针对煤矿井下采集图像可见度低、曝光不足和细节模糊的问题，提出了一种基于光照约束的煤矿井下低光照图像增强算法，该算法结构包括光照约束模块（ICM）、光照分解模块（IDM）和光照增强模块（IEM）。ICM通过捕捉图像的整体光分布，生成灰度注意力权图，减少光照信息的干扰；IDM将图像分解为光照分量和反射分量；IEM通过U-Net网络结构对光照分量进行增强，最后将增强后的光照分量与灰度注意力权图和反射分量结合，得到增强图像。ICM和IDM引入了高效通道注意力模块（ECA），分别用以调控光照分布和增强算法对光照分量和反射分量的特征捕捉能力。选取4组不同场景，使用TBEFN、RUAS、MBLLEN、KinD和Retinex-Net算法进行对比分析试验。结果表明：本文算法在视觉信息保真度（VIF）、结构相似性（SSIM）、峰值信噪比（PSNR）3项指标的均值分别达到0.58、0.61、16.58，均超过了其他对比算法，分别较原模型提升了约23.40%、16.07%、20.45%，图像增强效果最优。

To address the issues of low visibility, insufficient exposure, and blurred details in images collected in coal mine environments, this research introduces a low light image enhancement algorithm, which is grounded in illumination constraints. The algorithm structure consists of three main modules: the illumination constraint module (ICM), the illumination decomposition module (IDM), and the illumination enhancement module (IEM). The ICM captures the overall light distribution of the image, creates a grayscale attention map to minimize illumination interference, and the IDM decomposes the image into illumination and reflection components. The IEM uses a U-Net network structure to enhance the illumination component. The enhanced illumination component is then combined with the grayscale attention map and reflection component to produce the enhanced image. Both ICM and IDM incorporate an efficient channel attention module (ECA), which regulates light distribution and enhances the feature capture capability for illumination and reflection components. Experiments were conducted in four different scenarios, comparing this algorithm against TBEFN, RUAS, MBLLEN, KinD, and Retinex-Net algorithms. Results indicate that this algorithm surpasses others in visual information fidelity (VIF), structural similarity index metric (SSIM), and peak signal to noise ratio (PSNR), achieving averages of 0.58, 0.61, and 16.58 respectively. Compared to the original model, it showed improvements of approximately 23.40%, 16.07%, and 20.45% in these metrics, demonstrating optimal image enhancement effectiveness.