Salt dome identification method, device and computer equipment

Abstract

This specification provides a method, device and computer equipment for identifying salt domes. Based on the above method, after obtaining the target seismic data of the target area, first obtain multiple seismic attribute data bodies based on the screened sensitive seismic attributes and target seismic data; then use a convolutional autoencoder to process the multiple seismic attribute data bodies to obtain multiple local features; and fuse the multiple local features to obtain overall features; perform a first recognition based on the overall features to obtain a first recognition result; then, based on the first recognition result, construct an initial training set with labels; based on preset training rules, use the initial training set, and perform multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt domes in the target area; then use the target recognition model to perform a second recognition to determine the salt domes in the target area. In this way, the salt domes in the target area can be efficiently and accurately determined based on seismic data that does not carry labels.

Description

Technical Field

[0001] This specification belongs to the field of oil and gas seismic exploration technology, and in particular to salt dome identification methods, devices and computer equipment. Background Art

[0002] In the field of oil and gas seismic exploration technology, it is often necessary to identify and determine the specific location and boundaries of salt domes in the study area of ​​interest, so as to better guide subsequent oil and gas exploration and development in the study area.

[0003] Based on existing methods, it is often necessary to first obtain and use a large amount of labeled seismic data for model training, and then use the trained model to detect and identify salt domes.

[0004] However, when implementing existing methods, it is first necessary to acquire and rely on a large amount of labeled seismic data. However, in actual engineering scenarios, the cost of acquiring such labeled seismic data is relatively high, and the amount of labeled seismic data that can be directly acquired is relatively limited. This in turn affects the training of subsequent models, resulting in relatively weak generalization capabilities of the trained models, making it prone to errors when using such models for salt dome identification. Furthermore, when training models based on the above methods, the specificity is relatively poor, and the model training cycle is relatively long. As a result, the model accuracy of the trained models is relatively low, further affecting the accuracy of subsequent salt dome identification.

[0005] To address the above issues, no effective solutions have been proposed so far. Summary of the Invention

[0006] This specification provides a salt dome identification method, device, and computer equipment that do not rely on or use labeled seismic data. Salt domes in target areas can be efficiently and accurately identified based solely on unlabeled seismic data.

[0007] This manual provides a method for identifying salt domes, including:

[0008] Acquire target seismic data of the target area; and screen out sensitive seismic attributes of the local characteristics of the salt domes in the target area;

[0009] Acquiring corresponding multiple seismic attribute data volumes according to the sensitive seismic attributes and the target seismic data;

[0010] Processing the multiple seismic attribute data volumes separately by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area;

[0011] fusing the multiple local features to obtain an overall feature of the salt dome in the target area;

[0012] performing a first identification of the salt domes in the target area according to the overall characteristics to obtain a first identification result;

[0013] Constructing an initial training set with labels based on the first recognition result;

[0014] Based on a preset training rule, using the initial training set, the initial recognition model is trained through multiple rounds of recursive iterations to obtain a target recognition model for salt domes in the target area;

[0015] The target seismic data and the corresponding seismic attribute data are processed by using the target recognition model to perform a second recognition to determine the salt dome in the target area.

[0016] In one embodiment, the sensitive seismic attributes include at least two of the following seismic attributes: a root mean square attribute, an amplitude attribute, a variance attribute, and a Chaos attribute.

[0017] In one embodiment, the multiple seismic attribute data volumes are processed separately using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area, including:

[0018] The current seismic attribute data volume among the multiple seismic attribute data volumes is processed by using a convolutional autoencoder in the following manner to obtain the corresponding local features:

[0019] Using a convolutional autoencoder to process the current earthquake attribute data volume and extract multiple deep semantic features output by a specified intermediate network layer of the convolutional autoencoder;

[0020] Classifying the multiple deep semantic features to obtain multiple deep semantic feature groups;

[0021] Determining a main feature group through principal component analysis based on the multiple deep semantic feature groups;

[0022] According to the main feature group, K-means clustering is performed to obtain the corresponding local features.

[0023] In one embodiment, a first identification is performed on the salt dome in the target area based on the overall characteristics to obtain a first identification result, including:

[0024] The salt dome in the target area is characterized according to the overall characteristics and the target seismic data to determine a first position and a first boundary of the salt dome in the target area as a first identification result.

[0025] In one embodiment, after first identifying the salt dome in the target area based on the overall characteristics and obtaining a first identification result, the method further includes:

[0026] Detect whether the clarity of the salt dome boundary reflection of the seismic profile in the target area is greater than the preset clarity threshold, and whether there is strong phase axis interference;

[0027] When the clarity of the salt dome boundary reflection of the seismic profile in the target area is determined to be greater than a preset clarity threshold and there is no strong isotropic interference, the salt dome in the target area is determined according to the first recognition result.

[0028] In one embodiment, based on preset training rules and using the initial training set, the initial recognition model is subjected to multiple rounds of recursive iterative training to obtain a target recognition model for salt domes in the target area, including:

[0029] Perform the current round of recursive iterative training based on the preset training rules in the following manner:

[0030] Based on the training set of the previous round, the test data of the current round is determined from the target seismic data along the specified direction and at the specified step size;

[0031] Use the recognition model of the previous round to process the test data of the current round and determine the corresponding pseudo label;

[0032] Based on the test data and pseudo labels of the current round, the training set of the previous round is updated to obtain the training set of the current round;

[0033] The recognition model of the previous round is trained using the training set of the current round to obtain the recognition model of the current round.

[0034] In one embodiment, after obtaining the recognition model of the current round, the method further includes:

[0035] Check whether the preset end condition is currently met;

[0036] When it is determined that the preset end condition is currently satisfied, the recognition model of the current round is determined as the target recognition model for the salt dome in the target area.

[0037] In one embodiment, a second identification is performed by processing target seismic data and corresponding seismic attribute data using a target identification model to determine salt domes in the target area, including:

[0038] Utilize the target recognition model to process the target seismic data and the corresponding seismic attribute data volume to obtain the corresponding target recognition result;

[0039] Based on the target identification results, the salt domes in the target area are determined.

