← Back to Sessions

WITHDRAWN Domain Confusion in Dispersions Picking Based on Neural Network and the Learning Features From Dispersion Spectrograms

Description: 

WITHDRAWN Ambient noise tomography has been widely used to obtain the shear-wave velocity for underground structure of the Earth, and there is a key point in dispersion spectrogram where people need to pick dispersion curves manually. By the frequency-Bessel (F-J) transform, the dispersion spectrogram not only provides reliable fundamental dispersion curve, but also rich higher modes of dispersion curves. However, manual picking will consume much time and energy for researchers. Consequently, using neural network to pick dispersions automatically has great significance. In the field of computer vision, the feature of data from different source domains has a certain degree of difference. If we ignore these differences in the data, the trained neural network will have an issue of data limitation, and the performance will be weak. By analyzing lots of dispersion spectrograms, we have summarized the learning features of neural network from spectrograms. In addition, based on our findings we have applied domain confusion in training and testing neural network. After domain confusion, we can make one neural network learn dispersion spectrograms from multi sources, and its performance is stronger than before. The trained neural network can effectively process large number of test data and help us easily obtain more dispersion curves automatically. The proposed study can provide a deep insight into the denoising of dispersion spectrograms by neural network.

Session: From Earthquake Recordings to Empirical Ground-Motion Modelling [Poster Session]

Type: Poster

Date: 5/2/2024

Presentation Time: 08:00 AM (local time)

Presenting Author: Weibin

Student Presenter: No

Invited Presentation: 

Authors