Masked deep networks based on self-supervised learning for folk art image recognition and optimization of digital strategies for intangible cultural heritage preservation

Folk art is an essential component of traditional culture. Currently, the classification of folk art images still relies heavily on manual efforts, which are inefficient and costly, making it difficult to meet the demands of large-scale digitalization. With the advancement in the field of AI technologies, deep learning-based image classification models offer an effective strategy for enhancing the performance and reliability of folk art image classification, and have thus become a focal point of research. However, the lack of sufficient training samples remains a major constraint on model performance. In addition, the generally low quality of folk art images often interferes with how well the model can learn stable and representative features. To address these challenges, this paper introduces a novel model named FolkMaskNet, a masked deep network based on self-supervised learning, designed to enhance both recognition accuracy and feature learning robustness for folk art image classification. The model integrates reconstruction and masking strategies during training, effectively mitigating issues related to limited training data and image noise interference. Comprehensive testing on actual folk art datasets confirms that FolkMaskNet achieves better performance than existing methods. FolkMaskNet improves the current methods of folk art image recognition and provides technical support to support the digital safeguarding of non-material cultural traditions. It contributes to more efficient digital documentation of heritage resources and promotes the transmission and utilization of folk art in the digital era.