Training a 3D model for identifying meandering rivers, a method for identifying meandering rivers, and an apparatus.
By using a 3D recognition model training method, the label data volume is initialized using seismic data volume and interpretation results, and masking processing and image segmentation model training are performed. This solves the problem of insufficient continuity of 2D models and insufficient generalization ability of 3D models in meandering river recognition, and achieves efficient and accurate meandering river recognition.
Patent Information
- Application Number
- CN202411695062.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-25
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies for identifying meandering rivers suffer from problems such as the lack of continuity in the stitching dimension of two-dimensional deep learning models and poor generalization ability of three-dimensional deep learning models. Furthermore, they are affected by faults and noise interference in the study area, resulting in low-quality seismic data and thus poor identification performance.
A 3D recognition model training method is adopted. By acquiring the interpretation results and seismic data volume of the target area, the label data volume is initialized and masking is performed. The image segmentation model is used for training, and the parameters are iteratively adjusted until the loss function converges. Combined with data preprocessing such as standardization and rotation transformation, the continuity and generalization ability of the model are improved.
It achieves efficient and accurate identification of three-dimensional models of meandering rivers, with better continuity and generalization performance, and can accurately identify meandering rivers under complex geological conditions.
Smart Images
Figure CN122090256A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of oil and gas geophysical exploration technology, and in particular to a training method and apparatus for a three-dimensional identification model of meandering rivers. Background Technology
[0002] River channel identification is fundamental to river reservoir development, well control, and paleochannel geomorphology. Its main task is to accurately identify river channels within complex and noisy seismic bodies. Meandering rivers are a type of river with unique meandering morphology, and the abundant sand bodies formed during their multi-stage evolution provide favorable locations for oil and gas accumulation, making them an important target for river channel identification.
[0003] Existing technologies typically rely on attribute-based methods for identifying meandering rivers. These methods utilize seismic data to extract coherence volumes and other attributes to characterize continuity, while differentiating between the river channel and surrounding rock based on their discontinuities. However, these methods are susceptible to interference from factors unrelated to the river channel, such as faults and strong noise, and require high standards for fault development and seismic data quality in the study area.
[0004] While deep learning can enable intelligent identification of meandering rivers, current methods train deep learning models for this purpose using two-dimensional sample data. These models are then composed of two-dimensional data, resulting in a lack of continuity in the splicing dimension. Furthermore, three-dimensional deep learning models require fewer labels, are more costly to acquire, and are typically concentrated in a small area, leading to poor generalization ability. Therefore, there is an urgent need for an efficient and accurate method for identifying meandering rivers. Summary of the Invention
[0005] In the process of river channel identification, the training of 3D deep learning models is greatly limited due to interference from factors such as faults and strong noise in the study area, the low quality of seismic data in the study area, the small number of manually interpreted labels, and the high cost of obtaining them.
[0006] In view of the above problems, the present invention is proposed to provide a method and apparatus for training a three-dimensional model of meandering rivers, identifying meandering rivers, overcoming or at least partially solving the above problems.
[0007] In a first aspect, embodiments of the present invention provide a method for training a three-dimensional recognition model of a meandering river, comprising:
[0008] Acquire several interpretation results and the corresponding seismic data volume in the target area, wherein the interpretation results are characterized as meandering rivers or non-meandering rivers;
[0009] Based on the interpretation results and the earthquake data body, initialize the preset label data body, which has the same size as the earthquake data body;
[0010] The tag data body is masked in order to update the tag data body;
[0011] Each earthquake data point in the earthquake data volume is used as sample data and input into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and the label data in the updated label data volume, the value of a preset loss function is calculated. The parameters of the image segmentation model are then adjusted in reverse according to the loss function value. This iterative training process continues until the loss function reaches a preset convergence condition, at which point training stops, resulting in a trained 3D meandering river recognition model. The output of the 3D meandering river recognition model is the recognition result of the meandering river.
[0012] In one embodiment, initializing a preset label data body based on the interpretation results and the seismic data body includes:
[0013] The seismic data volume is divided into grids using sparse grids with equal spacing. The corresponding label data on the grids is then placed into the label data volume to obtain the initialized label data volume. The grid of the label data volume corresponds to the grid of the seismic data volume.
[0014] In one embodiment, masking the tag data body includes:
[0015] The data between the grids in the label data body is masked.
[0016] In one embodiment, the masking process involves assigning a preset value to the label data of the grid outside the cross-section, and the preset value is different from the label data of the grid on the cross-section.
