Surface Wave Multi-parameter Waveform Inversion Imaging Based on Partial Convolutional Siamese Network

Surface wave waveform inversion (SWI), recognized as a high-precision technique for shallow subsurface imaging, is extensively employed in the detection of underground structures. However, SWI faces several challenges that hinder its broader application in engineering, including the absence of low-frequency components, limited sensitivity to longitudinal waves and density parameters, and a pronounced reliance on initial models. To mitigate the Cycle-Skipping phenomenon stemming from low-frequency information loss and initial model dependency, this study introduces a novel multi-parameter waveform inversion approach for surface waves, leveraging a partial convolutional Siamese network (SIAMPCNN-SWI). This approach ingests observational and initial data into a shared-parameter partial convolutional Siamese network to extract multi-scale features, thereby enhancing the recovery of low-frequency information and reducing the dependency on initial models. Subsequently, the feature discrepancies extracted are utilized as loss functions for estimation, and automatic differentiation is harnessed for backpropagation, enabling the high-precision reconstruction of shallow subsurface structures. Numerical simulations and field data validations demonstrate that this method outperforms traditional approaches in terms of inversion accuracy, stability, and computational efficiency for multi-parameter inversion tasks. This method offers a robust solution for shallow subsurface exploration under complex geological conditions and holds significant potential for broader application.