Adaptive channel threshold segmentation-based low-light image enhancement method for fully mechanized coal mining faces

The safety monitoring of workers in fully mechanized coal mining faces (FMCMF) under the dust, water mist, and uneven lighting environment is important for the effective operation of coal mines, so it is essential to improve the image quality under the low-light environment for FMCMF. To solve the mentioned problem, this paper proposes a low-light enhancement method based on channel threshold segmentation of the physical model (LEM-CTSPM) by decreasing the brightness of over-exposure areas, increasing the brightness of the dark regions, and improving the image quality. The segmentation thresholds are refined to three adaptive values according to the relative differences between the darkest and the brightest regions of the input image by adaptive thresholds K1 and K2. A low-light image enhancement method is developed by combining the Conv-based dark channel thresholds and adaptive atmospheric light intensity estimation to distinguish bright and dark regions. An image quality enhancement method is established by integrating the haze formation model theory and saturation optimization in HSV. The performance of LEM-CTSPM is verified by three regular low-light image datasets and the FMCMF low-light image dataset, and the results show that LEM-CTSPM achieves the best performance for low-light image enhancement. Moreover, 11 low-light image enhancement methods are introduced into the YoLov8 to verify the performance of the proposed low-light enhancement method under the FMCMF workers detection scenario, and the results show that LEM-CTSPM can improve the detection accuracy by 6.93%. Therefore, LEM-CTSPM is suitable for the FMCMF environment to enhance the quality of low-light images.