Abstract: For the problems such as large-scale target variation, missed detection, false detection and target occlusion in the detection of small targets in drone images, an improved SDP-YOLOv8s algorithm is proposed. Firstly, a small target detection head is introduced in the detection head part and the large target detection head is removed, which improves the accuracy of small target detection while reducing the redundant computational load of the algorithm. Secondly, a parallel position-aware attention module (PPA) is introduced in the feature fusion layer to capture multi-scale feature information and enhance the feature fusion ability. A dynamic upsampling module X-DySample is designed to further optimize the algorithm's ability to process features of different scales and improve the algorithm's anti-interference ability, and the SPD-Conv module is introduced in the backbone network to improve the problem of feature loss during convolution. Experimental results on the public dataset VisDrone2019 show that compared with the YOLOv8s algorithm, the SDP-YOLOv8 algorithm has improved the detection accuracy by 8.2% and 5.8% in mAP@0.5 and mAP@0.5:0.95 respectively, and the algorithm parameters have been reduced by 22.5%. At the same time, the generalization and effectiveness of the proposed algorithm have been verified on the Tiny-Person dataset.