A method for earthquake fault identification based on dual-attention mechanism and multi-scale fusion network

Abstract

The present invention discloses a method for identifying earthquake faults based on a dual-attention mechanism and a multi-scale fusion network, which belongs to the field of seismic data interpretation technology and includes the following steps: S1, obtaining three-dimensional post-stack seismic data, slicing the seismic data volume, and obtaining multi-channel seismic data; S2, constructing a DAMFaultNet fault identification model; S3, completing fault feature extraction in the encoding stage; S4, obtaining intermediate aggregated features and original same-scale features from the encoder in the decoding stage; S5, adding a deep supervision mechanism to the terminal decoder; S6, splicing and restoring the predicted results to the original seismic profile size. The present invention adopts the above-mentioned method for identifying earthquake faults based on a dual-attention mechanism and a multi-scale fusion network, and demonstrates excellent accuracy and fault continuity recognition capabilities in fault identification tasks, realizing automatic, efficient, and accurate identification of faults in seismic images, and providing strong technical support for earthquake analysis and research.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic data interpretation, and in particular to a seismic fault identification method based on a dual-attention mechanism and a multi-scale fusion network. Background Art

[0002] Faults are important geological structures formed when rock layers fracture under tectonic stress. They are crucial factors controlling the formation and location of ore bodies and serve as storage spaces and migration pathways for underground resources. Accurate and efficient fault identification is of great significance and application value in a variety of fields, including oil and gas reservoir development, geophysical exploration, and engineering geology.

[0003] Current conventional fault interpretation methods primarily rely on automated or semi-automated identification methods that utilize computer-aided techniques combined with seismic attributes. Commonly used seismic attributes include coherence, variance, curvature, and texture, which effectively improve the accuracy and efficiency of fault identification and interpretation from seismic profiles. However, these methods often involve numerous parameters that must be set empirically and are sensitive to various noises present in the real data. Furthermore, as the amount of available seismic data increases, the computational cost of extracting seismic attributes also increases.

[0004] In the field of computer vision, seismic data interpretation tasks such as fault identification, horizon tracing, and salt dome identification can all be viewed as image classification or semantic segmentation problems. Given the advanced performance of convolutional neural networks (CNNs) in a wide range of computer vision tasks in recent years, an increasing number of researchers have begun applying CNN-based methods to seismic data interpretation. These new methods not only reduce reliance on professional seismic interpreters but also better adapt to the massive growth of data and effectively reveal patterns hidden within the data. Classic image segmentation models include FPN, U-Net, the DeepLab series, and PSPNet, which have achieved excellent results in seismic exploration. For example, Wu et al. (2019) proposed a U-Net-based 3D seismic fault model, FaultSeg3D, and trained and tested it on simulated and real data. Gao et al. (2022) built on FaultSeg3D by employing a nested U-Net structure and multi-scale fusion operations to improve the accuracy and reliability of fault detection in complex and noisy seismic images. Li et al. (2023) proposed a multi-scale residual block for mining fine-grained fault features from a low-dimensional feature space.

[0005] Although U-Net and its variants have demonstrated excellent performance in many applications, particularly in tomographic image interpretation tasks, they still face some inherent limitations. While previous improvements have alleviated these issues to some extent, they have also introduced additional complexity and computational burden. Specifically, increasing network parameters and the number of layers in pursuit of higher performance often makes the model more susceptible to overfitting, which not only impairs the model's generalization ability but also significantly increases computational complexity and time overhead. Furthermore, in terms of the selection of training data, while directly using 3D data for training can provide richer spatial information, this approach greatly increases the difficulty of model training and places higher demands on computing resources and algorithm optimization. Therefore, it is necessary to find a better balance between improving performance and maintaining model simplicity and reducing computational costs. Summary of the Invention

[0006] The purpose of this invention is to provide a seismic fault identification method based on the dual-attention mechanism and multi-scale fusion network, which demonstrates excellent accuracy and fault continuity recognition capabilities in fault identification tasks, realizes automatic, efficient and accurate identification of faults in seismic images, and provides strong technical support for earthquake analysis and research.

