Scholarlink

<p>This study introduces HoneyGAN, a new Generative Adversarial Network framework designed for improved image generation. It uses a three-phase method that is inspired by the natural properties of honey: preservation (fuzzy attention mechanism to retain global dependencies and structural information), refinement (contextual texture enhancement for fine-grained details), and enrichment (enhanced multiscale consistency to ensure structural coherence across resolutions). We introduce an adaptive metric approach for dynamic weight control for the generator and discriminator, enhancing training stability with a gradient penalty from Wasserstein GAN to enforce Lipschitz continuity. We assessed our model on images from the Mammographic Image Analysis Society (MIAS) dataset. Our results show that the proposed architecture generates high-quality images with well-defined features and realistic textures. Compared to baseline GANs, the suggested model, HoneyGAN, outperformed them all, achieving a maximum accuracy of 99.23%, precision of 99.50%, recall of 99.00%, and an F1-score of 99.00%.</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

Honeygan: perceptually enhanced generative adversarial network with fuzzy attention for breast cancer detection on mammography images

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Nirbhay Kumar Kashyap,
  • Anil Kumar Sagar

摘要

This study introduces HoneyGAN, a new Generative Adversarial Network framework designed for improved image generation. It uses a three-phase method that is inspired by the natural properties of honey: preservation (fuzzy attention mechanism to retain global dependencies and structural information), refinement (contextual texture enhancement for fine-grained details), and enrichment (enhanced multiscale consistency to ensure structural coherence across resolutions). We introduce an adaptive metric approach for dynamic weight control for the generator and discriminator, enhancing training stability with a gradient penalty from Wasserstein GAN to enforce Lipschitz continuity. We assessed our model on images from the Mammographic Image Analysis Society (MIAS) dataset. Our results show that the proposed architecture generates high-quality images with well-defined features and realistic textures. Compared to baseline GANs, the suggested model, HoneyGAN, outperformed them all, achieving a maximum accuracy of 99.23%, precision of 99.50%, recall of 99.00%, and an F1-score of 99.00%.