[0040] This specification also provides a salt dome identification device, comprising:

[0041] The first acquisition module is used to acquire target seismic data of the target area and screen out sensitive seismic attributes of local characteristics of salt domes in the target area;

[0042] A second acquisition module is configured to acquire a plurality of corresponding seismic attribute data volumes according to the sensitive seismic attributes and the target seismic data;

[0043] a third acquisition module, configured to process the plurality of seismic attribute data volumes respectively by using a convolutional autoencoder to acquire a plurality of local features of the salt dome in the target area;

[0044] A fusion module, configured to fuse the multiple local features to obtain an overall feature of the salt dome in the target area;

[0045] a first recognition module, configured to perform a first recognition on the salt dome in the target area according to the overall characteristics, and obtain a first recognition result;

[0046] A construction module, configured to construct an initial training set with labels according to the first recognition result;

[0047] A training module is used to perform multiple rounds of iterative training on the initial recognition model based on preset training rules and using the initial training set to obtain a target recognition model for salt domes in the target area;

[0048] The second recognition module is used to process the target seismic data and the corresponding seismic attribute data volume by using the target recognition model to perform a second recognition to determine the salt dome in the target area.

[0049] This specification also provides a computer device, comprising a processor and a memory for storing processor-executable instructions, wherein when the processor executes the instructions, the following steps are implemented: acquiring target seismic data of a target area; screening out sensitive seismic attributes for local features of salt domes in the target area; acquiring corresponding multiple seismic attribute data volumes based on the sensitive seismic attributes and the target seismic data; processing the multiple seismic attribute data volumes separately by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area; fusing the multiple local features to obtain an overall feature of the salt dome in the target area; performing a first recognition of the salt dome in the target area based on the overall feature to obtain a first recognition result; constructing an initial training set with a label based on the first recognition result; based on a preset training rule, using the initial training set, performing multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt dome in the target area; performing a second recognition by processing the target seismic data and the corresponding seismic attribute data volumes using the target recognition model to determine the salt dome in the target area.

[0050] Based on the salt dome identification method, device and computer equipment provided in this specification, after obtaining the target seismic data of the target area, multiple seismic attribute data bodies can be obtained based on the selected sensitive seismic attributes and the target seismic data; then, a convolutional autoencoder is used to process the multiple seismic attribute data bodies to obtain multiple local features; and the multiple local features are fused to obtain the overall features for the salt dome; then, a first recognition is performed based on the overall features to obtain a first recognition result; further, based on the first recognition result, an initial training set with labels can be constructed; based on a preset training rule, the initial training set is used to perform multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt dome in the target area; and then the target recognition model is used to perform a second recognition to finally determine the salt dome in the target area. Thus, it is possible to conveniently, efficiently and accurately identify and determine the salt dome in the target area by using only the unlabeled seismic data and guiding it by the seismic attributes without relying on and using the labeled seismic data. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of this specification, the following will briefly introduce the drawings required for use in the embodiments. The drawings described below are only some of the embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0052] Figure 1 This is a flow chart of a method for identifying a salt dome provided in one embodiment of this specification;

[0053] Figure 2 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0054] Figure 3 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0055] Figure 4 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0056] Figure 5 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0057] Figure 6 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0058] Figure 7 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0059] Figure 8 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0060] Figure 9 This is a schematic diagram of an embodiment of a salt dome identification method provided by an embodiment of this specification, in a scenario example;

[0061] Figure 10 This is a schematic diagram of the structure of a computer device provided by one embodiment of this specification;

[0062] Figure 11 This is a schematic diagram of the structure of a salt dome identification device provided in one embodiment of this specification. DETAILED DESCRIPTION

[0063] To help those skilled in the art better understand the technical solutions in this specification, the following will provide a clear and complete description of the technical solutions in the embodiments of this specification, in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of this specification, not all of them. All other embodiments derived by those skilled in the art based on the embodiments in this specification without creative effort shall fall within the scope of protection of this specification.

[0064] See Figure 1 and Figure 2 As shown, the embodiment of this specification provides a method for identifying salt domes. The specific implementation of this method may include the following:

[0065] S101: Acquire target seismic data of the target area; and screen out sensitive seismic attributes of local characteristics of salt domes in the target area;

[0066] S102: Acquire corresponding multiple seismic attribute data volumes according to the sensitive seismic attributes and the target seismic data;

[0067] S103: Processing the multiple seismic attribute data volumes separately by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area;

[0068] S104: fusing the multiple local features to obtain an overall feature of the salt dome in the target area;

[0069] S105: performing a first identification of the salt dome in the target area according to the overall characteristics to obtain a first identification result;

[0070] S106: Constructing an initial training set with labels based on the first recognition result;

[0071] S107: Based on a preset training rule, using the initial training set, the initial recognition model is subjected to multiple rounds of recursive iterative training to obtain a target recognition model for salt domes in the target area;

[0072] S108: Performing a second recognition by processing the target seismic data and the corresponding seismic attribute data volume using the target recognition model to determine the salt dome in the target area.

[0073] In some embodiments, the target area may be specifically understood as a stratum area containing a salt dome to be measured.

[0074] Specifically, a salt dome refers to a diapir structure formed when fluid, low-viscosity materials such as rock salt, clay, or gypsum flow upward under the influence of vertical compressive stress from overlying strata and lateral compressive stress from movement, compressing the overlying rock strata and causing them to arch and uplift. This unique structure provides space and trapping conditions for the accumulation and storage of oil and natural gas. Many of the high-yield oil and gas fields discovered so far are closely associated with salt domes. Specifically, a salt dome can be understood as a geological anomaly closely related to oil and gas resources. Salt domes differ from other geological bodies in terms of amplitude, frequency, phase, continuity, lithologic composition, and stratigraphic occurrence. Accurate delineation and identification of salt domes is fundamental and crucial in seismic data interpretation, playing a vital role in subsequent exploration of the subsurface geological environment and structure, seismic inversion, reservoir prediction, and accurate modeling and imaging of salt bodies. Therefore, accurately describing the spatial distribution, identification, and localization of salt domes is essential.

[0075] In some embodiments, the acquired target seismic data may be raw seismic data about the target area. Specifically, the target seismic data may be three-dimensional tensor data, as opposed to a two-dimensional pixel image.

[0076] In some embodiments, the sensitive seismic attribute can be understood as a seismic attribute that is sensitive to the local characteristics of the salt dome and can highlight the local characteristics of the salt dome geological anomaly. The seismic attribute can be derived through geometric calculation.

[0077] During specific implementation, sensitive seismic attributes of the local characteristics of salt domes in the target area can be screened out based on geological and geophysical knowledge.

[0078] In some embodiments, the sensitive seismic attributes include at least two of the following seismic attributes: root mean square attribute, amplitude attribute, variance attribute, Chaos attribute, etc.

[0079] The RMS and amplitude attributes can effectively highlight sharp reflections in seismic data and are relatively effective in depicting salt dome boundaries. The variance and Chaos attributes are more sensitive to areas with large variances in seismic profiles and are relatively effective in indicating chaotic reflection characteristics within the salt dome.

