Seismic Data Processing in a Compressed Domain using Constrained Dictionary Learning

Abstract

oday's seismic exploration surveys generate vast amounts of data as larger areas are covered, and an increased number of sensors are employed. Consequently, seismic data have grown exponentially in size. Hence, such large data poses significant challenges for traditional processing and imaging methods. Early-stage compression of seismic data can be key element to overcoming storage and data transfer barriers. Moreover, applying seismic processing steps on compressed data instead of large data would not only save storage and transfer costs but could also lead to cost-effective alternatives compared to standard seismic processing. This doctoral thesis demonstrates the effectiveness of dictionary learning-based methods in compressing large seismic data and enabling key processing steps, such as wavefield separation and deghosting, to be carried out directly in the compressed domain. While sparse transforms have previously been utilized for some seismic processing steps, there has been no prior proposal for methods aimed at compressing seismic data and simultaneously processing them in the compressed domain. Thus, this thesis shows that such methods can significantly reduce costs related to data storage and transfer, and also bring computational cost reduction. Future research may focus on allowing other key processing steps in the compressed domain.