<p>In aerial images, the small size, significant scale variation, and dense object distribution often result in low detection accuracy. Therefore, small object detection in aerial images remains a highly challenging task. This paper proposed EDet-YOLO, a small object detection algorithm designed to improve detection precision. Based on YOLO11, several innovations have been introduced. First, the original C3k2 module is restructured into an Efficient Convolutional Feature Extraction module (ECFE), which incorporates a novel Efficient Convolution Module (ECM) to enhance multi-scale feature extraction. Second, a Spatial Bidirectional Attention Module (SBAM) is proposed to establish a bidirectional attention-guided mechanism between high- and low-resolution feature layers, achieving complementary fusion of semantic and detail information. This design effectively enhanced the discriminability and localization accuracy of small objects in complex backgrounds. In addition, a dynamic head is employed to replace the original detection head, enabling adaptive feature enhancement and multi-level feature integration to boost detection performance. A new small object detection layer is also introduced to further improve accuracy. Experimental results on the VisDrone2019 and HIT-UAV datasets demonstrate that the proposed EDet-YOLO outperforms existing models. Compared to the baseline, EDet-YOLO achieves improvements of 10.2% and 1.7% in mAP@50, and 7.6% and 3.4% in mAP@50–95, respectively. Moreover, the detection speed of EDet-YOLO on Jetson Orin Nano reached 24.9 FPS, and this performance met the requirements of real-time detection.</p>