Seismic data sample set optimization method and device, equipment and medium

Abstract

The embodiment of the invention discloses a seismic data sample set optimization method and device, equipment and a medium. The method comprises the steps that a seismic data sample set is acquired, and at least one seismic data sample is copied from the seismic data sample set; processing the at least one seismic data sample in a target processing mode to obtain a newly added seismic data sample; determining an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, and training a machine learning model based on the augmented seismic data sample set; the target processing mode comprises at least one of the following: overturning processing, noise adding processing, high-pass filtering processing and low-pass filtering processing. According to the technical scheme, the number of samples in the seismic data sample set is increased, and after the machine learning model is trained based on the increased seismic data sample set, the generalization ability and other performances of the machine learning model are improved.

Description

Technical Field

[0001] The present invention relates to the field of machine learning technology, and in particular to a method, device, equipment and medium for optimizing a seismic data sample set. Background Art

[0002] In recent years, the research on artificial intelligence technology represented by deep learning in the field of seismic exploration data processing and interpretation has become increasingly in-depth. The quality of the seismic data sample set directly affects the accuracy of the model's interpretation of seismic data. How to optimize the seismic data samples is a technical problem that technical personnel in this field urgently need to solve.

[0003] Since different seismic data are collected under different surface conditions or underground structures, the characteristics of the collected seismic data samples are very different. How to augment the smaller-scale and less feature-rich seismic data sample sets, improve their scale and feature richness, and improve the generalization ability of the trained network model is a technical problem that needs to be solved urgently. Summary of the invention

[0004] The present invention provides a method, device, equipment and medium for optimizing a seismic data sample set, which can perform augmentation processing on seismic data samples to improve the quantity and quality of the seismic data sample set, so that the augmented seismic data sample set can train a machine learning model with better generalization ability.

[0005] According to one aspect of the present invention, a method for optimizing a seismic data sample set is provided, the method comprising:

[0006] Acquire a seismic data sample set, and copy at least one seismic data sample from the seismic data sample set;

[0007] Processing the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample;

[0008] Determine an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented seismic data sample set;

[0009] The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing;

[0010] If the seismic data sample set is a first seismic data sample set for solving the regional segmentation problem, then the copying of at least one seismic data sample from the seismic data sample set includes: copying at least one first seismic data sample from the first seismic data sample set;

[0011] If the seismic data sample set is a second seismic data sample set for solving a classification problem, then, copying at least one seismic data sample from the seismic data sample set includes: in the second seismic data sample set, if the number of second seismic data samples corresponding to a target category is less than a quantity threshold, then copying at least one second seismic data sample in the second seismic data samples corresponding to the target category.

[0012] According to another aspect of the present invention, there is provided a device for optimizing a seismic data sample set, comprising:

[0013] A seismic data sample duplication module, used to obtain a seismic data sample set and to duplicate at least one seismic data sample from the seismic data sample set;

[0014] A seismic data sample processing module, used for processing the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample;

[0015] An augmented seismic data sample set determination module, used to determine an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented seismic data sample set;

[0016] The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing;

[0017] The seismic data sample duplication module comprises: a first seismic data sample duplication unit, configured to, if the seismic data sample set is a first seismic data sample set for solving the region segmentation problem, copy at least one first seismic data sample from the first seismic data sample set;

[0018] The seismic data sample copying module includes: a second seismic data sample copying unit, which is used to copy at least one second seismic data sample in the second seismic data samples corresponding to the target category in the second seismic data sample set if the seismic data sample set is a second seismic data sample set used to solve a classification problem and if the number of second seismic data samples corresponding to a target category is less than a quantity threshold.

[0019] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0020] at least one processor; and

[0021] a memory communicatively connected to the at least one processor; wherein,

[0022] The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can perform the method for optimizing the seismic data sample set described in any embodiment of the present invention.

[0023] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for optimizing a seismic data sample set described in any embodiment of the present invention when executed.

[0024] The technical solution of the embodiment of the present application includes: obtaining a seismic data sample set, copying at least one seismic data sample from the seismic data sample set; processing the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample; determining an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data sample, so as to train a machine learning model based on the augmented seismic data sample set; the target processing manner includes at least one of the following: flipping processing, noise adding processing, high-pass filtering processing and low-pass filtering processing; if the seismic data sample set is a first seismic data sample set for solving a regional segmentation problem, then copying at least one seismic data sample from the seismic data sample set includes: copying at least one first seismic data sample from the first seismic data sample set; if the seismic data sample set is a second seismic data sample set for solving a classification problem, then copying at least one seismic data sample from the seismic data sample set includes: in the second seismic data sample set, if the number of second seismic data samples corresponding to a target category is less than a number threshold, then copying at least one second seismic data sample in the second seismic data samples corresponding to the target category. This technical solution obtains new seismic data samples by processing the seismic data samples in a target processing manner, thereby increasing the quantity and quality of the seismic data samples, so that after training the machine learning model based on the augmented seismic data sample set, the generalization ability and other performance of the machine learning model are improved.

[0025] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0027] Figure 1 is a flow chart of a method for optimizing a seismic data sample set provided according to Embodiment 1 of the present application;

[0028] Figure 2 is a flow chart of a method for optimizing a seismic data sample set provided according to Embodiment 2 of the present application;

[0029] Figure 3 is a screenshot of the augmented software of the first seismic data sample set provided in Example 2 of the present application;

[0030] Figure 4 This is a schematic diagram of a software interface for optimizing the imbalance provided in the second embodiment of the present application;

[0031] Figure 5 This is a network structure diagram provided according to Embodiment 2 of the present application;

[0032] Figure 6 This is a schematic diagram of samples with high imbalance and their labels provided according to the second embodiment of the present application;

[0033] Figure 7 is a flow chart of a method for optimizing a seismic data sample set provided according to Embodiment 3 of the present application;

[0034] Figure 8 It is a screenshot of the software for augmenting the second seismic data sample provided in the third embodiment of the present application;

[0035] Fig. 9 It is a schematic diagram of the flipping processing of the horizon interpretation sample provided in the third embodiment of the present application;

[0036] Fig.10 is a schematic diagram of high-pass filtering processing of a horizon interpretation sample provided in Example 3 of the present application;

[0037] Fig.11 is a schematic diagram of high-pass filtering processing of seismic phase division samples provided in Example 3 of the present application;

[0038] Fig.12 It is a structural schematic diagram of a seismic data sample set optimization device provided according to the fourth embodiment of the present application;

[0039] Fig.13 It is a structural schematic diagram of an electronic device for implementing a method for optimizing a seismic data sample set according to an embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only embodiments of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0041] It should be noted that the terms "first", "second", "target", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0042] Embodiment 1

[0043] Figure 1 A flowchart of a method for optimizing a seismic data sample set is provided for the first embodiment of the present application. The present embodiment of the present application is applicable to the case of augmenting seismic data samples. The method can be executed by an optimization device for a seismic data sample set. The optimization device for a seismic data sample set can be implemented in the form of hardware and / or software. The optimization device for a seismic data sample set can be configured in an electronic device with data processing capabilities. Figure 1 As shown, the method includes:

[0044] S110, obtaining a seismic data sample set, and copying at least one seismic data sample from the seismic data sample set.

