Automated arrival-time picking using a pixel-level network

Geophysics ◽  
2020 ◽  
Vol 85 (5) ◽  
pp. V415-V423
Author(s):  
Yuanyuan Ma ◽  
Siyuan Cao ◽  
James W. Rector ◽  
Zhishuai Zhang
Keyword(s):  
Seismic Data ◽  
Arrival Time ◽  
Spatial Information ◽  
Training Data ◽  
Convolutional Network ◽  
Data Set ◽  
Arrival Times ◽  
Data Volume ◽  
Borehole Seismic ◽  
Arrival Time Picking

Arrival-time picking is an essential step in seismic processing and imaging. The explosion of seismic data volume requires automated arrival-time picking in a faster and more reliable way than existing methods. We have treated arrival-time picking as a binary image segmentation problem and used an improved pixel-wise convolutional network to pick arrival times automatically. Incorporating continuous spatial information in training enables us to preserve the arrival-time correlation between nearby traces, thus helping to reduce the risk of picking outliers that are common in a traditional trace-by-trace picking method. To train the network, we first convert seismic traces into gray-scale images. Image pixels before manually picked arrival times are labeled with zeros, and those after are tagged with ones. After training and validation, the network automatically learns representative features and generates a probability map to predict the arrival time. We apply the network to a field microseismic data set that was not used for training or validation to test the performance of the method. Then, we analyze the effects of training data volume and signal-to-noise ratio on our autopicking method. We also find the difference between 1D and 2D training data with borehole seismic data. Microseismic and borehole seismic data indicate the proposed network can improve efficiency and accuracy over traditional automated picking methods.

First-arrival picking with a U-net convolutional network

Geophysics ◽  
2019 ◽  
Vol 84 (6) ◽  
pp. U45-U57 ◽  
Author(s):  
Lianlian Hu ◽  
Xiaodong Zheng ◽  
Yanting Duan ◽  
Xinfei Yan ◽  
Ying Hu ◽  
...  
Keyword(s):  
Seismic Data ◽  
Synthetic Data ◽  
Training Data ◽  
Convolutional Network ◽  
Data Set ◽  
Waveform Data ◽  
Near Surface ◽  
Arrival Times ◽  
Seismic Waveform ◽  
First Arrival

In exploration geophysics, the first arrivals on data acquired under complicated near-surface conditions are often characterized by significant static corrections, weak energy, low signal-to-noise ratio, and dramatic phase change, and they are difficult to pick accurately with traditional automatic procedures. We have approached this problem by using a U-shaped fully convolutional network (U-net) to first-arrival picking, which is formulated as a binary segmentation problem. U-net has the ability to recognize inherent patterns of the first arrivals by combining attributes of arrivals in space and time on data of varying quality. An effective workflow based on U-net is presented for fast and accurate picking. A set of seismic waveform data and their corresponding first-arrival times are used to train the network in a supervised learning approach, then the trained model is used to detect the first arrivals for other seismic data. Our method is applied on one synthetic data set and three field data sets of low quality to identify the first arrivals. Results indicate that U-net only needs a few annotated samples for learning and is able to efficiently detect first-arrival times with high precision on complicated seismic data from a large survey. With the increasing training data of various first arrivals, a trained U-net has the potential to directly identify the first arrivals on new seismic data.

Arrival-time picking of microseismic events based on MSNet

Geophysics ◽  
2021 ◽  
pp. 1-66
Author(s):  
Guanqun Sheng ◽  
Shuangyu Yang ◽  
Xiaolong Guo ◽  
Xingong Tang
Keyword(s):  
Arrival Time ◽  
Spatial Information ◽  
Synthetic Data ◽  
Microseismic Monitoring ◽  
Monitoring Data ◽  
Convolutional Network ◽  
Low Snr ◽  
Microseismic Data ◽  
Microseismic Events ◽  
Arrival Time Picking

Arrival-time picking of microseismic events is a critical procedure in microseismic data processing. However, as field monitoring data contain many microseismic events with low signal-to-noise ratios (SNRs), traditional arrival-time picking methods based on the instantaneous characteristics of seismic signals cannot meet the picking accuracy and efficiency requirements of microseismic monitoring owing to the large volume of monitoring data. Conversely, methods based on deep neural networks can significantly improve arrival-time picking accuracy and efficiency in low-SNR environments. Therefore, we propose a deep convolutional network that combines the U-net and DenseNet approaches to pick arrival times automatically. This novel network, called MSNet not only retains the spatial information of any input signal or profile based on the U-net, but also extracts and integrates more essential features of events and non-events through dense blocks, thereby further improving the picking accuracy and efficiency. An effective workflow is developed to verify the superiority of the proposed method. First, we describe the structure of MSNet and the workflow of the proposed picking method. Then, datasets are constructed using variable microseismic traces from field microseismic monitoring records and from the finite-difference forward modeling of microseismic data to train the network. Subsequently, hyperparameter tuning is conducted to optimize the MSNet. Finally, we test the MSNet using modeled signals with different SNRs and field microseismic data from different monitoring areas. By comparing the picking results of the proposed method with the results of U-net and short-term average and long-term average (STA/LTA) methods, the effectiveness of the proposed method is verified. The arrival picking results of synthetic data and microseismic field data show that the proposed network has increased adaptability and can achieve high accuracy for picking the arrival-time of microseismic events.

