Deep learning network optimization and hyperparameter tuning for seismic lithofacies classification

2021 ◽  
Vol 40 (7) ◽  
pp. 514-523
Author(s):  
Michael Jervis ◽  
Mingliang Liu ◽  
Robert Smith
Keyword(s):  
Deep Learning ◽  
Seismic Data ◽  
Network Performance ◽  
Random Search ◽  
Oil Field ◽  
Neural Network Approach ◽  
Data Representations ◽  
Very Fast Simulated Annealing ◽  
Deep Learning Network ◽  
Well Data

Deep learning is increasingly being applied in many aspects of seismic processing and interpretation. Here, we look at a deep convolutional neural network approach to multiclass seismic lithofacies characterization using well logs and seismic data. In particular, we focus on network performance and hyperparameter tuning. Several hyperparameter tuning approaches are compared, including true and directed random search methods such as very fast simulated annealing and Bayesian hyperparameter optimization. The results show that improvements in predictive capability are possible by using automatic optimization compared with manual parameter selection. In addition to evaluating the prediction accuracy's sensitivity to hyperparameters, we test various types of data representations. The choice of input seismic data can significantly impact the overall accuracy and computation speed of the optimized networks for the classification challenge under consideration. This is validated on a 3D synthetic seismic lithofacies example with acoustic and lithologic properties based on real well data and structure from an onshore oil field.

scholarly journals Deep Learning Network Intensification for Preventing Noisy-Labeled Samples for Remote Sensing Classification

2021 ◽  
Vol 13 (9) ◽  
pp. 1689
Author(s):  
Chuang Lin ◽  
Shanxin Guo ◽  
Jinsong Chen ◽  
Luyi Sun ◽  
Xiaorou Zheng ◽  
...  
Keyword(s):  
Remote Sensing ◽  
Deep Learning ◽  
Network Performance ◽  
Aerial Image ◽  
Label Noise ◽  
Remote Sensing Classification ◽  
Learning Network ◽  
Training Samples ◽  
Deep Learning Network ◽  
The Impact

The deep-learning-network performance depends on the accuracy of the training samples. The training samples are commonly labeled by human visual investigation or inherited from historical land-cover or land-use maps, which usually contain label noise, depending on subjective knowledge and the time of the historical map. Helping the network to distinguish noisy labels during the training process is a prerequisite for applying the model for training across time and locations. This study proposes an antinoise framework, the Weight Loss Network (WLN), to achieve this goal. The WLN contains three main parts: (1) the segmentation subnetwork, which any state-of-the-art segmentation network can replace; (2) the attention subnetwork (λ); and (3) the class-balance coefficient (α). Four types of label noise (an insufficient label, redundant label, missing label and incorrect label) were simulated by dilate and erode processing to test the network’s antinoise ability. The segmentation task was set to extract buildings from the Inria Aerial Image Labeling Dataset, which includes Austin, Chicago, Kitsap County, Western Tyrol and Vienna. The network’s performance was evaluated by comparing it with the original U-Net model by adding noisy training samples with different noise rates and noise levels. The result shows that the proposed antinoise framework (WLN) can maintain high accuracy, while the accuracy of the U-Net model dropped. Specifically, after adding 50% of dilated-label samples at noise level 3, the U-Net model’s accuracy dropped by 12.7% for OA, 20.7% for the Mean Intersection over Union (MIOU) and 13.8% for Kappa scores. By contrast, the accuracy of the WLN dropped by 0.2% for OA, 0.3% for the MIOU and 0.8% for Kappa scores. For eroded-label samples at the same level, the accuracy of the U-Net model dropped by 8.4% for OA, 24.2% for the MIOU and 43.3% for Kappa scores, while the accuracy of the WLN dropped by 4.5% for OA, 4.7% for the MIOU and 0.5% for Kappa scores. This result shows that the antinoise framework proposed in this paper can help current segmentation models to avoid the impact of noisy training labels and has the potential to be trained by a larger remote sensing image set regardless of the inner label error.

Application of deep learning for seismic horizon interpretation

2019 ◽  
Vol 59 (1) ◽  
pp. 426
Author(s):  
James Lowell ◽  
Jacob Smith
Keyword(s):  
Machine Learning ◽  
Deep Learning ◽  
Test Data ◽  
Seismic Data ◽  
Training Data ◽  
Learning Networks ◽  
Learning Network ◽  
Deep Learning Network

The interpretation of key horizons on seismic data is an essential but time-consuming part of the subsurface workflow. This is compounded when surfaces need to be re-interpreted on variations of the same data, such as angle stacks, 4D data, or reprocessed data. Deep learning networks, which are a subset of machine learning, have the potential to automate this reinterpretation process, and significantly increase the efficiency of the subsurface workflow. This study investigates whether a deep learning network can learn from a single horizon interpretation in order to identify that event in a different version of the same data. The results were largely successful with the target horizon correctly identified in an alternative offset stack, and was correctly repositioned in areas where there was misalignment between the training data and the test data.

