Deep reinforcement learning-supported innovative design method for intangible cultural heritage creative products

The living signs, capabilities, and traditions that reflect human creativity and individuality are represented by intangible cultural heritage (ICH). To address the challenges in the maintenance and development of ICH in the digital age, the research develops a deep reinforcement learning-supported innovative design method for ICH creative products. In the first place, the intangible cultural heritage design data, including images, symbols, and motifs, are digitalized to carry out image normalization and resizing in the development of the multimodal feature library, which comprises convolutional neural networks (CNNs). Data augmentation increases ICH sample diversity using image transformations. It improves model robustness and reduces overfitting while preserving cultural features. An adaptive Tasmanian devil tuned intelligent soft actor critic (ATD-ISAC) is used to extract the aesthetic and semantic properties of the heritage items and optimize the adaptive design. The DRL agent uses a reward system to evaluate the results iteratively and improve the quality of the design through a policy update. The system was developed for a region's textile and craft design datasets, and the results showed an innovation index, a cultural retention rate, and an average user satisfaction score. The ATD-ISAC model showed high and consistent results with 92.3% accuracy, 95.1% precision, 93.4% recall, 94.2% F1-score, and 97.6% AUC values. The experiments showed that the results are significantly better than traditional AI-based design methods. The results prove that deep reinforcement learning is a good approach to improve the balance between cultural and modern design.