Airport sign detection based on improved RT-DETR

To improve the detection accuracy and small object recognition capabilities of signage devices in airport scenarios, this paper proposes a target detection method based on an improved RT-DETR. First, to address the shortcomings of the original RT-DETR in feature extraction, the SPPELAN module is introduced. Through multi-scale pooling and an efficient feature fusion mechanism, it enhances the perception capabilities for targets of different sizes. Second, this paper introduces a lightweight downsampling structure, the PD-Conv module, which effectively retains fine-grained information while maintaining computational efficiency, further improving the detection performance of small targets. Experiments were conducted on a publicly available dataset containing four categories of airport signage devices. The results show that the proposed method outperforms the original RT-DETR model in terms of mAP, recall rate, and other metrics, validating its effectiveness and robustness in real-world scenarios.