Scholarlink

<p>Underground mine surveillance systems impose stringent requirements for low latency and high throughput, while abrupt depth discontinuities and strong light interference often lead to transmission-map misestimation and edge halos in dehazing results. To address these issues, we propose a structure-aware, guided, and computationally efficient dehazing algorithm. The method performs channel-wise color calibration using luminance statistics with contrast compensation. In the luminance domain, it integrates gradient cues with morphological constraints to construct low-confidence candidate regions for strong light interference, thereby suppressing overexposure-induced bias in atmospheric light estimation. Transmission estimation is further reformulated as a discriminability optimization between weak-texture regions and background areas; edge-aware weights and exponential-decay fusion are used to obtain a scale-adaptive transmission map, which is then refined via highlight-mask-guided filtering. Experiments on a self-built underground dataset show that, relative to the average performance of competing methods, the proposed approach reduces FADE/NIQE/BRISQUE by 46.4%/12.8%/29.3% and improves <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\sigma \)</EquationSource> <EquationSource Format="MATHML"><math> <mi>σ</mi> </math></EquationSource> </InlineEquation>/H/AG by 27.4%/11.8%/60.6%, respectively. At a resolution of 600<InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(\times \)</EquationSource> <EquationSource Format="MATHML"><math> <mo>×</mo> </math></EquationSource> </InlineEquation>400, the method processes a single frame in 0.011 s, corresponding to 91 FPS. The core computations are dominated by regular local sliding-window and morphological operations, yielding near-linear time complexity and high data parallelism. Consequently, the proposed algorithm is amenable to parallel acceleration for multi-stream online processing, providing a practical reference and implementation insight for the efficient use of HPC resources in underground intelligent-vision preprocessing.</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

An efficient structure-aware image dehazing algorithm for underground mining scenes

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Feng Tian,
  • Ruifen Zhang,
  • Liu Xiaopei

摘要

Underground mine surveillance systems impose stringent requirements for low latency and high throughput, while abrupt depth discontinuities and strong light interference often lead to transmission-map misestimation and edge halos in dehazing results. To address these issues, we propose a structure-aware, guided, and computationally efficient dehazing algorithm. The method performs channel-wise color calibration using luminance statistics with contrast compensation. In the luminance domain, it integrates gradient cues with morphological constraints to construct low-confidence candidate regions for strong light interference, thereby suppressing overexposure-induced bias in atmospheric light estimation. Transmission estimation is further reformulated as a discriminability optimization between weak-texture regions and background areas; edge-aware weights and exponential-decay fusion are used to obtain a scale-adaptive transmission map, which is then refined via highlight-mask-guided filtering. Experiments on a self-built underground dataset show that, relative to the average performance of competing methods, the proposed approach reduces FADE/NIQE/BRISQUE by 46.4%/12.8%/29.3% and improves \(\sigma \) σ /H/AG by 27.4%/11.8%/60.6%, respectively. At a resolution of 600 \(\times \) × 400, the method processes a single frame in 0.011 s, corresponding to 91 FPS. The core computations are dominated by regular local sliding-window and morphological operations, yielding near-linear time complexity and high data parallelism. Consequently, the proposed algorithm is amenable to parallel acceleration for multi-stream online processing, providing a practical reference and implementation insight for the efficient use of HPC resources in underground intelligent-vision preprocessing.