Optimizing the Digital Design Process for Lolita-Style Clothing Patterns Using the StyleGAN Image Generation Algorithm
M. Ren, J.J. Wang, Gu Zheng · Advanced Electromagnetics · 2026
With the increasing adoption of electromagnetic sensing, wireless imaging, and intelligent information acquisition technologies in digital fashion design and virtual manufacturing, high-quality pattern generation has become an important component of efficient visual information processing. To address the low efficiency of manual creativity and the difficulty of preserving cultural characteristics in the digital design of Lolita-style clothing patterns, this paper proposes an intelligent pattern generation framework based on an improved StyleGAN. A multi-scale style control architecture with a 12-layer mapping network and a hierarchical attention mechanism is developed to achieve fine-grained control of representative elements such as lace, ruffles, and decorative textures. Furthermore, a cultural feature constraint module combining VGG-19 feature alignment and Gram matrix style loss is introduced to preserve semantic consistency with Lolita aesthetic norms, while an improved adaptive instance normalization mechanism enhances local feature representation through grouped channel control and learnable normalization strength. A latent-space optimization strategy and seamless integration with CLO 3D further support efficient digital design and virtual garment development. Experimental results demonstrate that the proposed method achieves an average aesthetic score of 8.7, compresses the design cycle by 92.7%, and significantly reduces manual intervention while maintaining stable robustness under different noise conditions. The proposed framework provides an efficient and reliable solution for intelligent fashion pattern generation and offers methodological support for electromagnetic-enabled visual information acquisition, digital design systems, and multidisciplinary engineering applications involving advanced image processing and virtual modeling.