Vision-based small UAV target detection is an important part of anti-UAV technology, which has received extensive attention and research in recent years. Aiming at the problems of complex background, small UAV targets and limited computational resources in small UAV target detection, this paper proposes an improved UAV target detection algorithm and makes algorithmic progress. Based on YOLOv8, this paper adds a detection header from the P2 high-resolution feature layer of YOLOv8 to retain more detailed information about the target and improve the network's ability to detect small targets. In order to reduce the number of parameters of the model, accelerate the training process, and make it easier to be deployed to mobile applications, this paper is based on the GhostNetV2 lightweight network and improves the original convolutional module and C2f module. The experimental results on the Det-Fly dataset demonstrate that compared to the baseline model YOLOv8, the improved model achieves an 8.5% increase in mAP@0.5 while reducing the parameter count by 42.0%. Compared to YOLOv9, the improved model achieves a 4.1% increase in mAP@0.5. The enhanced model not only achieves significant model lightweighting, but also substantially improves the detection accuracy, making it more suitable for deployment on mobile devices.