To enhance the efficiency and effectiveness of cultural heritage scene rendering, enabling dynamic interaction and immersive experiences for cultural inheritance. The study improves the grey wolf optimization algorithm, integrates it into deep learning oversampling training, and optimizes rendering based on color attribute theory. Performance is evaluated through function testing and image reconstruction, with examples from Dunhuang murals and the Forbidden City. The improved algorithm achieved fast convergence (50 iterations for uni-modal, 80 for multi-modal functions) and high robustness (98.6% success rate, 0.35 s runtime). Rendering optimization increased contrast and color saturation significantly, enhancing visual appeal and cultural conveyance. This research innovatively combines improved optimization algorithms with deep learning and color theory, offering an efficient technical solution for digital cultural heritage preservation and interactive display, promoting cultural diversity and heritage appreciation.