In this paper, a magnification method based on deep learning was used for the study of astronomical images. An effective astronomical image magnification method was proposed based on the structural characteristics of the new vacuum solar telescope (NVST) images. First, the Binning technology was used to down-sample the data to obtain the corresponding low-resolution images. Second, an improved residual dense network was used to fully extract and utilize the multi-level features of low-resolution images, and thus to enable the reconstruction of high-resolution solar images. Finally, the residual, correlation, and power spectrum analysis methods were used to quantitatively evaluate the reconstruction errors in solar images. The experimental results show that compared with the conventional interpolation method, the proposed method can finely magnify the small-scale structures and effectively improve the signal-to-noise ratio in solar images.