Digital Restoration and Cultural and Creative Application of Ethnic Embroidery Textile Patterns Using U-Net and Texture Fusion

Ethnic embroidery textile patterns constitute important carriers of cultural heritage; however, long-term preservation often results in fading, structural damage, and texture loss, limiting their artistic value and digital utilization. To address these challenges, this study proposes a digital restoration framework that integrates an improved U-Net convolutional neural network with adaptive texture fusion for high-fidelity embroidery pattern reconstruction. The enhanced U-Net architecture, incorporating residual connections, attention-guided skip pathways, and multi-scale feature extraction, is employed to accurately identify defective regions and recover structural information, while a PatchMatch-based texture migration strategy and VGG-guided style consistency optimization are introduced to achieve seamless texture completion and visual coherence. Furthermore, the restored patterns are transformed into reusable digital assets for cultural and creative product design, promoting the sustainable preservation and value regeneration of ethnic embroidery resources. Experimental results demonstrate that the proposed method achieves an IoU of 0.89 and a Dice coefficient of 0.88, while attaining PSNR values of 33.5–34.4 dB and superior texture consistency and visual aesthetics compared with conventional approaches. From the perspective of intelligent digital heritage systems, the proposed framework also provides technical support for multimodal information acquisition, electromagnetic-enabled sensing platforms, and wireless visualization applications, offering potential references for future integration with antenna-assisted imaging and electromagnetic data transmission technologies in smart cultural preservation environments.