Abstract: Existing deep learning-based image super-resolution networks often lead to redundant computations and a huge amount of parameters, as well as the lack of high-level texture features in the super-resolution results. Aimed at the above problems, an image super-resolution based on attention and iterative feedback fusion network is proposed. The model used a super-resolution architecture of iterative up-and-down sampling. The network used the enhanced attention feedback module to efficiently obtain the corresponding weights of image feature channels by reducing the number of feature channels and enhancing the attention mechanism, ensuring the quality of super-score and reducing the amount of network parameters. In addition, the network model designed a feedback fusion network block, which used the bidirectional iterative feedback fusion of high-level feature information and low-level feature information to maximize information extraction and generate image with higher quality. The experimental results show that compared with the current advanced image super-resolution networks (SRFBN, SMSR, RFAN), the network model has certain advantages in quantitative indicators (PSNR, SSIM) and subjective visual effects.