Research On Intelligent Seismic First Break Picking Method Under Low SNR

As one of the most effective oil and gas exploration methods,seismic exploration generally includes three stages: seismic data acquisition,seismic data processing and seismic data interpretation.During the data acquisition,the fluctuation of surface and the change of low velocity zone cause signal interference,therefore the static correction of seismic data is taken firstly.Picking up the first break is a prerequisite for obtaining a reasonable static correction result,which provides the static correction quantity for subsequent calculation.The existing first break picking methods are mainly divided into automatic method and semi-automatic method.Generally,for the single and little noise data,most of automatic method can pick up the first break quickly and effectively.However,due to the existence of a large number of complex exploration areas,the number of seismic data with low signal-to-noise ratio is rapidly increasing.Automatic picking method cannot guarantee the quality of first break picking,that is to say,there are a large number of poor first breaks in the result.Therefore,human-computer interaction correction is taken in order to meet the needs,which called semi-automatic method.However,the processing process of semi-automatic method is very complicated and time-consuming,and the manual intervention has great subjectivity,so it is difficult to meet the accuracy and efficiency,which seriously restricts the process of static correction.In view of this,the purpose of this study is to explore an automatic first break picking method under low SNR seismic data.Considering the advantages of simple calculation and stable accuracy of energy ratio algorithm,In this study,the output of the algorithm is used as the result of picking up the first break,and then automatic poor first breaks detection and correction are carried out.The specific research contents are as follows:(1)Aiming at the poor first breaks which is initially picked up by energy ratio algorithm,an poor first breaks detection method based on CN-Fit algorithm is proposed.Firstly,considering that the existing data processing methods lead to the lack of context information,this study improves the data processing method by fusing seismic information and first break information;Secondly,considering that the poor first breaks are aggregated at most,the CN network is used to detect the continuous poor first breaks by the way of label sharing of multiple poor first breaks;Thirdly,aiming at the first break of the residual discontinuous poor first breaks,the means of nonlinear regression(Fitting)method is used for further detection.Finally,The poor first breaks from the above combination algorithm are further filtered to obtain the final correct first breaks.(2)After eliminating the poor first breaks,the gathers need to be picked up.Considering that in a seismic image,The first breaks is always shown a dividing line,so the picking problem is equivalent to the image segmentation problem in this study,and a light-weight encoder decoder network ED-light is proposed to pick up the first break.Secondly,considering the problem of limited amount of near shot data,data enhancement is used to optimize the model;Finally,a series of post-processing methods are used to further optimize the picking results.Verified by real seismic data,for the poor first breaks detection task,CN-Fit algorithm improves the F1 score by 3.91% and the detection efficiency by 5.78 times compared with the existing algorithms;For the poor first breaks correction task,ED-light improves the accuracy by 2.6% compared with the existing network,and the edge detail processing is better than other classical networks;The overall efficiency of the first breaks detection and correction process is about 24 times higher than the existing human-computer interaction method.In a comprehensive consideration,this algorithm is feasible and effective for automatic poor first breaks correction,which can replace manual detection and correction to a certain extent,and further improve the efficiency of seismic data processing.