[0080] It should be noted that the above-mentioned target seismic data can also be understood as a data body of amplitude attributes.

[0081] In this embodiment, see Figure 2 As shown, the combination of the earthquake attributes of amplitude attribute (corresponding to attribute 1) and variance attribute (corresponding to attribute 2) can be preferably used as the sensitive earthquake attribute. Of course, in specific implementation, according to specific circumstances and processing requirements, other combinations of earthquake attributes can also be used as sensitive earthquake attributes.

[0082] In some embodiments, during specific implementation, corresponding multiple seismic attribute data volumes may be acquired based on sensitive seismic attributes and target seismic data.

[0083] However, due to the high level of redundant information in the seismic attribute data bodies that are usually directly obtained, these sensitive seismic attributes react roughly to the local characteristics of salt domes and are easily affected by a lot of noise or other seismic events, resulting in relatively large errors. Therefore, further feature screening and purification of the above seismic attribute data bodies are required so that the salt domes in the target area can be more accurately determined in the future.

[0084] In some embodiments, see Figure 3 As shown, the above-mentioned method of respectively processing the multiple seismic attribute data volumes using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area may include the following steps: processing a current seismic attribute data volume among the multiple seismic attribute data volumes using a convolutional autoencoder in the following manner to obtain corresponding local features:

[0085] S1: Use the convolutional autoencoder to process the current earthquake attribute data volume and extract multiple deep semantic features output by the specified intermediate network layer of the convolutional autoencoder;

[0086] S2: classifying the multiple deep semantic features to obtain multiple deep semantic feature groups;

[0087] S3: Determine a main feature group based on the multiple deep semantic feature groups through principal component analysis;

[0088] S4: Perform K-means clustering based on the main feature group to obtain the corresponding local features.

[0089] Among them, the above-mentioned convolutional autoencoder (CAE) is usually used to reconstruct the original seismic data to obtain the corresponding reconstructed seismic data.

[0090] According to the above method, a plurality of different seismic attribute data volumes can be processed respectively to obtain a plurality of local features corresponding to the different seismic attribute data volumes.

[0091] Based on the above embodiment, the powerful feature extraction capabilities of convolutional neural networks can be first utilized to extract features from seismic attribute data using a convolutional autoencoder (CAE) to obtain features that are more effective for local salt domes. Furthermore, these features are refined through principal component analysis (PCA) and K-means clustering to obtain local features with high accuracy and low error. Subsequently, salt domes can be more accurately identified based on these local features.

[0092] In some embodiments, for example, see Figure 4 As shown in Figure 2, the “Conv 3×3+Leakrelu,64” structure in the autoencoder can be used as the intermediate network layer.

[0093] Taking the processing of earthquake attribute data corresponding to amplitude attributes as an example, after the earthquake attribute data is input into CAE, 64-dimensional deep semantic features are extracted from the intermediate network layer; then according to the characteristics of deep semantic features, the 64-dimensional deep semantic features are first divided into 3 deep semantic feature groups, corresponding to Figure 3 ; then, principal component analysis is performed on the three deep semantic feature groups to screen out the main feature group from the three deep semantic feature groups; then, K-means clustering with k=2 is performed on the main feature group to obtain the local feature corresponding to the amplitude attribute (corresponding to local feature 1).

[0094] For example, see Figure 5 As shown, by processing the seismic attribute data volume corresponding to the variance attribute in a similar manner, the local feature corresponding to the variance attribute (corresponding to the local feature 2) can be obtained.

[0095] See Figure 4 and Figure 5 It can be seen that the local features corresponding to the amplitude attribute and the local features corresponding to the variance attribute obtained in the above manner can respectively more clearly characterize the local boundary features and the internal local features of the salt dome.

[0096] In some embodiments, in order to further improve processing accuracy and reduce processing errors, refer to Figure 2 As shown, after obtaining the corresponding multiple seismic attribute data bodies, in specific implementation, the multiple seismic attribute data bodies (for example, seismic attribute data body one, seismic attribute data body two) can be cleaned and / or normalized respectively to eliminate data errors in the multiple seismic attribute data bodies.

[0097] In some embodiments, the above fusion of the multiple local features is used to obtain the overall features of the salt dome in the target area. Figure 6 As shown in the figure, the characteristics and advantages of the local features corresponding to the amplitude attribute and the local features corresponding to the variance attribute can be fully utilized, and the two local features can be combined and used for feature fusion in the following way to obtain a better overall feature for the salt dome in the target area: based on multiple local features, the salt dome boundary is first smoothed by KNN to remove abnormal points; then the smoothed boundary is linearly interpolated to obtain the interpolated local feature; based on the multiple local features after the above processing, the union and then the intersection operation are performed to retain the common part of multiple local features, and the abnormal reflection boundary and some high-variance non-salt dome pixels are removed, so that the overall feature with high accuracy and small error can be obtained.

[0098] The specific processing process can be expressed as follows:

[0099]

[0100] y1=kmeans(PCA(f1 c×h×w )),

[0101]

[0102]

[0103] y=y1+y2

[0104] Among them, CAE encode represents convolutional autoencoder, PCA represents principal component analysis, x1 h×w represents the seismic attribute data volume numbered 1 (for example, the seismic attribute data volume corresponding to the amplitude attribute), f1 c×h×w represents the deep semantic feature numbered 1, y1 represents the local feature numbered 1 (for example, the local feature corresponding to the amplitude attribute), x2 h×w represents the seismic attribute data volume numbered 2 (for example, the seismic attribute data volume corresponding to the variance attribute), f2 c×h×w represents the deep semantic feature numbered 2, y2 represents the local feature numbered 2 (for example, the local feature corresponding to the variance attribute), + represents the fusion operation, and y represents the overall feature obtained by fusing multiple local features.

[0105] In some embodiments, the above-mentioned first identification of the salt dome in the target area based on the overall characteristics to obtain a first identification result may include: characterizing the salt dome in the target area based on the overall characteristics and target seismic data to determine a first position and a first boundary of the salt dome in the target area as the first identification result.

[0106] In some embodiments, after first identifying the salt dome in the target area based on the overall characteristics and obtaining a first identification result, the method may further include the following steps when implemented:

[0107] S1: Detect whether the clarity of the salt dome boundary reflection of the seismic profile in the target area is greater than the preset clarity threshold, and whether there is strong event interference;

[0108] S2: When it is determined that the clarity of the salt dome boundary reflection of the seismic profile in the target area is greater than a preset clarity threshold and there is no strong isotropic interference, the salt dome in the target area is determined according to the first recognition result.

