Scholarlink

<p>This study proposes a sustainable and resource-efficient Artificial Intelligence Generated Content (AIGC) art design framework based on an Enhanced Generative Adversarial Network (GAN) integrated with Particle Swarm Optimization (PSO) for automated hyperparameter tuning. The proposed framework is designed to simultaneously enhance artistic image quality and computational efficiency through a multi-objective fitness function that jointly optimizes perceptual fidelity, training time, and memory consumption. Unlike conventional GAN models that rely on manual hyperparameter selection and prioritize only visual realism, the PSO-enhanced approach stabilizes adversarial training and minimizes redundant computational overhead. Experimental evaluation conducted on the WikiArt dataset demonstrates that the proposed PSO-GAN achieves superior reconstruction performance, with a PSNR of 19.5 dB and SSIM of 0.65 compared to 16.4 dB and 0.55 obtained by the baseline GAN, while reducing MSE from 1990 to 1520. Furthermore, the framework significantly decreases energy consumption from 15.4 kWh to 9.8 kWh, reduces memory usage from 4.2 GB to 2.8 GB, and shortens training time from 12.5&#xa0;h to 8.2&#xa0;h. Visual convergence analysis and heatmap-based error evaluation further confirm improved structural preservation and stable learning dynamics. Overall, the results validate that integrating PSO-based optimization within an enhanced GAN architecture enables high-quality artistic image generation while promoting sustainable and performance-optimized computing systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A sustainable and resource-efficient AIGC art design framework using an enhanced GAN for performance-optimized computing systems

  • 期刊论文

  • 发表日期：

    2026年5月27日

查看全文
  • Yunan Wu,
  • Haitao Zhang

摘要

This study proposes a sustainable and resource-efficient Artificial Intelligence Generated Content (AIGC) art design framework based on an Enhanced Generative Adversarial Network (GAN) integrated with Particle Swarm Optimization (PSO) for automated hyperparameter tuning. The proposed framework is designed to simultaneously enhance artistic image quality and computational efficiency through a multi-objective fitness function that jointly optimizes perceptual fidelity, training time, and memory consumption. Unlike conventional GAN models that rely on manual hyperparameter selection and prioritize only visual realism, the PSO-enhanced approach stabilizes adversarial training and minimizes redundant computational overhead. Experimental evaluation conducted on the WikiArt dataset demonstrates that the proposed PSO-GAN achieves superior reconstruction performance, with a PSNR of 19.5 dB and SSIM of 0.65 compared to 16.4 dB and 0.55 obtained by the baseline GAN, while reducing MSE from 1990 to 1520. Furthermore, the framework significantly decreases energy consumption from 15.4 kWh to 9.8 kWh, reduces memory usage from 4.2 GB to 2.8 GB, and shortens training time from 12.5 h to 8.2 h. Visual convergence analysis and heatmap-based error evaluation further confirm improved structural preservation and stable learning dynamics. Overall, the results validate that integrating PSO-based optimization within an enhanced GAN architecture enables high-quality artistic image generation while promoting sustainable and performance-optimized computing systems.