GAN-Based Model for Spatiotemporal and Detail-Preserving Digitization of Rural Traditional Culture

Abstract

In the context of the severe crisis of loss of traditional rural culture, this study focuses on the use of innovative models to achieve its digital protection. Through carefully designed experiments, the "Rural Heritage Image Dataset" (RHI-D) and the "Folk Activity Video Dataset" (FAV-D) were selected to compare the proposed model with the "Traditional 3D Reconstruction Model" (T3DRM), the "Model Based on Conventional Video Recording" (CVRM) and the "Simple Generative Adversarial Network Model" (SGM). The proposed model extends traditional GANs by incorporating an attention mechanism to enhance critical visual features and an LSTM-based temporal modeling component to preserve motion coherence in folk activity videos. These architectural improvements enable more accurate, detail-preserving, and temporally consistent digital representations of rural cultural elements. The experiment was evaluated using baseline indicators such as the structural similarity index (SSIM) and the video quality index (VQM). The results show that in the digitization of ancient building images, the average SSIM of the proposed model reached 0.85, the wood carving detail restoration was 88%, the color similarity was 92%, and the structural component integrity rate was 95%, exceeding the comparison model. In the digitization of folk activity videos, the average VQM of the proposed model was 0.82, the action coherence score was 85, the sound clarity score was 8.0, and the scene switching smoothness score was 8.8. These results demonstrate the proposed model’s clear advantages in digitally preserving rural traditional culture and its potential value for supporting cultural diversity and heritage.