[0045] Among them, the seismic data sample set includes seismic data samples, and the seismic data samples in the seismic data sample set are used to train the machine learning model to obtain the trained machine learning model. The seismic data samples can be pre-labeled seismic data samples. Obviously, for different situations, the labels of seismic data samples will also be different. For example, if a seismic data sample reflects seismic data collected in stratum A, the corresponding label is stratum A; if a seismic data sample reflects seismic data collected in three areas B, C, and D, the corresponding label includes the area corresponding to each seismic data in the sample. Exemplarily, the seismic data sample can be composed of seismic amplitude data, each data is arranged in a matrix form, corresponding to a seismic trace and an observation point, respectively. The seismic data sample can be a two-dimensional data sample or a three-dimensional data sample. If it is a two-dimensional data sample, it reflects the seismic data collected on the geological profile; if it is a three-dimensional data sample, it reflects the seismic data collected in a certain area.

[0046] Specifically, the seismic data sample set includes multiple seismic data samples, and at least one seismic data sample can be selected from the seismic data sample set randomly or according to a selection condition; the selecting at least one seismic data sample according to the selection condition can be: according to the label of the sample, selecting the seismic data sample corresponding to the target label.

[0047] S120: Process the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample.

[0048] The target processing method refers to a method of changing the seismic data sample. After the seismic data sample is processed based on the target processing method, the processed seismic data sample will become a new seismic data sample. The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing.

[0049] Furthermore, flip processing refers to flipping the seismic data samples, specifically: if the seismic data is two-dimensional data, the distribution of the two-dimensional data is rotated, for example, rotating 180 degrees along the center point, etc., to obtain new seismic data samples.

[0050] Noise processing refers to adding noise to each seismic data in at least one seismic data sample, thereby obtaining a newly added seismic data sample.

[0051] High-pass filtering refers to filtering each seismic data in at least one seismic data sample, retaining the high-frequency part and removing the low-frequency part, so as to obtain a new seismic data sample.

[0052] Low-pass filtering refers to filtering each seismic data in at least one seismic data sample, retaining the low-frequency part and removing the high-frequency part to obtain a new seismic data sample.

[0053] Specifically, after obtaining at least one seismic data sample, the at least one seismic data sample is processed in different target processing modes, so that the newly added seismic data sample remains different from the original seismic data sample and other newly added seismic data samples.

[0054] S130, determining an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented seismic data sample set.

[0055] After obtaining the newly added seismic data samples, the seismic data sample set and the newly added seismic data samples are merged to obtain an augmented seismic data sample set; for example, the seismic data sample set includes 10 samples, the newly added seismic data samples include 5 samples, and the augmented seismic data sample set includes 15 samples. Further, a machine learning model is trained based on the augmented seismic data sample set to improve various performances of the machine learning model, such as the generalization ability.

[0056] In an embodiment of the present application, if the seismic data sample set is a first seismic data sample set for solving a regional segmentation problem, then, copying at least one seismic data sample from the seismic data sample set includes: copying at least one first seismic data sample from the first seismic data sample set; if the seismic data sample set is a second seismic data sample set for solving a classification problem, then, copying at least one seismic data sample from the seismic data sample set includes: in the second seismic data sample set, if the number of second seismic data samples corresponding to a target category is less than a quantity threshold, copying at least one second seismic data sample in the second seismic data samples corresponding to the target category.

[0057] The technical solution of the embodiment of the present application includes: obtaining a seismic data sample set, copying at least one seismic data sample from the seismic data sample set; processing the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample; determining an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data sample, so as to train a machine learning model based on the augmented seismic data sample set; the target processing manner includes at least one of the following: flipping processing, noise adding processing, high-pass filtering processing and low-pass filtering processing; if the seismic data sample set is a first seismic data sample set for solving a regional segmentation problem, then copying at least one seismic data sample from the seismic data sample set includes: copying at least one first seismic data sample from the first seismic data sample set; if the seismic data sample set is a second seismic data sample set for solving a classification problem, then copying at least one seismic data sample from the seismic data sample set includes: in the second seismic data sample set, if the number of second seismic data samples corresponding to a target category is less than a number threshold, then copying at least one second seismic data sample in the second seismic data samples corresponding to the target category. This technical solution obtains new seismic data samples by processing the seismic data samples in a target processing manner, thereby increasing the quantity and quality of the seismic data samples, so that after training the machine learning model based on the augmented seismic data sample set, the generalization ability and other performance of the machine learning model are improved.

[0058] Embodiment 2

[0059] Figure 2 This is a flowchart of a method for optimizing a seismic data sample set provided in the second embodiment of the present application. The present embodiment is optimized based on the above embodiment. The seismic data sample set in the embodiment of the present application is a first seismic data sample set for solving the regional segmentation problem.

[0060] like Figure 2 As shown, the method of the embodiment of the present application specifically includes the following steps:

[0061] S210, obtaining a first seismic data sample set, and copying at least one first seismic data sample from the first seismic data sample set.

[0062] In an embodiment of the present application, optionally, the seismic data sample set is a first seismic data sample set for solving the regional segmentation problem, the first seismic data sample is composed of seismic data arranged in a matrix form, the seismic data arranged in a matrix form corresponds to each region on a geological profile, and the label corresponding to the first seismic data sample includes the segmented region corresponding to each seismic data.

[0063] The first seismic data sample is a sample used to solve the regional segmentation problem, that is, the machine learning model trained based on the first seismic data sample is able to identify which segmentation regions are included in the seismic data input into the model. Exemplarily, taking the first seismic data sample in two-dimensional form as an example, the first seismic data sample is composed of a plurality of seismic data arranged in a matrix form (the seismic data are sample points of the sample), and each seismic data in the horizontal and vertical directions of the matrix can be seismic amplitude data collected at each observation point of each seismic channel. The seismic data arranged in matrix form in the first seismic data sample corresponds to each area on the geological profile, and the label corresponding to the first seismic data sample includes the segmented area corresponding to each seismic data. For example, a first seismic data sample includes 6 sample points at positions (1,1), (1,2), (1,3), (2,1), (2,2) and (2,3) (obviously there will be many actual sample points, and a small number of sample points are used here for example). The label of the first seismic data sample reflects that the sample points at positions (1,1) and (1,2) correspond to area A, the sample points at positions (1,3) and (2,1) correspond to area B, and the sample points at positions (2,2) and (2,3) correspond to area C.

[0064] Specifically, copying at least one first seismic data sample from the first seismic data sample set includes: randomly copying at least one first seismic data sample from the first seismic data sample set.

[0065] S220: Process the at least one first seismic data sample in a target processing manner to obtain a newly added seismic data sample.

[0066] The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing.

[0067] S230: Determine an augmented seismic data sample set according to the first seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented first seismic data sample set.

[0068] Specifically, training a machine learning model based on the augmented first seismic data sample set includes: training a machine learning model for identifying segmented areas based on the augmented first seismic data sample set, so that the machine learning model for identifying segmented areas can more accurately identify the segmented areas.

