Geologic body identification method, device, apparatus, and storage medium

Abstract

The application relates to the technical field of seismic data interpretation, in particular to a geological body identification method, device and equipment and a storage medium, wherein the method comprises the following steps: acquiring a seismic profile graph sequence, the seismic profile graph sequence comprising a benchmark seismic profile graph manually labeled and continuous seismic profile graphs without manual labeling; inputting the seismic profile graph sequence into a target geological body identification model for processing to obtain a geological body identification result; wherein the target geological body identification model is fused with a self-attention mechanism, and the self-attention mechanism is used for fusing image features of each seismic profile graph in the seismic profile graph sequence to obtain global features. The application facilitates improving the identification effect of the geological body.

Description

Technical Field

[0001] This application relates to the field of seismic data interpretation technology, and in particular to a method, apparatus, equipment and storage medium for identifying geological bodies. Background Technology

[0002] Seismic exploration often involves identifying geological bodies, which can be composed of various rocks and minerals. For example, granite geological bodies are mainly composed of minerals such as quartz, feldspar, and mica. In addition, geological bodies can also include soil.

[0003] Currently, a commonly used method for geological body identification is as follows: first, seismic images of geological bodies are acquired; then, geological bodies in the seismic images are manually labeled to obtain training sample data; next, a pre-set geological body identification model is trained based on the training sample data to obtain a trained geological body identification model; furthermore, the seismic images of geological bodies are processed through this geological body identification model to identify the geological bodies in the seismic images.

[0004] However, on the one hand, manual annotation of geological bodies in seismic images can only achieve two-dimensional annotation, making it difficult for geological body recognition models to extract the three-dimensional spatial features of geological bodies. On the other hand, in order to improve the diversity of the labeled data, manual annotation of geological bodies is generally required on seismic images with a certain distance distribution. However, due to the quality and complexity of seismic data, the labels are prone to bias. These two reasons lead to the current geological body recognition models trained on two-dimensional training data often exhibiting a "patchy phenomenon" when recognizing geological bodies. Figure 1 As shown, the boundary of the identified geological body (red area in the figure) will fluctuate, resulting in poor identification of the geological body. Summary of the Invention

[0005] To improve the identification of geological bodies, this application provides a method, apparatus, equipment, and storage medium for identifying geological bodies.

[0006] In a first aspect, this application provides a method for identifying geological bodies, including:

[0007] Obtain a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps;

[0008] The seismic profile sequence is input into the target geological body identification model for processing to obtain the geological body identification result;

[0009] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0010] Secondly, this application provides a geological body identification device, comprising:

[0011] The sequence acquisition module is used to acquire a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0012] The result recognition module is used to input the seismic profile sequence into the target geological body recognition model for processing to obtain the geological body recognition result;

[0013] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0014] Thirdly, this application provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the method described above.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method.

[0016] Fifthly, this application also provides a computer program product. The computer program product includes a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0017] The aforementioned geological body identification method, apparatus, equipment, and storage medium acquire a sequence of seismic profile images, including manually annotated baseline seismic profile images and unannotated continuous seismic profile images. The sequence is then input into a target geological body identification model for processing to obtain the geological body identification result. The target geological body identification model incorporates a self-attention mechanism, which fuses image features from each seismic profile image in the sequence to obtain global features. Through this implementation, because the self-attention mechanism of the target geological body identification model can fuse image features from each seismic profile image in the sequence, it enables the model to extract three-dimensional spatial features from two-dimensional seismic profile images. Furthermore, manual annotation of only the baseline seismic profile image in the sequence is required, reducing the likelihood of errors in the annotations. This prevents the geological body identification model from exhibiting a "sheet-like" phenomenon when identifying geological bodies, thereby improving the identification effect.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is an example diagram provided in the background art of this application to illustrate the "sheet-like phenomenon";

[0021] Figure 2 This is a flowchart of a geological body identification method provided in the embodiments of this application;

[0022] Figure 3 This is a schematic diagram of a seismic profile sequence provided in an embodiment of this application;

[0023] Figure 4 This is an example image provided in the embodiments of this application for improving the identification effect of example geological bodies;

[0024] Figure 5 This is a schematic diagram of the structure of a geological body identification device provided in the embodiments of this application;

[0025] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application;

[0026] Figure 7 This is an internal structural diagram of a computer-readable storage medium provided in an embodiment of this application. Detailed Implementation

[0027] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this disclosure.

