<p>Target detection constitutes a pivotal technology for UAVs to achieve autonomous environmental perception, precise operation, and real-time decision-making. To address the challenges of significant target scale variations, high miss rates, and low detection accuracy in UAV aerial imagery object detection tasks, we propose an end-to-end target detection network featuring multi-scale feature extraction and fusion, designated as EA-DETR. Specifically, we replace the deep residual modules in the backbone network with the proposed Dual-Path Scaled Attention (DPSA) Block. By incorporating a single-head attention mechanism into the feature extraction module and dynamically modulating the contribution of the attention branch, this approach enhances feature representation in the backbone while reducing model parameter complexity. Furthermore, to enhance small target detection capability, we incorporate shallow feature information. During small target feature fusion, we propose a Dual-Path Multi-scale Fusion Module (DPMK) based on channel-wise and spatial attention mechanisms. This effectively strengthens feature representation for small targets while suppressing noise interference in complex backgrounds. Experimental results on the VisDrone2019 dataset demonstrate that EA-DETR achieves improvements of 0.92% in mAP50 and 5.76% in precision compared to the baseline model. These findings validate that EA-DETR enhances detection accuracy while maintaining a low false positive rate. The project page: <a href="https://github.com/qc4059/EA-DETR/tree/master">https://github.com/qc4059/EA-DETR/tree/master</a></p>