scholarly journals Fault Detection Based on Fully Convolutional Networks (FCN)

2021 ◽  
Vol 9 (3) ◽  
pp. 259
Author(s):  
Jizhong Wu ◽  
Bo Liu ◽  
Hao Zhang ◽  
Shumei He ◽  
Qianqian Yang
Keyword(s):  
Fault Detection ◽  
Seismic Data ◽  
Binary Image ◽  
Training Data ◽  
Real Field ◽  
Convolutional Network ◽  
Efficient Manner ◽  
Data Set ◽  
Seismic Image ◽  
End To End

It is of great significance to detect faults correctly in continental sandstone reservoirs in the east of China to understand the distribution of remaining structural reservoirs for more efficient development operation. However, the majority of the faults is characterized by small displacements and unclear components, which makes it hard to recognize them in seismic data via traditional methods. We consider fault detection as an end-to-end binary image-segmentation problem of labeling a 3D seismic image with ones as faults and zeros elsewhere. Thus, we developed a fully convolutional network (FCN) based method to fault segmentation and used the synthetic seismic data to generate an accurate and sufficient training data set. The architecture of FCN is a modified version of the VGGNet (A convolutional neural network was named by Visual Geometry Group). Transforming fully connected layers into convolution layers enables a classification net to create a heatmap. Adding the deconvolution layers produces an efficient network for end-to-end dense learning. Herein, we took advantage of the fact that a fault binary image is highly biased with mostly zeros but only very limited ones on the faults. A balanced crossentropy loss function was defined to adjust the imbalance for optimizing the parameters of our FCN model. Ultimately, the FCN model was applied on real field data to propose that our FCN model can outperform conventional methods in fault predictions from seismic images in a more accurate and efficient manner.

Deep learning-driven velocity model building workflow

2019 ◽  
Vol 38 (11) ◽  
pp. 872a1-872a9 ◽  
Author(s):  
Mauricio Araya-Polo ◽  
Stuart Farris ◽  
Manuel Florez
Keyword(s):  
Deep Learning ◽  
Seismic Data ◽  
Model Building ◽  
Ground Truth ◽  
Velocity Model ◽  
Training Data ◽  
Quality Data ◽  
Generative Adversarial Network ◽  
Data Set ◽  
Velocity Models

Exploration seismic data are heavily manipulated before human interpreters are able to extract meaningful information regarding subsurface structures. This manipulation adds modeling and human biases and is limited by methodological shortcomings. Alternatively, using seismic data directly is becoming possible thanks to deep learning (DL) techniques. A DL-based workflow is introduced that uses analog velocity models and realistic raw seismic waveforms as input and produces subsurface velocity models as output. When insufficient data are used for training, DL algorithms tend to overfit or fail. Gathering large amounts of labeled and standardized seismic data sets is not straightforward. This shortage of quality data is addressed by building a generative adversarial network (GAN) to augment the original training data set, which is then used by DL-driven seismic tomography as input. The DL tomographic operator predicts velocity models with high statistical and structural accuracy after being trained with GAN-generated velocity models. Beyond the field of exploration geophysics, the use of machine learning in earth science is challenged by the lack of labeled data or properly interpreted ground truth, since we seldom know what truly exists beneath the earth's surface. The unsupervised approach (using GANs to generate labeled data)illustrates a way to mitigate this problem and opens geology, geophysics, and planetary sciences to more DL applications.

Structural crack detection using deep learning–based fully convolutional networks

2019 ◽  
Vol 22 (16) ◽  
pp. 3412-3419 ◽  
Author(s):  
Xiao-Wei Ye ◽  
Tao Jin ◽  
Peng-Yu Chen
Keyword(s):  
Edge Detection ◽  
Structural Damage ◽  
Crack Detection ◽  
Training Data ◽  
Detection Methods ◽  
Crack Identification ◽  
Potential Threat ◽  
Online Data ◽  
Convolutional Network ◽  
Data Set

Cracks are a potential threat to the safety and endurance of civil infrastructures, and therefore, careful and regular structural crack inspection is needed during their long-term service periods. Many image-processing approaches have been developed for structural crack detection. However, like traditional edge detection algorithms, these methods are easily disturbed by the environmental effect. Convolutional neural networks are newly developed methods and have excellent performances in the image-classification tasks. This study proposes a fully convolutional network called Ci-Net for structural crack identification. Pixel-level labeled image training data are obtained from the online data set. Four indices are adopted to evaluate the performance of the trained Ci-Net. Crack images from an indoor concrete beam test are adopted for validation of its structural crack recognition capacity. The recognition results are also compared with those obtained by the edge detection methods. It indicates that Ci-Net exhibits a better performance over the edge detection methods in structural damage detection.

