The high-resolution processing method of seismic signals based on CCU-Net
In the process of seismic data acquisition,it is affected by a variety of infections and instruments and often presents the characteristics of low resolution and low signal-to-noise ratio,which affects the subsequent seismic interpretation process.Conventional deconvolution,Q compensation and other high resolution processing methods are limited by the difficulty of wavelet extraction and Q value estimation,and can not be applied to the actual seismic data processing.Because deep learning can quickly learn the mapping relationship between input and target,it has recently gained attention in the field of resolution processing,but the resolution methods based on deep learning often fail to consider multi-scale and spatial information.Therefore,a processing method based on CCU-Net is proposed.The main body of the network consists of encoding and decoding structures.Integrating CBAM and CA modules,the network can pay more attention to the spatial information and channel information of the feature layer.L1,SSIM and MS-SSIM are used as combined loss functions to improve the ability of the network to learn structural information at different scales and to recover the details better.The seismic data with different resolution and noise level are generated by forward simulation method for network training.In the synthetic seismic data and actual data,the CCU-Net based processing method has been significantly improves the main frequency of seismic data,widens the frequency band,and displays the interlayer information better.Compared with other methods,the resolution of seismic data is effectively improved,and the noise is also suppressed significantly.
deep learningseismic dataimprove resolutiondenoising
万方数据
维普