[0007] To achieve the above objectives, the present invention provides an earthquake fault identification method based on a dual-attention mechanism and a multi-scale fusion network, comprising the following steps:

[0008] S1. Obtain 3D post-stack seismic data, slice the seismic data volume, and obtain multi-channel seismic data suitable for model input;

[0009] S2. Based on the encoder-decoder model, we introduce ResNeSt as the backbone network in the encoder part of the U-Net++ network, and introduce multiple dual attention modules in the decoder part to build the DAMFaultNet fault recognition model;

[0010] S3. In the encoding stage, the seismic slice data obtained in step S1 undergoes a series of downsampling operations through convolution and maximum pooling to complete the fault feature extraction;

[0011] S4, in the decoding stage, the output of S3 is processed by multi-level embedded decoding units and then converged to the end decoder of the same layer. With dense connection, each encoder node fuses the aggregated features of the same scale to obtain the intermediate aggregated features and the original same-scale features from the encoder;

[0012] S5. Add a deep supervision mechanism to the end decoder, and use a different upsampling rate for each path;

[0013] S6. The prediction results are spliced ​​and restored to the original earthquake profile size.

[0014] Preferably, the seismic data in S1 includes synthetic seismic data and actual seismic data, multi-channel two-dimensional seismic image slices are extracted according to the main survey line or the connecting survey line, and the seismic data set is scaled to the interval [0,1] using a normalization method.

[0015] Preferably, during the training phase of the model, a hybrid loss function based on cross entropy, Tversky loss, and Focal loss is used as the optimization objective, and the AdamW algorithm is used as the optimizer to iteratively update the model parameters. The loss function L is specifically defined as:

[0016] L=ω1L bce +ω2L tversky +ω3L focal ;

[0017] Where ω1, ω2, and ω3 are the weight factors of each part of the loss, L bce is the binary cross entropy loss, L tversky For Tversky loss and L focal for Focal loss;

[0018] Given the true class label y and the predicted class label y', L bce The binary cross loss function is used, which is defined as:

[0019]

[0020] In the formula, μ is the sample weight coefficient, N is the number of sample points, and y i is the true category of the i-th sample point, y' i is the prediction result of the i-th sample point;

[0021] The Tversky loss function is defined as follows:

[0022]

[0023] In the formula, α and β are balance parameters. By adjusting α and β, the model's attention to positive and negative samples can be adjusted. N is the number of samples.

[0024] The Focal loss is defined as follows:

[0025]

[0026] Where ω and γ are both adjustable parameters, ω ≥ 0 is the focusing parameter, and γ is the adjustment factor.

[0027] Preferably, DAMFaultNet in S4 fuses the high-resolution feature maps from the encoder network with the corresponding semantically rich feature maps from the decoder network through a hierarchical nested network.

[0028] Preferably, the specific operation steps of the dual attention module in S4 are as follows:

[0029] S41, use 1×1 convolution and softmax function to obtain attention weights, and then obtain global context features through attention pooling operation;

[0030] S42, Conv1 and Conv2 through two bottleneck layers of 1×1 convolution obtain the dependencies between the channels and adjust the feature map according to the dependencies between the channels;

[0031] Its expression is as follows:

[0032]

[0033] Where i∈{1,2,…,H×W}, where H is the feature map height, W is the feature map width, and α j Represents the global attention weight, Conv1, Conv2 represent two different convolutional layers, LN is the layer normalization operation, ReLU is the activation function, is the module output, x i is the input feature value, j represents the position index;

[0034] S43, compress the spatial features of the feature map, and calculate the importance of spatial information at each position in the feature map through 1×1 convolution and sigmoid function;

[0035] S44, feature subgraph that concatenates global context features and important spatial features.

[0036] Preferably, in DAMFaultNet, the Swish activation function is used instead of the ReLU activation function. The Swish activation function is expressed as follows:

[0037] swish(x)=x*sigmoid(ln(1+e x ));

[0038] Where x is the independent variable and sigmoid is the activation function.

[0039] Preferably, DAMFaultNet in S5 introduces a deep supervision mechanism and adds supervision signals in each layer of the nested network.

[0040] Therefore, the present invention adopts the above-mentioned earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network, which has the following beneficial effects:

[0041] (1) Based on the advanced U-Net++ network architecture, ResNeSt is integrated as the core backbone network in the encoding stage, significantly enhancing the network's ability to extract earthquake fault characteristics;

[0042] (2) In the feature decoding stage, a dual attention mechanism is introduced, which can intelligently identify and give greater weight to information with higher weight, effectively suppressing semantic noise while highlighting the key features of the fault;

[0043] (3) A deep supervision mechanism is added to the network. This mechanism ensures that the network can accurately capture fault features at different scales by embedding supervision signals at each level of the network, thereby greatly enhancing the generalization performance of the model.

