The rapid development of large scale sky survey project has produced a large amount of stellar spectral data, which makes the automatic classification of stellar spectral data a challenging task. We select F5, G5, and K5 types of spectra data, which are derived from LAMOST Data Release 5 (DR5), one dimensional convolution network and short-time Fourier transform (STFT) are used to transform the one-dimensional spectra data into two-dimensional Fourier spectrum images. And then the two-dimensional Fourier spectrum images are classified automatically by the capsule network. Because the capsule network preserves the hierarchical pose relationships between the entities in the image and the advantages of removing the pooled layers, the experimental results show that for the classification accuracy of F5, G5, and K5 stellar spectra, capsule Network has better classification performance and is superior to other classification methods.