SC-PSNET: A deep neural network for automatic P- and S-phase detection and arrival-time picker using 1C recordings

Geophysics ◽  
2020 ◽  
Vol 85 (4) ◽  
pp. U87-U98
Author(s):  
Jing Zheng ◽  
Jerry M. Harris ◽  
Dongzhuo Li ◽  
Badr Al-Rumaih
Keyword(s):  
Neural Network ◽  
Data Processing ◽  
Seismic Data ◽  
Arrival Time ◽  
Time Lapse ◽  
S Phase ◽  
Training Data ◽  
Phase Detector ◽  
Phase Detection ◽  
Seismic Data Processing

It is important to autopick an event’s arrival time and classify the corresponding phase for seismic data processing. Traditional arrival-time picking algorithms usually need 3C seismograms to classify event phase. However, a large number of borehole seismic data sets are recorded by arrays of hydrophones or distributed acoustic sensing elements whose sensors are 1C and cannot be analyzed for particle motion or phase polarization. With the development of deep learning techniques, researchers have tried data mining with the convolutional neural network (CNN) for seismic phase autopicking. In the previous work, CNN was applied to process 3C seismograms to detect phase and pick arrivals. We have extended this work to process 1C seismic data and focused on two main points. One is the effect of the label vector on the phase detection performance. The other is to propose an architecture to deal with the challenge from the insufficiency of training data in the coverage of different scenarios of [Formula: see text] ratios. Two novel points are summarized after this analysis. First, the width of the label vector can be designed through signal time-frequency analysis. Second, a combination of CNN and recurrent neural network architecture is more suitable for designing a P- and S-phase detector to deal with the challenge from the insufficiency of training data for 1C recordings in time-lapse seismic monitoring. We perform experiments and analysis using synthetic and field time-lapse seismic recordings. The experiments show that it is effective for 1C seismic data processing in time-lapse monitoring surveys.

scholarly journals A review and appraisal of arrival-time picking methods for downhole microseismic data

Geophysics ◽  
2016 ◽  
Vol 81 (2) ◽  
pp. KS71-KS91 ◽  
Author(s):  
Jubran Akram ◽  
David W. Eaton
Keyword(s):  
Arrival Time ◽  
Hybrid Methods ◽  
Optimal Parameter ◽  
Signal To Noise Ratio ◽  
Information Criterion ◽  
High Signal ◽  
Data Set ◽  
Single Level ◽  
Microseismic Data ◽  
Arrival Time Picking

We have evaluated arrival-time picking algorithms for downhole microseismic data. The picking algorithms that we considered may be classified as window-based single-level methods (e.g., energy-ratio [ER] methods), nonwindow-based single-level methods (e.g., Akaike information criterion), multilevel- or array-based methods (e.g., crosscorrelation approaches), and hybrid methods that combine a number of single-level methods (e.g., Akazawa’s method). We have determined the key parameters for each algorithm and developed recommendations for optimal parameter selection based on our analysis and experience. We evaluated the performance of these algorithms with the use of field examples from a downhole microseismic data set recorded in western Canada as well as with pseudo-synthetic microseismic data generated by adding 100 realizations of Gaussian noise to high signal-to-noise ratio microseismic waveforms. ER-based algorithms were found to be more efficient in terms of computational speed and were therefore recommended for real-time microseismic data processing. Based on the performance on pseudo-synthetic and field data sets, we found statistical, hybrid, and multilevel crosscorrelation methods to be more efficient in terms of accuracy and precision. Pick errors for S-waves are reduced significantly when data are preconditioned by applying a transformation into ray-centered coordinates.

