In the field of unmanned aerial vehicle (UAV) target detection, the significant vertical fluctuations of UAVs pose a considerable challenge to image detection, particularly in detecting small targets, due to the large variation in the size of the main subjects. To overcome this challenge, we propose a novel algorithm architecture based on YOLO-v8, named MSD-YOLO. Firstly, we design a more innovative network for feature extraction and integration (MUBIFPN) to replace the original Neck part, enabling the model to better fuse features. Secondly, we also design a Feature Pyramid Pooling structure (SPPFCSPC-SM) to replace the original SPPF, enhancing the receptive field of this part. Finally, we introduce an advanced multi-dimensional perception detection head (DyHead) as the detection head of this network, significantly enhancing the expression ability of the detection head. Experiments show that the proposed method achieves a 4.7% improvement in recall rate and a 5.8% improvement in mAP50 on the VisDrone2019 dataset compared to the original YOLO-v8n model. The mAP50-90 is improved by 4.0%. Compared to the larger YOLO-v8s, not only is there a slight improvement in recall rate and accuracy, but also the parameters is reduced by 58.6%, and GFLOPs are reduced by 54.7%.