[0044] (4) In terms of data preparation, the present invention implements comprehensive data enhancement processing on the synthetic seismic data to generate a high-quality training data set.

[0045] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 This is a flow chart of an earthquake fault identification method based on a dual-attention mechanism and a multi-scale fusion network according to the present invention;

[0047] Figure 2 This is a schematic diagram of synthetic seismic data and fault labels of an earthquake fault identification method based on a dual-attention mechanism and a multi-scale fusion network according to the present invention;

[0048] Figure 3 This is a network architecture diagram of the DAMFaultNet fault identification model, an earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network of the present invention;

[0049] Figure 4 Schematic diagram of a ResNeSt unit module of an earthquake fault identification method based on a Dual-Attention mechanism and a multi-scale fusion network according to the present invention;

[0050] Figure 5 Schematic diagram of a decoder unit module of an earthquake fault identification method based on a dual-attention mechanism and a multi-scale fusion network according to the present invention;

[0051] Figure 6 Schematic diagram of a dual-attention mechanism module of an earthquake fault identification method based on a dual-attention mechanism and a multi-scale fusion network according to the present invention;

[0052] Figure 7 It is a loss and IoU curve diagram of the model training process of the earthquake fault identification method based on the Dual-Attention mechanism and multi-scale fusion network of the present invention; Figure 7 (a) in the figure is the loss graph of the model training process, where Figure 7 (b) is the IoU curve diagram of the model training process;

[0053] Figure 8 This is a diagram showing the fault identification results of the DAMFaultNet fault identification model, an earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network, on the validation set; Figure 8 (a) in the figure is the original seismic section with fault labels. Figure 8 (b) in the figure is the PSP prediction result. Figure 8 (c) in the figure is the FPN prediction result. Figure 8 (d) in the figure is the prediction result of DeepLabV3++. Figure 8 (e) in the figure is the U-Net prediction result. Figure 8 (f) in the figure is the prediction result of DAMFaultNet;

[0054] Figure 9 This is a prediction result diagram of an earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network on the GSB dataset; Figure 9 (a) is a seismic section with fault markers. Figure 9 (b) is the DAMFaultNet prediction result. Figure 9 (c) in the figure is the U-Net prediction result. Figure 9 (d) in the figure is the CNN prediction result;

[0055] Figure 10 This is a prediction result diagram of an earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network on the main survey line of the F3 dataset; Figure 10 (a) is a seismic section with fault markers. Figure 10 (b) is the DAMFaultNet prediction result. Figure 10 (c) in the figure is the U-Net prediction result;

[0056] Figure 11This is a prediction result diagram of the earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network in the F3 data set contact line and Z-axis direction;

[0057] Figure 11 (a) in the figure is the prediction result of DAMFaultNet; Figure 11 (b) in the figure is the U-Net prediction result. DETAILED DESCRIPTION

[0058] The technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0059] Unless otherwise defined, technical or scientific terms used in the present invention shall have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

[0060] Example 1

[0061] like Figure 1 As shown, the present invention provides an earthquake fault identification method based on the dual-attention mechanism and the multi-scale fusion network, comprising the following steps:

[0062] S1. Obtain 3D post-stack seismic data, slice the seismic data volume, and obtain multi-channel seismic data suitable for model input;

[0063] A convolution model using Ricker wavelets is used to calculate 3D synthetic seismic responses. A moderate amount of noise is incorporated to ensure the results closely resemble actual earthquake events. To introduce the necessary randomness, model parameters, including fault location, dip, dip, and strike, are randomly set. The fundamental purpose of this method is to accurately simulate the rich diversity and complexity exhibited in seismic data.

[0064] like Figure 2 As shown, Figure 2 A schematic diagram of synthetic seismic data and fault labels according to an embodiment of the present invention is shown. The first and third columns in the figure are seismic slice data, and the corresponding fault interpretation is defined by the binary image corresponding to the slice data, with the same resolution as the seismic data for pixels identified as faults (1) or non-faults (0).

[0065] Seismic data includes synthetic seismic data and actual seismic data. Multi-channel two-dimensional seismic image slices can be extracted according to the main survey line or the connecting survey line, and the seismic data set can be scaled to the interval [0,1] using the normalization method.

