Long-Short-Term Memory in Active Wavefield Geophysical Methods

The PhD thesis discusses the application of Long Short-Term Memory (LSTM) networks in active wavefield geophysical methods. In this work we emphasizes the advantages of Deep Learning (DL) techniques in geophysics, such as improved accuracy, handling complex datasets, and reducing subjectivity. The work explores the suitability of LSTM networks compared to Convolutional Neural Networks (CNNs) in some geophysical applications. The research aims to comprehensively investigate the strengths, limitations, and potential of recurrent neurons, particularly LSTM, in active wavefield geophysics. LSTM networks have the ability to capture temporal dependencies and are well-suited for analyzing geophysical data with non-stationary behavior. They can process both time and frequency domain information, making them valuable for analyzing Seismic and Ground Penetrating Radar (GPR) data. The PhD thesis consists of five main chapters covering methodological development, regression, classification, data fusion, and frequency domain signal processing.