[0028] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings herein are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0029] In this article, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0030] Example 1

[0031] Figure 2 This is a flowchart of a geological body identification method provided in Embodiment 1 of this application, with reference to... Figure 2 The method can be executed by a device that performs the method, which can be implemented in software and / or hardware, and the method includes:

[0032] S110. Obtain a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0033] In seismic exploration, to obtain seismic data and achieve the exploration objectives, shot points and multiple receiver points are typically set up. A seismic source is installed at the shot point to generate seismic waves, and a geophone is installed at the receiver points to detect the seismic wave signals generated by the seismic waves. To facilitate understanding of the meaning of the seismic profile sequence, in this embodiment, multiple receiver points are evenly distributed along a horizontal straight line; in other embodiments, the specific distribution is not limited. The seismic wave signals detected by the receiver points can be used to generate seismic profiles. For example, a seismic profile can be referenced... Figure 1 The seismic wave signal from each receiver point can generate a corresponding seismic profile map, and the seismic profile maps generated from multiple receiver points together form a seismic profile map sequence.

[0034] The data volume of each receiver point is set to an odd number. The receiver point located at the center of multiple receiver points is designated as the reference receiver point, and the seismic profile corresponding to this reference receiver point is recorded as the reference seismic profile. The seismic profiles corresponding to the remaining receiver points are recorded as continuous seismic profiles. The index of the reference seismic profile in the seismic profile sequence is set to l. The indexes of each seismic profile in the seismic profile sequence are: lk*s, l-(k-1)*s, ..., l, ..., l-(k+1)*s, l+k*s. The seismic profile sequence comes from... The specific process of generating the seismic profile sequence, derived from the seismic profiles corresponding to each receiver point, is as follows: First, obtain the seismic profile corresponding to the reference receiver point to obtain the reference seismic profile. Then, based on the reference seismic profile, continuously acquire the same number of seismic profiles in the directions before and after it to obtain continuous seismic profiles in both directions. Alternatively, after obtaining the reference seismic profile, acquire seismic profiles once in the directions before and after it with one or more seismic profiles at intervals, based on the reference seismic profile, to obtain seismic profiles in both directions.

[0035] In this embodiment, a reference seismic profile is first obtained by acquiring the seismic profile corresponding to the reference receiver point. Then, the same number of seismic profiles are continuously acquired in both directions from the reference seismic profile to obtain a continuous seismic profile sequence. The 's' in the sequence number of each seismic profile represents the step size. In this embodiment, when acquiring the seismic profile sequence, s = 1, indicating that the same number of seismic profiles are continuously acquired in both directions from the reference seismic profile. In other embodiments, if s = 2, it indicates that a seismic profile is acquired once every one seismic profile in both directions from the reference seismic profile, and so on. The 'k' in the sequence number of each seismic profile is generally 1 or 2 in this embodiment, but is not specifically limited in other embodiments.

[0036] In this embodiment, after the acquisition of the baseline seismic profile is completed, the geological bodies on the baseline seismic profile will be manually labeled; it should be noted that continuous seismic profiles will not be manually labeled.

[0037] Specifically, a baseline seismic profile is obtained, and then manually annotated. Continuous seismic profiles along the directions preceding and following the baseline profile are also obtained. The manually annotated baseline seismic profile and each continuous seismic profile without annotations are obtained, and then arranged sequentially according to the order in which the corresponding receiver points lie on a straight line, resulting in a seismic profile sequence. This seismic profile sequence can be used as an example for reference. Figure 3 .

[0038] S120. Input the seismic profile sequence into the target geological body identification model for processing to obtain the geological body identification result.