[0066] The original seismic data were cropped using an overlapping cropping strategy (16 pixels overlap). Finally, a total of 11,108 pairs of samples of size 5×64×64 were obtained, and then the training samples were randomly divided into training set, validation set and test set according to the ratio of 8:1:1.

[0067] The prepared training seismic dataset is enhanced through data augmentation techniques including two variants of random rotation, grid distortion, horizontal and vertical flipping, transposition, a combination of vertical flipping and random rotation, random brightness, random contrast, random brightness contrast, and random gamma to increase the image diversity during training.

[0068] During the training phase of the model, a hybrid loss function based on cross entropy, Tversky loss, and Focal loss is used as the optimization target, and the AdamW algorithm is used as the optimizer to iteratively update the model parameters. The loss function L is specifically defined as:

[0069] L=ω1L bce +ω2L tversky +ω3L focal ;

[0070] Where ω1, ω2, and ω3 are the weight factors of each part of the loss, L bce is the binary cross entropy loss, L tversky For Tversky's loss, L focal for Focal loss;

[0071] Given the true class label y and the predicted class label y', L bce The binary cross loss function is used, which is defined as:

[0072]

[0073] In the formula, μ is the sample weight coefficient, N is the number of sample points, and y i is the true category of the i-th sample point, y' i is the prediction result of the i-th sample point;

[0074] The Tversky loss function is defined as follows:

[0075]

[0076] In the formula, α and β are balance parameters. By adjusting α and β, the model's attention to positive and negative samples can be adjusted. N is the number of samples.

[0077] The Focal loss is defined as follows:

[0078]

[0079] Where ω and γ are both adjustable parameters, ω ≥ 0 is the focusing parameter, and γ is the adjustment factor.

[0080] S2. Based on the encoder-decoder model, ResNeSt is introduced as the backbone network in the encoder part of the U-Net++ network, and multiple dual-attention modules are introduced in the decoder part to build the DAMFaultNet fault recognition model;

[0081] like Figure 3 As shown, Figure 3 This figure shows the network architecture of the DAMFaultNet fault recognition model, based on an embodiment of the present invention. Building on the encoding-decoding model, the network utilizes multi-scale information through a nested structure and dense skip connections, overcoming the impact of semantic differences between shallow and deep features on segmentation results. Unlike conventional symmetric encoding and decoding structures, DAMFaultNet uses a hierarchical nested network to fuse high-resolution feature maps from the encoder network with corresponding semantically rich feature maps from the decoder network, gradually enriching the model. Furthermore, dense skip connections are introduced between the encoder and decoder to reduce the feature map gap between the encoding and decoding subnetworks.

[0082] S3. In the encoding stage, the seismic slice data obtained in step S1 undergoes a series of downsampling operations through convolution and maximum pooling to complete the fault feature extraction;

[0083] The specific encoder structure is as follows Figure 4 As shown in Figure 1, the input is divided into K groups along the channel dimension. Each base group is further divided and a different feature transformation is applied to each group. A voting mechanism is then applied to form attention. Finally, all outputs are accumulated to form the final output.

[0084] S4, in the decoding stage, the output of S3 is processed by multi-level embedded decoding units and then converged to the end decoder of the same layer. With dense connection, each encoder node fuses the aggregated features of the same scale to obtain the intermediate aggregated features and the original same-scale features from the encoder;

[0085] DAMFaultNet utilizes multi-scale information through a nested structure and dense skip connections, overcoming the impact of semantic differences between shallow and deep features on segmentation results. Unlike conventional symmetric encoder-decoder structures, DAMFaultNet uses a layered nested network to fuse high-resolution feature maps from the encoder network with corresponding semantically rich feature maps from the decoder network, gradually enriching the model.

[0086] Through the horizontal output layer, the feature maps of different scales constructed by the encoder are fused layer by layer. i,jrepresents the jth node X in the i-th layer i,j The output of , then the corresponding mathematical expression is:

[0087]

[0088] Where DConv represents the decoder convolution block, Concat represents the feature concatenation function, [*] represents the object list, and Up represents the upsampling method.

[0089] like Figure 5 As shown, Figure 5 The internal structure of the decoder unit of an embodiment of the present invention is shown. In the decoder unit, a dual attention mechanism is used to automatically select information with a larger weight ratio to suppress semantic differences and highlight significant features.

