A generative adversarial network with enhanced feature extraction for intelligent interface generation

Automatic intelligent interface generation is a critical task in human-computer interaction, yet traditional generative models suffer from issues such as poor semantic consistency, insufficient component-level structural accuracy, and a trade-off between generation quality and computational efficiency. To address these challenges, this paper proposes DCS-GAN, a novel generative adversarial network tailored for intelligent interface generation. The model integrates three core modules with complementary functions: a ConvNeXt-Tiny-based feature extraction module for hierarchical interface representation learning, a DG parallel generation module to balance global layout optimization and component-level structural preservation, and an SE-AdaIN perceptual calibration module to enhance the consistency between generated interfaces and real samples in terms of perception and semantics. Comprehensive evaluations on the RICO and WebUI datasets demonstrate that DCS-GAN outperforms mainstream comparative models in generation quality, noise robustness, and computational efficiency. The model exhibits strong generalization ability across datasets and maintains stable performance under noisy input conditions, while achieving a favorable balance between generation efficiency and resource consumption. This work provides a feasible solution for high-quality and efficient automatic intelligent interface generation, and offers practical reference for the application of generative models in interface design scenarios.