Abstract:As a new approach of blind sources separation (BSS), the independent component analysis (ICA) is powerful tool of decomposing independent information in surveyed data. Former ICA algorithm generally supposes that the noises can be neglected, but practically, the additional noises are often included in surveyed data. Under the condition that the additional noises can not be neglected, the improved ICA algorithm first uses non-zero time-lapse covariance and two-way eigenvalue decomposition (EVD) to successfully eliminate the influence of part of additional noises;then the original signals can be better separated by using the ICA algorithm again.The paper demonstrated by the tests of seismic theoretical model and practical data that improved ICA algorithm can effectively overcome the influence of additional noises on ordinary ICA algorithm and separate the useful signals in seismic data,realizing the goal by using independent component analysis to suppress the noises in seismic data.