[0039] The target geological body recognition model is based on a two-dimensional convolutional network used for semantic segmentation. In this embodiment, the two-dimensional convolutional network can be one of U-net, FCN, or Deeplab networks; in other embodiments, the specific implementation is not limited. The target geological body recognition model is a pre-trained geological body recognition model used to process seismic profile sequence to obtain the geological body recognition result corresponding to the seismic profile sequence. The geological body recognition result is the geological body profile image displayed on the seismic profile, which can be exemplarily referenced. Figure 1 The red area in the middle.

[0040] Specifically, the acquired seismic profile sequence is input into the target geological body recognition model, which is in a completed training state, for processing. The target geological body recognition model outputs the geological body recognition result.

[0041] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0042] The self-attention mechanism is designed based on the seismic profile sequence and embedded (fused) into the target geological body recognition model. It is used to process the seismic profile sequence to fuse the image features of each seismic profile in the sequence. In this embodiment, the image features include color features, texture features, shape features, and spatial relationship features. The color features specifically include color histogram and color distance. The texture features specifically include gray-level co-occurrence matrix and local binary pattern. The shape features specifically include contour features and region features. The spatial relationship features include relative positional relationship and topological relationship.

[0043] It should be noted that this embodiment obtains a sequence of seismic profile images, including manually annotated baseline seismic profile images and unannotated continuous seismic profile images. The sequence is then input into a target geological body recognition model for processing to obtain geological body recognition results. This target geological body recognition model incorporates a self-attention mechanism, which fuses image features from each seismic profile image in the sequence to obtain global features. Through this implementation, because the self-attention mechanism of the target geological body recognition model can fuse image features from each seismic profile image in the sequence, it enables the model to extract three-dimensional spatial features from two-dimensional seismic profile images. Furthermore, manual annotation of the baseline seismic profile image in the sequence is sufficient, reducing the likelihood of errors in the annotations. This prevents the geological body recognition model from exhibiting a "sheet-like phenomenon" when recognizing geological bodies. Figure 1 The identified geological body images exhibit jagged edges, or discontinuous edges. This improved recognition process enhances the identification of geological bodies, essentially eliminating jagged edges (discontinuous edges) in the final image. (See reference for more information.) Figure 4 This means that the "sheet-like phenomenon" in the geology identified by existing technologies has been resolved.

[0044] Example 2

[0045] This application provides a geological body identification method in Embodiment 2, which refines the step in Embodiment 1 of "inputting the seismic profile sequence into the target geological body identification model for processing to obtain the geological body identification result." It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0046] S210. Obtain a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0047] S221. Determine the tensor size of the seismic profile sequence to obtain the first sequence tensor size; the first sequence tensor size includes the batch size, image height, image width, and number of images.

[0048] Wherein, the tensor size is used to describe the information of the seismic profile sequence, and the tensor size of the first sequence is the tensor size of the seismic profile sequence; the batch size B is the number of seismic profiles input into the target geological body identification model from the seismic profile sequence in a single operation; each seismic profile in the seismic profile sequence has the same height and width, the image height H is the height of the seismic profile in the seismic profile sequence, and the image width W is the width of the seismic profile in the seismic profile sequence; the number of images is the number of seismic profiles in the seismic profile sequence, and in this embodiment, the number of images is 2k+1; the tensor size of the first sequence can be represented as (B, H, W, 2k+1).

[0049] Specifically, the number of seismic profiles input into the target geological body identification model from the seismic profile sequence in a single step is determined to obtain the batch number B; the height of the seismic profiles in the seismic profile sequence is determined to obtain the image height H; the width of the seismic profiles in the seismic profile sequence is determined to obtain the image width W; and the number of seismic profiles in the seismic profile sequence is determined to be 2k+1, resulting in the first sequence tensor size (B,H,W,2k+1) of the seismic profile sequence.

[0050] S222. Fix the batch size in the first sequence tensor size to obtain the second sequence tensor size.

[0051] In this embodiment, the number of seismic profiles input into the target geological body identification model from the seismic profile sequence at one time is 1, that is, the batch number B is fixed at 1; the tensor size of the second sequence is the tensor size obtained after the batch number B in the tensor size of the first sequence is fixed at 1, and the tensor size of the second sequence is denoted as (1,H,W,2k+1).

