Classification of Traditional Handmade Papers from China, Japan, and Korea Using NIR Hyperspectral Imaging

Traditional handmade papers from China, Japan, and Korea, including Xuan paper, Washi, and Hanji, are difficult to distinguish visually because they share cellulose-rich compositions and similar appearances. This study applied near-infrared hyperspectral imaging (NIR-HSI) and machine-learning classifiers to identify selected traditional handmade papers by country and product type. Spectra in the 1250-1700 nm region were analyzed using k-nearest neighbors, support vector machines, and artificial neural networks. The models achieved high classification performance, with F1-scores of up to 1.000, and Y-scrambling confirmed that the results were not attributable to random class assignment. SHAP analysis identified important wavelength regions near 1256, 1360, 1404, 1449, 1537, 1576, 1635, and 1685 nm, which were associated with C-H, O-H, phenolic, hydrogen-bonded polysaccharide, and lignin-related vibrations. These bands varied among paper groups and provided chemically meaningful information for classification, while SAM visualization revealed pixel-level spectral similarity. These results show that NIR-HSI provides a compact, nondestructive, and interpretable approach for classifying selected East Asian handmade papers.