Automatic extraction of surface wave dispersion curves using unsupervised learning

Abstract

Extraction of the dispersion curves is an important step in surface wave imaging of subsurface shear wave velocity distribution. Accurate extraction of the dispersion curve is crucial for the precision of the inverted velocity, but this step requires experience and is time consuming. Therefore, automatic and accurate picking of dispersion curves can significant enhance the efficiency for surface wave prospecting. This paper presents an approach based on unsupervised learning algorithms to achieve automatic picking of surface wave dispersion curves of the fundamental mode. With this method, K-means clustering is used to generate the clustering points on the dispersion spectrum; clustering density analysis and principal component analysis (PCA) is used to exclude the false clustering points introduced by noise; then a curve fitting using the remaining clustering points generates the final dispersion curve. This method is applied to a noisy surface wave dataset acquired in an urban area, and the results verify its effectiveness.