An Enhanced YOLOv13 Framework for Multi-target Recognition with Adaptive Feature Enhancement

Accurate recognition of locomotive identifiers is essential for intelligent railway transportation and autonomous perception systems. However, this task remains challenging due to complex outdoor environments, illumination variations, and the visual ambiguity of character features. To address these issues, this paper presents an enhanced YOLOv13-based framework for multi-target recognition with adaptive feature enhancement. Specifically, a High–Low Frequency Adaptive Enhancement Module (HLAEM) is embedded in the backbone network to capture both fine-grained local details and global contextual semantics, while a lightweight Dynamic Upsampling Module (DySample) is employed to improve the accuracy and efficiency of feature map reconstruction. Experiments on a real-world locomotive identifier dataset demonstrate that the proposed method achieves a mean Average Precision (mAP@50) of 99.50%, significantly outperforming the baseline of 95.90%. These results confirm the effectiveness of the framework in locomotive identifier recognition under complex railway environments. Furthermore, the adaptive perception mechanism provides potential reference value for broader applications in intelligent transportation and ground–air cooperative autonomous sensing.