A seismic to simulation unconventional workflow using automated fault-detection attributes

2017 ◽  
Vol 5 (3) ◽  
pp. SJ41-SJ48 ◽  
Author(s):  
Jesse Lomask ◽  
Luisalic Hernandez ◽  
Veronica Liceras ◽  
Amit Kumar ◽  
Anna Khadeeva
Keyword(s):  
Seismic Data ◽  
Spatial Information ◽  
Spatial Location ◽  
Statistical Information ◽  
Natural Fracture ◽  
Fracture Networks ◽  
Data Set ◽  
Fault Indicator ◽  
Well Data ◽  
Additional Constraints

Natural fracture networks (NFNs) are used in unconventional reservoir simulators to model pressure and saturation changes in fractured rocks. These fracture networks are often derived from well data or well data combined with a variety of seismic-derived attributes to provide spatial information away from the wells. In cases in which there is a correlation between faults and fractures, the use of a fault indicator can provide additional constraints on the spatial location of the natural fractures. We use a fault attribute based on fault-oriented semblance as a secondary conditioner for the generation of NFNs. In addition, the distribution of automatically extracted faults from the fault-oriented semblance is used to augment the well-derived statistics for natural fracture generation. Without the benefit of this automated fault-extraction solution, to manually extract the fault-statistical information from the seismic data would be prohibitively tedious and time consuming. Finally, we determine, on a 3D field unconventional data set, that the use of fault-oriented semblance results in simulations that are significantly more geologically reasonable.

The use of drill‐bit energy as a downhole seismic source

Geophysics ◽  
1991 ◽  
Vol 56 (5) ◽  
pp. 628-634 ◽  
Author(s):  
J. W. Rector ◽  
B. P. Marion
Keyword(s):  
Seismic Energy ◽  
Seismic Source ◽  
Drilling Process ◽  
Drill Bit ◽  
Data Set ◽  
Pilot Signal ◽  
Data Volume ◽  
Vsp Data ◽  
Borehole Seismic ◽  
Incoherent Noise

A new wellbore seismic technique uses the vibrations produced by a drill bit while drilling as a downhole seismic energy source. The technique is described as “inverse” VSP because the source and receiver positions of conventional VSP are reversed. No downhole instrumentation is required to obtain the data and the data recording does not interfere with the drilling process. These characteristics offer a method by which borehole seismic data can be acquired, processed, and interpreted while drilling. Interchanging the conventional VSP source and receiver positions improves the efficiency of recording multioffset surveys for imaging a 3-D data volume in the borehole vicinity. The continuous signals generated by the drill bit are recorded by a pilot sensor attached to the top of the drillstring and by receivers located at selected positions around the borehole. The pilot signal is crosscorrelated with the receiver signals to compute traveltimes of the arrivals and to attenuate incoherent noise. Deconvolution and time shifts of the pilot signal compensate for the effects of propagation from the drill bit to the top of the drillstring. By repeating this process for an interval of the well, a VSP‐equivalent data set is generated. Results from a test well demonstrate that the processed drill‐bit data are comparable to conventional VSP data.

A method for adequate selection of training data sets to reconstruct seismic field data using a convolutional U-Net

Geophysics ◽  
2021 ◽  
pp. 1-103
Author(s):  
Jiho Park ◽  
Jihun Choi ◽  
Soon Jee Seol ◽  
Joongmoo Byun ◽  
Young Kim
Keyword(s):  
Data Analysis ◽  
Seismic Data ◽  
Field Data ◽  
Training Data ◽  
Data Reconstruction ◽  
Data Sets ◽  
Training Set ◽  
Data Set ◽  
Seismic Data Reconstruction ◽  
Target Data

Deep learning (DL) methods are recently introduced for seismic signal processing. Using DL methods, many researchers have adopted these novel techniques in an attempt to construct a DL model for seismic data reconstruction. The performance of DL-based methods depends heavily on what is learned from the training data. We focus on constructing the DL model that well reflect the features of target data sets. The main goal is to integrate DL with an intuitive data analysis approach that compares similar patterns prior to the DL training stage. We have developed a two-sequential method consisting of two stage: (i) analyzing training and target data sets simultaneously for determining target-informed training set and (ii) training the DL model with this training data set to effectively interpolate the seismic data. Here, we introduce the convolutional autoencoder t-distributed stochastic neighbor embedding (CAE t-SNE) analysis that can provide the insight into the results of interpolation through the analysis of both the training and target data sets prior to DL model training. The proposed method were tested with synthetic and field data. Dense seismic gathers (e.g. common-shot gathers; CSGs) were used as a labeled training data set, and relatively sparse seismic gather (e.g. common-receiver gathers; CRGs) were reconstructed in both cases. The reconstructed results and SNRs demonstrated that the training data can be efficiently selected using CAE t-SNE analysis and the spatial aliasing of CRGs was successfully alleviated by the trained DL model with this training data, which contain target features. These results imply that the data analysis for selecting target-informed training set is very important for successful DL interpolation. Additionally, the proposed analysis method can also be applied to investigate the similarities between training and target data sets for another DL-based seismic data reconstruction tasks.

Sign in / Sign up

Export Citation Format

Share Document