Automatic picking of surface-wave dispersion curves with an image segmentation method

Abstract

The surface-wave method is a widely used technique for shallow subsurface exploration, and the extraction of the dispersion curve is one of the most important steps in the surface-wave method. Traditionally, this extraction of surface-wave dispersion curves heavily relies on manual or semi-manual picking, which is both time-consuming and prone to human error, especially when dealing with large datasets. Recent developments in machine learning algorithms have provided a promising way for the automated extraction of surface-wave dispersion curves. We present a random forest (RF) algorithm designed for the automatic extraction of surface-wave dispersion curves. In this approach, the extraction task is conceptualized as an image segmentation problem, enabling a rapid and accurate extraction of dispersion curves from dispersion energy images. We generate a dataset of 1800 models and their corresponding dispersion images. The proposed method is tested on both the noise-free and noisy datasets contaminated by Gaussian noise. Synthetic results demonstrate that our proposed method achieves relatively high accuracy and efficiency in the automatic extraction of surface-wave dispersion curves. We further analyze the impact of tuning parameters, including the number and depth of random-forest trees in the proposed algorithm on its performance and choose the best parameters in our study. Finally, the trained RF model is applied to two field datasets, which confirms the validity of our proposed RF method.