Aiming at the characteristics of large amount of remote sensing image data,large terrain fluctuations and wide coverage,this paper proposes a method for super-resolution reconstruction of remote sensing i mage based on convolutional neural network.This method combines dense network and deep back projection network to form dense projection .The unit forms a deep dense projection network, which solves the problems of insufficient texture representation,insufficient d etail extraction,and difficult training in traditional algorithms in the super-resolution reconstruction of remote sensing images.The experiment al results show that on multiple remote sensing image data sets,compared with other comparison methods,the PSNR and SSIM are significantl y improved,and the reconstructed remote sensing image texture signs and details are more abundant.