[0052] Specifically, the batch size B in the first sequence tensor size (B,H,W,2k+1) is fixed to 1 to obtain the second sequence tensor size (1,H,W,2k+1).

[0053] S223. Perform a dimensionality transformation on the batch size and the number of images in the second sequence tensor to obtain a feature tensor.

[0054] In this embodiment, the tensor size includes four dimensions: batch size B, image height H, image width W, and number of images. Dimension transformation involves swapping the positions of any two dimensions in the tensor size. The feature tensor is obtained by performing a dimension transformation on the batch size and the number of images in the second sequence tensor size. The feature tensor can be represented as (2k+1,H,W,1).

[0055] Specifically, the number of batches and the number of images in the second sequence tensor are transformed to obtain the feature tensor (2k+1,H,W,1).

[0056] It should be noted that the tensor size input into the traditional 2D convolutional network is in the format of the first sequence tensor size (B, H, W, 2k+1). The traditional 2D convolutional network superimposes image features on the third dimension (the dimension containing 2k+1, and the dimension containing B is the 0th dimension) of the first sequence tensor size (B, H, W, 2k+1). However, this method of image feature superposition leads to the loss of spatial information between different seismic profiles, which in turn results in the geological body image being subject to the "..." described in the background technology. The "sheet-like phenomenon" is addressed in this embodiment. Based on the first sequence tensor size (B, H, W, 2k+1), the batch size B is further fixed to 1, and the batch size B and the number of images 2k+1 are transformed to obtain the feature tensor (2k+1, H, W, 1). In this way, the two-dimensional convolutional network can avoid the image feature superposition and confusion between different seismic profiles during the subsequent processing of the feature tensor (2k+1, H, W, 1), thereby avoiding the above-mentioned "sheet-like phenomenon" and improving the recognition effect of geological bodies.

[0057] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0058] Example 3

[0059] This application provides a geological body identification method in Embodiment 3, which refines the "fusion of image features of each seismic profile in a seismic profile sequence to obtain global features" method in Embodiment 2. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0060] S310. Obtain a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0061] S321. Determine the tensor size of the seismic profile sequence to obtain the first sequence tensor size; the first sequence tensor size includes the batch size, image height, image width, and number of images.

[0062] S322. Fix the batch size in the first sequence tensor size to obtain the second sequence tensor size.

[0063] S323. Perform a dimensionality transformation on the batch number and the number of images in the second sequence tensor to obtain a feature tensor.

[0064] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0065] Specifically, global features are obtained by fusing image features from each seismic profile sequence, including:

[0066] A310. Perform convolution operation on the feature tensor corresponding to the seismic profile sequence to obtain a first query vector, a first key vector, and a first value vector.

[0067] To improve the identification of geological bodies, this embodiment integrates a self-attention mechanism into the target geological body identification model. The self-attention mechanism can be used to extract global features from seismic profiles, thereby facilitating the integration of local features extracted by the convolutional layers in a two-dimensional convolutional network, as well as other distant contour features or important texture features that are highly related to these local features, thus improving the identification of geological bodies. To extract global features from seismic profiles, it is necessary to first perform convolution operations on the feature tensor I corresponding to the seismic profile sequence. In this embodiment, a 1×1 convolution kernel is used to perform three convolution operations on the feature tensor I to obtain the first query vector Q1, the first key vector K1, and the first value vector V1, respectively.

[0068] Specifically, the feature tensor I corresponding to the seismic profile sequence is obtained, and the feature tensor I is convolved three times using a preset convolution kernel to obtain the first query vector Q1, the first key vector K1, and the first value vector V1, respectively.

[0069] A320. Perform dimensional transformation on the first query vector, the first key vector, and the first value vector to obtain the second query vector, the second key vector, and the second value vector.

