Scholarlink

<p>The underground coal mine environment, characterised by high dust concentrations and substantial water mist, often results in image degradation, manifesting as detail loss, severe artefacts, and reduced contrast. To address these challenges, this study proposes an Efficient Multi-scale Transformation Cycle Generative Adversarial Network (EMT-CycleGAN), designed for image enhancement in underground coal mines. First, to overcome the difficulty of acquiring paired underground image data, the EMT-CycleGAN framework is used as the training framework. Second, a self-attention mechanism is incorporated into the encoder to enhance the encoder architecture, thereby improving the model’s capacity to capture global features and facilitating effective restoration of image details. Furthermore, the EMA attention mechanism is integrated into residual blocks to design the Efficient Multi-scale Transformation (EMT) module, which improves high-quality information representation, strengthens global feature extraction, and preserves fine-grained details. In addition, the pixel shuffle operation is adopted to replace deconvolution operations, effectively mitigating artefacts and improving detail reconstruction, thereby enhancing overall visual quality. Finally, a multiscale discriminator is introduced to strengthen the discriminator’s ability to capture both global and local features, thereby effectively mitigating the issue of reduced contrast. Experimental results demonstrate that the proposed method achieves significant improvements in PSNR, SSIM, Entropy, EPI, FSIM, NIQE and LPIPS, achieving average improvements of 37.2%, 32.8%, 4.1%, 13.34%, 27.6%, 17.44% and 11.62%, respectively, compared with other enhancement algorithms. Moreover, experiments using the YOLOv8 object recognition framework further corroborate the effectiveness of the proposed approach in improving recognition accuracy.</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

Coal mine underground image enhancement method based on efficient multi-scale transformation cycle generative adversarial network

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • Yuanbin Wang,
  • Zhaoxu Yan

摘要

The underground coal mine environment, characterised by high dust concentrations and substantial water mist, often results in image degradation, manifesting as detail loss, severe artefacts, and reduced contrast. To address these challenges, this study proposes an Efficient Multi-scale Transformation Cycle Generative Adversarial Network (EMT-CycleGAN), designed for image enhancement in underground coal mines. First, to overcome the difficulty of acquiring paired underground image data, the EMT-CycleGAN framework is used as the training framework. Second, a self-attention mechanism is incorporated into the encoder to enhance the encoder architecture, thereby improving the model’s capacity to capture global features and facilitating effective restoration of image details. Furthermore, the EMA attention mechanism is integrated into residual blocks to design the Efficient Multi-scale Transformation (EMT) module, which improves high-quality information representation, strengthens global feature extraction, and preserves fine-grained details. In addition, the pixel shuffle operation is adopted to replace deconvolution operations, effectively mitigating artefacts and improving detail reconstruction, thereby enhancing overall visual quality. Finally, a multiscale discriminator is introduced to strengthen the discriminator’s ability to capture both global and local features, thereby effectively mitigating the issue of reduced contrast. Experimental results demonstrate that the proposed method achieves significant improvements in PSNR, SSIM, Entropy, EPI, FSIM, NIQE and LPIPS, achieving average improvements of 37.2%, 32.8%, 4.1%, 13.34%, 27.6%, 17.44% and 11.62%, respectively, compared with other enhancement algorithms. Moreover, experiments using the YOLOv8 object recognition framework further corroborate the effectiveness of the proposed approach in improving recognition accuracy.