A partial convolution-based deep-learning network for seismic data regularization1

2020 ◽  
Vol 145 ◽  
pp. 104609
Author(s):  
Shulin Pan ◽  
Kai Chen ◽  
Jingyi Chen ◽  
Ziyu Qin ◽  
Qinghui Cui ◽  
...  
Keyword(s):  
Deep Learning ◽  
Seismic Data ◽  
Learning Network ◽  
Deep Learning Network

scholarly journals High-resolution reservoir characterization using deep learning-aided elastic full-waveform inversion: The North Sea field data example

Geophysics ◽  
2020 ◽  
Vol 85 (4) ◽  
pp. WA137-WA146
Author(s):  
Zhen-dong Zhang ◽  
Tariq Alkhalifah
Keyword(s):  
Deep Learning ◽  
High Resolution ◽  
North Sea ◽  
Seismic Data ◽  
Field Data ◽  
Reservoir Characterization ◽  
Waveform Inversion ◽  
Oil Field ◽  
Full Waveform Inversion ◽  
Full Waveform

Reservoir characterization is an essential component of oil and gas production, as well as exploration. Classic reservoir characterization algorithms, deterministic and stochastic, are typically based on stacked images and rely on simplifications and approximations to the subsurface (e.g., assuming linearized reflection coefficients). Elastic full-waveform inversion (FWI), which aims to match the waveforms of prestack seismic data, potentially provides more accurate high-resolution reservoir characterization from seismic data. However, FWI can easily fail to characterize deep-buried reservoirs due to illumination limitations. We have developed a deep learning-aided elastic FWI strategy using observed seismic data and available well logs in the target area. Five facies are extracted from the well and then connected to the inverted P- and S-wave velocities using trained neural networks, which correspond to the subsurface facies distribution. Such a distribution is further converted to the desired reservoir-related parameters such as velocities and anisotropy parameters using a weighted summation. Finally, we update these estimated parameters by matching the resulting simulated wavefields to the observed seismic data, which corresponds to another round of elastic FWI aided by the a priori knowledge gained from the predictions of machine learning. A North Sea field data example, the Volve Oil Field data set, indicates that the use of facies as prior knowledge helps resolve the deep-buried reservoir target better than the use of only seismic data.

scholarly journals A deep learning network for estimation of seismic local slopes

2020 ◽  
Author(s):  
Wei-Lin Huang ◽  
Fei Gao ◽  
Jian-Ping Liao ◽  
Xiao-Yu Chuai
Keyword(s):  
Deep Learning ◽  
Seismic Data ◽  
Seismic Profile ◽  
Structural Features ◽  
Seismic Velocities ◽  
Local Slope ◽  
Learning Network ◽  
Local Window ◽  
Deep Learning Network ◽  
Fully Connected

AbstractThe local slopes contain rich information of the reflection geometry, which can be used to facilitate many subsequent procedures such as seismic velocities picking, normal move out correction, time-domain imaging and structural interpretation. Generally the slope estimation is achieved by manually picking or scanning the seismic profile along various slopes. We present here a deep learning-based technique to automatically estimate the local slope map from the seismic data. In the presented technique, three convolution layers are used to extract structural features in a local window and three fully connected layers serve as a classifier to predict the slope of the central point of the local window based on the extracted features. The deep learning network is trained using only synthetic seismic data, it can however accurately estimate local slopes within real seismic data. We examine its feasibility using simulated and real-seismic data. The estimated local slope maps demonstrate the successful performance of the synthetically-trained network.

scholarly journals UAVs in rail damage image diagnostics supported by deep-learning networks

2021 ◽  
Vol 11 (1) ◽  
pp. 339-348
Author(s):  
Piotr Bojarczak ◽  
Piotr Lesiak
Keyword(s):  
Deep Learning ◽  
Head Width ◽  
Learning Networks ◽  
Image Brightness ◽  
Learning Network ◽  
Rail Head ◽  
Image Recording ◽  
Python Language ◽  
Deep Learning Network ◽  
Efficiency Rate

Abstract The article uses images from Unmanned Aerial Vehicles (UAVs) for rail diagnostics. The main advantage of such a solution compared to traditional surveys performed with measuring vehicles is the elimination of decreased train traffic. The authors, in the study, limited themselves to the diagnosis of hazardous split defects in rails. An algorithm has been proposed to detect them with an efficiency rate of about 81% for defects not less than 6.9% of the rail head width. It uses the FCN-8 deep-learning network, implemented in the Tensorflow environment, to extract the rail head by image segmentation. Using this type of network for segmentation increases the resistance of the algorithm to changes in the recorded rail image brightness. This is of fundamental importance in the case of variable conditions for image recording by UAVs. The detection of these defects in the rail head is performed using an algorithm in the Python language and the OpenCV library. To locate the defect, it uses the contour of a separate rail head together with a rectangle circumscribed around it. The use of UAVs together with artificial intelligence to detect split defects is an important element of novelty presented in this work.

Sign in / Sign up

Export Citation Format

Share Document