Aiming at the problem that it is difficult for the existing remote sensing image denoising algorithms to fuse shallow image features into deep image information,a remote sensing image denoising network that fuses multi-scale features is proposed,which consists of an asymmetric convolution block,a dilated attention block,a residual projection block and a residual fusion block.First,asymmetric convolution is used to preliminarily extract features and reduce a large amount of information redundancy in the network.Then,the multi-scale features are extracted through the dilated attention block to learn rich context information, and the fusion of the extracted multi-scale features is more conducive to denoising and retaining more edge texture details of the image.The residual projection block collects a large amount of contextual and spatial information from the multi-scale features,and finally the residual fusion block generates a residual image to remove noise.Experimental results show that the proposed network outperforms several advanced image denoising algorithms in both quantitative and qualitative evaluations on the NWPU-RESISC45 and UCMerced_LandUse remote sensing image datasets.