高质量矿井影像为矿山安全生产提供保障，也有利于提高后续图像分析技术的性能。矿井影像受低照度环境的影响，易出现亮度低，照度不均，颜色失真，细节信息丢失严重等问题。针对上述问题，提出一种基于Transformer和自适应特征融合的矿井低照度图像亮度提升和细节增强方法。基于生成对抗思想搭建生成对抗式主体模型框架,使用目标图像域而非单一参考图像驱动判别器监督生成器的训练，实现对低照度图像的充分增强；基于特征表示学习理论搭建特征编码器，将图像解耦为亮度分量和反射分量，避免图像增强过程中亮度与颜色特征相互影响从而导致颜色失真问题；设计CEM-Transformer Encoder通过捕获全局上下文关系和提取局部区域特征，能够充分提升整体图像亮度并消除局部区域照度不均；在反射分量增强过程中，使用结合CEM-Cross-Transformer Encoder的跳跃连接将低级特征与深层网络处特征进行自适应融合，能够有效避免细节特征丢失，并在编码网络中添加ECA-Net，提高浅层网络的特征提取效率。制作矿井低照度图像数据集为矿井低照度图像增强任务提供数据资源。试验显示，在矿井低照度图像数据集和公共数据集中，与5种先进的低照度图像增强算法相比，该算法增强图像的质量指标PSNR、SSIM、VIF平均提高了16.564%，10.998%，16.226%和14.438%，10.888%，14.948%，证明该算法能够有效提升整体图像亮度，消除照度不均，避免颜色失真和细节丢失，实现矿井低照度图像增强。

High quality mine images can provide guarantee for mine safety production, and improve the performance of subsequent image analysis technologies. Affected by low illuminance environment, mine images suffer low brightness, uneven brightness, color distortion, and serious loss of details. Aiming at the above problems, an illuminance improvement and details enhancement method on coal mine low-light images based on Transformer and adaptive feature fusion was proposed to enhance the brightness and detail of mine low illuminance images. Based on the idea of generative confrontation, a framework of generative adversary agent model was built, and the target image domain was used instead of a single reference image to drive discriminator to supervise the training of generator, so as to achieve adaptive enhancement of low illuminance images; The feature encoder was built based on the feature representation learning theory to decouple the image into illuminance component and reflection component, the method can avoid the interaction between illuminance and color features during image enhancement to solve the color distortion; the CEM-Transformer Encoder was designed to enhance the brightness component, the method can improve the overall image brightness and eliminate the local area brightness unevenness, by capturing the global context and extracting the local area features; In the process of reflection component enhancement, the skip connection combined with CEM-Cross-Transformer Encoder was used to adaptively fuse low-level features with features at the deep CNN layers, which can effectively avoid the loss of detailed features, and ECA-Net was added to the encode network to improve the feature extraction efficiency of the shallow CNN layers. The low illuminance mine image dataset was produced to provide data resources for the low illuminance mine image enhancement task. The experiments show that, compared with five advanced low illuminance image enhancement algorithms, the quality indicators PSNR, SSIM and VIF of the images enhanced by the algorithm are improved by 16.564%, 10.998%, 16.226% and 14.438%, 10.888% and 14.948% on average on the low illuminance mine image dataset and the public dataset. And the algorithm also perform well in subjective visual evaluation. The above results prove that the algorithm can effectively improve the overall image brightness and eliminate the uneven brightness, thus to achieve mine low illuminace image enhancement.