Aiming at the problems of complex background, dense small targets, and large scale changes in UAV aerial images, this paper proposes a target detection algorithm based on improved YOLOv5s. First, ShuffleNetV2 is used to reconfigure the backbone network, which improves the detection speed while reducing the computational complexity and realizes the model lightweight. Second, the ASPP module is introduced to enhance the multi-scale feature extraction capability, reduce the small target feature loss, and improve the detection accuracy. Further, the CA attention mechanism is introduced into ShuffleNetV2 and YOLOv5s feature fusion network to strengthen the extraction and expression of key features. Finally, the SIoU loss function is substituted for the original CIoU to better adapt to multi-scale targets and improve the bounding box regression accuracy.Experiments conducted on the VisDrone2019 dataset demonstrate that the mAP0.5 and mAP0.5:0.95 of the improved model reach 39.5% and 21.5%, respectively, which are 5.0% and 3.1% higher than the original model, and the amount of parameters is reduced to 54%. The results show that the algorithm effectively improves the performance of target detection in complex scenes under the premise of ensuring the lightweight of the model, and has high application value.