<p>Detecting foreign object intrusions in long-range railway scenarios is a critical yet challenging task for ensuring operational safety. The primary difficulties stem from the extremely small pixel footprint of distant objects and the severe degradation of their structural and textural details during the imaging process, which are often irreversibly lost by conventional detection networks. Here we introduce SaS-Det, a unified framework that synergistically integrates spatial slicing, structure-aware super-resolution, and an optimized detector to address these challenges. The framework first employs a sliced segmentation integration module with a SAM-based validity check to preserve target structures and reduce background redundancy. Subsequently, a novel long-range spatial-texture-enhanced ESRGAN (LRST-ESRGAN) recovers fine-grained edges and textures specifically for small object regions. Finally, a Swin Transformer and GSConv-based YOLO11 (STG-YOLO11) detector leverages enhanced global context and efficient feature extraction for robust localization. On our newly constructed Small Object Railway Dataset, here we show that SaS-Det achieves a mean average precision (mAP@0.5) of 95.4%, which is a substantial improvement of approximately 21% over the baseline YOLO11s model, with consistent gains across precision and recall. This significant performance enhancement, further validated on the public BDD100K, SRSDD-V1.0, and AI-TOD datasets, demonstrates the framework’s robust generalization capability across diverse scenes, offering a more reliable solution for intelligent railway perception systems. The code will be available at: https://github.com/thailand88/SaS-Det.</p>