[0069] The technical solution of the embodiment of the present application includes: obtaining a first seismic data sample set, copying at least one first seismic data sample from the first seismic data sample set; processing the at least one first seismic data sample in a target processing manner to obtain a newly added seismic data sample; determining an augmented seismic data sample set based on the first seismic data sample set and the newly added seismic data sample, so as to train a machine learning model based on the augmented first seismic data sample set. The technical solution augments the first seismic data sample set to obtain an augmented first seismic data sample set, thereby achieving the effect of expanding the number of samples.

[0070] In a specific implementation process, based on the corresponding sample augmentation software ( Figure 3 The following are screenshots of the augmentation software for the first seismic data sample set:

[0071] 1) Select the sample set directory and label directory: All sample files and corresponding label files of the first seismic data sample set are stored in two folders named "train" and "label" in the same directory. You need to select the specific folders where the sample files and label files are located.

[0072] 2) Select sample dimension: Select whether the samples in the sample set to be augmented are two-dimensional samples or three-dimensional samples.

[0073] 3) Set the sample size: For two-dimensional samples, you need to set the number of channels and sample points; for three-dimensional samples, you need to set the number of lines, channels, and sample points.

[0074] 4) Select the label data type: Select whether the sample labels in the sample set to be augmented are integer or floating point.

[0075] 5) Select the sample augmentation processing method and set the execution times and parameters: This step first randomly selects one or more processing methods from the four sample augmentation processing methods of rotation (corresponding to flip processing), noise addition (corresponding to noise addition processing), high-frequency extraction (corresponding to high-pass filtering processing) and low-frequency extraction (corresponding to low-pass filtering processing), and then sets the execution times of each processing method, that is, the number of samples augmented by each processing method, and finally sets the sample augmentation processing parameters. If high-frequency extraction or low-frequency extraction is selected, the high-pass threshold frequency range and low-pass threshold frequency range need to be set.

[0076] The specific implementation process of the conventional random sample augmentation method to augment the small-scale seismic data sample set is as follows: first, randomly sort all the samples in the sample set, then randomly select one or more processing methods from the four sample augmentation processing methods of rotation, noise addition, high-frequency extraction, and low-frequency extraction, set the number of executions of each processing method, that is, the number of samples augmented by each processing method, and set the parameters of each processing method according to the specified rules, and finally, the corresponding samples and their labels in the randomly sorted sample set are augmented in turn according to the set augmentation processing method and the number of executions to obtain the augmented sample set. If the total number of sample augmentation processing is greater than or equal to the number of samples in the original sample set, the number of samples in the sample set will double after augmentation; if the total number of sample augmentation processing is less than the number of samples in the original sample set, the number of samples in the sample set after augmentation is the sum of the number of samples in the original sample set and the total number of sample augmentation processing.

[0077] The results of sample collection and sample augmentation for a certain work area are as follows: the stratigraphic interpretation sample set of a certain work area contains 225 samples, each sample size is 80*160 (80 channels, 160 sampling points). Due to the small number of samples in the sample set, the deep learning network model training is prone to overfitting, resulting in limited prediction and generalization capabilities of the network model. The first seismic data sample set used to solve the regional segmentation problem is augmented. The target processing methods of sample augmentation are flip processing, noise processing and high-pass filtering processing. The execution times of the three processing methods are set to 2, 111 and 112 respectively, and the augmented first seismic data sample set containing 450 samples is obtained.

[0078] The present invention designs a U-shaped fully convolutional neural network structure, uses the first seismic data sample set and the first seismic data sample set after augmentation to respectively train the network model and compare the improvement in accuracy. The network designed by the present invention adopts a left-right symmetrical structure, consisting of an encoder network containing 5 convolution blocks on the left and a decoder network containing 5 convolution blocks on the right, named FCN-U5. Each convolution block of the encoder network and the decoder network contains 2 to 3 groups of "Conv+BN+ReLU" operations. The direct connection between the convolution blocks corresponding to the two networks can fuse multi-scale first-arrival wave features and transfer maximum pooling indexes to achieve better first-arrival upper and lower area segmentation performance, and the upsampling operation of the decoder network uses simple de-pooling instead of deconvolution that requires complex calculations, which speeds up the convergence speed of network model training. The network structure diagram designed by the present invention is shown in the figure. Figure 5 In the present invention, the hardware environment for training and testing the FCN-U5 network model is a Tesla P100 GPU with 16GB of memory, and the deep learning framework is Keras 2.3.1 installed on a RedHat Linux 9.0 operating system.

[0079] 20 samples were randomly selected from the first seismic data sample set as the validation set for evaluating network performance, and the other samples were used as the training set for training the network model. The FCN-U5 network model training adopts the small batch gradient descent method, the batch size is set to 50, the loss function used is the multi-classification cross entropy, and it is minimized during the training process. The gradient descent optimization algorithm is Adam, where the parameters β1 and β2 take the default values ​​of 0.9 and 0.999. A dynamic learning rate is used in the model training process, and the learning rate is halved every 20 iterations to prevent the gradient from falling into the local minimum and stopping the training. The early stopping strategy is to stop the training when the loss value of the validation set increases within 5 consecutive iterations.

[0080] The training process of the FCN-U5 network model for stratigraphic interpretation was completed within 32 iterations using the first seismic data sample set, and the final verification accuracy was 92.21%. The training process of the FCN-U5 network model for stratigraphic interpretation was completed within 56 iterations using the augmented first seismic data sample set, and the final verification accuracy was 95.77%. After the sample set was optimized, the verification set accuracy of the FCN-U5 network model was improved by 3.56%.

[0081] In an embodiment of the present application, optionally, the method further includes: calculating the imbalance of each initial seismic data sample for an initial seismic data sample set used to solve the regional segmentation problem; the imbalance reflects the proportion of each segmented region in the sample; deleting the samples in the initial seismic data sample set whose imbalance is greater than the imbalance threshold, and using the remaining samples as the first seismic data sample set.

[0082] In the embodiment of the present application, for the initial seismic data sample set used to solve the regional segmentation problem, since the detailed processing and interpretation of seismic data requires a lot of time and energy of the staff, the large-scale initial seismic data sample set is often not of very high quality and contains samples with high imbalance, which will affect the regional segmentation ability of the deep learning network. This method targets a two-dimensional or three-dimensional initial seismic data sample set for the regional segmentation problem. According to the label of the initial seismic data sample, the number of sample points corresponding to each segmentation area in the label is counted, and then the proportion of each segmentation area in the total area is counted, that is, the proportion of the number of sample points corresponding to the segmentation area in the total number of sample points (for example, in an initial seismic data sample, 100×80 sample points are arranged in a matrix form, and the label of the sample indicates that there are 2000 sample points for segmentation area A, 4000 sample points for segmentation area B, and 2000 sample points for segmentation area C, then the A segmentation area accounts for 25%, the B segmentation area accounts for 50%, and the C segmentation area accounts for 25%). According to the proportion of the number of sample points corresponding to the segmentation area in the total number of sample points, the variance of the sample, that is, the imbalance of the sample, is calculated, and finally the samples with an imbalance greater than the imbalance threshold are eliminated, which can significantly reduce the overall imbalance of the sample set and improve the regional segmentation ability of the deep learning network model.

