To address the challenges of large model size, high computational complexity, and notable accuracy degradation when deploying railway track obstacle detection models on edge devices, this paper proposes a lightweight re-parameterized network, YOLOv11s-Slim-Rep. The network achieves synergistic optimization of accuracy and efficiency through two core improvements: adjusting the network width scaling factor from 0.50 to 0.35, resulting in a 44.8% reduction in model size and a 45.4% decrease in parameter count; and replacing all stride=2 downsampling convolutional layers with a RepConvCustom module, which is fused into a single convolution via structural re-parameterization, incurring zero additional inference overhead. On two railway track obstacle datasets (NewData and Railway), compared to the YOLOv11s baseline, the mAP50 decreases by only approximately 1.1% while achieving nearly 45% model compression. Compared to YOLOv11n, it improves mAP50 by 4.7% in complex scenarios, with a cross-dataset mAP50 fluctuation of only 0.6%. The proposed method maintains high detection accuracy while significantly compressing the model, demonstrating strong potential for edge deployment.