[0109] In some embodiments, when the clarity of the salt dome boundary reflection of the seismic profile of the target area is determined to be less than or equal to a preset clarity threshold, or when there is strong coaxial interference, it can be determined that the salt dome in the target area cannot be accurately determined based on the above-mentioned first recognition result alone. Then, based on the preset training rules, the first recognition result can be used to perform multiple rounds of unsupervised recursive iterative training to continuously update the training set and the model at the same time to obtain a target recognition model with better salt dome recognition effect for the target area; and then the target recognition model can be used to perform a second recognition to accurately and meticulously determine the salt dome in the target area.

[0110] In some embodiments, when implementing Figure 2 As shown, a neural network model based on the residual Unet structure can be constructed as an initial recognition model; at the same time, corresponding labels can be set for part of the seismic data in the target seismic data according to the first recognition result; and the above-mentioned part of the seismic data carrying the labels can be combined to obtain an initial training set.

[0111] Furthermore, this specification also takes into account that since sedimentary movement occurs continuously and the underground geological structure also has strong continuity, the salt dome categories of geological bodies with similar spatial locations will also show similarity to a certain extent, with almost no frequent mutations. Due to the above characteristics, the salt dome classification results of seismic data with similar spatial locations are often basically consistent during the seismic data interpretation process. Furthermore, considering that the above geological laws can be fully utilized, the model can be trained through unsupervised learning by using recursive training and iterative gradual label updates for seismic profiles with complex boundary reflections and strong phase axis interference, so as to identify and determine salt domes based on the final recognition model.

[0112] In some embodiments, see Figure 7 As shown, the above-mentioned method, based on the preset training rules, utilizes the initial training set, and performs multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for salt domes in the target area. In specific implementation, the method may include: performing the current round of recursive iterative training based on the preset training rules in the following manner:

[0113] S1: Based on the training set of the previous round, the test data of the current round is determined from the target seismic data along the specified direction and at the specified step size;

[0114] S2: Use the recognition model of the previous round to process the test data of the current round and determine the corresponding pseudo labels;

[0115] S3: Update the training set of the previous round based on the test data and pseudo labels of the current round to obtain the training set of the current round;

[0116] S4: Use the current round training set to train the recognition model of the previous round to obtain the current round recognition model.

[0117] When training the model, an initial labeled training set can be constructed based on the first recognition result. For example, a small amount of seismic data corresponding to the target seismic data can be labeled based on the first recognition result to obtain a portion of labeled seismic data. Seismic attribute data (e.g., variance attribute data) corresponding to this portion of labeled seismic data can also be obtained and combined to obtain initial training data. Based on this initial training data, an initial training set can be constructed. Furthermore, a neural network model based on a residual Unet structure can be constructed as the initial recognition model.

[0118] During the current round of recursive iterative training, considering that the inline section of the dataset generally does not experience sudden changes in data distribution, and based on expert experience, the data in this direction provides a clear contrast during training, the inline direction was chosen as the designated direction. Furthermore, considering that geological features at adjacent locations exhibit a certain degree of continuity, 10 inline lengths were used as the designated length to determine the test data to be added to the training set for the current round. This ensures that when the recognition model from the previous round is used to process the current round of test data, the recognition results obtained are highly accurate.

[0119] Next, the recognition model from the previous round can be used to process the test data for the current round to obtain the corresponding recognition results (or prediction results). Based on this recognition result, the label for the test data in the current round is determined. Because this label is different from the label of the training data in the initial training set and is determined based on the model, it is recorded here as a pseudo-label.

[0120] Specifically, it can be expressed as follows:

[0121] net(x h,w,m...n ,y h,w,m...n )→y h,w,n+10

[0122] net(x h,w,m...n+10 ,y h,w,m...n+10 )→y h,w,n+20

[0123] Wherein, h represents the height of the input seismic data (eg, grayscale matrix), w represents the width of the input seismic data, m represents the starting inline number of the training data set, and n represents the ending inline number of the training data set.

[0124] Then, the test data of the current round and the corresponding pseudo labels are combined with the corresponding earthquake attribute data as a new set of training data and added to the training set of the previous round to update the training set of the previous round and obtain the training set of the current round.

[0125] The current round of training set can then be used to further train the previous round of recognition model to obtain a current round of recognition model with higher accuracy and better generalization ability, thereby completing the current round of recursive iterative training.

[0126] In some embodiments, after obtaining the recognition model of the current round, the method may further include the following steps when implemented:

[0127] S1: Check whether the preset end condition is met;

[0128] S2: When it is determined that the preset end condition is currently satisfied, the recognition model of the current round is determined as a target recognition model for the salt dome in the target area.

[0129] In some embodiments, after completing the recursive iterative training of the current round, the recognition model of the current round can be used to process the test data of the current round to obtain the corresponding recognition results, which can be saved and recorded.

[0130] In some embodiments, when specifically detecting whether the preset end condition is currently met, you can refer to Figure 2 As shown, according to the recognition results accumulated from the current round of recursive iterative training and combined with the first recognition result, it is detected whether the salt dome interpretation of the entire three-dimensional data volume of the target area has been completed.

[0131] If it is determined that it is not completed, the above process will be repeated to carry out the next round of recursive iterative training. On the contrary, if it is determined that it has been completed and the preset end conditions are currently met, the recursive iterative training will be stopped, and then the salt dome interpretation results of the three-dimensional data volume can be summarized and the second recognition can be carried out to accurately determine the specific location and specific boundaries of the salt dome in the target area. For details, please refer to Figure 8 shown.

[0132] In some embodiments, when specifically detecting whether a preset termination condition is currently satisfied, the determination of whether the preset termination condition is currently satisfied can be made by detecting whether the number of rounds in the current round has reached a specified number. Furthermore, a validation test can be performed on the recognition model in the current round using a validation set to obtain a validation test result. Based on the validation test result, the model accuracy of the recognition model in the current round is detected to determine whether it meets the requirements. If it is determined that the model accuracy of the recognition model in the current round meets the requirements, the preset termination condition is determined to be currently satisfied.

[0133] Based on the above method, when it is determined that the preset end condition is currently met, the recognition model of the current round is determined as the target recognition model for the salt dome in the target area.

[0134] Furthermore, the target seismic data and the corresponding seismic attribute data volume can be processed by using the target recognition model to perform a second recognition to determine the salt domes in the target area.

