Scholarlink

<p>Blank bands in electrical imaging logging constitute a major source of uncertainty in fracture identification and sedimentary structure interpretation. They frequently cause biased extraction of fracture parameters, introduce artificial discontinuities in sedimentary structure continuity, and hinder facies classification, thereby substantially reducing the accuracy of reservoir effectiveness assessment and geological model construction. Existing blank band filling methods suffer from prominent issues such as texture distortion, insufficient geological constraints, and pronounced artifacts in large-scale missing regions. To address this, this paper proposes the HR-FFC algorithm—combining Highway-Fast Fourier Convolution residual enhancement with relativistic adversarial learning—to efficiently fill electrical imaging blank bands. This algorithm employs two synergistic optimization strategies to achieve effective expansion of the receptive field and enhanced recovery of high-frequency details. 1) A parallel Highway-FFC residual branch is designed, dynamically fusing the original input with the output from the Fast Fourier Convolution (FFC) branch via the introduction of a gating coefficient o. This effectively strengthens gradient backpropagation and significantly suppresses artifacts generated during the inpainting of large-scale missing regions. 2) The Relativistic Hinge discriminator loss function enables adversarial training between the generator and discriminator around the mean decision boundary of “image authenticity,” thereby enhancing the visual realism and naturalness of reconstructed textures. Experimental results demonstrate that the HR-FFC algorithm significantly outperforms classical algorithms in terms of FID (55.98), compared to OpenCV (251.18), Criminisi (124.97), and Filtersim-SLB (68.03). The restored images exhibit excellent quality with continuous textural details and well-preserved structural continuity, offering a reliable basis for high-precision geophysical analysis and quantitative characterisation of complex reservoirs.</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

Efficient Blank Filling Method for Electrical Imaging Based on the HR-FFC Algorithm

  • 期刊论文

  • 发表日期：

    2026年4月7日

查看全文
  • Jun-lei Su,
  • Xu Dong,
  • Ze-yi Jia,
  • Yu Zeng,
  • Xue-ying Shi,
  • Jin-ku Dai

摘要

Blank bands in electrical imaging logging constitute a major source of uncertainty in fracture identification and sedimentary structure interpretation. They frequently cause biased extraction of fracture parameters, introduce artificial discontinuities in sedimentary structure continuity, and hinder facies classification, thereby substantially reducing the accuracy of reservoir effectiveness assessment and geological model construction. Existing blank band filling methods suffer from prominent issues such as texture distortion, insufficient geological constraints, and pronounced artifacts in large-scale missing regions. To address this, this paper proposes the HR-FFC algorithm—combining Highway-Fast Fourier Convolution residual enhancement with relativistic adversarial learning—to efficiently fill electrical imaging blank bands. This algorithm employs two synergistic optimization strategies to achieve effective expansion of the receptive field and enhanced recovery of high-frequency details. 1) A parallel Highway-FFC residual branch is designed, dynamically fusing the original input with the output from the Fast Fourier Convolution (FFC) branch via the introduction of a gating coefficient o. This effectively strengthens gradient backpropagation and significantly suppresses artifacts generated during the inpainting of large-scale missing regions. 2) The Relativistic Hinge discriminator loss function enables adversarial training between the generator and discriminator around the mean decision boundary of “image authenticity,” thereby enhancing the visual realism and naturalness of reconstructed textures. Experimental results demonstrate that the HR-FFC algorithm significantly outperforms classical algorithms in terms of FID (55.98), compared to OpenCV (251.18), Criminisi (124.97), and Filtersim-SLB (68.03). The restored images exhibit excellent quality with continuous textural details and well-preserved structural continuity, offering a reliable basis for high-precision geophysical analysis and quantitative characterisation of complex reservoirs.