Double-peaked emission lines may reflect some rare scenes, such as binary active nucleus (AGNs), double supermassive black holes (SMBHs), interaction between jet and narrow line regions, or be low-quality spectra. Generally, at least 2 peaks are included in double-peaked emission lines, which are useful in searching rare objects such as AGNs, SMBHs, galaxypairs. It is meaningful in researching double-peaked emission lines for further studying formation and revolution of galaxies and even the universe. In this paper, a new method based on relevant subspace for extracting and analyzing of double-peaked emission line spectra is proposed. There are 3 parts in this paper: (1) Sparse difference factor δ is defined to measure attribute difference degree in spectra with double-peaked emission lines. And KNN algorithm is employed to restrain the involved spectral data. Then, characteristics extraction method for low solution spectra is proposed based on relevant subspace. (2) To test the efficient of parameters of sparse difference factor δ and k of KNN algorithm, 664 spectra are selected from LAMOST as training set, including 332 positive samples and 332 negative ones. To ensure 8 lines (Hα, Hβ, ［OⅢ］λλ4 959, 5 007, ［NⅡ］λλ6 548, 6 584, ［SⅡ］λλ6 717, 6 731) are in wavelength coverage of LAMOST, the redshift region is z<0.3. And then influence of experiment result about two parameters k and α of δ threshold is analyzed. The result indicates that distribution of relevant attributes is dense with less sparse points when k=18 and α=0.6. (3) Wavelength coverage, red/blue shift interval and line strength ratio of double-peaked emission lines in subspace are analyzed and measured theoretically. Then characteristic description of double peaks based on relevant subspace is given. Finally profiles of double peaks are analyzed according to emission excitation mechanism and line strength relationship.