[0083] The imbalance threshold can be determined according to actual conditions, and the present application embodiment does not limit this. The imbalance of the initial seismic data sample is determined by the following formula:

[0084]

[0085] Among them, σ 2 is the imbalance of the initial seismic data sample, X i is the proportion of the samples in the ith segmentation area to the total sample points, μ is the average proportion of the samples in each segmentation area to the total sample points, and n is the number of segmentation areas.

[0086] In a specific implementation example, for an initial seismic data sample set used to solve the regional segmentation problem, the imbalance check and optimization functions are implemented by the following steps:

[0087] 1) Select the label directory of the candidate earthquake data sample set: All sample files and corresponding label files of the candidate earthquake data sample set are stored in two folders in the same directory, named "train" and "label" respectively. You need to select the specific folder where the label file is located.

[0088] 2) Select sample dimension: Select the sample dimension of the sample set to be optimized.

[0089] 3) Select the type of label data: The sample set imbalance check and optimization function supports integer and floating-point labels. You need to select the specific data type.

[0090] 4) Set the sample size: For two-dimensional samples, you need to set the number of channels and sample points; for three-dimensional samples, you need to set the number of lines, channels, and sample points. Figure 4 Schematic diagram of the software interface optimized for imbalance)

[0091] 5) Perform sample set imbalance check: According to the sample dimension, sample size and label value in the label (different segmented areas correspond to different label values), first traverse each label value of all label files in the sample set label directory, count the number of label values, that is, the number of areas, and then count the proportion of each label value in each label file, that is, the proportion of the number of sample points corresponding to the segmented area in the total number of sample points, and calculate the variance of the proportion of each segmented area, that is, the imbalance of the sample, and finally calculate the imbalance corresponding to each sample. In this function, the proportion of each segmented area in each sample and its imbalance, and the proportion of each segmented area in the entire sample set and its imbalance will be scrolled in the result display area;

[0092] 6) Sample set optimization: First, set the imbalance threshold, then traverse the imbalance value of each sample, and finally remove each sample file with an imbalance value greater than the threshold and its corresponding label file from the sample set, thus completing the imbalance optimization process of the sample set.

[0093] Implementation example of imbalance optimization for the initial seismic data sample set used to solve the regional segmentation problem:

[0094] The initial seismic data sample set of a certain work area contains 12,000 samples, with a size of 80*160 (80 channels, 160 sampling points). Since the samples in the sample set are randomly picked, there are many samples with high imbalance, and the overall quality of the sample set is not high enough. The imbalance threshold is set to 0.2, and the sample set is optimized. After removing samples with imbalance greater than 0.2, the overall imbalance of the sample set is reduced from 0.0196 to 0.0174.

[0095] In order to verify the quality improvement level of the sample set after the imbalance optimization, the first seismic data sample set and the optimized first seismic data sample set are used to train the FCN-U5 network model designed by the present invention and compare the improvement in accuracy. 200 samples are randomly selected from the two seismic data sample sets as the validation set for evaluating network performance, and the other samples are used as the training set for training the network model. The FCN-U5 network model training adopts the small batch gradient descent method, the batch size is set to 70, the loss function used is the binary cross entropy, and it is minimized during the training process. The gradient descent optimization algorithm, dynamic learning rate and early stopping strategy are set to be the same as the training of the previous horizon interpretation FCN-U5 network model.

[0096] The first seismic data sample set was used to train the first-arrival picking FCN-U5 network model, and the training process was completed within 54 iterations. The final verification accuracy was 97.28%. The first seismic data sample set after imbalance optimization was used to train the first-arrival picking FCN-U5 network model, and the training process was completed within 42 iterations. The training process converged faster, and the final verification accuracy was 99.31%. After the sample set was optimized, the verification accuracy of the FCN-U5 network model increased by 2.03%, which is closer to 100%.

[0097] Figure 6 This is a schematic diagram of a highly unbalanced sample and its labels. In the figure: the label is on the right, the sample value of 1 accounts for 99.35%, the sample value of 0 accounts for 0.65%, and the variance of the proportion of each type of sample is 0.24, that is, the imbalance of the sample is 0.24.

[0098] Embodiment 3

[0099] Figure 7 This is a flowchart of a method for optimizing a seismic data sample set provided in the third embodiment of the present application. The present embodiment is optimized based on the above embodiment. The seismic data sample set in the embodiment of the present application is a second seismic data sample set used to solve the classification problem.

[0100] like Figure 7 As shown, the method of the embodiment of the present application specifically includes the following steps:

[0101] S310, obtaining a second seismic data sample set. In the second seismic data sample set, if the number of second seismic data samples corresponding to the target category is less than a quantity threshold, copying at least one second seismic data sample in the second seismic data samples corresponding to the target category.

[0102] The target category may be any category, for example, any category except the category including the largest number of samples.

[0103] The copying of at least one second seismic data sample in the second seismic data samples corresponding to the target category may be copying a target number of second seismic data samples, where the target number is determined according to the number of samples of the target category and a number threshold.

[0104] In an embodiment of the present application, optionally, the label of the second seismic data sample reflects the category corresponding to the second seismic data sample; the quantity threshold is determined based on the number of samples included in the category whose sample quantity reaches a preset requirement in the second seismic data sample set.

[0105] (Exemplarily, the sum of the target quantity and the number of samples of the target category is equal to the quantity threshold; the quantity threshold is equal to the number of samples included in the category with the largest number of samples in the second seismic data sample set)

[0106] In an embodiment of the present application, optionally, the seismic data sample set is a second seismic data sample set for solving a classification problem, and the label of the second seismic data sample reflects the category corresponding to the second seismic data sample.

[0107] Specifically, in the second seismic data sample set, for the two-dimensional samples, each sample is still composed of sample points arranged in a matrix form, and each second seismic data sample corresponds to a category, which can be a stratum category. For example, if a second seismic data sample is collected in stratum A, the label of the sample is pre-marked as the category corresponding to stratum A.

[0108] S320, randomly and non-repetitively selecting a target processing method to process the copied second seismic data samples to obtain newly added second seismic data samples corresponding to the target category.

[0109] The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing.

[0110] Exemplarily, if in the second seismic data sample set, the category with the largest number includes 10 samples, Category A includes 8 samples, and Category B includes 5 samples, then Category A and Category B can both be target categories, and then for Category A, 2 samples are copied from Category A; for Category B, 5 samples are copied from Category B; and the target processing method is randomly and non-repetitively selected for the copied samples to obtain newly added second seismic data samples corresponding to the target category.

[0111] It should be noted that when processing in a target processing mode, the target processing mode can be randomly selected for each sample; or a target processing mode that has not been used can be selected for each sample. For example, four samples can be processed respectively using flipping processing, noise adding processing, high-pass filtering processing, and low-pass filtering processing.

[0112] S330: Determine an augmented second seismic data sample corresponding to the target category according to the second seismic data sample corresponding to the target category and the newly added second seismic data sample corresponding to the target category.

[0113] Specifically, after obtaining the newly added second seismic data samples corresponding to the target category, the number of samples corresponding to the target category reaches the quantity threshold, and then each category is traversed so that the number of samples corresponding to each category reaches the quantity threshold.