[0017] In one embodiment,
[0018] Each seismic data point in the seismic data volume is used as sample data and input into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and the updated label data in the label data volume, a preset loss function is calculated. The parameters of the image segmentation model are then adjusted in reverse according to the loss function value. This iterative training process continues until the loss function reaches a preset convergence condition, at which point training stops, resulting in a trained 3D recognition model of the meandering river, including:
[0019] The seismic data volume is sampled according to a preset size to obtain several sub-seismic data volumes. For each sub-seismic data volume, each seismic data in the sub-seismic data volume is used as sample data to obtain a sample dataset corresponding to the sub-seismic data volume. The sub-seismic data volumes have overlapping parts.
[0020] Each sample dataset is input into the image segmentation model for training, and the image segmentation model outputs the predicted probability of each sample data in each sample dataset; the predicted probability represents the probability that the sample data is a meandering river;
[0021] Based on the predicted probability of each sample data and the corresponding label data, the value of the loss function is calculated, and the parameters of the image segmentation model are adjusted in reverse according to the value of the loss function. The training process is iterated until the loss function reaches the preset convergence condition, and then the training is stopped to obtain the trained meandering river 3D recognition model.
[0022] In one embodiment, before inputting each sample dataset into the image segmentation model for training, the method further includes:
[0023] For each sample dataset, each sample data in the sample dataset is standardized.
[0024] The standardized sample data and their corresponding label data are then subjected to enhanced transformation.
[0025] In one embodiment, the enhancement transformation of the standardized sample data and the corresponding label data includes:
[0026] By rotating the data around the coordinate axis by a preset angle, the standardized sample data and the corresponding label data are expanded respectively.
[0027] Secondly, embodiments of the present invention provide a method for identifying meandering rivers, comprising:
[0028] Obtain a seismic data volume with the same size as the area to be identified, and divide the seismic data volume into several sub-seismic data volumes of the same size; there is data overlap between the sub-seismic data volumes;
[0029] Standardize the seismic data in each sub-seismic data volume;
[0030] The standardized sub-seismic data volumes are input into the meandering river 3D recognition model to obtain the predicted probability of each seismic data in each sub-seismic data volume. Based on the predicted probability and the preset probability threshold, it is determined whether the seismic data is characterized as a meandering river.
[0031] Based on seismic data characterized as meandering rivers, meandering rivers in the area to be identified are identified;
[0032] The three-dimensional recognition model of the meandering river is obtained by training using the aforementioned training method for the three-dimensional recognition model of the meandering river.
[0033] Thirdly, embodiments of the present invention provide a training device for a three-dimensional recognition model of a meandering river, comprising:
[0034] The acquisition module is used to acquire several interpretation results and the seismic data volume corresponding to the target area in the target area, wherein the interpretation results are characterized as meandering rivers or non-meandering rivers.
[0035] An initialization module is used to initialize a preset label data body based on the interpretation results and the seismic data body. The label data body has the same size as the seismic data body.
[0036] An update module is used to mask the label data body in order to update the label data body;
[0037] The training module is used to input each earthquake data in the earthquake data volume as sample data into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and each label data in the updated label data volume, the module calculates the value of a preset loss function and adjusts the parameters of the image segmentation model in reverse according to the loss function value. The training process is iterated until the loss function reaches a preset convergence condition, at which point the training stops, and a trained meandering river 3D recognition model is obtained. The output of the meandering river 3D recognition model is the recognition result of the meandering river.
[0038] Fourthly, embodiments of the present invention provide a device for identifying meandering rivers, comprising:
[0039] The acquisition module is used to acquire seismic data volumes of the same size as the area to be identified, and to divide the seismic data volumes into several sub-seismic data volumes of the same size; there is data overlap between the sub-seismic data volumes;
[0040] The standardization module is used to standardize the seismic data in each sub-seismic data volume;
[0041] The prediction module is used to input the standardized sub-seismic data volumes into the meandering river 3D recognition model, obtain the prediction probability of each seismic data in each sub-seismic data volume, and determine whether the seismic data is characterized as a meandering river based on the prediction probability and a preset probability threshold.
[0042] Based on seismic data characterized as meandering rivers, meandering rivers in the area to be identified are identified;
[0043] The three-dimensional recognition model of the meandering river is obtained by training using the aforementioned training method for the three-dimensional recognition model of the meandering river.