[0135] The specific implementation may include: using the target recognition model to process the target seismic data and the corresponding seismic attribute data volume to obtain the corresponding target recognition result; and determining the salt dome in the target area according to the target recognition result.

[0136] For details, please refer to Figure 9As shown in FIG, the target seismic data and the corresponding seismic data volume (for example, the seismic attribute data volume corresponding to the variance attribute) can be combined and inputted into the target recognition model, and the target recognition model is run to obtain and output the corresponding target recognition result. Then, based on the target recognition result, the salt domes in the target area can be accurately determined. Among them, Conv can specifically represent a convolution layer, Batchnorm can specifically represent a batch normalization layer, Crop and copy can specifically represent feature fusion, Maxpool can specifically represent maximum pooling, Upsample can specifically represent upsample, Residual convolution can specifically represent residual convolution, and weight layer can specifically represent a parameter layer.

[0137] As can be seen from the above, based on the salt dome identification method provided by the embodiment of this specification, after obtaining the target seismic data of the target area, multiple seismic attribute data bodies can be obtained based on the screened sensitive seismic attributes and the target seismic data; then, a convolutional autoencoder is used to process the multiple seismic attribute data bodies to obtain multiple local features; and the multiple local features are fused to obtain an overall feature; a first recognition is performed based on the overall feature to obtain a first recognition result; then, based on the first recognition result, an initial training set with a label is constructed; based on a preset training rule, the initial training set is used to perform multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt dome in the target area; then, the target recognition model is used to perform a second recognition to determine the salt dome in the target area. Thus, there is no need to rely on and use seismic data with labels, and only seismic data without labels can be used to conveniently, efficiently, and accurately identify and determine the salt dome in the target area.

[0138] In a specific scenario example, the salt dome identification method provided in the embodiments of this specification can be applied to accurately identify and determine the salt domes in the study area. For the specific implementation process, please refer to the following content.

[0139] In the first step (corresponding to the first identification), geological and geophysical knowledge can be combined to screen out some seismic attributes that are more sensitive to the local characteristics of salt domes. Then, the powerful feature extraction capability of the convolutional autoencoder (CAE) and the powerful feature screening capability of cluster analysis can be combined to extract and screen features of multiple attributes that are more sensitive to the local characteristics of salt domes, thereby reconstructing the overall characteristics of salt domes and characterizing salt domes based on seismic data.

[0140] In this scenario, the significant density and velocity differences between the salt rock within the salt dome and the overlying strata result in sharp reflections on the seismic profile at the salt dome boundary. Furthermore, the complex lithologic structure within the salt dome often presents chaotic or blank seismic reflections. Seismic attributes are derived through geometric calculations and can highlight the local characteristics of geological anomalies. For example, root mean square (RMS) and amplitude attributes highlight sharp reflections in seismic data and are effective in delineating salt dome boundaries. Variance and chaos attributes are more sensitive to areas of large variance on the seismic profile and are indicative of the chaotic reflection characteristics of the salt dome. Due to the high level of redundant information in seismic data, these attributes provide a rough representation of the local characteristics of the salt dome and are susceptible to noise or other seismic events. Therefore, we considered leveraging the powerful feature extraction capabilities of convolutional neural networks to extract salt dome attributes using CAE. Principal component analysis and K-means clustering were then used to refine the resulting features. Finally, these local features sensitive to salt domes were aggregated to form a set of global features sensitive to salt domes.

[0141] In this scenario, amplitude and variance attributes can be used as input data for the first step of the device. These attributes can roughly reflect the local characteristics of the salt dome. After CAE, PCA, and K-means clustering are used to obtain the local characteristics of the salt dome, these characteristics need to be smoothed using KNN to filter out non-target areas that are classified as the same as the target characteristics. The resulting salt dome boundary is then linearly interpolated, and the two feature maps are unioned and then intersected. The common portion of the two features is retained, while the anomalous reflective boundary and some high-variance non-salt dome pixels are removed.

[0142] The above steps are the first step in salt dome identification, and are most effective when the salt dome boundary reflections on the seismic profile are clear and free of strong interfering events. This step not only accurately delineates salt domes from seismic data but also provides high-quality labels for deep AI data-driven models. For seismic data with complex salt dome boundary reflections and strong interfering events, the second step of salt dome identification is required to achieve good salt dome identification results.

[0143] In the second step (corresponding to the second identification), recursive training is used to iteratively update the labels and training sets. Specifically, based on the labels obtained in the first step, pseudo labels with certain distribution differences can be continuously produced with a smaller iteration step size and added to the next iteration, thereby realizing the interpretation of salt domes in seismic profiles with complex salt dome boundaries and strong phase axis interference, and then interpreting the salt domes of the entire three-dimensional data volume.

[0144] Given the continuous nature of sedimentary movement and the strong continuity of underground geological structures, salt dome classifications are generally similar between spatially close geological bodies, with few frequent mutations. This characteristic leads to essentially identical salt dome classifications during seismic interpretation for spatially close seismic data. Taking advantage of this geological principle, we employ recursive training to iteratively update labels for salt dome prediction in seismic profiles with complex boundary reflections and strong event interference.

[0145] In this scenario, based on the salt dome interpretation obtained in the first step, a residual Unet structure is used to fit a nonlinear mapping between seismic data and salt dome labels. At each iteration, test results near the training data are treated as pseudo-labels and added to the original training dataset until salt domes are predicted across the entire 3D seismic data set. To minimize the impact of error accumulation, the added pseudo-labels must be as accurate as possible. The device uses the smallest possible iteration step size for pseudo-label addition. For a 385 (Depth) × 768 (Crossline) × 542 (Inline) seismic data set, the device identifies salt domes in the inline direction. During each iteration, the test set is spaced 10 inline lengths away from the training set in the inline direction. Because the test set is spatially close to the training set in each iteration, the resulting pseudo-labels are largely consistent with the true labels. Within the acceptable range of pseudo-label error, this does not significantly impact subsequent iterations.

[0146] Deep neural networks are capable of end-to-end, image-by-image, and pixel-by-pixel learning. In the field of salt dome identification in seismic interpretation, pixel-by-pixel labeling is prohibitively expensive, and accurate labeling requires sophisticated data acquisition equipment, which is difficult to achieve with real-world geological data. Consequently, this makes it difficult to meet the needs of artificial intelligence for large amounts of seismic data labeled with salt domes. Neural networks often struggle to map the most common relationships between seismic data and salt dome labels, leading to overfitting or even underfitting of the trained weak neural networks on the test dataset. This is a fundamental problem with existing deep learning salt dome identification techniques. Due to the significant geological structural differences between training and test datasets across different regions, neural networks are unable to bridge this gap. This often requires feeding the neural network with more labeled data from test regions to improve its generalization ability. By employing CAE-PCA-K-means and recursive training to iteratively update labels and training sets, it is possible to automatically interpret salt domes using artificial intelligence based on their characteristics across different seismic attributes, without manual labeling.