[0114] S340, determining an augmented second seismic data sample set according to the augmented second seismic data samples corresponding to each category, so as to train a machine learning model based on the augmented second seismic data sample set.

[0115] Specifically, a machine learning model for solving classification problems is trained based on the augmented seismic data sample set, so that the machine learning model for solving classification problems can accurately identify the category corresponding to the input seismic data.

[0116] In an embodiment of the present application, optionally, in the second seismic data sample set, if the number of second seismic data samples corresponding to the target category is less than the quantity threshold, then at least one second seismic data sample is copied in the second seismic data sample corresponding to the target category, including: calculating the number of second seismic data samples corresponding to each category in the second seismic data sample set, and taking the maximum value of the number of second seismic data samples corresponding to each category as the quantity threshold; if the number of second seismic data samples corresponding to the target category is less than the quantity threshold, determining the serial numbers corresponding to each second seismic data sample of the target category; the serial numbers are integers starting from 0; establishing a target array corresponding to the target category; the target array includes randomly arranged integers, and the values ​​of each integer range from 0 to a first value; the first value is equal to the quantity threshold minus 1; dividing each value in the target array by the second value and taking the remainder; the second The numerical value is equal to the number of second seismic data samples included in the target category; the remainder is used as a serial number, and if the same remainder appears twice or more, the corresponding second seismic data sample is copied each time the remainder appears repeatedly; accordingly, the at least one seismic data sample is processed in a target processing manner to obtain an additional seismic data sample, including: the target processing manner is randomly and non-repetitively selected for processing the copied second seismic data sample to obtain an additional second seismic data sample corresponding to the target category; accordingly, an augmented seismic data sample set is determined based on the seismic data sample set and the additional seismic data samples, including: based on the second seismic data samples corresponding to the target category and the additional second seismic data samples corresponding to the target category, the augmented second seismic data samples corresponding to the target category are determined; based on the augmented second seismic data samples corresponding to each category, an augmented second seismic data sample set is determined.

[0117] In a specific embodiment, if the second seismic data sample set includes 3 categories, the samples of each category are numbered starting from 0, for example, the first category corresponds to three samples with IDs 0, 1 and 2; the second category corresponds to two samples with IDs 0 and 1; the third category corresponds to five samples with IDs 0, 1, 2, 3 and 4; then the five numbers 0, 1, 2, 3, 4 are randomly arranged three times to obtain three arrays; then the numbers in each array are divided by the number of samples included in the corresponding category to obtain three groups of remainders; obviously, each group has 5 remainders (for example, in the first array, the five numbers are randomly arranged as 0, 4, 2, 1, 3, and each number is divided by 3 to obtain the remainder, and the remainders obtained are 0, 1, 2, 1, 0); for a group of remainders, the remainders are used as IDs to obtain samples corresponding to each remainder. If the remainder appears twice or more, the samples corresponding to the repeated remainders are the samples that need to be processed in a target manner (for example, when the remainders are 0, 1, 2, 1, 0, the samples corresponding to the last two remainders are the samples that need to be processed in a target manner); the samples that need to be processed in a target manner are processed in different target processing methods, or the processing methods (flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing) are randomly and non-repetitively selected for processing to obtain a newly added second seismic data sample corresponding to the target category.

[0118] In a specific implementation process, based on the corresponding sample augmentation software ( Figure 8 The following are screenshots of software for augmenting the second seismic data sample set:

[0119] 1) Select the sample set description file: The sample set description file is an Excel file containing two columns. The first column is the second earthquake data sample file name containing the path, and the second column is the label value, that is, the category of the sample.

[0120] 2) Select sample dimension: Select whether the samples in the sample set to be augmented are two-dimensional samples or three-dimensional samples.

[0121] 3) Set the sample size: For two-dimensional samples, you need to set the number of channels and sample points; for three-dimensional samples, you need to set the number of lines, channels, and sample points.

[0122] 4) Select the label data type: Select whether the sample labels in the sample set to be augmented are integer or floating point.

[0123] 5) Select the sample augmentation processing method and set the execution times and parameters: This step first randomly selects one or more processing methods from the four sample augmentation processing methods of rotation (corresponding to flip processing), noise addition (corresponding to noise addition processing), high-frequency extraction (corresponding to high-pass filtering processing) and low-frequency extraction (corresponding to low-pass filtering processing), and then sets the execution times of each processing method, that is, the number of samples augmented by each processing method, and finally sets the sample augmentation processing parameters. If high-frequency extraction or low-frequency extraction is selected, the high-pass threshold frequency range and low-pass threshold frequency range need to be set.

[0124] 6) Select the sample set description file to be augmented (including the target number of second earthquake data samples). The sample set description file to be augmented has the same format as the original sample set description file. In addition to the original sample file name and label value, it also contains the newly generated sample file name and label value after augmentation.

[0125] 7) Perform random augmentation of class-balanced samples on the sample set to be augmented (the target number of second seismic data samples) (i.e., the specific process of S310-S330): This step first performs the following 4 steps: ① Assume that there are N categories of samples in the sample set, count the number of samples of each category in the sample set, and obtain the array [#C0, #C1,…, #C N-1 ], the maximum number of samples in the array is represented by M; ② randomly arrange the sequence (0,1,…,M-1) to generate N groups of sequences; ③ respectively use the array [#C0,#C1,…,#C N-1 ] to obtain the remainder and obtain N groups of new sequences representing sample indices; ④ According to the N groups of new sequences, repeated samples are selected from the N categories of samples to obtain N categories of samples with M sample numbers; then, according to the augmentation processing method and its execution times and parameters set in step 5), the repeated samples in the N categories of samples are randomly augmented without repeated samples; finally, the N categories of samples with the same number of samples after augmentation are merged into the final sample set, and the augmented sample set description file is generated.

[0126] When performing random non-repetitive sample augmentation on the second seismic data sample set, the parameter setting rules of the four processing methods are the same as those of the conventional sample random augmentation method, but the total number of sample processing executions is required to be greater than or equal to the product of the maximum number of samples in the second seismic data sample set by category and the number of sample categories minus the number of the second seismic data sample set, otherwise the number of samples in each category cannot reach the maximum number of samples in the second seismic data sample set, that is, random augmentation of class-balanced samples cannot be achieved.

[0127] Implementation example of sample augmentation by category for a second set of earthquake data samples used to solve a classification problem:

[0128] The seismic phase classification sample set of a certain work area (the second seismic data sample set) contains 153 samples with a size of 80*80*80 (80 lines, 80 traces, 80 sampling points), including samples of 5 seismic phases, named seismic phase 1, seismic phase 2, seismic phase 3, seismic phase 4 and seismic phase 5, with the number of samples included being 81, 17, 8, 12 and 35 respectively, and the label values ​​of the 5 types of seismic phases are represented by 0, 1, 2, 3 and 4 respectively. Due to the small number of samples in the sample set and the large difference in the number of samples of different types, the categories are seriously unbalanced, resulting in weak prediction and generalization capabilities of the trained network model. According to the above invention, the sample set is augmented, and the number of samples of the four types of seismic phase 2, seismic phase 3, seismic phase 4 and seismic phase 5 are augmented to 64, 73, 69 and 46 respectively. The sample augmentation processing methods include sample rotation, sample noise addition, sample high frequency extraction and sample low frequency extraction. The execution times of the four processing methods are set to 6, 80, 84 and 82 respectively, and finally an augmented sample set containing 405 samples is obtained.