[0090] The specific operation steps of the dual attention module are as follows:

[0091] S41, use 1×1 convolution and softmax function to obtain attention weights, and then obtain global context features through attention pooling operation;

[0092] In DAMFaultNet, the Swish activation function is used instead of the ReLU activation function. The Swish activation function is expressed as follows:

[0093] swish(x)=x*sigmoid(ln(1+e x ));

[0094] Where x is the independent variable and sigmoid is the activation function.

[0095] S42, Conv1 and Conv2 through two bottleneck layers of 1×1 convolution obtain the dependencies between the captured channels and adjust the feature map according to the dependencies between the channels;

[0096] Its expression is as follows:

[0097]

[0098] Where i∈{1,2,…,H×W}, where H is the feature map height, W is the feature map width, and α j Represents the global attention weight, Conv1, Conv2 represent two different convolutional layers, LN is the layer normalization operation, ReLU is the activation function, is the module output, x i is the input feature value, j represents the position index;

[0099] S43, compress the spatial features of the feature map, and calculate the importance of spatial information at each position in the feature map through 1×1 convolution and sigmoid function;

[0100] S44, feature subgraph that concatenates global context features and important spatial features.

[0101] like Figure 6 As shown, Figure 6 A diagram showing the internal structure of the dual attention module in an embodiment of the present invention. This module calculates the importance of global and spatial features in feature maps, enhancing the network's ability to learn long-range dependencies and spatial features in feature maps, thereby optimizing target detection.

[0102] S5. Add a deep supervision mechanism to the end decoder, and use a different upsampling rate for each path;

[0103] DAMFaultNet introduces a deep supervision mechanism, which improves the multi-scale detection performance of the network by adding supervision signals in each layer of the nested network. It also enhances its adaptability to images of different sizes and improves the generalization of the network.

[0104] S6. The prediction results are spliced ​​and restored to the original earthquake profile size. The prediction result size is 128×128.

[0105] In a specific example, the training process of the fault recognition model is as follows:

[0106] In this example, considering the uneven distribution of tomographic pixels, a stratified sampling method was used, with the proportion of tomographic pixels as the sampling basis. The dataset was then randomly divided into three subsets, designated as training, validation, and test sets, with a ratio of 8:1:1. The AdamW optimizer was used for model training. Due to GPU memory limitations, the batch size was set to 64. The optimizer's initial learning rate was set to 0.0008, and the weight decay coefficient was set to 0.01.

[0107] During training, the learning rate is dynamically adjusted by monitoring the performance metric, Intersection over Union (IoU). Once this value remains constant for six training epochs, the learning rate is halved. Furthermore, an early stopping mechanism is employed to prevent model overfitting. Specifically, the training process is monitored to see if the validation set loss is decreasing. Training is terminated if the loss does not decrease after 12 epochs. The maximum number of training epochs in the experiment is set to 200, and model evaluation metrics are recorded for each epoch on both the training and validation sets.

[0108] like Figure 7 As shown, Figure 7 The loss and IoU curves of the model training process of the embodiment of the present invention are shown. Figure 7(a) in the figure is the loss graph of the model training process. Figure 7 (b) in the figure is the IoU curve of the model training process, as shown in Figure 7 As shown, since the validation loss did not increase, the early stopping mechanism was triggered and the training stopped after 90 iterations.

[0109] To quantitatively compare the prediction results, we use accuracy, precision, F1 score, and intersection over union (IoU) as four main metrics. In the task of fault prediction, we consider fault pixels as positive (1) and non-fault pixels as negative (0). As shown in Table 1, true positives (TP) refer to the number of pixels accurately predicted as faults, while true negatives (TN) refer to the number of pixels accurately predicted as non-faults. False positives (FP) and false negatives (FN) refer to fault pixels and non-fault pixels that are incorrectly predicted, respectively.

[0110] Table 1 Confusion matrix

[0111]

[0112] The four indicators are defined as follows:

[0113]

[0114] Among them, Recall is the recall rate.

[0115] The embodiment of the present invention verifies the performance of the fault identification model, as follows:

[0116] First, we conducted a comparative analysis on a synthetic earthquake dataset with existing state-of-the-art methods. To ensure a fair comparison, we used the same parameters throughout, initializing both the experimental and model parameters using pre-trained ImageNet parameters. The test set consisted of 1,000 slices, each measuring 5 × 64 × 64. This dataset was used to quantitatively evaluate the model, and the results are shown in Table 2.

