Abstract: The gradual progress of coal mine intelligent construction has led to an increased use of intelligent video surveillance systems in underground coal mines. However, the images collected by these systems are often affected by low brightness, uneven illumination, information loss, blurred details and other issues caused by dust, water mist and light sources. This has a detrimental effect on the subsequent image analysis and intelligent decision-making, as well as on the overall visual effect of the video surveillance in underground coal mines. It is therefore of great significance to study image enhancement methods in underground coal mines. In order to address the issues of low local brightness and the loss of detail features in underground images under non-uniform illumination, an improved Retinex-Net underground image enhancement algorithm based on adaptive estimation has been proposed. A decomposition network is designed to separate the illumination component and reflection component of the image. In the reflection component processing, an attention module CBAM (Convolutional Block Attention Module) is introduced to enhance the details and contrast of the image, thereby improving clarity. Furthermore, a gradual light estimation network is constructed in the light optimisation process. The illumination estimation network is constructed in a process that gradually optimises the estimation of the illumination component through the cascade of multiple network layers. A self-calibration module is introduced to automatically adjust the estimated value of the illumination component to make it closer to the real illumination conditions. Finally, the optimised illuminance component and the reflection component are combined to obtain the enhanced downhole image. The improved algorithm, constructed on the basis of a self-constructed downhole image dataset, has been demonstrated to enhance the average gradient, peak signal-to-noise ratio, structural similarity, and information entropy by 25%, 17%, 24%, and 8%, respectively.