[0044] Fifthly, embodiments of the present invention provide a computer storage medium storing computer-executable instructions, which, when executed by a processor, implement the aforementioned training method for the three-dimensional recognition model of a meandering river or the aforementioned recognition method for a meandering river.
[0045] In a sixth aspect, embodiments of the present invention provide a terminal device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the aforementioned training method for the three-dimensional recognition model of a meandering river or the aforementioned recognition method for a meandering river.
[0046] The beneficial effects of the above-described technical solutions provided in the embodiments of the present invention include at least the following:
[0047] The training method for a three-dimensional meandering river recognition model provided in this invention involves determining a corresponding label data volume containing masked data based on the manually interpreted labels and seismic data volume obtained in the target area. A pre-constructed image segmentation model is trained using the seismic data volume. A loss function is calculated using the output of the image segmentation model and the corresponding label data. When the loss function reaches a preset convergence condition, it indicates that the error between the output of the image segmentation model and the corresponding label data is very small. After training, a three-dimensional meandering river recognition model is obtained. The model is trained using continuous seismic data from the seismic data volume, resulting in better continuity. Furthermore, since the label data volume corresponds to the entire target area, the trained three-dimensional meandering river recognition model has higher genericity, enabling it to accurately and efficiently identify meandering rivers in the target area.
[0048] The label data volume is determined based on the selected seismic data volume profile. Specifically, the seismic data volume is divided using a sparse grid with equal spacing. Then, the label data corresponding to the grid on the profile is retained, and the grid outside the profile is masked. The advantage of this is that the masking process is applied to the grid outside the profile that does not have actual labels, so that the label data for training the meandering river 3D recognition model can be distributed throughout the entire study area, making the meandering river 3D recognition model more generic.
[0049] The inventors of this invention also preprocessed the training data by rotating it around the coordinate axis by a preset angle to simultaneously enhance the transformation of the seismic data and the label data. This increased the number of samples while ensuring the correspondence between the seismic data and the label data.
[0050] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.
[0051] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0052] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0053] Figure 1 This is a flowchart of the training method for a three-dimensional recognition model of a meandering river according to an embodiment of the present invention;
[0054] Figure 2 This is a flowchart of the data preprocessing method in Embodiment 1 of the present invention;
[0055] Figure 3 This is a schematic diagram of the image segmentation model in Embodiment 1 of the present invention;
[0056] Figure 4 This is a flowchart of the training method for the three-dimensional recognition model of the meandering river in Embodiment 1 of the present invention;
[0057] Figure 5 This is a schematic diagram of the grid used to illustrate the results in Embodiment 1 of the present invention;
[0058] Figure 6 This is a flowchart of the method for identifying meandering rivers in Embodiment 2 of the present invention;
[0059] Figure 7 This is a comparison chart of the meandering river identification results and the manual interpretation results in Embodiment 2 of the present invention;
[0060] Figure 8 This is a schematic diagram of the training device for the three-dimensional recognition model of the meandering river in Embodiment 1 of the present invention;
[0061] Figure 9 This is a schematic diagram of the structure of the meandering river identification device in Embodiment 1 of the present invention. Detailed Implementation
[0062] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
[0063] To address the problem of inaccurate and rapid river identification in existing technologies, this invention provides a method and apparatus for training a three-dimensional identification model of meandering rivers and for identifying meandering rivers.
[0064] Example
[0065] This invention provides a method for training a 3D recognition model of a meandering river, the process of which is as follows: Figure 1 As shown, it includes the following steps:
[0066] Step S1: Obtain several interpretation results and the corresponding seismic data volume in the target area. The interpretation results are characterized as meandering rivers or non-meandering rivers.
[0067] Step S2: Based on the interpretation results and the seismic data volume, initialize the preset label data volume. The label data volume has the same size as the seismic data volume.
[0068] Step S3: Mask the label data body to update the label data body;
[0069] Step S4: Take each earthquake data in the earthquake data volume as sample data and input it into the pre-constructed image segmentation model for training. Calculate the value of the preset loss function based on the output of the image segmentation model and the updated label data in the label data volume. Adjust the parameters of the image segmentation model in reverse according to the loss function value. Iterate the training process until the loss function reaches the preset convergence condition, then stop training to obtain the trained meandering river 3D recognition model. The meandering river 3D recognition model outputs the predicted probability of each sample data.
