About this webinar
Artificial Intelligence (AI) has become very popular for solving problems in seismic data processing. This involves the creation of a predictive model trained on labeled data. However, model outputs are as good as their training set. Poor quality training data results in poor performance of the AI model. Therefore, work by a geoscientist is critical for data preprocessing. This involves significant efforts in labeling the data and developing complex signal or image processing algorithms.
We propose the use of wavelets to simplify these workflows. For this talk, we demonstrate the use of wavelet transforms to enhance two common seismic processing tasks.
(1) Automatically pick the arrival times and duration of P- and S-waves: We use the continuous wavelet transform (CWT) to generate 2-D time-frequency maps of time series data, which can then be used as image inputs for deep convolutional neural networks (CNN). The ability of the CWT to simultaneously capture high frequency shorter duration and low frequency longer duration components in time series data makes the wavelet-based time-frequency representation particularly robust when paired with deep CNNs.
(2) Seismic trace analysis: Some key features used to define seismic attributes can be difficult to extract from the seismic traces due to noise, clutter, trends etc. Wavelet based multiresolution analysis is a powerful framework for extracting these relevant features from complex traces without adding any spatial distortions or artifacts. These features can be paired with AI models for classification. Our recent results obtained from this approach are promising for an automated seismic interpretation model, increasing the productivity of the interpreter by ~10x.