With the rapid growth of the drone industry, the number of illegal flights is on the rise, threatening public safety. Thus, the task of efficient and accurate drone detection is crucial. In response to the limitations of existing deep learning-based object detection methods, such as the insufficient balance between accuracy and speed, and model sizes being too large to be suitable for edge devices, an improved unmanned aerial vehicle(UAV) target detection method called YOLO-Drone was proposed based on YOLOV8n. Firstly, the C2f module in the backbone was replaced with the bneck module from MobileNetv3 to achieve more lightweight feature extraction, meanwhile, the PConv mode was used for some convolution operations to reduce redundant calculations. Secondly, aiming to better retain the detailed features of drones and enhance the model's ability to recognize small targets, SPD-Conv was introduced into the backbone network for downsampling. Thirdly, focusing on the detection of small targets, the large target detection head was removed, and the network width was reduced to lower the model complexity. Lastly, WIoU was used as the loss function to improve the accuracy of localization. The experimental results based on public datasets show that the proposed model achieves an mAP@0.5 of 94.1%, which is 0.8% higher than that of YOLOv8n. Meanwhile, the computational load and the number of parameters are reduced by 40% and 70% respectively. While achieving further lightweight design and maintaining detection accuracy, it makes the proposed model a more suitable option to be deployed in the edge computing devices.