<p>Effective detection of foreign object intrusions on railways is essential for ensuring railway safety. A complex and variable railway environment leads to high miss and false alarm rates, particularly in detecting small-scale foreign objects. To address this challenge, we propose a railway foreign intrusion detection method based on multi-scale feature enhancement. Central to our method is the Dimension-aware Diffusion Fusion network, which enhances the capture and utilization of multi-scale feature information through diffusion and fusion in the intermediate layers of the backbone network. Additionally, we integrate the Variable Size and Stride module into the backbone network, enabling adaptive multi-scale feature extraction, which is crucial for detecting objects of varying sizes. Furthermore, we employ the High–Low (HiLo) attention mechanism, which significantly enhances feature representation by focusing on both high and low-frequency information, thereby improving the network’s ability to discern subtle object details amidst complex backgrounds. Moreover, we introduce the EMA slide loss, a flexible loss function capable of dynamically adjusting parameters for multi-scale detection tasks. The proposed method was tested on the Railway Foreign Intrusion (RFI) dataset and the public CityPersons dataset. Specifically, on the RFI dataset, our method achieves mAP@50 of 81.4% and mAP@50–95 of 59.4%. Additionally, on the CityPersons dataset, we achieve mAP@50 of 68.6% and mAP@50–95 of 42.8%. Experimental results demonstrate that our method outperforms comparative detectors in detection performance, highlighting its potential to significantly enhance railway safety and operational efficiency.</p>