Abstract: In view of the problems such as complex background interference, weak small targets and multi-scale variations existing in the infrared target detection task of unmanned aerial vehicles, a method named DPE-YOLOv8 based on the improved YOLOv8n is proposed. Firstly, the re-parameterized C2f_DBB module was introduced, which enhances the model's ability to extract features of the target in complex backgrounds through diversified branch paths; simultaneously, the multi-scale self-attention mechanism PSM was designed to strengthen the model's perception of multi-scale information; secondly, the SDC2f module was proposed, which significantly enhances the network's ability to express features of small targets while reducing computational costs. Finally, an efficient group convolution decoupling head structure is proposed in this paper, using group convolution to balance accuracy and processing speed. Experimental results show that the proposed algorithm achieves an mAP@50 of 83.4% on the HIT-UAV dataset, which is 3.4% higher than the original YOLOv8n and 4.6% higher in recall rate, while the model's computational complexity is 8.6% lower. Generalization tests were conducted on the CTIR dataset, further verifying the effectiveness and robustness of the algorithm proposed in this paper. Compared with the existing state-of-the-art methods, the improved model provides a good balance between accuracy and efficiency, fully demonstrating that the improved algorithm outperforms other mainstream algorithms and can better complete the infrared target detection task of UAVs.