Aiming at the problems of high resource occupancy rate and poor real-time performance when detecting UAV on embedded platform, a target detection algorithm based on improved YOLOv5 network is proposed. Based on YOLOv5s network, MobileNetV3 network is used to replace CSP-Darknet53 as the backbone network for feature extraction, and the feature enhancement network and the regression box loss function of the algorithm are optimized and improved. The improved YOLOv5 algorithm is tested on PC and embedded platform RK3399 based on the self-built UAV data set, and the experimental results show that compared with the original algorithm, the improved YOLOv5 algorithm improves the detection speed by 38% and reduces the model size by 45% while maintaining a high detection accuracy, which effectively improves the detection performance of the algorithm. Meet the actual needs of embedded devices.