<p>Addressing the problems of fuzzy target edges, low information of color channels, and complex background interference in UAV infrared target detection tasks, we propose a lightweight and efficient detection method GA-DETR. Firstly, in order to improve the computational efficiency of the model and the feature extraction capability, we propose a cross-stage gated feature extraction module, which effectively reduces the redundant computation of the model and optimizes the multi-scale target detection performance. Second, the cascaded group attention module is introduced to further improve the model’s ability to perceive small targets and sparse features. In addition, the composite loss function FMPIoU is designed to significantly improve the classification and localization performance in response to the imbalance in the distribution of positive and negative samples and the low accuracy of bounding box regression. Compared with RT-DETR, GA-DETR reduces the parameters by 36%, the GFLOPs by 24%, and the model size by 35%, while improving the FPS by 10.9% and achieving better accuracy (+6%), recall (+1.9%), and mAP (mAP50 +4.8%, mAP50:95 +5.6%). These results validate its benefits in UAV infrared detection tasks.</p>