Abstract: The data mining research of large-scale survey is focused on handling, processing and extracting information from massive astronomical data. In this paper, we try to apply the extensible R programming language in LAMOST spectral analysis, and make full use of its capability of integrated data analysis and visualization methods. We mainly study the functions and characteristics of the RFITSIO package for reading and writing FITS format files in R. We then group the LAMOST DR2 data according to the released classification result, and the PCA package in R is applied in each group to extract spectral features from the large amount of noisy spectra. The result shows that, the spectral features are well kept through PCA reconstruction. By extracting the FLUX eigenvalues of the spectral signal description capability of each band in the spectrum, the PCA is used to extract the characteristic value of LAMOST. Rotating coordinate system to eliminate the correlation between the characteristics of the spectral resolution of the data, to reduce the dimensionality of data and remove the effect of noise. This dimensional reduction based feature extraction method can be a very efficient pre-processing approach for the follow-up data mining in LAMOST dataset.