Research on Pattern Style Transfer and Intelligent Generation Based on Multimodal Feature Fusion
Gaosheng Luo · Procedia Computer Science · 2026
The digital preservation and transmission of intangible cultural heritage face severe challenges such as insufficient stylistic representation and loss of cultural connotations. Taking grass cloth embroidery as a case study, this research constructs an intelligent pattern generation framework based on multimodal feature fusion. It proposes a three-dimensional modeling system integrating visual, craftsmanship, and cultural dimensions, employing deep learning techniques to achieve collaborative modeling and unified encoding of heterogeneous features. Building upon this foundation, a culturally-semantically guided style transfer algorithm and a multi-constraint intelligent generation model were designed to balance modern adaptation with the preservation of traditional cultural symbols. Furthermore, a multidimensional evaluation mechanism integrating aesthetic, craftsmanship, and cultural attributes was established, forming a closed-loop technical framework: “Feature Modeling → Style Transfer → Intelligent Generation → Quality Evaluation → Iterative Optimization.”