Abstract: Aiming at the problem that the super-resolution reconstruction algorithms based on deep learning increase the complexity of the network model and cause a large number of parameters by increasing the depth of neural network to improve the super-resolution reconstruction performance,an inverted N-type lightweight network based on back projection for image super-resolution reconstruction is proposed.First,the initial convolutional block is used to extract the shallow features of low-resolution images.Secondly,the deep features of low-resolution images are extracted through two rounds of gradual model compression based on the inverted N-type network of back projection and two rounds of gradual model restoration of the inverted N-type network.Then,the extracted shallow features and deep features are combined by global residual learning and scaled up to the desired output reconstructed image size by using the upsampling module.Finally,the reconstruction module is used to reconstruct the super-resolution image.The results of experiments on the Set5,Set14,BSDS100 and Urban100 testing sets show that the proposed algorithm has a lighter network structure than other algorithms,and the reconstructed super-resolution images not only have higher peak signal-to-noise ratio(PSNR)and structural similarity(SSIM),but also have better visual effects.