[0070] Unlike current deep learning models for meandering river recognition based on two-dimensional image segmentation, the recognition method provided in this embodiment is based on three-dimensional image segmentation, resulting in better continuity of recognition performance. Furthermore, by using labeled data distributed throughout the entire study area, the deep learning model for meandering river recognition provided in this embodiment exhibits higher generalization performance, thereby making the recognition method provided in this embodiment more practical.
[0071] In some optional embodiments, the interpretation results in the target area are obtained manually, and the interpretation results are either meandering rivers or non-meandering rivers. Correspondingly, the labels marked according to the interpretation results are only divided into two types: background (non-meandering river) and foreground (meandering river). The labels in the target area do not only mark rivers. For any location in the target area, the meandering river depositional characteristics of that location can be characterized by labels. That is, according to the interpretation results of each river in the target area, the corresponding seismic data in the seismic data volume can be labeled, that is, the seismic data is labeled. For example, if a certain place is a river channel and meets the characteristics of meandering river deposition, the interpretation result of that place is "meandering river", and the corresponding label is foreground (meandering river). If a certain place is not a river channel, or a certain place is a river channel but does not meet the characteristics of meandering river deposition, the interpretation result of that place is "non-meandering river", and the corresponding label is background (non-meandering river).
[0072] The seismic data volume corresponding to the target area can be obtained through surveying. Obtaining seismic data volume is an existing technology, and this embodiment of the present invention will not elaborate on it.
[0073] Based on the above method of labeling earthquake data, the above step S2 is performed to initialize the preset label data body, which can be done in the following way:
[0074] The seismic data volume is divided into grids using sparse grids with equal spacing. The corresponding label data on the grids is then placed into the label data volume to obtain the initialized label data volume. The grid of the label data volume corresponds to the grid of the seismic data volume.
[0075] Specifically, an empty label data volume of the same size as the seismic data volume is pre-constructed. The seismic data volume is divided using a sparse grid with equal spacing. Since the seismic data volume is three-dimensional, any grid line along the horizontal or vertical direction corresponds to a profile of the seismic data volume. Several interpretation results (including meandering and non-meandering rivers) from the grid lines of each profile are placed into this empty label data volume. Then, this label data volume with the interpretation results is also gridded, and the grid division of the label data volume is consistent with that of the seismic data volume. Then, each label data in the label data volume is digitally encoded (for example, using one-hot encoding) to initialize the label data volume. For example, after encoding, if an interpretation result is a meandering river, the encoding result of the corresponding label data in the label data volume is 1; conversely, if an interpretation result is a non-meandering river, the encoding result of the corresponding label data in the label data volume is 0. In this way, the data on each grid line corresponding to the seismic data volume in the label data volume are all labeled.
[0076] In some optional embodiments, determining a profile of the tag data body in step S3 above can be achieved in the following way:
[0077] Mask the data between the grids in the label data body.
[0078] In some optional embodiments, the masking process assigns a preset value to the data between the grids, and the preset value is different from the label data on the grid lines on the profile. For example, if the label data is 1 to represent a meandering river and the label data is 0 to represent a non-meandering river, then the preset value can be set to -1.
[0079] In some optional embodiments, if the quantity or quality of sample data is small, preprocessing, including enhancement transformation, can be performed on the sample data before each sample dataset is input into the image segmentation model for training, in order to ensure the quantity or quality of the sample data. Specifically, refer to... Figure 2 As shown, this can be achieved in the following way:
[0080] Step S21: For each sample dataset, standardize the data of each sample in the sample dataset;
[0081] The following formula can be used to standardize the sample data:
[0082]
[0083] In formula (1), A ′ Let A be the standardized sample, μ be the unstandardized sample, and σ be the sample mean.
[0084] Step S22: Perform enhancement transformation on the standardized sample data and the corresponding label data.
[0085] Specifically, by rotating the data around the coordinate axis by a preset angle, the standardized sample data and the corresponding label data are expanded. For example, for each sample dataset, the sample data and the corresponding label data in each sample dataset are rotated four times around the vertical axis, each time by 90 degrees, to expand the sample by four times to achieve data augmentation. While increasing the number of samples, the correspondence between the seismic data and the label data is maintained.
