Intangible cultural heritage paintings often suffer from fading, structural damage, and detail loss due to environmental and material aging, making high-fidelity digital reconstruction essential for cultural preservation.To address the limitations of red-green-blue imaging (RGB) and insufficient spectral-spatial modelling, this study proposes context-aware multi-spectral imaging network (CA-MSI-Net), a reconstruction method integrating multi-spectral imaging (MSI) with an improved U-Net architecture.Spatial and channel transformer modules are embedded to enhance long-range spatial dependencies and cross-band spectral interactions, while contextual modelling and multi-scale attention mechanisms strengthen texture perception and boundary restoration.Experiments on multispectral datasets demonstrate that CA-MSI-Net achieves superior reconstruction accuracy, with mean intersection over union (mIoU), dice similarity coefficient (DSC), and F1 score reaching 81.3%, 92.1%, and 0.94, respectively, outperforming UCTransNet and dynamic optimisation vision network (DovNet).The method shows robust performance across painting styles, lighting conditions, and spectral configurations.
Bin Chen (Thu,) studied this question.