[0117] Table 2 Comparison of experimental results on the test set

[0118]

[0119] As shown in Table 2, DAMFaultNet performs best across all four evaluation metrics, achieving an accuracy of 0.9843, an F1 score of 0.7742, an IoU of 0.6024, and an AUROC of 0.8759. It should be noted that DAMFaultNet outperforms the other four models in many aspects, achieving significant improvements over the state-of-the-art models. Although U-Net performs similarly to the proposed model in terms of accuracy, its lower F1 score and IoU indicate that it misses many fault structures and produces a significant amount of FN in its predictions. Furthermore, after introducing the dual attention mechanism, the traditional U-Net model shows significant improvements. This demonstrates the effectiveness of the dual attention mechanism in enhancing feature fusion and fault identification within the U-Net network.

[0120] In order to more intuitively understand the differences in fault identification performance between various models, the proposed model is qualitatively evaluated by visualizing the fault identification results using a test dataset. Figure 8 As shown, Figure 8 Figure 4 shows the fault recognition results of the DAMFaultNet fault recognition model on the validation set according to an embodiment of the present invention; wherein, Figure 8 (a) in the figure is the original seismic section with fault labels. Figure 8 (b) in the figure is the PSP prediction result. Figure 8 (c) in the figure is the FPN prediction result. Figure 8 (d) in the figure is the prediction result of DeepLabV3++. Figure 8 (e) in the figure is the U-Net prediction result. Figure 8 (f) in the figure is the prediction result of DAMFaultNet.

[0121] To train and evaluate the proposed model, two public real-world earthquake datasets were used in the experiments. The first dataset is the GSB dataset, which is from the Great South Basin in New Zealand and has a size of 5 × 76 × 484. The second dataset is the F3 dataset, which is provided by TNO and dGB Earth Sciences and has a size of 415 × 96 × 384.

[0122] like Figure 9 As shown, Figure 9 This is a comparison of the prediction results of different models on the GSB dataset in the embodiment of the present invention. Figure 9 As shown in (a) of Figure 1, the GSB seismic data includes various earthquake faults, most of which are vertical. These faults are particularly obvious on the reflecting surface. Figure 9Subfigures (b)-(d) in the figure show the predicted faults using the DAMFaultNet model, the U-Net model proposed by An et al. (2021), and the method used by Cunha et al. (2020). Compared to the actual labels, several unlabeled faults were found and exhibited good continuity, as indicated by the red boxes. Furthermore, in some areas without fault features, possible faults were found, as indicated by the green arrows.

[0123] like Figure 10 As shown, Figure 10 This is the prediction result on the main measurement line of the F3 dataset according to an embodiment of the present invention. Figure 10 (a) in the figure depicts the seismic data of 160 main survey lines and the true value of fault distribution. Figure 10 (b) shows the results obtained when using our proposed DAMFaultNet network. Figure 10 (c) in the figure is the prediction result of the U-Net method using ResNeSt101 as the encoder. Figure 10 As shown in (c) of Figure 3, the faults extracted by U-Net exhibit discontinuous features, as indicated by the green arrows. Compared with U-Net, the prediction results produced by DAMFaultNet are more consistent with the labels. Although DAMFaultNet can predict a wider range of faults than U-Net, as shown in Figure 3, Figure 10 As shown in the red box in (b), its predictions do not result in more fault predictions than labels. This highlights the significant impact of the training set on the network's recognition capabilities. Both U-Net and DAMFaultNet demonstrate practical potential for application in data domains. However, due to DAMFaultNet's enhanced multi-scale fusion capabilities, its predicted fault results are more complex than U-Net, and the fault lines are more complete.

[0124] like Figure 11 As shown, Figure 11 This is the prediction result of the connection line and Z-axis direction in the F3 data set according to an embodiment of the present invention. Figure 11 (a) is the prediction result of DAMFaultNet. Figure 11 (b) in the figure is the U-Net prediction result. Figure 11 (c) in the figure is the prediction result of DAMFaultNet in the Z-axis direction; Figure 11 (d) in the figure shows the prediction result of U-Net in the Z-axis direction. The results verify the effectiveness of our model and highlight the significant advantages of DAMFaultNet in improving model performance and prediction accuracy.