[0070] In this context, the tensor dimensions corresponding to the first query vector Q1, the first key vector K1, and the first value vector V1 are all (2k+1, H, W, C), where C is the number of feature channels. The dimensional transformation here involves placing the image height H in the 0th dimension, the image width W in the 1st dimension, the number of images 2k+1 in the 2nd dimension, and the number of feature channels C in the 3rd dimension. The second query vector Q2, the second key vector K2, and the second value vector V2 are the new query vector, key vector, and value vector obtained after the dimensional transformation of the tensor dimensions corresponding to the first query vector Q1, the first key vector K1, and the first value vector V1, respectively. The tensor dimensions corresponding to the second query vector Q2, the second key vector K2, and the second value vector V2 are all (H, W, 2k+1, C).

[0071] Specifically, the tensor dimensions (2k+1, H, W, C) corresponding to the first query vector Q1, the first key vector K1, and the first value vector V1 are obtained. The image height H dimension is placed in the 0th dimension, the image width W dimension is placed in the 1st dimension, the number of images 2k+1 dimension is placed in the 2nd dimension, and the number of feature channels C dimension is placed in the 3rd dimension, thereby obtaining the second query vector Q2, the second key vector K2, and the second value vector V2.

[0072] A330. Calculate global features based on the second query vector, the second key vector, and the second value vector.

[0073] The global features are calculated based on the second query vector Q2, the second key vector K2, and the second value vector V2.

[0074] It should be noted that while convolutional layers in a two-dimensional convolutional network can extract local features from seismic profiles, the self-attention mechanism integrated into the two-dimensional convolutional network in this embodiment can further extract global features from seismic profiles. The local features provided by the convolutional layers provide a targeted information basis for the self-attention mechanism, enabling it to more effectively capture important information related to local features within the global scope.

[0075] Example 4

[0076] This application provides a geological body identification method in Embodiment 4, which refines the "calculation of global features based on the second query vector, the second key vector, and the second value vector" in Embodiment 3. It should be noted that for parts not detailed in this embodiment, please refer to the descriptions in other embodiments. The method includes:

[0077] S421. Determine the tensor size of the seismic profile sequence to obtain the first sequence tensor size; the first sequence tensor size includes the batch size, image height, image width, and number of images.

[0078] S422. Fix the batch size in the first sequence tensor size to obtain the second sequence tensor size.

[0079] S423. Perform a dimensionality transformation on the batch number and the number of images in the second sequence tensor to obtain a feature tensor.

[0080] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0081] Specifically, global features are obtained by fusing image features from each seismic profile sequence, including:

[0082] A410. Perform convolution operation on the feature tensor corresponding to the seismic profile sequence to obtain a first query vector, a first key vector, and a first value vector.

[0083] A420. Perform dimensional transformation on the first query vector, the first key vector, and the first value vector to obtain the second query vector, the second key vector, and the second value vector.

[0084] A431. Calculate the tensor matrix based on the second query vector and the second key vector.

[0085] The tensor matrix is ​​calculated based on the second query vector Q2 and the second key vector K2.

[0086] Specifically, the second dimension (2k+1) and the third dimension (C) of the tensor size (H,W,2k+1,C) of the second key vector K2 are transposed to obtain the first matrix formed by the second dimension (2k+1) and the third dimension (C). The second dimension (2k+1) and the third dimension (C) of the tensor size (H,W,2k+1,C) of the second query vector Q2 are obtained, and then the second matrix is ​​constructed based on the second dimension (2k+1) and the third dimension (C) of the second query vector Q2. Further, the product of the first matrix and the second matrix is ​​calculated to obtain the tensor matrix Q2K2. T The tensor size of the tensor matrix is ​​(H, W, 2k+1, 2k+1).

[0087] A432. The tensor matrix is ​​scaled to obtain the scaling result.

[0088] Scale adjustment, in this context, involves calculating the tensor matrix Q2K2. T The quotient of the preset value, wherein the preset value in this embodiment is Where C is the number of feature channels; the adjustment result is the quotient mentioned above.

[0089] Specifically, calculate the tensor matrix Q2K2. T Compared with preset value The quotient value is used to obtain the adjustment result.

