Quality Control of Thermally Modified Western Hemlock Wood Using Near-Infrared Spectroscopy and Explainable Machine...

Area of Science:

• Materials Science
• Wood Science
• Spectroscopy

Background:

• Quality control of thermally modified wood is crucial.
• Nondestructive testing methods are needed to assess heat treatment intensity.
• Near-infrared (NIR) spectroscopy offers potential for wood characterization.

Purpose of the Study:

• To classify thermally modified wood using NIR spectroscopy and machine learning.
• To identify critical NIR wavelengths for distinguishing heat treatment intensities.
• To develop an explainable machine learning framework for wood quality control.

Main Methods:

• Collected NIR spectra from untreated and thermally treated western hemlock samples (170°C, 212°C, 230°C).
• Employed a TreeNet gradient boosting machine for classification without dimensionality reduction.
• Analyzed feature importance to understand critical wavelengths and their contribution to model performance.

Main Results:

• Achieved high classification accuracies (94.35% with 1100-2500 nm range).
• Identified wood color changes as a primary driver of NIR reflectance variation among treatments.
• Found that models incorporating wood color information were more accurate than those focusing solely on water or wood chemistry.

Conclusions:

• The developed NIR spectroscopy and machine learning framework effectively classifies thermally modified wood.
• Wood color is a key indicator for assessing heat treatment intensity via NIR.
• This approach provides valuable insights for wood characterization and quality control applications using NIR spectroscopy.