Paper Summary
Noise attenuation is a key processing step in a typical seismic imaging flow. Assessing the quality of the denoise process to ensure that noise has been sufficiently attenuated without signal distortion is important. With the increase in the volume of seismic data, the QC is becoming a bottleneck for many processing project. This abstract proposes a machine learning solution that can automate the QC. Given the availability of some training seismic data lines that have been through optimal, mild and harsh filtering, the task of identifying where in the data the filtering is suboptimal is similar to a supervised classification problem. Attributes are computed from the training lines and are then used to train a support vector machine (SVM) classifier. The trained classifier is used to find the filtering class for any seismic ensemble in the survey leading to a fast QC report. The solution is tested on a full-scale processing project to QC the denoise step prior to wavefield separation in marine seismic data. Tests show encouraging results and the solution was able to predict locations in the data with clear residual noise. false positives are an issue but their rate can be reduced by using informative attributes.