Lifting wavelet transform for Vis-NIR spectral data optimization to predict wood density.

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Forestry Science

Background:

• Visible and near-infrared (Vis-NIR) spectroscopy is a powerful analytical technique.
• The 'curse of dimensionality' in Vis-NIR data increases computational complexity.
• Variable optimization is crucial for enhancing prediction accuracy by removing noise.

Purpose of the Study:

• To analyze the performance of 2D correlation spectroscopy with spectral transformations.
• To optimize spectral variables for improved wood density prediction.
• To reduce the dimensionality of Vis-NIR spectral data.

Main Methods:

• Spectral transformations including reflectance (R), reciprocal (1/R), and logarithm (log(1/R)).
• Decomposition using biorthogonal wavelet family (3rd-8th level) via lifting wavelet transform (LWT).
• Variable selection using uninformative variable elimination (UVE) and competitive adaptive reweighted sampling (CARS).
• Partial least squares (PLS) for optimal wavelet coefficient selection.

Main Results:

• Dimensionality of the spectral matrix was reduced from 2048 to 16 variables.
• Optimal spectral transformation and decomposition were identified.
• Best wood density prediction for Siberian elm (Ulmus pumila L.) achieved (Rp² = 0.899, RMSEP = 0.016) using LWT.

Conclusions:

• Lifting wavelet transform (LWT) effectively reduces spectral data dimensionality.
• Optimized Vis-NIR spectral analysis significantly improves wood density prediction accuracy.
• This approach enhances the application of Vis-NIR spectroscopy in forestry and wood quality assessment.