Detection of foreign object encroachment limit in railroad based on improved YOLOv11 algorithm

In order to cope with the challenges of difficult detection of small targets and heavy interference from complex backgrounds in railroad foreign object encroachment detection, this paper proposes a novel railroad foreign object encroachment detection method by improving on the basis of the YOLOv11 model. The method introduces the Eiestem module in the backbone network to replace part of the original standard convolutional layer, thus strengthening the feature extraction capability of the model for details and small targets; meanwhile, the ReCalibrationFPN module is used in the feature fusion stage to replace the traditional feature pyramid fusion strategy, which effectively improves the dynamic calibration effect of multi-scale features. The experimental results show that the original YOLOv11 model has a mAP50 of 92.0% on all categories, while the overall mAP50 reaches 92.7% after adding only the Eiestem module, and the overall mAP50 is further improved to 94.6% after adding both the Eiestem and ReCalibrationFPN modules. These data prove that the improved model has stronger detection capability and higher robustness in the face of small targets and complex backgrounds, which significantly reduces the phenomenon of false alarms and missed alarms, and meets the practical needs of real-time detection of foreign object encroachment limits in railroads.