[0086] In some optional embodiments, the image segmentation model consists of an encoder, a decoder, and an output section. The encoder comprises convolutional layers, pooling layers, and multiple residual modules; the decoder comprises multiple residual modules; and the output section comprises multiple convolutional layers with batch normalization. The decoder uses a trilinear interpolation algorithm for upsampling. There are short connections between the encoder and decoder. A schematic diagram of the image segmentation model is shown below. Figure 3As shown, the input data is a batch of 3D data volumes. The encoder ResNet part extracts low-level and mid-level features, which are then fed into the decoder and encoder ASPP part, respectively. The encoder ASPP outputs high-level features, which are then fed into the decoder. Finally, the decoder obtains the output results. The loss is calculated using the combined loss function, and the optimizer (the optimizer is the Adam optimizer) is used to update the model's intrinsic parameters. This process is repeated multiple times until the termination condition is met.
[0087] Next, the loss function is constructed as follows:
[0088] Loss=λL Focal +(1-λ)L Dice (2)
[0089] In formula (2), λ is an adjustable weight, for example, the preset value of λ is 20 / 21; L Focal Let L be the Focal loss function. Dice The Dice loss function;
[0090] The Focal loss function and the Dice loss function are shown in Equation (3) and Equation (4), respectively:
[0091] L Focal (P t )=-α(1-P t ) γ log(P t (3)
[0092] In formula (3), P t For the model to predict probabilities, α and γ are two parameters of Focal, for example, the values of α and γ are preset to 0.25 and 2 respectively;
[0093]
[0094] In formula (4), TP is the number of foreground class samples predicted by the model as foreground class (meandering river), FP is the number of background class samples predicted as foreground class, and FN is the number of foreground class samples predicted as background class.
[0095] A 3D recognition model of a meandering river can be trained using the following method, for reference. Figure 4 As shown, it includes the following steps:
[0096] Step S31: Sample the seismic data volume according to a preset size to obtain several sub-seismic data volumes. For each sub-seismic data volume, use each seismic data in the sub-seismic data volume as sample data to obtain a sample dataset corresponding to the sub-seismic data volume; there is an overlap between the sub-seismic data volumes.
[0097] Step S32: Input each sample dataset into the image segmentation model for training. The image segmentation model outputs the predicted probability of each sample data in each sample dataset. The predicted probability represents the probability that the sample data is a meandering river.
[0098] Step S33: Calculate the value of the loss function based on the predicted probability of each sample data and the corresponding label data, and adjust the parameters of the image segmentation model in reverse according to the loss function value. Iterate the training process until the loss function reaches the preset convergence condition, stop training, and obtain the trained meandering river 3D recognition model.
[0099] In step S31 above, the size of the sampled sub-seismic data volume is not less than the grid spacing, and the sampled seismic data volumes have overlapping portions, which avoids the boundary effect in the prediction.
[0100] Multiple sub-seismic data volumes are divided into training sample sets and validation sample sets. For example, they can be divided into training sample sets and validation sample sets in a ratio of 8:2. In fact, "each sample dataset is input into the image segmentation model for training" in step S32 means that each sample dataset in the training sample set is input into the image segmentation model for training. The image segmentation model outputs the predicted probability that the sample data is a meandering river.
[0101] For each sample data, the image segmentation model outputs a probability value predicting that it is a meandering river. This probability value, along with the one-hot encoded label data corresponding to that sample data, is input into the loss function to calculate the value of the loss function. Based on the value of the loss function, the parameters of the image segmentation model are adjusted in reverse. After each round of training, the newly trained model is validated using a validation set. After multiple rounds of training, the calculated loss function value reaches the preset convergence condition, and the training is completed, or the training prediction round is stopped, resulting in a trained 3D meandering river recognition model.
[0102] During the training phase, a decaying learning rate step strategy was used, where the step size for updating parameters according to the loss gradient decreased as the training epochs increased. A learning rate step warm-up strategy was also used, gradually increasing the learning rate step size from its minimum value to the initial learning rate in the few epochs before formal training using the decaying learning rate step size. The optimizer was the Adam adaptive moment estimator, which updated all parameters that propagate the loss gradient within the model. The mean intersection-over-union (MIOU) parameter and classification accuracy were used as image segmentation evaluation parameters, and training was completed when the evaluation parameters on the model validation set reached preset values. A maximum number of iterations was set, and training terminated when the maximum number of iterations was reached. An early stopping mechanism was also implemented, which terminated training if the training loss did not decrease for several consecutive epochs.
[0103] refer to Figure 5The grid diagram shown is a representation of the interpretation results of the extracted profiles, derived from a horizontal slice of the label data volume. The seismic data volume for the work area has dimensions of 1024inline×2048crossline×1024sample. Using a grid spacing of 100, 10 inline profiles and 20 crossline profiles were extracted to obtain the interpretation results for the meandering river. These results were then placed into a label data volume of the same size as the seismic data volume. Special values were assigned to the grid regions between the interpretation profiles to achieve masking. The seismic data volume and label data volume were sampled in three dimensions according to 128inline×128xline×128sample, with some overlap between samples. A total of 6975 pairs of samples were obtained, which were divided into training and validation sets at an 8:2 ratio.