[0147] This can improve the accuracy of salt dome interpretation while ensuring its efficiency, and effectively and accurately identify and determine the salt domes in the study area.

[0148] Through the above scenario examples, the salt dome identification method provided in this specification has been verified. Under the guidance of seismic attribute knowledge, it combines the powerful feature extraction and purification capabilities of convolutional autoencoders and K-means to extract local features of salt domes from different attribute volumes, and then aggregate these features into overall features sensitive to salt domes. On this basis, for more complex seismic data volumes, based on the strong continuity of salt dome data in underground geological structures, a recursive training iterative method of gradually updating labels and training sets is used to perform incremental knowledge interpretation of salt domes in the entire three-dimensional data volume. This can effectively alleviate the problem of traditional deep artificial intelligence salt dome identification devices training on small data sets and having poor generalization performance in different blocks. Moreover, under the guidance of seismic attribute knowledge, it can give full play to the advantages of the powerful feature extraction and purification of artificial intelligence data-driven models, achieve improved generalization performance on discontinuous test sets when the amount of seismic sample data is small, shorten the cycle of manual salt dome interpretation, and improve the efficiency of salt dome interpretation. In addition, considering the serious problem of overfitting of small sample training neural networks on discontinuous data, based on the powerful feature learning function of artificial intelligence for seismic big data, rich seismic data are continuously added to the neural network. When the pseudo-label error range allows, the neural network is made to fit the more general relationship mapping between seismic data and salt dome labels as much as possible, thereby improving its generalization ability on discontinuous seismic data and providing strong technical support for the development of refined and efficient seismic data interpretation.

[0149] An embodiment of the present specification also provides a computer device, including a processor and a memory for storing processor-executable instructions. When specifically implemented, the processor can perform the following steps according to the instructions: obtaining target seismic data of a target area; and screening out sensitive seismic attributes for local features of salt domes in the target area; obtaining corresponding multiple seismic attribute data volumes based on the sensitive seismic attributes and the target seismic data; processing the multiple seismic attribute data volumes separately by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area; fusing the multiple local features to obtain overall features of the salt dome in the target area; performing a first identification on the salt dome in the target area based on the overall features to obtain a first identification result; constructing an initial training set with a label based on the first identification result; based on preset training rules, using the initial training set, performing multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt dome in the target area; performing a second identification by processing the target seismic data and the corresponding seismic attribute data volumes using the target recognition model to determine the salt dome in the target area.

[0150] In order to complete the above instructions more accurately, refer to Figure 10 The embodiments of this specification also provide another specific computer device, wherein the computer device includes a network communication port 1001, a processor 1002 and a memory 1003, and the above structures are connected through internal cables so that each structure can perform specific data interaction.

[0151] The network communication port 1001 can be used to obtain target seismic data of a target area and screen out sensitive seismic attributes of local features of salt domes in the target area.

[0152] The processor 1002 can be specifically used to obtain corresponding multiple seismic attribute data volumes based on the sensitive seismic attributes and the target seismic data; process the multiple seismic attribute data volumes separately by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area; fuse the multiple local features to obtain the overall features of the salt dome in the target area; perform a first identification on the salt dome in the target area based on the overall features to obtain a first identification result; construct an initial training set with a label based on the first identification result; based on a preset training rule, use the initial training set to perform multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt dome in the target area; perform a second identification by using the target recognition model to process the target seismic data and the corresponding seismic attribute data volumes to determine the salt dome in the target area.

[0153] The memory 1003 may be specifically used to store corresponding instruction programs.

[0154] In this embodiment, the network communication port 1001 can be a virtual port that is bound to different communication protocols, thereby being capable of sending or receiving different data. For example, the network communication port can be a port responsible for web data communication, a port responsible for FTP data communication, or a port responsible for email data communication. Furthermore, the network communication port can also be a physical communication interface or communication chip. For example, it can be a wireless mobile network communication chip, such as GSM or CDMA; it can also be a Wi-Fi chip; or it can be a Bluetooth chip.

[0155] In this embodiment, the processor 1002 may be implemented in any suitable manner. For example, the processor may take the form of a microprocessor or a processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application-specific integrated circuit (ASIC), a programmable logic controller, an embedded microcontroller, etc. This specification is not intended to limit this.

[0156] In this embodiment, the memory 1003 may include multiple levels. In a digital system, anything that can store binary data can be a memory. In an integrated circuit, a circuit with a storage function that has no physical form is also called a memory, such as RAM, FIFO, etc. In a system, a storage device with a physical form is also called a memory, such as a memory stick, TF card, etc.

[0157] The present specification also provides a computer storage medium based on the above-mentioned salt dome identification method, wherein the computer storage medium stores computer program instructions, which, when executed, implement the following steps: acquiring target seismic data of a target area; screening out sensitive seismic attributes for local features of salt domes in the target area; acquiring corresponding multiple seismic attribute data volumes based on the sensitive seismic attributes and the target seismic data; processing the multiple seismic attribute data volumes separately using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area; fusing the multiple local features to obtain an overall feature of the salt dome in the target area; performing a first identification on the salt dome in the target area based on the overall feature to obtain a first identification result; constructing an initial training set with labels based on the first identification result; performing multiple rounds of recursive iterative training on the initial identification model using the initial training set based on preset training rules to obtain a target identification model for the salt dome in the target area; and performing a second identification by processing the target seismic data and the corresponding seismic attribute data volumes using the target identification model to determine the salt dome in the target area.

[0158] In this embodiment, the storage medium includes, but is not limited to, random access memory (RAM), read-only memory (ROM), cache, hard disk drive (HDD), or memory card. The memory can be used to store computer program instructions. The network communication unit can be an interface configured in accordance with the standards specified by the communication protocol for network connection communication.

[0159] In this embodiment, the functions and effects specifically implemented by the program instructions stored in the computer storage medium can be explained in comparison with other implementations and will not be repeated here.

