To address the challenges of low detection accuracy caused by large-scale variations in targets, severe background interference, and low-quality samples in UAV aerial traffic object detection tasks, this paper proposes an efficient small object detection algorithm based on an improved RT-DETR framework. The key contributions include: 1) Backbone Optimization: Replacing the baseline backbone with a lightweight YOLOv8 network and introducing a multi-stage gated attention mechanism and channel MLP to enhance the C2f module, named C2f_MogaBlock, for improved feature extraction capability; 2) Frequency-Spatial Pyramid Attention Neck (FS-PAN): Designing a novel neck network with a lightweight shallow feature fusion module (LSFF) to recover fine-grained information lost during downsampling, and embedding a frequency-spatial attention block (FSA) to strengthen multi-scale feature fusion; 3) Adaptive Context Fusion Block (ACF): Constructing a context-aware feature interaction module to enhance contextual perception; 4) Loss Function Enhancement: Integrating the W-ShapeIoU loss function to suppress interference from low-quality samples. Experimental results on the VisDrone2019 dataset demonstrate significant improvements, with gains of 3.9 and 3.2 percentage points in mAP@0.5 and mAP@0.5:0.95, respectively. Compared to state-of-the-art models with similar computational complexity, the proposed method exhibits superior performance in balancing accuracy and efficiency for UAV-based traffic object detection.