The study of high-performance generation methods for rural plan based on generative adversarial network

In China, traditional village layouts are dynamic, harmoniously integrated with the natural environment, and rich in unique cultural characteristics. However, rapidly constructed villages often lack professional design, resulting in overly simple layouts and causing the villages to lose their traditional characteristics. Artificial intelligence holds the potential to alleviate this specific challenge. This study employs CGAN to generate comprehensive village layouts based on archetypal traditional villages, while also exploring parameters and network architectures to enhance result quality. The research address on traditional villages in southwestern Hubei, refining generative factors, introducing image-based geographic scales, and employing machine vision to address data scarcity. The key findings of this study includes: 1) The research explores a class of AI-generated evaluation metrics suitable for village layout generation. 2) It confirms that the combination of the Unet_256 generator with the LSGAN architecture yields the best results in image generation. 3) It is observed that the optimal generation results are achieved when the equivalent geographic scale of the image is 150 m × 150 m. The study validates that GANs can be effectively applied in the village layout, producing layout results that incorporate traditional local experiences. This provides a novel approach to village layout.