Abnormal working conditions in the production process of electricity melt magnesium furnace will lead to local overheating of furnace wall and even leakage of furnace, which seriously affects equipment safety and product quality. Because the traditional monitoring methods rely on manual observation and electrical parameter detection, it is difficult to meet the needs of high precision and real-time in modern industry. Aiming at the shortcomings of insufficient detection accuracy of small targets in infrared images, this paper designs a YOLO v11-MSD model based on YOLO v11 model to identify abnormal working conditions of electricity melt magnesium furnace. Firstly, the lightweight network MobileNetv3 is used to replace the original backbone network, and combined with the lightweight SDCB convolution module, the complexity and computational burden of the model can be effectively reduced. The DFformer attention mechanism is introduced to enhance the recognition ability of the model to small target areas in infrared images. Finally, through experimental verification, the FPS of the model in this paper is 110, which is 20% higher than that of YOLOv11 (baseline). Although mAP decreases by 0.2%, it meets the real-time constraint. Compared with similar models, this model achieves a better balance between accuracy and speed. It can be seen from the comparison of ablation experiments that this design sacrifices a small amount of accuracy (mAP decreases by 0.2%) in exchange for a significant increase in speed (+ 22% FPS).