Scholarlink

<p>In the age of advanced digital imaging, removing raindrops from images has become a crucial and practical challenge. Traditional methods often fall short and require significant computational resources. To address this issue, we have developed an improved neural network for image de-raining, leveraging the CBAM. This method employs CBAM network to accurately pinpoint key features from input images affected by rain, generating a collection of vital characteristics. By combining L1 loss, perceptual loss, and multi-scale loss functions, Enhanced DeRainNet adopts an encoder decoder architecture to balance pixel level accuracy and perceptual quality of images. At the same time, hybrid automatically mixed precision (AMP) is used to improve training efficiency and reduce memory consumption. We assessed our model’s performance using the DID-MDN dataset and found that it significantly outperformed the baseline, with improvements of 1.5% in peak signal-to-noise ratio (PSNR) and 0.3% in structural similarity (SSIM) scores, respectively. In terms of gradient magnitude similarity bias (GMSD), conscious learning of image block similarity (LPIPS), and reference-based supervised learning (RSR), it also greatly surpasses the basic model. The ablation study also validated the importance of channel attention and spatial attention mechanisms in improving the performance of CBAM models. Our experiments demonstrate that our algorithm excels at eliminating rain from individual images while preserving their inherent details.</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

Enhanced rain removal network with convolutional block attention module (CBAM): a novel approach to image de-raining

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年3月20日

查看全文
  • Ping Jiang,
  • Junzi Zhang,
  • Jiejie Chen

摘要

In the age of advanced digital imaging, removing raindrops from images has become a crucial and practical challenge. Traditional methods often fall short and require significant computational resources. To address this issue, we have developed an improved neural network for image de-raining, leveraging the CBAM. This method employs CBAM network to accurately pinpoint key features from input images affected by rain, generating a collection of vital characteristics. By combining L1 loss, perceptual loss, and multi-scale loss functions, Enhanced DeRainNet adopts an encoder decoder architecture to balance pixel level accuracy and perceptual quality of images. At the same time, hybrid automatically mixed precision (AMP) is used to improve training efficiency and reduce memory consumption. We assessed our model’s performance using the DID-MDN dataset and found that it significantly outperformed the baseline, with improvements of 1.5% in peak signal-to-noise ratio (PSNR) and 0.3% in structural similarity (SSIM) scores, respectively. In terms of gradient magnitude similarity bias (GMSD), conscious learning of image block similarity (LPIPS), and reference-based supervised learning (RSR), it also greatly surpasses the basic model. The ablation study also validated the importance of channel attention and spatial attention mechanisms in improving the performance of CBAM models. Our experiments demonstrate that our algorithm excels at eliminating rain from individual images while preserving their inherent details.