[0160] See Figure 11 As shown, at the software level, the embodiments of this specification also provide a salt dome identification device, which may specifically include the following structural modules:

[0161] The first acquisition module 1101 may be used to acquire target seismic data of a target area and filter out sensitive seismic attributes of local features of salt domes in the target area;

[0162] The second acquisition module 1102 may be specifically configured to acquire a plurality of corresponding seismic attribute data volumes according to the sensitive seismic attributes and the target seismic data;

[0163] The third acquisition module 1103 may be specifically configured to process the multiple seismic attribute data volumes respectively by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area;

[0164] A fusion module 1104 may be used to fuse the multiple local features to obtain an overall feature of the salt dome in the target area;

[0165] A first identification module 1105 may be specifically configured to perform a first identification on the salt dome in the target area according to the overall characteristics to obtain a first identification result;

[0166] A construction module 1106 may be specifically configured to construct an initial training set with labels based on the first recognition result;

[0167] The training module 1107 may be specifically configured to perform multiple rounds of iterative training on the initial recognition model based on a preset training rule and using the initial training set to obtain a target recognition model for salt domes in the target area;

[0168] The second identification module 1108 may be specifically configured to process the target seismic data and the corresponding seismic attribute data volume using the target identification model to perform a second identification to determine the salt domes in the target area.

[0169] In some embodiments, the sensitive seismic attributes may specifically include at least two of the following seismic attributes: root mean square attribute, amplitude attribute, variance attribute, Chaos attribute, etc.

[0170] In some embodiments, when the above-mentioned third acquisition module 1103 is implemented, the current seismic attribute data body among multiple seismic attribute data bodies can be processed by using a convolutional autoencoder to obtain corresponding local features in the following manner: the current seismic attribute data body is processed by using a convolutional autoencoder, and multiple deep semantic features output by a specified intermediate network layer of the convolutional autoencoder are extracted; the multiple deep semantic features are classified to obtain multiple deep semantic feature groups; based on the multiple deep semantic feature groups, a main feature group is determined through principal component analysis; based on the main feature group, K-means clustering processing is performed to obtain corresponding local features.

[0171] In some embodiments, when the first identification module 1105 is specifically implemented, the salt dome in the target area can be first identified based on the overall characteristics in the following manner to obtain a first identification result: the salt dome in the target area is characterized based on the overall characteristics and target seismic data to determine the first position and first boundary of the salt dome in the target area as the first identification result.

[0172] In some embodiments, after the salt domes in the target area are first identified based on the overall characteristics and a first identification result is obtained, the device can also be used to detect whether the clarity of the salt dome boundary reflection of the seismic profile of the target area is greater than a preset clarity threshold and whether there is strong event axis interference; when it is determined that the clarity of the salt dome boundary reflection of the seismic profile of the target area is greater than the preset clarity threshold and there is no strong event axis interference, the salt dome in the target area is determined according to the first identification result.

[0173] On the contrary, when the clarity of the salt dome boundary reflection of the seismic profile in the target area is determined to be less than or equal to the preset clarity threshold, or when there is strong isotropic interference, it is necessary to call the second identification module 1108 to accurately determine the salt dome in the target area by performing a second identification.

[0174] In some embodiments, when the above-mentioned training module 1107 is implemented, the recursive iterative training of the current round can be performed in the following manner based on the preset training rules: based on the training set of the previous round, along the specified direction and at intervals of specified steps, the test data of the current round is determined from the target seismic data; the recognition model of the previous round is used to process the test data of the current round to determine the corresponding pseudo-labels; according to the test data and pseudo-labels of the current round, the training set of the previous round is updated to obtain the training set of the current round; the recognition model of the previous round is trained using the training set of the current round to obtain the recognition model of the current round.

[0175] In some embodiments, after obtaining the recognition model of the current round, the training module 1107 can also be used to detect whether a preset end condition is currently met during implementation; if it is determined that the preset end condition is currently met, the recognition model of the current round is determined as the target recognition model for the salt dome in the target area.

[0176] In some embodiments, when the second identification module 1108 is specifically implemented, the target seismic data and the corresponding seismic attribute data volume can be processed by using the target identification model to perform second identification to determine the salt dome in the target area in the following manner: the target seismic data and the corresponding seismic attribute data volume are processed by using the target identification model to obtain a corresponding target identification result; and the salt dome in the target area is determined based on the target identification result.

[0177] It should be noted that the units, devices or modules described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. For the convenience of description, the above devices are described in terms of functions and are divided into various modules and described separately. Of course, when implementing this specification, the functions of each module can be implemented in the same or multiple software and / or hardware, or the module that implements the same function can be implemented by a combination of multiple sub-modules or sub-units. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0178] As can be seen from the above, the salt dome identification device provided by the embodiment of this specification, after obtaining the target seismic data of the target area, can first obtain multiple seismic attribute data bodies based on the screened sensitive seismic attributes and the target seismic data; then use the convolutional autoencoder to process the multiple seismic attribute data bodies to obtain multiple local features; and fuse the multiple local features to obtain the overall feature; perform a first recognition based on the overall feature to obtain a first recognition result; then, based on the first recognition result, construct an initial training set with a label; based on the preset training rules, use the initial training set, and perform multiple rounds of recursive iterative training on the initial recognition model to obtain a target recognition model for the salt dome in the target area; then use the target recognition model to perform a second recognition to determine the salt dome in the target area. Thus, it is possible to conveniently, efficiently, and accurately identify and determine the salt dome in the target area by using only the seismic data without relying on and using the seismic data with labels.

[0179] Although this specification provides the method operation steps as described in the embodiments or flow charts, more or fewer operation steps may be included based on conventional or non-creative means. The order of steps listed in the embodiments is only one way of executing the order of many steps and does not represent the only execution order. When the device or client product in practice is executed, it can be executed in sequence or in parallel according to the method shown in the embodiments or the drawings (for example, a parallel processor or a multi-threaded processing environment, or even a distributed data processing environment). The term "comprise", "include" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, product or device including a series of elements includes not only those elements, but also includes other elements that are not explicitly listed, or also includes elements inherent to such process, method, product or device. In the absence of more restrictions, it is not excluded that there are other identical or equivalent elements in the process, method, product or device including the elements. Words such as first and second are used to represent names and do not represent any particular order.

[0180] Those skilled in the art will also appreciate that, in addition to implementing the controller in pure computer-readable program code, it is entirely possible to implement the same functionality by logically programming the method steps in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, embedded microcontrollers, and the like. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be considered structures within the hardware component. Alternatively, the devices for implementing various functions can be considered both software modules implementing the method and structures within the hardware component.

[0181] This specification may be described in the general context of computer-executable instructions, such as program modules, executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, classes, and the like that perform specific tasks or implement specific abstract data types. This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media, including storage devices.

