Saw a practical example using open source DeepLab V3+ applied to transformed surface wave data (i.e., in frequency-wavenumber and Radon domain), followed by inversion to derive the subsurface shear-wave velocity structure.
Takeaway? Automated "picking" of dispersion curves in appropriate feature-engineering space, using deep learning algo can minimize the pain points out of data prep and processing (otherwise time-consuming and labor-intensive).
Possibility? DeepLab next-gen can be valuable pathway to extend the auto-picking scheme for multi-component P-S data quality assurance. Unlock the door to automation in harnessing multicomponent data, perform joint P-S processing and inversion for CCUS and energy storage projects in a cost-effective manner.
To stimulate further exchange, here is the DeepLab Github reference:
https://github.com/tensor...ter/research/deeplab .