[0090] A433. Normalize the adjustment result to obtain the normalized result.

[0091] In this embodiment, the Softmax function is used to normalize the adjustment result. The normalized result is the result output by the Softmax function after processing the adjustment result. In other embodiments, the specific function used for normalization is not limited.

[0092] Specifically, the adjustment results The output is fed into the Softmax function for normalization, thus obtaining the normalized result.

[0093] A434. Calculate global features based on the normalization result and the second value vector.

[0094] Among them, the global feature is the normalized result. The product of the second value vector V2.

[0095] Specifically, calculate the normalization result. The product of the second value vector V2 and the global feature is obtained.

[0096] Example 5

[0097] This application provides a geological body identification method in Embodiment 5, which refines the training process of the "target geological body identification model" in Embodiment 1. The method includes:

[0098] S510. Obtain a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0099] S520. Input the seismic profile sequence into the target geological body identification model for processing to obtain the geological body identification result.

[0100] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0101] Wherein, if the target geological body identification model is used for identifying a single geological body, the training process of the target geological body identification model includes:

[0102] B510. Input the seismic profile training map sequence into a primary geological body recognition model that incorporates a self-attention mechanism for processing to obtain prediction results corresponding to the benchmark seismic profile training map in the seismic profile training map sequence.

[0103] Among them, single geological body identification means identifying only one type of geological body in the seismic profile map; the seismic profile training map sequence is similar to the seismic profile map sequence, except that the seismic profile training map sequence is used to train the primary geological body identification model that incorporates a self-attention mechanism. The primary geological body identification model is an untrained geological body identification model; the seismic profile training map sequence class includes manually annotated benchmark seismic profile training maps and unannotated continuous seismic profile training maps.

[0104] Specifically, the seismic profile training image sequence is input into a primary geological body recognition model that incorporates a self-attention mechanism for processing. The primary geological body recognition model outputs prediction results corresponding to the baseline seismic profile training image in the seismic profile training image sequence. Each pixel on the baseline seismic profile training image has a corresponding prediction result. The prediction results are binary classification results, which are used to characterize whether the corresponding pixel on the baseline seismic profile training image belongs to a geological body.

[0105] B520. Construct a binary cross-entropy loss based on the prediction results and the corresponding label data of the benchmark seismic profile training map.

[0106] Among them, the benchmark seismic profile training map can be manually labeled to obtain corresponding label data. The label data is used together with the corresponding prediction results to construct the binary classification cross-entropy loss; the formula for calculating the binary classification cross-entropy loss is: Where y represents the label data. This is the predicted result.

[0107] Specifically, the label data y is compared with the corresponding prediction results. Substituting into the above formula for calculating the cross-entropy loss in binary classification... The calculation is performed to obtain the binary cross-entropy loss L1.

[0108] B530. Optimize the primary geological body identification model based on the binary cross-entropy loss to obtain the target geological body identification model.

[0109] Among them, the binary cross-entropy loss L1 is used to backpropagate in the primary geological body recognition model with self-attention mechanism to optimize the model parameters of the primary geological body recognition model, thereby obtaining a new primary geological body recognition model; further, the new primary geological body recognition model is iteratively optimized in the above manner until the preset number of iterations is reached, thereby completing the training of the primary geological body recognition model to obtain the target geological body recognition model.

[0110] Specifically, the primary geological body recognition model with self-attention mechanism is iteratively optimized based on the binary cross-entropy loss L1, thereby obtaining the target geological body recognition model in the completed training state.

[0111] Example 6

[0112] This application provides a geological body identification method in Embodiment Six, which refines the training process of the "target geological body identification model" in Embodiment One. The method includes:

[0113] S610. Obtain a sequence of seismic profile maps, the sequence of seismic profile maps including manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0114] S620. Input the seismic profile sequence into the target geological body identification model for processing to obtain the geological body identification result.

[0115] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0116] Wherein, if the target geological body identification model is used for multi-geological body identification, the training process of the target geological body identification model includes:

[0117] B610. Input the seismic profile training map sequence into a primary geological body recognition model that incorporates a self-attention mechanism for processing to obtain prediction results corresponding to the benchmark seismic profile training map in the seismic profile training map sequence.

