Abstract:In seismic data processing,the KL filtering is often used for elimination of coherent and random noises in order to improve S/N ratio in seismic records.The concentrating algorithms such a-seigenvalue decomposition (EVD) and singular value decomposition (SVD) are generally adopted in ordinary KL transform, but the computational cost is higher when there is a large amount of seismic traces,which is difficult to use in practice.The paper first introduced the artificial neural network (ANN) to KL filtering of seismic signal that obtained projection vectors with orthogonal decomposition by self-organizing (without monitoring) studying weights connected in adaptive computational networks.The correlation of ordinary KL transform method with the method introduce in the paper by the test of theoretical records showed the consistence of the results by two methods,but fast in the method introduced in the paper in comparison with ordinary method when having more input traces of records, which greatly reduced computational cost in filtering processing that is rid of the large time-consuming shortage of KL filtering and more practical.The paper finally gave two computational cases of KL filtering of real seismic data.