In this study, a CycleGAN framework integrating traditional pattern gene network and multi-objective evolutionary computation MOEAs is proposed, aiming at realizing the intelligent style migration and innovative design of decorative patterns of Central Plains culture. The association rules and spatial structure features of pattern design genes are extracted by constructing a traditional pattern gene network model, and MOEAs are introduced to optimize the generator weight parameters, which are combined with the channel attention mechanism to enhance the detail capturing ability. The experiments use the PASCAL VOC 2020 public dataset and the self-constructed Central Plains culture tattoo dataset ZYWY for double-benchmark validation, and the results show that the model’s tattoo segmentation accuracy MIoU on PASCAL VOC 2020 reaches 86.67%, which is significantly better than that of DeepLab V3+ at 83.46%. The generated image quality FID=6.83, IS=5.34 with diversity are better than the comparison model, e.g., FID=17.92, IS=3.65 for DeepLab V3+.Ablation experiments show that removing the traditional texture gene network leads to an 8.46% decrease in MIoU and a 177% deterioration in FID, which verifies its central role in the constraints on the structure of the tattoos. The subjective evaluation showed that the application value rating P=3.52 and the innovation rating Cr=2.98 of the generated tattoos by the art design practitioners were significantly higher than that of the traditional AI generation method P=1.95 and Cr=2.74. The study provides a solution that combines both technological feasibility and artistic practicability for the digitization of traditional tattoos for inheritance and innovation.
