In geospatial remote sensing images,the target detection accuracy is low due to dense target distribution,large target scale variation range and too little feature information of small targets.To solve the problem,a multi-scale remote sensing small target detection algorithm cosSTR-YOLOv7 based on Swin Transformer (STR) and YOLOv7 is proposed.YOLOv7 is taken as the baseline network.Firstly,STR module is used to replace E-ELAN module in the backbone network，and it is improved to be cosSTR module by using cosine attention mechanism and post-regularization method，so as to improve the stability of model training.Secondly,a new feature fusion layer is constructed in the Neck part to reduce feature information loss.A small target prediction layer is added to the prediction part to improve the model’s ability of small target detection.Finally,a new SIoU loss function is used to calculate the positioning loss to accelerate the convergence rate of the model.The remote sensing dataset DIOR is used for the experiment,and the experimental results show that the mean Average Precision (mAP) of the proposed algorithm reaches 92.63%which is 3.73 percentage points higher than that of the original YOLOv7 algorithmand the performance of multi-scale small target detection has been significantly improved.