To improve the rapid identification and response capabilities of coal-fired power enterprises during emergencies such as gas explosions, equipment failures, or collapses, we propose a hazardous object detection method tailored for coal power scenarios based on an enhanced YOLOv11 framework. The method is designed to detect hazardous objects that may trigger or indicate emergencies, including abnormal equipment, unsafe personnel behaviors, and environmental risks. It incorporates a staged data augmentation strategy together with Contrast Limited Adaptive Histogram Equalization (CLAHE) preprocessing to mitigate low-light and dust interference in industrial environments. Furthermore, an ADown dual-path downsampling module is integrated into the backbone to better preserve multi-scale features. We conduct systematic evaluations on a public dataset of underground coal mine drilling scenes, where progressively combining ADown, data augmentation, and CLAHE improves mAP@0.5 from 0.935 to 0.983. The results demonstrate significant gains in detecting small and occluded objects, such as irregular drill rods or partially hidden miners, under low illumination and complex backgrounds. The proposed method achieves robust real-time performance, showing strong potential for deployment in coal-fired power enterprises to reduce risks and enhance emergency response efficiency.