Abstract: With the continuous advancement of intelligent mine construction,images play a crucial role in mine safety monitoring,equipment identification,and operation assistance. However,mine images often encounter complex environmental issues such as low illumination,uneven lighting,and noise interference,leading to blurred details and imbalanced brightness, which seriously affect the accuracy of subsequent image recognition and intelligent analysis. To address these problems,a mine image enhancement algorithm that integrates a multi-scale enhancement mechanism with the hue,saturation,and value (HSV) color space is proposed. This method is based on a deep enhancement network constructed by the Retinex theory. Firstly,the mine image is decomposed into two components:illumination and reflection. For the illumination component,a multi-scale convolutional network is designed to extract brightness information at different spatial scales,enhancing the global illumination modeling capability. For the reflection component,a bilateral filtering mechanism is introduced to suppress noise and preserve edge structures. Subsequently,the optimized illumination and reflection components are fused and reconstructed to form a preliminary enhanced image. Finally,in the HSV color space,the brightness channel of the preliminary enhanced image is separated,and an exposure adjustment and detail enhancement module is introduced to further achieve joint optimization of brightness compensation and texture restoration. Experimental results show that the proposed method achieves a peak signal-to-noise ratio of up to 28. 9 dB and a structural similarity index of up to 0. 87 on the public DIV2K dataset. On the self-made mine image dataset,the feature similarity index of the algorithm is increased to a maximum of 0. 902,and the universal image quality index reaches a maximum of 0. 847. Under different lighting conditions,this method demonstrates excellent detail recovery and brightness balancing capabilities,verifying its effectiveness in mine image enhancement. 

Key words: image enhancement,HSV,deep learning,Retinex,multi-scale features,Retinex-Net