Non-destructive prediction and pixel-level visualization of polysaccharide-based properties in ancient paper using SWNIR hyperspectral imaging and machine learning

Ancient documents and artworks are invaluable cultural heritage artworks that require careful preservation. Traditional methods for assessing their physical and chemical properties-such as tearing index, tensile index, water absorption, and pH-are often destructive, risking irreversible damage. This study introduces a novel, non-destructive approach using Short-Wave Near-Infrared (SWNIR) hyperspectral imaging (HSI) combined with advanced machine learning models. By integrating spectral preprocessing, feature selection, and machine learning techniques-including Back Propagation Neural Networks (BPNN), Long Short-Term Memory Networks (LSTM), and Convolutional Neural Networks (CNN)-with Sparrow Search Algorithm (SSA) optimization and Gray Level Co-occurrence Matrix (GLCM) texture feature extraction, the resulting SSA-BP-UVE-GLCM model achieved high predictive accuracy (R² ≥ 0.98). This framework enables precise, pixel-level predictions of paper properties, influenced by polysaccharides like cellulose, offering a non-invasive analysis that supports targeted restoration strategies and advances the conservation of cultural heritage. The findings enhance non-invasive testing and classification methods for polysaccharide-based materials, providing a foundation for further exploration of environmental impacts on artwork integrity using sophisticated machine learning algorithms.