There are several types of seismic waves in in-seam seismic exploration, including P-wave, S-wave, linear polarization Love wave and elliptical polarization Rayleigh wave. Since channel wave dispersion is serious, those waves composite together in seismic data and are difficult to differentiate. It's very important to find wave field separation method for in-seam seimic data processing. Depending on seismic vibration polarization characteristic, the paper proposed an adaptive covariance matrix method for polarization filter based on S transform in time-frequency domain. It processed horizontal dual component of channel waves signal and used eigenvalues and eigenvectors of this matrix to get ellipticity and azimuth polarization distribution, establish polarization filtering function by these parameters and process wave field separation of in-seam seimic signal. Theoretical synthetic signal and practical data tests showed that this method can effectively extract Love channel wave from in-seam seismic record. It can avoid interference with other types of seismic waves in channel waves and provide a new technology for in-seam seismic data processing.