[0104] Example 2
[0105] This invention provides a method for identifying meandering rivers, the process of which is as follows: Figure 6 As shown, it includes the following steps:
[0106] Step S51: Obtain a seismic data volume with the same size as the area to be identified, and divide the seismic data volume into several sub-seismic data volumes with the same size; there is data overlap between the sub-seismic data volumes;
[0107] Step S52: Standardize the seismic data in each sub-seismic data volume;
[0108] Step S53: Input the standardized sub-seismic data volumes into the meandering river 3D recognition model to obtain the predicted probability of each seismic data in each sub-seismic data volume. Based on the predicted probability and the preset probability threshold, determine whether the seismic data is represented as a meandering river.
[0109] Step S54: Based on the seismic data characterized as meandering rivers, identify the meandering rivers in the area to be identified;
[0110] The 3D recognition model of the meandering river was trained using the same method described above.
[0111] The meandering river 3D recognition model is used to predict a batch of sample data (each sub-seismic data volume contains multiple seismic data), and the block prediction results are stitched back to the size of the seismic data volume of the area to be identified. For overlapping parts, the average value is taken and compared with a preset probability threshold (e.g., the probability threshold is set to 0.5) to obtain the meandering river recognition result. Figure 7 As shown, Figure 7 The image shows a horizontal slice comparing the results of the meandering river identification with the results of the human interpretation. The left side represents the human interpretation, and the right side represents the prediction result. The yellow area represents the predicted meandering river channel. It can be seen that the two have a high degree of similarity.
[0112] Based on the same inventive concept, embodiments of the present invention also provide a training device for a three-dimensional recognition model of a meandering river, the structure of which is as follows: Figure 8 As shown, it includes:
[0113] The acquisition module 71 is used to acquire several interpretation results and the corresponding seismic data volume in the target area. The interpretation results are characterized as meandering rivers or non-meandering rivers.
[0114] Initialization module 72 is used to initialize a preset label data body based on each interpretation result and the seismic data body. The label data body has the same size as the seismic data body.
[0115] The update module 73 is used to determine a profile of the label data volume based on the seismic data volume, and to perform masking processing on the label data volume according to the label data on the profile in order to update the label data volume.
[0116] Training module 73 is used to input each earthquake data in the earthquake data volume as sample data into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and each label data in the updated label data volume, the value of the preset loss function is calculated, and the parameters of the image segmentation model are adjusted in reverse according to the loss function value. The training process is iterated until the loss function reaches the preset convergence condition, and the training stops, resulting in a trained meandering river 3D recognition model. The meandering river 3D recognition model outputs the predicted probability of each sample data.
[0117] Regarding the training device for the three-dimensional recognition model of the meandering river in the above embodiments, the specific methods of each module performing operations have been described in detail in the embodiments of the relevant method, and will not be elaborated here.
[0118] Based on the same inventive concept, embodiments of the present invention also provide a device for identifying meandering rivers, the structure of which is as follows: Figure 9 As shown, it includes:
[0119] The acquisition module 81 is used to acquire seismic data volumes with the same size as the area to be identified, and to divide the seismic data volumes into several sub-seismic data volumes with the same size; there is data overlap between the sub-seismic data volumes;
[0120] Standardization module 82 is used to standardize the seismic data in each sub-seismic data volume;
[0121] The prediction module 83 is used to input the standardized sub-seismic data volumes into the meandering river 3D recognition model, obtain the prediction probability of each seismic data in each sub-seismic data volume, and determine whether the seismic data is represented as a meandering river based on the prediction probability and the preset probability threshold.
[0122] Identification module 84 is used to identify meandering rivers in the area to be identified based on seismic data characterized as meandering rivers;
[0123] The 3D recognition model of the meandering river was trained using the same method described above.
[0124] Regarding the meandering river identification device in the above embodiments, the specific way in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated here.
