Accurate paper fiber identification is essential for cultural heritage conservation to select appropriate repair materials. However, conventional methods are often destructive, and macroscopic data-driven approaches lack physicochemical interpretability. This study introduces a non-destructive, interpretable approach to discriminate three traditional Japanese bast fibers (Kozo, Mitsumata, Gampi) using micro-hyperspectral imaging (Micro-HSI) combined with chemometric analysis.

Using well-documented, undyed, single-raw-material samples from an authenticated archive, we established baseline spectral characteristics. Reflectance spectra from microscopic regions were collected in the 400–1000 nm range. Data were preprocessed using Savitzky–Golay first derivative filtering and Z-score normalization to mitigate baseline drift and intensity variations.

While principal component analysis (PCA) yielded limited separation due to the fibers' similar cellulose compositions, linear discriminant analysis (LDA) achieved clear classification. LDA highlighted subtle spectral differences, with key discriminative features linked to visible-range baseline variations, lignin-related absorption (400–450 nm), and near-infrared O–H and C–H overtone vibrations (~835 nm).

As this is currently a feasibility study, the model relies on pristine samples. Future work will expand the dataset to include aged, mixed-fiber, and geographically diverse papers, incorporating additional preprocessing methods like standard normal variate (SNV) to enhance robustness. These findings validate Micro-HSI as an effective, non-destructive tool for fiber identification, providing a crucial baseline for the future conservation of historical paper artifacts.