Scholarlink

<p>Compressed sensing (CS) technology plays an important role in the efficient wireless transmission of large volumes of high-resolution satellite images. However, the CS image reconstruction process is often affected by noise, resulting in degraded reconstruction accuracy. To address this issue, we propose a feature-space anti-noise compressed sensing image reconstruction network (RFSCS-Net), which eliminates the block effect in the feature space by applying a de-blocking mechanism. In addition, the jump residual connection improves feature transfer and gradient flow. Multi-scale denoising reduces noise at different levels, while a spatial attention mechanism further enhances noise suppression in critical regions. Experimental results show that the network achieves a peak signal-to-noise ratio (PSNR) of 39.64 dB when Gaussian noise with a mean of 0 and an SNR of 30 dB is added to the measurements at a 50% sampling rate.</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

Feature Space-Based Anti-Noise Compressed Sensing Image Reconstruction Network

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Jianhong Xiang,
  • Tianyi Song,
  • Wei Wang

摘要

Compressed sensing (CS) technology plays an important role in the efficient wireless transmission of large volumes of high-resolution satellite images. However, the CS image reconstruction process is often affected by noise, resulting in degraded reconstruction accuracy. To address this issue, we propose a feature-space anti-noise compressed sensing image reconstruction network (RFSCS-Net), which eliminates the block effect in the feature space by applying a de-blocking mechanism. In addition, the jump residual connection improves feature transfer and gradient flow. Multi-scale denoising reduces noise at different levels, while a spatial attention mechanism further enhances noise suppression in critical regions. Experimental results show that the network achieves a peak signal-to-noise ratio (PSNR) of 39.64 dB when Gaussian noise with a mean of 0 and an SNR of 30 dB is added to the measurements at a 50% sampling rate.