Abstract:Mast methods for seismic noise removal use the characteristic information of single parameter to separate out useful signals or noises,thus achieving usual noise removal. A nonlinear filter for nonlinear transform of seismic data can be designed by taking high dimensional nonlinear mapping character of neural network. Such a filter,differing from other common ones,comprehensively analyses many differences between signal and noise to separate off them effectively. With the use of this noise removal method (in space domain),the seismic section retains its original resolution,and noise removal effect suffers little damage which results from low signal/noise ratio and the dip angle of reflection leg. The paper only displays the processing results in space domain. This noise removal method is also applicable to other domains.