[0125] Therefore, the present invention adopts the above-mentioned earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network, which demonstrates excellent accuracy and fault continuity recognition capabilities in fault identification tasks, and realizes automatic, efficient and accurate identification of faults in seismic images, providing strong technical support for earthquake analysis and research.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for earthquake fault identification based on a dual-attention mechanism and a multi-scale fusion network, characterized by: The following steps are involved: S1. Obtain 3D post-stack seismic data, slice the seismic data volume, and obtain multi-channel seismic data suitable for model input; S2. Based on the encoder-decoder model, we introduce ResNeSt as the backbone network in the encoder part of the U-Net++ network, and introduce multiple dual attention modules in the decoder part to build the DAMFaultNet fault recognition model; S3. In the encoding stage, the seismic slice data obtained in step S1 undergoes a series of downsampling operations through convolution and maximum pooling to complete the fault feature extraction; S4, in the decoding stage, the output of S3 is processed by multi-level embedded decoding units and then converged to the end decoder of the same layer. With dense connection, each encoder node fuses the aggregated features of the same scale to obtain the intermediate aggregated features and the original same-scale features from the encoder; S5. Add a deep supervision mechanism to the end decoder, and use a different upsampling rate for each path; S6. The prediction results are spliced ​​and restored to the original earthquake profile size.

2. The earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network according to claim 1 is characterized by: The seismic data in S1 include synthetic seismic data and actual seismic data. Multi-channel two-dimensional seismic image slices are extracted according to the main survey line or the connecting survey line, and the seismic data set is scaled to the interval [0,1] using the normalization method.

3. The earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network according to claim 2 is characterized by: During the training phase of the model, a hybrid loss function based on cross entropy, Tversky loss, and Focal loss is used as the optimization target, and the AdamW algorithm is used as the optimizer to iteratively update the model parameters. The loss function L is specifically defined as: L=ω1L bce +ω2L tversky +ω3L focal ; Where ω1, ω2, and ω3 are the weight factors of each part of the loss, L bce is the binary cross entropy loss, L tversky For Tversky's loss, L focal for Focal loss; Given the true class label y and the predicted class label y', L bce The binary cross loss function is used, which is defined as: In the formula, μ is the sample weight coefficient, N is the number of sample points, and y i is the true category of the i-th sample point, y' i is the prediction result of the i-th sample point; The Tversky loss function is defined as follows: In the formula, α and β are balance parameters. By adjusting α and β, the model's attention to positive and negative samples can be adjusted. N is the number of samples. The Focal loss is defined as follows: Where ω and γ are both adjustable parameters, ω ≥ 0 is the focusing parameter, and γ is the adjustment factor.

4. The earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network according to claim 1, characterized in that: In S4, DAMFaultNet fuses the high-resolution feature maps from the encoder network with the corresponding semantically rich feature maps from the decoder network through a hierarchical nested network.

5. The earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network according to claim 4 is characterized in that: The specific operation steps of the dual attention module in S4 are as follows: S41, use 1×1 convolution and softmax function to obtain attention weights, and then obtain global context features through attention pooling operation; S42, Conv1 and Conv2 through two bottleneck layers of 1×1 convolution obtain the dependencies between the channels and adjust the feature map according to the dependencies between the channels; Its expression is as follows: Where i∈{1,2,…,H×W}, where H is the feature map height, W is the feature map width, and α j Represents the global attention weight, Conv1, Conv2 represent two different convolutional layers, LN is the layer normalization operation, ReLU is the activation function, is the module output, x i is the input feature value, j represents the position index; S43, compress the spatial features of the feature map, and calculate the importance of spatial information at each position in the feature map through 1×1 convolution and sigmoid function; S44, feature subgraph that concatenates global context features and important spatial features.

6. The earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network according to claim 5, characterized in that: In DAMFaultNet, the Swish activation function is used instead of the ReLU activation function. The Swish activation function is expressed as follows: swish(x)=x*sigmoid(ln(1+e x )); Where x is the independent variable and sigmoid is the activation function.

7. The earthquake fault identification method based on the dual-attention mechanism and multi-scale fusion network according to claim 1, characterized in that: In S5, DAMFaultNet introduces a deep supervision mechanism and adds supervision signals to each layer of the nested network.

Images