<p>Object detection in UAV aerial images faces challenges such as complex backgrounds, dense small objects, and mutual occlusion, resulting in blurred object-background boundaries and reduced localization accuracy. In this paper, we propose REA-YOLO, an object detection algorithm designed to enhance small object detection accuracy and achieve model lightweight. YOLOv11 serves as the base model for our improvements. We introduce the REA module, which addresses the blurring of small object-background boundaries by aggregating multi-scale information and recalibrating edge features. Subsequently, the small object detection layer and dynamic detection head are integrated into the model to adaptively optimize the detection strategy and enhance small object extraction. The requirements for real-time inference of the model on the UAV platform impose a computational load. After training, the model is pruned using a pruning algorithm to reduce its size. Finally, REA-YOLO is deployed and tested on embedded devices, improving small object detection accuracy while ensuring real-time performance. Extensive experiments on both public and self-built datasets demonstrate the effectiveness and sophistication of REA-YOLO for small object detection in aerial images.</p>