A hybrid zero-reference and dehazing network for joint low-light underground image enhancement

Area of Science:

• Computer Vision
• Image Processing
• Mining Engineering

Background:

• Underground mine vision sensors face challenges like low illumination, high dust, and noise.
• Existing image enhancement methods often fail in real-world underground conditions due to reliance on synthetic data.

Purpose of the Study:

• To develop a novel image enhancement approach for real underground mine environments.
• To improve the quality of images captured by underground vision sensors for better supervision and object detection.

Main Methods:

• Proposed the Zero-Reference Depth Curve Estimation-Dehazing Network (Z-DCE-Net) combining low-light enhancement and dehazing.
• Incorporated higher-order loss curves and a new loss function into DCE-Net for low-light enhancement.
• Utilized AOD-Net for post-processing to address color distortion and blur, enhancing clarity.

Main Results:

• The Z-DCE-Net method significantly enhances the visual quality of underground mine images.
• Enhanced images demonstrated improved performance in subsequent object detection tasks compared to original images.

Conclusions:

• The proposed Z-DCE-Net effectively addresses the challenges of low illumination and fog distortion in underground mine images.
• The method provides visually superior enhanced images and improves the accuracy of object detection, benefiting mine supervision.