一种基于神经网络的地震信号去噪的处理方法

在众多的去噪处理方法中，Ｋ－Ｌ变换是去除随机噪声的有效方法．Ｋ－Ｌ变换是利用地震剖面各道在空间方向的相关性来取信号的共同特征，以实现去除剖面中的随机噪声．传统的Ｋ－Ｌ变换是通过矩阵的Ｈｏｕｓｈｏｌｄ变换实现的．其运算量及空间占用量比较大．笔者用一种Ｈｏｐ－ｆｉｅｌｄ网实现对地震剖面进行Ｋ－Ｌ变换的去噪处理，并且给出了该网络的详尽描述及稳定性证明．由于Ｈｏｐｆｉｅｌｄ网能实时进行处理，固而具有实时性．模拟实验的结果表明，这种方法对消除地震剖面的随机噪声是一种行之有效的方法．

A Method Denoising from Seismic Data Based on Nerve Network

The dcnoising is a very important and essential process in the processing of seismic data seciion. Of the denoising methods now available, K-L iransform is the most effective nieihod for removingstochastic noises. By means of the correlativities among signals in all channels of seismic section in all the spatial directions, the common useful components in them are extracted,so stochastic noises are removed from the seismic section. Traditionally,the realization of K-L transform is by Houshold transform,which needs a lot of storage space and computation. The authors put forward a K-L transform method by using Hopfield nerve network. In the paper,this network is stated in detail ,and its stability is also proven. The method has the real time property. The result of simulation test shows that this method is feasible in re-moving stochastic noises from seismic section. SubJect beadings/