[0129] The present invention uses the public VGG-16 deep learning network to perform multi-label classification on the second seismic data sample set and the augmented second seismic data sample set, with classification accuracy rates of 93.78% and 97.52% respectively. After the seismic phase classification sample set is augmented, the classification accuracy of the VGG-16 network is improved by 3.74%.

[0130] In an embodiment of the present application, optionally, the target processing method is a flipping processing; accordingly, the seismic data sample is processed in a target processing method, including: if the seismic data sample is a two-dimensional seismic data sample, the two-dimensional seismic data sample is horizontally flipped and / or vertically flipped; if the seismic data sample is a three-dimensional seismic data sample, the two-dimensional seismic data sample is horizontally flipped and / or vertically flipped based on any of the following planes; the planes include: InLine-Xline plane, InLine-time plane and XLine-time plane.

[0131] Accordingly, the seismic data samples are processed in a target processing manner, including:

[0132] If the seismic data sample is a two-dimensional seismic data sample, the j-th column and the (mj)-th column data in the two-dimensional seismic data sample are swapped; and / or the i-th row and the (pi-th row data) are swapped; m is the total number of columns of the two-dimensional seismic data sample, and p is the total number of rows of the two-dimensional seismic data sample;

[0133] If the seismic data sample is a three-dimensional seismic data sample, any one or more of the following processing is performed on the three-dimensional seismic data sample:

[0134] Interchange the data of the ath and (ba)th surfaces of the InLine-Xline plane;

[0135] Interchange the data of the cth and (dc)th planes of the InLine-time plane;

[0136] Interchange the data of the e-th and (fe)-th planes of the XLine-time plane;

[0137] Wherein, b is the total number of faces of the InLine-Xline plane, d is the total number of faces of the InLine-Time plane, and f is the total number of faces of the XLine-Time plane.

[0138] It should be noted that since the two-dimensional seismic data sample flipping is horizontal flipping and vertical flipping, the number of sample rotations performed on the two-dimensional seismic data must be within 2 times (for example, if three samples are flipped, two samples will have the same flipping, and the effect of these two samples on model training is similar to that of one sample. In the whole article, the same applies to limiting the number of executions of each processing method); three-dimensional seismic data samples can be horizontally flipped and vertically flipped in three planes: InLine-XLine plane, InLine-time plane and XLine-time plane, so the number of sample rotations performed on the three-dimensional seismic data samples must be within 6 times.

[0139] In the embodiment of the present application, optionally, the target processing method is noise addition processing; accordingly, the seismic data sample is processed in the target processing method, including: using the following formula to determine the value of each sample point (each seismic data) in the i-th seismic data sample:

[0140]

[0141] Among them, A n is the value of the sample point after adding noise, A r is the value of the sample point before adding noise, A max is the maximum value of all sample points in the seismic data sample, A min is the minimum value of all sample points in the seismic data sample, and i is the execution order of the noise addition process, that is, the noise addition process is performed for the i-th time. This setting can ensure that the random noise value added to each sample is different when adding noise.

[0142] In the embodiment of the present application, optionally, the target processing method is high-pass filtering; accordingly, the seismic data samples are processed in the target processing method, including: according to the part of the target sample point whose frequency is greater than or equal to the high-pass threshold frequency threshold, the value of the target sample point after high-pass filtering is determined; the target processing method is low-pass filtering; accordingly, the seismic data samples are processed in the target processing method, including: according to the part of the target sample point whose frequency is less than or equal to the low-pass threshold frequency threshold, the value of the target sample point after low-pass filtering is determined; wherein, the high-pass threshold frequency threshold is determined by the following formula:

[0143] ω hc =ω hmin +(ω hmax -ω hmin ) / (N h )*i;

[0144] Among them, ω hc is the high-pass threshold frequency, ω hmin is the minimum value of the high-pass threshold frequency range, ω hmax is the maximum value of the high-pass threshold frequency range, N h is the number of times the sample extracts high frequency, i is the execution order of the sample extracting high frequency; (for example, a total of 5 samples are processed with high-pass filtering 5 times, N h is the number of times the high frequency is extracted from the sample, that is, 5, i represents the i-th extraction of high frequency)

[0145] The low-pass threshold frequency threshold is determined by the following formula:

[0146] ω lc =ω lmin +(ω lmax -ω lmin ) / (N l )*i;

[0147] Among them, ω lc is the low-pass threshold frequency, ω lmin is the minimum value of the low-pass threshold frequency range, ω lmax is the maximum value of the low-pass threshold frequency range, N l is the number of times the sample is extracted with low frequency, and i is the execution order of the sample extraction with low frequency.

[0148] Exemplarily, each channel in the seismic data sample is subjected to high-pass filtering to extract high-frequency information. The frequency response of the high-pass filter is:

[0149]

[0150] Where H(ω) is the frequency response function of the high-pass filter, ω is the frequency, and ω ch is the high-pass threshold frequency.

[0151] For low-pass filtering: low-pass filter is used for sample extraction and low-frequency processing. Low-pass filtering is performed on each channel of the seismic data sample to extract low-frequency information. The frequency response of the low-pass filter is:

[0152]

[0153] Where: L(ω) is the frequency response function of the low-pass filter, ω is the frequency, ω cl is the low-pass threshold frequency.

[0154] Fig. 9 Schematic diagram of the flipping processing of the stratigraphic interpretation sample. In the figure, (a) is the stratigraphic interpretation sample A; (b) is the label of sample A; (c) is the horizontal flipping of sample A to obtain sample B; (d) is the label of sample B.

[0155] Fig.10 Schematic diagram of high-pass filtering processing of stratigraphic interpretation samples; (a) is stratigraphic interpretation sample C; (b) is the label of sample C; (c) is high-frequency extraction processing of sample C to obtain sample D (high-pass filtering threshold frequency: 48 Hz); (d) is the label of sample D.

[0156] Fig.11 Schematic diagram of high-pass filtering processing of seismic phase classification samples; (a) is seismic phase classification sample A (label is 0, indicating a certain seismic phase); (b) is the high-frequency extraction processing of sample A to obtain sample B (sample B label is also 0, high-pass filtering threshold frequency: 76Hz).

[0157] Embodiment 4

[0158] Fig.12 This is a schematic diagram of the structure of a seismic data sample set optimization device provided in the fourth embodiment of the present application. The device can execute the seismic data sample set optimization method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method. Fig.12 As shown, the device comprises:

[0159] A seismic data sample duplication module 410 is used to obtain a seismic data sample set and to duplicate at least one seismic data sample from the seismic data sample set;

[0160] A seismic data sample processing module 420, configured to process the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample;

[0161] An augmented seismic data sample set determination module 430, configured to determine an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented seismic data sample set;

[0162] The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing;

[0163] The seismic data sample copying module 410 includes: a first seismic data sample copying unit, configured to copy at least one first seismic data sample from the first seismic data sample set if the seismic data sample set is a first seismic data sample set for solving the region segmentation problem;

[0164] The seismic data sample copying module 410 includes: a second seismic data sample copying unit, which is used to copy at least one second seismic data sample in the second seismic data samples corresponding to the target category in the second seismic data sample set if the seismic data sample set is a second seismic data sample set used to solve a classification problem and if the number of second seismic data samples corresponding to the target category is less than a quantity threshold.

