Most existing image super-resolution reconstruction algorithms have an extremely deep network structure, which leads to excessive parameters and an inability to fully extract features. To solve these problems, this study proposes an image super-resolution reconstruction algorithm based on an enhanced multi-scale residual network (EMSRN). The network consists of serial enhanced multi-scale residual blocks (EMSRB), and the backbone structure of the EMSRB is constructed using a residual block and parallel multi-dilation rate dilated convolution group, which effectively reduce the network parameters while obtaining the local and global multi-scale features of the image. The channel attention mechanism is used at the end of the block to adaptively weight extracted features, which enables the network to pay more attention to high-frequency information. Experiments show that, compared with the basic multi-scale residual network, the proposed algorithm improves the peak signal-to-noise ratio (PSNR) by 0.53 dB, and the structural similarity (SSIM) reaches 0.9782. Compared with the enhanced deep super-resolution network, the proposed algorithm achieves similar reconstruction performance with only 31.7% of its parameters.