To address the issue of low detection accuracy caused by low illumination and occlusion in underground coal mines, this study proposes an innovative miner detection method. A large dataset encompassing complex environments, such as low-light conditions, partial strong light interference, and occlusion, was constructed. The Efficient Channel Attention (ECA) mechanism was integrated into the YOLOv11 model to enhance the model’s ability to focus on key features, thereby significantly improving detection accuracy. Additionally, a new weighted Complete Intersection over Union (CIoU) and adaptive confidence loss function were proposed to enhance the model’s robustness in low-light and occlusion scenarios. Experimental results demonstrate that the proposed method outperforms various improved algorithms and state-of-the-art detection models in both detection performance and robustness, providing important technical support and reference for coal miner safety assurance and intelligent mine management.

The images in underground coal mines suffer from low contrast and poor visibility. While diffusion model-based approaches for low-light image enhancement have shown considerable promise, these methods are often characterized by high resource consumption, slow processing speeds, and a tendency towards instability. To address these challenges, this paper presents a diffusion model that integrates Retinex and Wavelet transform (RW-DM). This methodology harnesses the generative capabilities of diffusion models to produce visually appealing outcomes in the enhancement of low-light images. This approach decomposes low-light images into reflectance and illumination components, utilizing the advantages of wavelet transformation to significantly enhance processing speed and reduce computational resource consumption while preserving data integrity. Additionally, the study introduces a Multiple Residual Module (MRM) to preserve details in both the decomposing network (MR-Decompose) and the denoising network (MRU-Denoise). Extensive experiments on low-light images from coal mines demonstrate that RW-DM outperforms existing state-of-the-art methods qualitatively. Moreover, the application of the proposed method in pedestrian detection using underground coal mine images resulted in a 3.1% increase in \(\hbox {mAP}_{50}\). This improvement not only showcases the effectiveness of the proposed approach but also highlights its potential practical value.