[0165] The technical solution of the embodiment of the present application includes: a seismic data sample copying module 410, which is used to obtain a seismic data sample set and copy at least one seismic data sample from the seismic data sample set; a seismic data sample processing module 420, which is used to process the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample; an augmented seismic data sample set determination module 430, which is used to determine an augmented seismic data sample set based on the seismic data sample set and the newly added seismic data sample, so as to train a machine learning model based on the augmented seismic data sample set; the target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing, and low-pass filtering processing. The technical solution obtains newly added seismic data samples by processing the seismic data samples in a target processing manner, thereby increasing the number of seismic data samples, so that after training the machine learning model based on the augmented seismic data sample set, the generalization ability and other performances of the machine learning model are improved.

[0166] In an embodiment of the present application, optionally, the first seismic data sample is composed of seismic data arranged in a matrix form, the seismic data arranged in a matrix form corresponds to each region on a geological profile, and the label corresponding to the first seismic data sample includes a segmented region corresponding to each seismic data.

[0167] In an embodiment of the present application, optionally, the label of the second seismic data sample reflects the category corresponding to the second seismic data sample; the quantity threshold is determined based on the number of samples included in the category whose sample quantity reaches a preset requirement in the second seismic data sample set.

[0168] In the embodiment of the present application, optionally, the target processing method is flip processing;

[0169] Accordingly, the seismic data sample processing module 420 includes:

[0170] A two-dimensional seismic data sample processing unit, configured to perform horizontal flipping and / or vertical flipping processing on the two-dimensional seismic data sample if the seismic data sample is a two-dimensional seismic data sample;

[0171] The three-dimensional seismic data sample processing unit is used to perform horizontal flipping and / or vertical flipping processing based on any of the following planes if the seismic data sample is a three-dimensional seismic data sample; the planes include: InLine-Xline plane, InLine-time plane and XLine-time plane.

[0172] In the embodiment of the present application, optionally, the target processing method is noise addition processing;

[0173] Accordingly, the seismic data sample processing module 420 includes:

[0174] The sample point value determination unit is used to determine the value of each sample point in the i-th seismic data sample using the following formula:

[0175]

[0176] Among them, A n is the value of the sample point after adding noise, A r is the value of the sample point before adding noise, A max is the maximum value of all sample points in the seismic data sample, A min is the minimum value of all sample points in the seismic data sample, and i is the sequence number of the seismic data sample.

[0177] In the embodiment of the present application, optionally, the target processing method is high-pass filtering;

[0178] Accordingly, the seismic data sample processing module 420 includes:

[0179] A high-pass filtering unit, used to determine the value of the target sample point after high-pass filtering according to the part of the target sample point whose frequency is greater than or equal to the high-pass threshold frequency threshold;

[0180] The target processing method is low-pass filtering; accordingly, the seismic data sample processing module 420 includes:

[0181] A low-pass filtering unit, used to determine the value of the target sample point after low-pass filtering according to the part of the target sample point whose frequency is less than or equal to the low-pass threshold frequency threshold;

[0182] The high-pass threshold frequency threshold is determined by the following formula:

[0183] ω hc =ω hmin +(ω hmax -ω hmin ) / (N h )*i;

[0184] Among them, ω hc is the high-pass threshold frequency, ω hmin is the minimum value of the high-pass threshold frequency range, ω hmax is the maximum value of the high-pass threshold frequency range, N h is the number of times the sample extracts high frequency, i is the execution order of the sample extracting high frequency;

[0185] The low-pass threshold frequency threshold is determined by the following formula:

[0186] ω lc =ω lmin +(ω lmax -ω lmin ) / (N l )*i;

[0187] Among them, ω lc is the low-pass threshold frequency, ω lmin is the minimum value of the low-pass threshold frequency range, ω lmax is the maximum value of the low-pass threshold frequency range, N l is the number of times the sample is extracted with low frequency, and i is the execution order of the sample extraction with low frequency.

[0188] In the embodiment of the present application, optionally, the device further includes:

[0189] An imbalance calculation unit, for calculating the imbalance of each initial seismic data sample for an initial seismic data sample set used to solve the region segmentation problem; the imbalance reflects the proportion of each segmented region in the sample;

[0190] The imbalance optimization unit is used to delete the samples whose imbalance is greater than the imbalance threshold in the initial seismic data sample set, and use the remaining samples as the first seismic data sample set.

[0191] In the embodiment of the present application, optionally, the second seismic data sample copying unit includes:

[0192] a sample quantity calculation subunit, configured to calculate the quantity of second seismic data samples corresponding to each category in the second seismic data sample set, and take the maximum value among the quantities of second seismic data samples corresponding to each category as a quantity threshold;

[0193] A sample sequence number determination subunit, for determining the sequence numbers corresponding to the second seismic data samples of the target category if the number of the second seismic data samples corresponding to the target category is less than the quantity threshold; the sequence number is an integer starting from 0;

[0194] A target array establishment subunit is used to establish a target array corresponding to the target category; the target array includes randomly arranged integers, and the value of each integer ranges from 0 to a first value; the first value is equal to the quantity threshold minus 1;

[0195] A remainder determination subunit, used for dividing each value in the target array by a second value and taking the remainder; the second value is equal to the number of second seismic data samples included in the target category;

[0196] A second seismic data sample copying subunit is used to use the remainder as a sequence number, and if the same remainder appears twice or more, copy the corresponding second seismic data sample each time the remainder appears repeatedly;

[0197] Accordingly, the seismic data sample processing module 420 includes:

[0198] A second seismic data sample processing unit, configured to randomly and non-repetitively select a target processing method to process the copied second seismic data sample, so as to obtain a newly added second seismic data sample corresponding to the target category;

[0199] Accordingly, the augmented seismic data sample set determination module 430 includes:

[0200] An augmented second seismic data sample determination unit corresponding to a target category, configured to determine an augmented second seismic data sample corresponding to a target category according to the second seismic data sample corresponding to the target category and the newly added second seismic data sample corresponding to the target category;

[0201] The augmented second seismic data sample set determining unit is used to determine the augmented second seismic data sample set according to the augmented second seismic data samples corresponding to each category.

[0202] An optimization device for a seismic data sample set provided in an embodiment of the present application can execute an optimization method for a seismic data sample set provided in any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0203] Embodiment 5

[0204] Fig.13 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0205] like Fig.13 As shown, the electronic device 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the electronic device 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0206] A number of components in the electronic device 10 are connected to the I / O interface 15, including: an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0207] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the optimization method of the seismic data sample set.

[0208] In some embodiments, the optimization method of the seismic data sample set may be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as a storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 10 via the ROM 12 and / or the communication unit 19. When the computer program is loaded into the RAM 13 and executed by the processor 11, one or more steps of the optimization method of the seismic data sample set described above may be performed. Alternatively, in other embodiments, the processor 11 may be configured to execute the optimization method of the seismic data sample set in any other appropriate manner (e.g., by means of firmware).

