Single Image Dehazing for Coal Mines via Bright-Dark Channel Fusion
Xinliang Wang, Lei Zhu, Yan Huo, Cuifang Li · IEEE Access · 2026
Images in coal mine often suffer from marked deterioration due to complicated environments. Traditional dehazing methods, often grounded in prior theories and atmospheric scattering models, exhibit limited effectiveness. While more advanced, deep learning-based approaches typically require extensive paired datasets and high hardware configurations. To address these limitations, this paper proposes a single image dehazing method for coal mines based on bright-dark channel fusion (BDCF). Utilizing reference-based image quality assessment, the optimal dual-channel fusion strategy was determined through comparative analysis on a limited dataset of hazy and clear image pairs. The hue histogram is employed to automatically identify the color characteristics of an input image. Color images are processed using the dark channel prior, whereas low-light grayscale images are enhanced with the bright channel prior. Ambient light is estimated separately for each channel, and the estimation performance is improved by adding color variance compensation. Transmission estimation is derived from the dark channel prior, and a weighted guided filter is designed via dual-window mode and adaptive regularization weights to refine image details. Finally, the restored and enhanced results are obtained through the hazy image degradation model. To evaluate the efficacy of the proposed method, some experiments were conducted based on the self-constructed coal mine hazy image dataset and some public benchmarks. The proposed method was compared with other state-of-the-art methods. Experimental results demonstrate that our method effectively removes haze interference, enhances both image features and edge details, and exhibits favorable adaptability, stability, and computational efficiency across different situations.