[0125] Unless otherwise specifically stated, terms such as processing, calculation, operation, determination, display, etc., may refer to the actions and / or processes of one or more processing or computing systems or similar devices that represent the manipulation and conversion of data representing physical (e.g., electronic) quantities within the registers or memory of the processing system into other data similarly representing physical quantities within the memory, registers, or other such information storage, transmission, or display devices of the processing system. Information and signals can be represented using any of a variety of different techniques and methods. For example, data, instructions, commands, information, signals, bits, symbols, and chips mentioned throughout the above description can be represented by voltage, current, electromagnetic waves, magnetic fields or particles, light fields or particles, or any combination thereof.
[0126] It should be understood that the specific order or hierarchy of steps in the disclosed process is an example of an exemplary method. Based on design preferences, it should be understood that the specific order or hierarchy of steps in the process may be rearranged without departing from the scope of this disclosure. The appended method claims provide elements of various steps in an exemplary order and are not intended to limit the scope to the specific order or hierarchy described.
[0127] In the detailed description above, various features are combined together in a single embodiment to simplify this disclosure. This approach to disclosure should not be construed as reflecting an intention that embodiments of the claimed subject matter require more features than are explicitly stated in each claim. Rather, as reflected in the appended claims, the invention is presented with fewer features than all of the features in a single disclosed embodiment. Therefore, the appended claims are hereby explicitly incorporated into the detailed description, with each claim representing a separate preferred embodiment of the invention.
[0128] Those skilled in the art will also understand that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments herein can be implemented as electronic hardware, computer software, or a combination thereof. To clearly illustrate the interchangeability between hardware and software, the various illustrative components, blocks, modules, circuits, and steps described above are generally described in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. Those skilled in the art can implement the described functionality in alternative ways for each specific application; however, such implementation decisions should not be construed as departing from the scope of this disclosure.
[0129] The steps of the methods or algorithms described in conjunction with the embodiments herein can be directly embodied in hardware, software modules executed by a processor, or a combination thereof. The software modules can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disks, removable disks, CD-ROMs, or any other form of storage medium well known in the art. An exemplary storage medium is connected to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and storage medium can reside in an ASIC. The ASIC can reside in a user terminal. Alternatively, the processor and storage medium can exist as discrete components in the user terminal.
[0130] For software implementation, the techniques described in this application can be implemented using modules (e.g., procedures, functions, etc.) that perform the functions described in this application. This software code can be stored in memory units and executed by a processor. The memory units can be implemented within the processor or outside the processor; in the latter case, they are communicatively coupled to the processor via various means, as is well known in the art.
[0131] The foregoing description includes examples of one or more embodiments. It is certainly impossible to describe all possible combinations of components or methods in order to describe the above embodiments, but those skilled in the art will recognize that further combinations and arrangements of the various embodiments are possible. Therefore, the embodiments described herein are intended to cover all such changes, modifications, and variations that fall within the scope of the appended claims. Furthermore, the term "comprising" as used in the specification or claims is interpreted in a manner similar to the term "including," as interpreted when used as a conjunction in the claims. Additionally, the use of any term "or" in the specification of the claims is intended to mean "non-exclusive or."
Claims
1. A training method for a three-dimensional recognition model of a meandering river, characterized in that, include: Acquire several interpretation results and the corresponding seismic data volume in the target area, wherein the interpretation results characterize meandering rivers or non-meandering rivers; Based on the interpretation results and the earthquake data body, initialize the preset label data body, which has the same size as the earthquake data body; The tag data body is masked in order to update the tag data body; Each earthquake data point in the earthquake data volume is used as sample data and input into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and the label data in the updated label data volume, the value of a preset loss function is calculated. The parameters of the image segmentation model are then adjusted in reverse according to the loss function value. This iterative training process continues until the loss function reaches a preset convergence condition, at which point training stops, resulting in a trained 3D meandering river recognition model. The output of the 3D meandering river recognition model is the recognition result of the meandering river.
2. The method as described in claim 1, characterized in that, The initialization of a preset label data body based on the interpretation results and the seismic data body includes: The seismic data volume is divided into grids using sparse grids with equal spacing. The corresponding label data on the grids is then placed into the label data volume to obtain the initialized label data volume. The grid of the label data volume corresponds to the grid of the seismic data volume.
3. The method as described in claim 2, characterized in that, Masking the tag data body includes: The data between the grids in the label data body is masked.
4. The method as described in claim 3, characterized in that, The masking process involves assigning preset values to the label data of the grid outside the cross-section, and the preset values are different from the label data of the grid on the cross-section.
