YOLO-WSR: A wavelet feature and structure guided diffusion model for detection and automated restoration of cultural heritage artifact images

In response to the complexity and diversity involved in damage detection and automated restoration of cultural heritage artifact images associated with intangible cultural heritage preservation, this study proposes a YOLO-based restoration framework that integrates wavelet features with structure-guided diffusion, termed YOLO-WSR. The proposed method enhances detection accuracy through wavelet feature upsampling and employs a structure-guided diffusion generation mechanism to achieve effective restoration of texture and structural details in damaged artifact images. In addition, a CLIP-IQA-based feedback optimization module is introduced to perform no-reference quality assessment on the restored outputs and iteratively refine the restoration results, thereby forming a closed-loop collaborative framework that integrates detection, restoration, and evaluation. Experiments conducted on multiple publicly available datasets demonstrate the effectiveness of YOLO-WSR. The results show that the proposed method consistently improves performance over existing approaches, achieving approximately 5–6% improvement in mAP@50 for damage detection, while obtaining enhanced PSNR, SSIM, LPIPS, and SRCC performance for restoration quality evaluation. Further ablation studies demonstrate the contribution of each component to the overall framework. The proposed method provides an effective and quantifiable technical solution for the digital preservation of cultural heritage artifact images, offering practical support for accurate damage identification, automated restoration, and intelligent management of visual cultural heritage resources.