Wood you believe it-Detecting copper with hyperspectral imaging.

Area of Science:

• Materials Science
• Environmental Science
• Data Science

Background:

• Wood resources are underutilized due to non-recyclability, often caused by inorganic preservatives like copper.
• Identifying and managing these preservatives is key to enhancing wood circularity and cascade use.
• Current methods for detecting wood preservatives are limited in scope and accuracy.

Purpose of the Study:

• To develop and evaluate AI-driven hyperspectral imaging techniques for detecting and quantifying copper-based wood preservatives.
• To assess the performance of different machine learning models in analyzing hyperspectral data from treated wood.
• To investigate the generalization capabilities of AI models across different wood species and preservative treatments.

Main Methods:

• Collected a dataset of 1320 wood samples across six species, treated with various inorganic copper-based and organic preservatives at different concentrations.
• Utilized hyperspectral imaging in the visible and near-infrared (VNIR) and short-wave infrared (SWIR) ranges.
• Trained and evaluated three AI models: Support Vector Classifier (SVC), Partial Least Squares Discriminant Analysis (PLS-DA), and a Convolutional Neural Network (ResNet18).

Main Results:

• SVC and ResNet18 models achieved over 90% accuracy in detecting copper preservatives, outperforming the linear PLS-DA model.
• Non-linear models effectively captured complex spectral patterns in hyperspectral data.
• Out-of-Distribution generalization tests revealed that model performance is highly dependent on the diversity of the training data, requiring broad coverage of wood species and preservative types.

Conclusions:

• AI combined with hyperspectral imaging offers a highly accurate method for identifying copper-based wood preservatives.
• Non-linear AI models are superior for analyzing complex hyperspectral signals from treated wood.
• Robust model generalization for wood circularity applications necessitates comprehensive training datasets encompassing diverse species and treatments.