Abstract: Various large-scale sky survey plans release massive astronomical data. It is necessary to study the spectral automatic processing methods for large-scale data. It is difficult to apply the traditional galaxy spectral classification methods based on spectral line measurement or BPT diagram to automatic galaxy spectra classification pipeline directly. In contrast, machine learning method is more suitable for the classification and analysis of massive astronomical data. This paper proposes a galaxy spectral analysis method based on double hierarchical clustering. The first layer uses K-means clustering method to classify galaxy spectra into absorption line galaxies and emission line galaxies; the second layer uses Clustering Large Applications clustering algorithm to gather emission line galaxies into five subtypes. We experiment with galaxy spectral data from LAMOST DR5 and analyze the result in detail by spectral line detection, BPT diagram and color magnitude map.The experimental results show that:(1) The first layer K-means clustering can classify Galaxy spectra into absorption line galaxies and emission line galaxies successful, which are consistent with the classification results based on spectral line detection. (2) The results of CLARA cluster in the second layer can reflect different galaxy types in BPT diagram. (3) There is an expected correlation between spectral clustering results and color magnitude classification. (4) The two-layers clustering can be applied to large-scale data automatic analysis and processing. The clustering results can reflect the physical properties and evolution process of the galaxies. And the cluster centers can provide templates for automatic spectral classification pipeline.