For UAV aerial photography, the background is complex, the detection target is small and dense. A lightweight UAV aerial photography target detection algorithm SDS-YOLO based on YOLOv5 is proposed. Firstly, SDS-YOLO algorithm reconstructs the lightweight network structure, the feature extraction network and feature fusion network are reconstructed. Adjusts the detection layer and receptive field architecture, establishes the multi-scale detection information dependence between deep semantics and shallow semantics, increases the weight of shallow network feature layer, and improves the detection ability of small targets; Secondly, the pre selection box is adjusted by clustering and genetic learning algorithm to realize the optimal pre selection box selection mechanism of reconstructed network and accelerate the convergence speed of the model. Finally, SDS-YOLO was trained with varifocal loss to make IACS regression to improve the detection ability of the model to dense objects. The results show that the accuracy of the model is improved by 7.64%; The volume of the model is 4.25MB, which is significantly lower than that of the original model; The speed and amount of reasoning are improved. Compared with the current mainstream algorithms, SDS-YOLO has made good improvements in all aspects to meet the requirements of real-time target detection in UAV aerial photography.