Abstract: To accurately address the challenge of molten steel level detection in VD （vacuum degassing） furnace under vacuum environment, such as obstruction by smoke, low visibility, and interference from multi-scale targets, a VD furnace molten steel level detection model based on improved YOLOv12n was proposed. This model enhanced feature extraction and object detection performance through three core improvements. First, it was integrating the efficient channel attention（ECA） module, which adaptively strengthened critical channel features while reducing computation and mitigating interference from smoke, dust, and other disturbances. Second, it was embedding the enhance multi-scale dynamic region attention（MRA） module, which precisely focused on multi-scale feature responses in the molten steel level region to improve feature resolution in complex backgrounds. Third, it was utilizing the improved residual efficient layer aggregation（R-ELAN） module DElan to optimize the feature fusion structure, enhancing the correlation of deep features and gradient propagation efficiency, thereby improving the model's ability to capture fine details of molten steel level edge.Experimental validation was performed on a dataset comprising 16 000 molten steel level images captured from VD furnaces, encompassing diverse vacuum levels and obstruction scenarios. Results demonstrate that the proposed EMD-YOLO model outperforms the baseline YOLOv12n and YOLOv12n-seg models in key detection capabilities. Specifically, the model achieves respective improvements of 4.7 and 3.1 percentage point in mean average precision（mAP@50%） as well as 6.9 and 9.1 percentage point in comprehensive confidence（CC） relative to YOLOv12n and YOLOv12n-seg, while maintaining a moderate computational complexity of approximately 6.0 GB. Furthermore, the model's robustness under complex scenarios, such as smoke obstruction and molten steel level fluctuation, was significantly enhanced. This method achieves precise detection of molten steel liquid levels in a VD furnace under vacuum conditions, providing reliable technical support for automated control and safety monitoring in the metallurgical refining process.