Occlusion-aware aircraft target detection via a global information tracking network

In complex scenes, fuselage-induced occlusion often erodes key target features, posing a persistent challenge and active research focus in aircraft target detection. Building on RT-DETR, we introduce a global information tracking network that integrates a multi-dimensional cross-domain dynamic refinement module, a dynamic information fusion module, and a context-gating module. Evaluated on an in-house aircraft dataset, the model attains 56.3% mAP@[0.5:0.95], 80.6% mAP@0.5, 89.0% precision, and 79.7% recall. Relative to the baseline, these figures represent gains of 1.9%, 4.0%, 5.7%, and 0.7%, respectively, indicating a clear improvement in detection performance under occlusion.