[0182] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that this specification can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of this specification can essentially be embodied in the form of a software product. This computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a mobile terminal, a server, or a network device, etc.) to execute the methods described in various embodiments or certain parts of the embodiments of this specification.

[0183] The various embodiments in this specification are described in a progressive manner. References to the common or similar parts of the various embodiments are sufficient. Each embodiment focuses on the differences from the other embodiments. This specification can be used in a variety of general-purpose or specialized computer system environments or configurations. For example, personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable electronic devices, network PCs, minicomputers, mainframe computers, and distributed computing environments that include any of the above systems or devices.

[0184] Although the present specification has been described through embodiments, those skilled in the art will appreciate that there are many modifications and variations to the present specification without departing from the spirit of the present specification. It is intended that the appended claims include these modifications and variations without departing from the spirit of the present specification.

Claims

1. A method for identifying salt domes, characterized in that: include: Acquiring target seismic data in a target area; and screen out sensitive seismic attributes for the local characteristics of salt domes in the target area; Acquiring corresponding multiple seismic attribute data volumes according to the sensitive seismic attributes and the target seismic data; Processing the multiple seismic attribute data volumes separately by using a convolutional autoencoder to obtain multiple local features of the salt dome in the target area; fusing the multiple local features to obtain an overall feature of the salt dome in the target area; performing a first identification of the salt domes in the target area according to the overall characteristics to obtain a first identification result; Constructing an initial training set with labels based on the first recognition result; Based on a preset training rule, using the initial training set, the initial recognition model is trained through multiple rounds of recursive iterations to obtain a target recognition model for salt domes in the target area; Performing a second recognition by processing the target seismic data and the corresponding seismic attribute data volume using the target recognition model to determine the salt dome in the target area; The method comprises the following steps: processing the multiple seismic attribute data volumes by using a convolutional autoencoder respectively to obtain multiple local features of the salt dome in the target area, the method comprising: processing a current seismic attribute data volume among the multiple seismic attribute data volumes by using a convolutional autoencoder to obtain corresponding local features: processing the current seismic attribute data volume by using a convolutional autoencoder and extracting multiple deep semantic features output by a specified intermediate network layer of the convolutional autoencoder; classifying the multiple deep semantic features to obtain multiple deep semantic feature groups; determining a main feature group by principal component analysis based on the multiple deep semantic feature groups; and performing K-means clustering processing based on the main feature group to obtain corresponding local features. Furthermore, the method further includes: detecting whether the clarity of the salt dome boundary reflection of the seismic profile of the target area is greater than a preset clarity threshold, and whether there is strong event interference; and determining the salt dome in the target area according to the first recognition result when it is determined that the clarity of the salt dome boundary reflection of the seismic profile of the target area is greater than the preset clarity threshold and there is no strong event interference.

2. The method according to claim 1, characterized in that The sensitive seismic attributes include at least two of the following seismic attributes: a root mean square attribute, an amplitude attribute, a variance attribute, and a chaos attribute.

3. The method according to claim 1, characterized in that Based on the overall characteristics, a first identification is performed on the salt domes in the target area to obtain a first identification result, including: The salt dome in the target area is characterized according to the overall characteristics and the target seismic data to determine a first position and a first boundary of the salt dome in the target area as a first identification result.

4. The method according to claim 1, wherein Based on the preset training rules and using the initial training set, the initial recognition model is trained through multiple rounds of recursive iterations to obtain a target recognition model for salt domes in the target area, including: Perform the current round of recursive iterative training based on the preset training rules in the following manner: Based on the training set of the previous round, the test data of the current round is determined from the target seismic data along the specified direction and at the specified step size; Use the recognition model of the previous round to process the test data of the current round and determine the corresponding pseudo label; Based on the test data and pseudo labels of the current round, the training set of the previous round is updated to obtain the training set of the current round; The recognition model of the previous round is trained using the training set of the current round to obtain the recognition model of the current round.

5. The method according to claim 4, characterized in that After obtaining the recognition model of the current round, the method further includes: Check whether the preset end condition is currently met; When it is determined that the preset end condition is currently satisfied, the recognition model of the current round is determined as the target recognition model for the salt dome in the target area.

6. The method according to claim 1, characterized in that The target seismic data and the corresponding seismic attribute data are processed using the target recognition model to perform a second recognition to determine the salt domes in the target area, including: Utilize the target recognition model to process the target seismic data and the corresponding seismic attribute data volume to obtain the corresponding target recognition result; Based on the target identification results, the salt domes in the target area are determined.

7. A salt dome identification device, characterized in that: include: A first acquisition module is used to acquire target seismic data of a target area; and screen out sensitive seismic attributes for the local characteristics of salt domes in the target area; A second acquisition module is configured to acquire a plurality of corresponding seismic attribute data volumes according to the sensitive seismic attributes and the target seismic data; a third acquisition module, configured to process the plurality of seismic attribute data volumes respectively by using a convolutional autoencoder to acquire a plurality of local features of the salt dome in the target area; A fusion module, configured to fuse the multiple local features to obtain an overall feature of the salt dome in the target area; a first recognition module, configured to perform a first recognition on the salt dome in the target area according to the overall characteristics, and obtain a first recognition result; A construction module, configured to construct an initial training set with labels according to the first recognition result; A training module is used to perform multiple rounds of iterative training on the initial recognition model based on preset training rules and using the initial training set to obtain a target recognition model for salt domes in the target area; A second identification module is used to process the target seismic data and the corresponding seismic attribute data volume by using the target identification model to perform a second identification to determine the salt dome in the target area; The third acquisition module is specifically used to process the current earthquake attribute data volume among the multiple earthquake attribute data volumes by using a convolutional autoencoder to obtain the corresponding local features in the following manner: using the convolutional autoencoder to process the current earthquake attribute data volume and extracting multiple deep semantic features output by a specified intermediate network layer of the convolutional autoencoder; classifying the multiple deep semantic features to obtain multiple deep semantic feature groups; determining a main feature group based on the multiple deep semantic feature groups through principal component analysis; and performing K-means clustering processing based on the main feature group to obtain corresponding local features; Furthermore, the device is further configured to detect whether the clarity of the salt dome boundary reflection of the seismic profile in the target area is greater than a preset clarity threshold, and whether there is strong event interference; when it is determined that the clarity of the salt dome boundary reflection of the seismic profile in the target area is greater than the preset clarity threshold, and there is no strong event interference, the salt dome in the target area is determined according to the first recognition result.

8. A computer device, characterized in that: The method comprises a processor and a memory for storing processor-executable instructions, wherein the processor implements the steps of the method according to any one of claims 1 to 6 when executing the instructions.

Images