Research On Denoising Method Of Urban Shallow Seismic Data Based On Deep Learning

Urban shallow reflection wave seismic exploration methods are affected by many factors such as urban industry,traffic,human activities and shallow anisotropy,so that seismic data often contains complex noise interference.The interference wave in shallow seismic record is always linear,and under the condition of various noise interference,the morphological change of recording waveform is complex,overlapping with the effective waves in both time and frequency domains.Among them,the surface wave is a typical interference wave in the seismic reflection of urban shallow reflection waves,which is characterized by low frequency,strong energy and low speed coherent noise on the seismic record.In shallow seismic exploration,the surface wave develops in the near-surface low-velocity zone.Due to the dispersion characteristics,it causes a strong velocity gradient in the near-surface region where the medium anisotropy is strong(<30 m),and propagates along the free surface at low group velocity,which seriously destroys the shallow reflection information in the small offset distance.Suppression of surface wave is critical to improving the signal-to-noise ratio of shallow seismic survey data.With the development of the instrument,the urban shallow seismic exploration method can obtain a large amount of shallow seismic data in high-density acquisition mode,but the complex urban interference makes it difficult to process and analyze these data.The current data-driven deep convolution neural network method is an effective solution for processing massive data,and it has achieved great success in the fields of image recognition,image super-resolution reconstruction,and speech enhancement and so on.It can extract massive data features through deep network structure and realize feature expression.Taking the surface wave as the noise to be suppressed,this paper proposes to study the data characteristics of a line in the study area based on the depth learning method,and applies the trained deep learning noise reduction model to the surface wave suppression of other seismic data obtained under the same period and the same acquisition parameters.The experimental results show that the deep convolution neural network algorithm can learn the mapping relationship between noise data and noiseless data through one line seismic data,and successfully suppress the surface wave interference in other line seismic data,and achieve the efficient processing of data in high density acquisition mode.The results also show that in the case of other noise samples,deep convolution neural network algorithm can also be extended to linear noise or more complex noise interference suppression processing.This paper first analyzes the similarities and differences between seismic data and images,speech,and proposes a seismic data preprocessing method for deep noise reduction convolutional neural networks.Secondly,the influence of different mapping objects on the noise reduction performance of the noise reduction neural network is compared by numerical simulation of seismic records.Finally,the proposed deep convolutional neural network denoising method is applied to practical multiple on the line data,it is proved that the deep convolution neural network method can learn the noise and effective wave characteristics,and achieve high-efficiency,high-fidelity suppression of the seismic data collected in the same study area,the same period,and the same observation system.