[0209] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0210] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0211] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0212] To provide interaction with a user, the systems and techniques described herein may be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices may also be used to provide interaction with the user; for example, the feedback provided to the user may be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form (including acoustic input, voice input, or tactile input).

[0213] The systems and techniques described herein may be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0214] A computing system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The client and server relationship is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services.

[0215] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0216] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing a seismic data sample set, characterized in that: include: Acquire a seismic data sample set, and copy at least one seismic data sample from the seismic data sample set; Processing the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample; Determine an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented seismic data sample set; The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing; If the seismic data sample set is a first seismic data sample set for solving the regional segmentation problem, then the copying of at least one seismic data sample from the seismic data sample set includes: copying at least one first seismic data sample from the first seismic data sample set; If the seismic data sample set is a second seismic data sample set for solving a classification problem, then, copying at least one seismic data sample from the seismic data sample set includes: in the second seismic data sample set, if the number of second seismic data samples corresponding to a target category is less than a quantity threshold, then copying at least one second seismic data sample in the second seismic data samples corresponding to the target category.

2. The method according to claim 1, characterized in that The first seismic data sample is composed of seismic data arranged in a matrix form, the seismic data arranged in a matrix form corresponds to each area on a geological profile, and the label corresponding to the first seismic data sample includes a segmented area corresponding to each seismic data.

3. The method according to claim 1, characterized in that The label of the second seismic data sample reflects the category corresponding to the second seismic data sample; the quantity threshold is determined based on the number of samples included in the category whose sample quantity meets the preset requirement in the second seismic data sample set.

4. The method according to claim 1, characterized in that: The target processing method is flip processing; Accordingly, the seismic data samples are processed in a target processing manner, including: If the seismic data sample is a two-dimensional seismic data sample, performing horizontal flipping and / or vertical flipping processing on the two-dimensional seismic data sample; If the seismic data sample is a three-dimensional seismic data sample, horizontal flipping and / or vertical flipping processing is performed based on any of the following planes; the planes include: InLine-Xline plane, InLine-time plane and XLine-time plane.

5. The method according to claim 1, characterized in that The target processing method is noise addition processing; Accordingly, the seismic data samples are processed in a target processing manner, including: The following formula is used to determine the value of each sample point in the i-th seismic data sample: Among them, A n is the value of the sample point after adding noise, A r is the value of the sample point before adding noise, A max is the maximum value of all sample points in the seismic data sample, A min is the minimum value of all sample points in the seismic data sample, and i is the sequence number of the seismic data sample.

6. The method according to claim 1, characterized in that The target processing method is high-pass filtering; Accordingly, the seismic data samples are processed in a target processing manner, including: Determine the value of the target sample point after high-pass filtering according to the part of the target sample point whose frequency is greater than or equal to the high-pass threshold frequency threshold; The target processing method is low-pass filtering; accordingly, the seismic data samples are processed in the target processing method, including: Determine the value of the target sample point after low-pass filtering according to the part of the target sample point whose frequency is less than or equal to the low-pass threshold frequency threshold; The high-pass threshold frequency threshold is determined by the following formula: oh hc =ω hmin +(ω hmax -oh hmin ) / (N h )*i; Among them, ω hc is the high-pass threshold frequency, ω hmin is the minimum value of the high-pass threshold frequency range, ω hmax is the maximum value of the high-pass threshold frequency range, N h is the number of times the sample extracts high frequency, i is the execution order of the sample extracting high frequency; The low-pass threshold frequency threshold is determined by the following formula: oh lc =ω lmin +(ω lmax -oh lmin ) / (N l )*i; Among them, ω lc is the low-pass threshold frequency, ω lmin is the minimum value of the low-pass threshold frequency range, ω lmax is the maximum value of the low-pass threshold frequency range, N l is the number of times the sample is extracted with low frequency, and i is the execution order of the sample extraction with low frequency.

7. The method according to claim 2, characterized in that The method further comprises: For the initial seismic data sample set used to solve the regional segmentation problem, the imbalance degree of each initial seismic data sample is calculated; the imbalance degree reflects the proportion of each segmented region in the sample; In the initial seismic data sample set, the samples whose imbalance degree is greater than the imbalance degree threshold are deleted, and the remaining samples are used as the first seismic data sample set.

8. The method according to claim 3, characterized in that In the second seismic data sample set, if the number of second seismic data samples corresponding to the target category is less than a quantity threshold, copying at least one second seismic data sample in the second seismic data samples corresponding to the target category, comprising: Calculating the number of second seismic data samples corresponding to each category in the second seismic data sample set, and taking the maximum value among the numbers of second seismic data samples corresponding to each category as a quantity threshold; If the number of second seismic data samples corresponding to the target category is less than the quantity threshold, determining the serial numbers corresponding to the second seismic data samples of the target category respectively; the serial numbers are integers starting from 0; Establishing a target array corresponding to the target category; the target array includes randomly arranged integers, each integer having a value ranging from 0 to a first value; the first value is equal to the quantity threshold minus 1; Divide each value in the target array by a second value and take the remainder; the second value is equal to the number of second seismic data samples included in the target category; The remainder is used as a serial number. If the same remainder appears twice or more, the corresponding second seismic data sample is copied each time the remainder appears repeatedly. Accordingly, the at least one seismic data sample is processed in a target processing manner to obtain a newly added seismic data sample, including: The copied second seismic data samples are randomly and non-repetitively processed in a target processing manner to obtain a newly added second seismic data sample corresponding to the target category; Accordingly, the augmented seismic data sample set is determined according to the seismic data sample set and the newly added seismic data sample, including: Determining an augmented second seismic data sample corresponding to the target category according to the second seismic data sample corresponding to the target category and the newly added second seismic data sample corresponding to the target category; An augmented second seismic data sample set is determined according to the augmented second seismic data samples corresponding to each category.

9. An optimization device for a seismic data sample set, characterized in that: include: A seismic data sample duplication module, used to obtain a seismic data sample set and to duplicate at least one seismic data sample from the seismic data sample set; A seismic data sample processing module, used for processing the at least one seismic data sample in a target processing manner to obtain a newly added seismic data sample; An augmented seismic data sample set determination module, used to determine an augmented seismic data sample set according to the seismic data sample set and the newly added seismic data samples, so as to train a machine learning model based on the augmented seismic data sample set; The target processing method includes at least one of the following: flip processing, noise adding processing, high-pass filtering processing and low-pass filtering processing; The seismic data sample duplication module comprises: a first seismic data sample duplication unit, configured to, if the seismic data sample set is a first seismic data sample set for solving the region segmentation problem, copy at least one first seismic data sample from the first seismic data sample set; The seismic data sample copying module includes: a second seismic data sample copying unit, which is used to copy at least one second seismic data sample in the second seismic data samples corresponding to the target category in the second seismic data sample set if the seismic data sample set is a second seismic data sample set used to solve a classification problem and if the number of second seismic data samples corresponding to a target category is less than a quantity threshold.

10. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the method for optimizing the seismic data sample set according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the method for optimizing a seismic data sample set according to any one of claims 1-8 when executed.

Images