Abstract: Due to complex geological conditions and unevenly artificial lighting in underground coal mines, surveillance video images often exhibit non-uniform illumination, detail loss, and low contrast. Moreover, existing enhancement algorithms frequently introduce color distortion and halo artifacts during processing. To address these issues, a mine image enhancement algorithm based on multi-scale fast bilateral filtering and wavelet fusion is proposed. First, homomorphic filtering is applied to preliminarily enhance the image and convert it into HSV space, where the hue component remains unchanged. A multi-scale fast bilateral filter is then constructed to extract the illumination component from the brightness channel, and a dual gamma correction function is employed to enhance the illumination component. The reflection component, estimated according to Retinex theory, is further enhanced using a grayscale adjustment function and the Constrained Contrast Adaptive Histogram Equalization (CLAHE) algorithm. Illumination and reflection components are subsequently fused by wavelet transform to obtain the enhanced brightness channel. In addition, a saturation correction function is designed to improve the saturation component and enhance the overall color representation of the mine image. Finally, the enhanced brightness and saturation components are combined with the hue component and transformed back from HSV to RGB space. Experimental results demonstrate that, compared with BPDHE, CLAHE, NPE, SRIE, BIMEF, and PnPRetinex algorithms, the proposed method achieves respective improvements of 25.31%, 42.75%, 9.59%, 1.60%, and 41.26% in objective evaluation metrics including mean, average gradient, standard deviation, information entropy, and spatial frequency. The method effectively enhances illumination, details, and contrast of mine images while suppressing halo artifacts and color distortion. Moreover, when extracting illumination components, multi-scale fast bilateral filtering achieves an average speed improvement of 87.29% compared with the classical bilateral filter. When YOLOv8 is applied to the enhanced images of mine workers, an average detection accuracy of 90% is obtained, representing a 40% increase compared with the original images and significantly improving the accuracy of intelligent detection.