The object detection models with transformers have pioneered a new direction in object detection by utilizing the transformer architecture. However, there are still limitations in performance, particularly when processing objects in aerial images. In this paper, the object detection model is proposed to enhance the efficiency of aerial object detection based on the detection transformers (DETR) technique. The CrossEntropy loss function is replaced with the Focal Loss to mitigate the impact of class imbalance and improve classification performance for small objects. Additionally, the model’s backbone is replaced with Swin Transformer, a more robust architecture for capturing multi-scale image features. Finally, static object queries are replaced with Dynamic Object Queries, enabling the model to initialize and automatically adjust bounding box information during training. The model is trained and evaluated on the VisDrone dataset, showing significant improvements in accuracy and convergence speed compared to the original version. Specifically, the improved model achieves an mAP of 34.7 at an IoU threshold of 0.5, a 50% increase, and significantly reduces the number of training epochs compared to the original DETR model. These results demonstrate that these improvements not only enhance detection efficiency but also pave the way for new applications of DETR in aerial image object detection tasks.