5. The method as described in claim 1, characterized in that, Each seismic data point in the seismic data volume is used as sample data and input into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and the updated label data in the label data volume, a preset loss function is calculated. The parameters of the image segmentation model are then adjusted in reverse according to the loss function value. This iterative training process continues until the loss function reaches a preset convergence condition, at which point training stops, resulting in a trained 3D recognition model of the meandering river, including: The seismic data volume is sampled according to a preset size to obtain several sub-seismic data volumes. For each sub-seismic data volume, each seismic data in the sub-seismic data volume is used as sample data to obtain a sample dataset corresponding to the sub-seismic data volume. The data between two geographically adjacent sub-seismic data volumes have overlapping parts. Each sample dataset is input into the image segmentation model for training, and the image segmentation model outputs the predicted probability of each sample data in each sample dataset; the predicted probability represents the probability that the sample data is a meandering river; Based on the predicted probability of each sample data and the corresponding label data, the value of the loss function is calculated, and the parameters of the image segmentation model are adjusted in reverse according to the value of the loss function. The training process is iterated until the loss function reaches the preset convergence condition, and then the training is stopped to obtain the trained meandering river 3D recognition model.
6. The method as described in claim 5, characterized in that, Before inputting each sample dataset into the image segmentation model for training, the process also includes: For each sample dataset, each sample data in the sample dataset is standardized. The standardized sample data and their corresponding label data are then subjected to enhanced transformation.
7. The method as described in claim 6, characterized in that, The enhancement transformation process performed on the standardized sample data and the corresponding label data includes: By rotating the data around the coordinate axis by a preset angle, the standardized sample data and the corresponding label data are expanded respectively.
8. A method for identifying meandering rivers, characterized in that, include: Obtain a seismic data volume with the same size as the area to be identified, and divide the seismic data volume into several sub-seismic data volumes of the same size; Standardize the seismic data in each sub-seismic data volume; The standardized sub-seismic data volumes are input into the meandering river 3D recognition model to obtain the predicted probability of each seismic data in each sub-seismic data volume. Based on the predicted probability and the preset probability threshold, it is determined whether the seismic data is characterized as a meandering river. Based on seismic data characterized as meandering rivers, meandering rivers in the area to be identified are identified; The meandering river 3D recognition model is obtained by training the meandering river 3D recognition model according to any one of claims 1-7.
9. The method as described in claim 8, characterized in that, Data between two geographically adjacent sub-seismic data volumes have overlapping portions.
10. A training device for a three-dimensional recognition model of a meandering river, characterized in that, include: The acquisition module is used to acquire several interpretation results and the seismic data volume corresponding to the target area in the target area, wherein the interpretation results represent meandering rivers or non-meandering rivers; An initialization module is used to initialize a preset label data body based on the interpretation results and the seismic data body. The label data body has the same size as the seismic data body. An update module is used to mask the label data body in order to update the label data body; The training module is used to input each earthquake data in the earthquake data volume as sample data into a pre-constructed image segmentation model for training. Based on the output of the image segmentation model and each label data in the updated label data volume, the module calculates the value of a preset loss function and adjusts the parameters of the image segmentation model in reverse according to the loss function value. The training process is iterated until the loss function reaches a preset convergence condition, at which point the training stops, and a trained meandering river 3D recognition model is obtained. The output of the meandering river 3D recognition model is the recognition result of the meandering river.
11. A device for identifying meandering rivers, characterized in that, include: The acquisition module is used to acquire seismic data volumes of the same size as the area to be identified, and to divide the seismic data volumes into several sub-seismic data volumes of the same size; There is data overlap between the various sub-seismic data volumes; The standardization module is used to standardize the seismic data in each sub-seismic data volume; The prediction module is used to input the standardized sub-seismic data volumes into the meandering river 3D recognition model, obtain the prediction probability of each seismic data in each sub-seismic data volume, and determine whether the seismic data is characterized as a meandering river based on the prediction probability and a preset probability threshold. Based on seismic data characterized as meandering rivers, meandering rivers in the area to be identified are identified; The meandering river 3D recognition model is obtained by training the meandering river 3D recognition model according to any one of claims 1-7.
12. A computer storage medium, characterized in that, The computer storage medium stores computer-executable instructions, which, when executed by a processor, implement the training method for the three-dimensional recognition model of the meandering river as described in any one of claims 1-7 or the recognition method for the meandering river as described in any one of claims 8-9.
13. A terminal device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the training method for the three-dimensional recognition model of the meandering river according to any one of claims 1-7 or the recognition method for the meandering river according to any one of claims 8-9.