Convolutional neural networks for microseismic waveform classification and arrival picking

Сверточные нейронные сети для классификации микросейсмических сигналов и определения времен вступления
Guoyin Zhang, Chengyan Lin, Yangkang Chen
2020-06-02

arrival pickingcontinuous wavelet transformconvolutional neural networklow signal-to-noise ratiomicroseismic waveform classification
ABSTRACT Microseismic data have a low signal-to-noise ratio (S/N). Existing waveform classification and arrival-picking methods are not effective enough for noisy microseismic data with low S/N. We have adopted a novel antinoise classifier for waveform classification and arrival picking by combining the continuous wavelet transform (CWT) and the convolutional neural network (CNN). The proposed CWT-CNN classifier is applied to synthetic and field microseismic data sets. Results show that CWT-CNN classifier has much better performance than the basic deep feedforward neural network (DNN), especially for microseismic data with low S/N. The CWT-CNN classifier has a shallow network architecture and small learning data set, and it can be trained quickly for different data sets. We have determined why CWT-CNN has better performance for noisy microseismic data. CWT can decompose the microseismic data into time-frequency spectra, where effective signals and interfering noise are easier to distinguish. With the help of CWT, CNN can focus on the specific frequency components to extract useful features and build a more effective classifier.
1
A CWT-CNN classifier was developed for microseismic waveform classification and arrival picking under low signal-to-noise conditions.
2
CWT improves noise robustness by separating signal and interference in time-frequency space, allowing CNNs to extract informative frequency-specific features.
3
On synthetic and field datasets, CWT-CNN substantially outperformed a basic deep feedforward neural network, particularly for low-S/N data.
4
The approach uses a shallow architecture and small training dataset, enabling rapid retraining for different microseismic datasets.
5
The method combines continuous wavelet transforms with convolutional neural networks to represent waveforms through time-frequency spectra.

synthetic and field microseismic data

noise-robust microseismic waveform classification and arrival picking, including extraction of discriminative time–frequency features under low signal-to-noise conditions

Publication Details
Publication Date
2020-06-02
Journal
Publisher
ISSN
Access Type
Author Information
Authors
Guoyin Zhang
Chengyan Lin
Yangkang Chen
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%