Multichannel Analysis of Surface Waves (MASW) is commonly used to determine theshallow subsurface velocity structure. The obtained velocities provide valuable information for foundation design and other geotechnical applications. The conventional method of picking surface-wave dispersion curves, which are subsequently inverted to obtain the desired velocity profiles, can be a labour-intensive task. In this thesis, we make use of recent advances in Deep Learning (DL), utilizing Convolutional Neural Networks (CNNs) to estimate dispersion curves directly from shot gathers or their corresponding dispersion spectra. For this, we first test the proposed approach on various synthetic data, successfully addressing challenges such as geological models with velocity inversion, different noise levels, and the estimation of the first higher mode. When applied to field data, our results indicate that training the CNN on dispersion spectra provides more accurate and reliable predictions compared to training on shot gathers directly. We improve the predictions using shot gathers as training data by modifying the original CNN architecture and applying transfer learning techniques. The enhanced CNN models demonstrate significant improvements, indicating that this approach could aid the analysis of MASW data. To fully realize the potential of this method, future efforts should focus on increasing the similarity between synthetic and field data, for example by incorporating realistic noise.
... Read more