[0118] Among them, multi-geological-body identification refers to the identification of multiple geological bodies in the seismic profile map; in this embodiment, it is assumed that the seismic profile map may contain F types of geological bodies, and the prediction result is used to characterize which geological body the pixel point corresponding to the benchmark seismic profile training map belongs to.

[0119] Specifically, the seismic profile training image sequence is input into a primary geological body recognition model that incorporates a self-attention mechanism for processing. This primary geological body recognition model outputs the integer prediction results corresponding to the baseline seismic profile training images in the seismic profile training image sequence.

[0120] B620. Encode the label data corresponding to the benchmark seismic profile training map to obtain the label code.

[0121] In this training image of the baseline seismic profile, each pixel has manually labeled data, and the label data is in integer form. The label data is used to represent which geological body the corresponding pixel belongs to, as determined by the human. For example, if the human determines that a certain pixel belongs to the 7th geological body, the corresponding label data is 7. In order to facilitate the subsequent construction of the model loss for optimizing the primary geological body identification model, the label data generated by the human annotation of the baseline seismic profile training image also needs to be encoded. In this embodiment, in order to facilitate the representation of the geological body type corresponding to the label data, one-hot encoding is used. For example, if the seismic profile may contain 7 geological bodies, the encoding bit is 7 bits. If a label data is 7, the label data is encoded as "0000001", and so on.

[0122] Specifically, a tag code is obtained by performing one-hot encoding on each tag data corresponding to the benchmark seismic profile training map, which corresponds one-to-one with each tag data.

[0123] B630. Construct a multi-class cross-entropy loss based on the label code and the prediction result.

[0124] The multi-class cross-entropy loss is based on the label code and the corresponding prediction result. The formula for calculating the multi-class cross-entropy loss is as follows: Among them, y i For tag codes, The prediction result corresponds to the tag code, where m represents the total number of tag codes.

[0125] Specifically, the tag code y i and the corresponding prediction results Substituting into the above formula for calculating multi-class cross-entropy loss The calculation is performed to obtain the multi-class cross-entropy loss.

[0126] B640. Optimize the primary geological body identification model based on the multi-class cross-entropy loss to obtain the target geological body identification model.

[0127] Among them, the multi-class cross-entropy loss L2 is used to iteratively optimize the primary geological body identification model, and the target geological body identification model is the geological body identification model obtained after the primary geological body identification model has been iteratively optimized.

[0128] Specifically, the primary geological body identification model is optimized by multi-class cross-entropy loss L2 iteration to obtain the target geological body identification model.

[0129] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0130] Example 7

[0131] Based on the same inventive concept, this embodiment also provides a geological body identification device for implementing the geological body identification method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more embodiments of the geological body identification device provided below can be found in the limitations of the geological body identification method described above, and will not be repeated here.

[0132] In this embodiment, as Figure 5 As shown, a geological body identification device is provided, comprising:

[0133] The sequence acquisition module is used to acquire a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps.

[0134] The result recognition module is used to input the seismic profile sequence into the target geological body recognition model for processing to obtain the geological body recognition result;

[0135] The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

[0136] Each module in the aforementioned geological body identification device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0137] It should be noted that this embodiment obtains a sequence of seismic profile images, including manually annotated baseline seismic profile images and unannotated continuous seismic profile images. The sequence is then input into a target geological body recognition model for processing to obtain geological body recognition results. This target geological body recognition model incorporates a self-attention mechanism, which fuses image features from each seismic profile image in the sequence to obtain global features. Through this implementation, because the self-attention mechanism of the target geological body recognition model can fuse image features from each seismic profile image in the sequence, it enables the model to extract three-dimensional spatial features from two-dimensional seismic profile images. Furthermore, manual annotation of the baseline seismic profile image in the sequence is sufficient, reducing the likelihood of errors in the annotations. This prevents the geological body recognition model from exhibiting a "sheet-like" phenomenon when recognizing geological bodies, thereby improving the recognition effect.

