<p>Recently, UAV aerial images show many different application potentials in many fields such as agricultural monitoring, public safety, and disaster management. However, target objects in UAV images are typically a little small, the complex and variable background, and the limited platform resources, which can result in lower accuracy of most embedded detection models for UAV. Therefore, it is challenging to achieve a suitable balance between detection effectiveness and usage of resources. To address the aforementioned problems, we present a small target detection algorithm based on improved YOLOv8 for UAV aerial images scene, in which we optimize YOLOv8 and introduce an object identification framework based on UAV aerial photos. First, we introduce the ContextAggregation module into the Neck part of the YoloV8 model. ContextAggregation can aggregate contextual information from different regions around the target, providing a richer feature representation. Thus, it helps YoloV8 to capture and utilize the global visual information in the image more effectively, especially for the problems of severe scale variation, low contrast, and dense distribution in remote sensing images. Second, we use GhostConv instead of PartialConv to cut down on the model’s computation and the number of parameters, which can improve the detection effectiveness of the model, and enable the model to run under limited computational resources. Finally, WIoU is introduced, which takes into account the weights of different targets when calculating the bounding box overlap degree, assigning higher weights to targets with smaller sizes or targets with specific importance. Experimental results on the VisDrone2019 dataset show that the improved model achieves an average detection accuracy Precision of 56.6%, which is a 2.6% improvement over the original YOLOv8 algorithm, with a reduction of 2.1 GFLOPS in the amount of parameters.</p>