requires no prior assumptions on the number of iterations or decomposition layers.
• The method demonstrates strong adaptability and preserves crucial spectral information for quantitative analysis.