[0138] Example 8

[0139] In this embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 6 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a geological body identification method.

[0140] Those skilled in the art will understand that Figure 6 The structure shown is merely a block diagram of a portion of the structure related to the present disclosure and does not constitute a limitation on the computer device to which the present disclosure is applied. A specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0141] Example 9

[0142] In this embodiment, a computer-readable storage medium is provided, such as... Figure 7 As shown, a computer program is stored thereon, and when the computer program is executed by the processor, it implements the steps in the above-described method embodiments.

[0143] Example 10

[0144] In this embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0145] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this disclosure are all information and data authorized by the user or fully authorized by all parties.

[0146] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this disclosure can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this disclosure may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this disclosure may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0147] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0148] The embodiments described above are merely illustrative of several implementations of this disclosure, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent disclosure. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this disclosure, and these all fall within the protection scope of this disclosure. Therefore, the protection scope of this disclosure should be determined by the appended claims.

Claims

1. A method for identifying geological bodies, characterized in that, include: Obtain a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps; The seismic profile sequence is input into the target geological body identification model for processing to obtain the geological body identification result; The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

2. The method according to claim 1, characterized in that, The process of inputting the seismic profile sequence into the target geological body identification model for processing to obtain the geological body identification result includes: The tensor dimensions of the seismic profile sequence are determined to obtain the first sequence tensor dimensions; the first sequence tensor dimensions include the batch size, image height, image width, and number of images; The second sequence tensor size is obtained by fixing the batch size in the first sequence tensor size; The feature tensor is obtained by performing a dimensionality transformation on the batch size and the number of images in the second sequence tensor size.

3. The method according to claim 2, characterized in that, The image features of each seismic profile in the fused seismic profile sequence are used to obtain global features, including: The first query vector, the first key vector, and the first value vector are obtained by performing a convolution operation on the feature tensor corresponding to the seismic profile sequence. The first query vector, the first key vector, and the first value vector are transformed to obtain the second query vector, the second key vector, and the second value vector. Global features are calculated based on the second query vector, the second key vector, and the second value vector.

4. The method according to claim 3, characterized in that, The calculation of global features based on the second query vector, the second key vector, and the second value vector includes: Calculate the tensor matrix based on the second query vector and the second key vector; The scaling result is obtained by scaling the tensor matrix. The adjustment result is normalized to obtain the normalized result; Global features are calculated based on the normalization result and the second value vector.

5. The method according to claim 1, characterized in that, If the target geological body identification model is used for identifying a single geological body, the training process of the target geological body identification model includes: The seismic profile training image sequence is input into a primary geological body recognition model that incorporates a self-attention mechanism for processing, and prediction results corresponding to the benchmark seismic profile training images in the seismic profile training image sequence are obtained. A binary classification cross-entropy loss is constructed based on the prediction results and the corresponding label data of the benchmark seismic profile training map. The primary geological body identification model is optimized based on the binary cross-entropy loss to obtain the target geological body identification model.

6. The method according to claim 1, characterized in that, If the target geological body identification model is used for multi-geological body identification, the training process of the target geological body identification model includes: The seismic profile training image sequence is input into a primary geological body recognition model that incorporates a self-attention mechanism for processing, and prediction results corresponding to the benchmark seismic profile training images in the seismic profile training image sequence are obtained. The label data corresponding to the benchmark seismic profile training map is encoded to obtain a label code; A multi-class cross-entropy loss is constructed based on the label code and the prediction result; The primary geological body identification model is optimized based on the multi-class cross-entropy loss to obtain the target geological body identification model.

7. A geological body identification device, characterized in that, The device includes: The sequence acquisition module is used to acquire a sequence of seismic profile maps, which includes manually annotated baseline seismic profile maps and unannotated continuous seismic profile maps. The result recognition module is used to input the seismic profile sequence into the target geological body recognition model for processing to obtain the geological body recognition result; The target geological body identification model incorporates a self-attention mechanism, which is used to fuse the image features of each seismic profile in the seismic profile sequence to obtain global features.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images