Scholarlink

<p>Scanning electron microscopy (SEM) is widely employed in diverse fields for high-resolution imaging. However, a trade-off exists between image quality and resolution, with higher resolutions often entailing longer scanning times and increased operational complexity. To address this challenge, we propose a deep channel-spatial attention network (DCSAN) for SEM image super-resolution. DCSAN integrates channel attention and spatial attention mechanisms to efficiently capture high-frequency features, leveraging channel-wise and spatial contextual information to enhance SEM image resolution, particularly at high magnifications. Experimental results demonstrate that DCSAN outperforms existing methods in terms of image clarity and fidelity, achieving a PSNR of 31.01 dB and an SSIM of 0.7849 at a scaling factor of x4. Here, we show that our approach not only improves the visual quality of SEM images but also has the potential to advance precision instrumentation and measurement systems. This work presents a novel deep learning-based method for enhancing SEM image analysis in various scientific and industrial applications. The code is available at <a href="https://github.com/sxdyyds/DCSAN">https://github.com/sxdyyds/DCSAN</a>.</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

Deep channel-spatial attention networks for enhancing super-resolution of high-magnification SEM images

  • 期刊论文

  • 发表日期：

    2025年5月20日

查看全文
  • Yizhi Cong,
  • Haoran Zhu,
  • Longyu Guo,
  • Wei Zhang,
  • Zhongqing Zhang

摘要

Scanning electron microscopy (SEM) is widely employed in diverse fields for high-resolution imaging. However, a trade-off exists between image quality and resolution, with higher resolutions often entailing longer scanning times and increased operational complexity. To address this challenge, we propose a deep channel-spatial attention network (DCSAN) for SEM image super-resolution. DCSAN integrates channel attention and spatial attention mechanisms to efficiently capture high-frequency features, leveraging channel-wise and spatial contextual information to enhance SEM image resolution, particularly at high magnifications. Experimental results demonstrate that DCSAN outperforms existing methods in terms of image clarity and fidelity, achieving a PSNR of 31.01 dB and an SSIM of 0.7849 at a scaling factor of x4. Here, we show that our approach not only improves the visual quality of SEM images but also has the potential to advance precision instrumentation and measurement systems. This work presents a novel deep learning-based method for enhancing SEM image analysis in various scientific and industrial applications. The code is available at https://github.com/sxdyyds/DCSAN.