Abstract: Small uncontrolled drones pose threats to public safety. In this study, we designed a lightweight unmanned aerial vehicle target detection algorithm based on infrared detection technology. This method achieves the effect of auxiliary training by defining two backbone networks with the same model structure and initialization parameters. This paper improves the structure on the basis of Ghost PAN to build a multi-scale feature fusion structure that is more suitable for UAV target detection. The ablation experiment results show that each module involved in this algorithm improves the UAV target detection accuracy (), and the results of the multi-algorithm comparison experiments show that the algorithm proposed in this study can adapt to a variety of UAV flight scenarios. Compared to Nanodet Plus-m, the detection time was unchanged, and the mAP increased by 11.4%, 4.2%, 16%, and 4.2% on the Sea, Sky, Mountain, and City datasets, respectively.