Earthquake fault detection method and system based on N-type network

Abstract

The invention relates to the technical field of seismic exploration data processing, provides a seismic fault detection method and system based on an N-type network, and aims to solve the problems that in a seismic fault detection result in the prior art, continuity of some areas is insufficient, and a seismic fault structure is not complete enough; and the technical problem that the GPU memory is greatly consumed in the existing prediction process is solved. Comprising the following steps: inputting a seismic cross-sectional image into an input layer for down-sampling processing; then inputting the down-sampling fault feature map into an encoder, and performing semantic extraction on the down-sampling fault feature map by using a plurality of convolutional layers; respectively inputting the down-sampling fault feature map and the plurality of semantic extraction feature maps into a first decoder and a second decoder, and decoding the input feature maps by using the first decoder and the second decoder according to an opposite sequence; and then feature map input and output layers output by the encoder and the two decoders are subjected to splicing processing, and finally fault detection is carried out to obtain a fault detection result.

Description

Technical Field

[0001] The present invention relates to the technical field of seismic exploration data processing, and in particular to a seismic fault detection method and system based on an N-type network. Background Art

[0002] Oil and gas resources are not only important chemical raw materials, but also important energy bases, and are known as the "food and blood of industry". The interpretation of seismic faults has always been a very important link in oil and gas exploration and development. Seismic faults are not only the boundaries of oil and gas fields, but also the channels for oil and gas migration and accumulation. The distribution and morphology of seismic faults play a key role in the identification and description of oil and gas, and have a great impact on the exploitation and distribution of oil and gas. In underground structures, seismic faults exist in various forms and sizes. In the transportation of oil and gas, seismic faults have the possibility of serving as channels or blockages. Therefore, clarifying the location and morphological distribution of seismic faults is an indispensable link in the entire process of oil and gas exploration. With the increasing scale of seismic exploration, the shortening of the cycle, and the in-depth development of exploration theory, seismic faults, as one of the three major transmission systems, have increasingly higher requirements for their interpretation accuracy. Therefore, the research on seismic fault identification has attracted much attention from the industry and academia.

[0003] With the continuous improvement of seismic exploration technology, the volume of seismic data is also increasing. Manual interpretation of seismic faults is not only very trivial and time-consuming, but also difficult and cannot be repeatedly verified. The complex interpretation process also requires a high level of professionalism from the interpreters. Manual interpretation can no longer meet today's production needs. Traditional methods identify seismic faults by calculating the coherence, curvature, variance and other attributes of the seismic data body. However, the seismic fault detection method based on attribute calculation is very sensitive to noise and formation characteristics, and has the disadvantages of being greatly interfered by noise and insufficient accuracy.

[0004] In the prior art, a large amount of earthquake fault data is obtained as a training set, and then the training set is preprocessed, the preprocessed earthquake data is input into a deep learning model for training, and the trained model is applied to new earthquake fault data to perform earthquake fault detection.

[0005] The prior art has the following technical problems:

[0006] 1. The earthquake fault detection results generated by the existing depth model have problems such as insufficient continuity in some areas and incomplete earthquake fault structure.

[0007] 2. The training of deep learning models is easily limited by hardware level and requires the use of high-performance graphics cards. Due to the large volume of 3D seismic data, the prediction process will greatly consume GPU memory. Summary of the invention

[0008] The present invention provides an N-type network-based earthquake fault detection method and system, aiming to solve the technical problems existing in the above-mentioned prior art, that is, in some areas of the earthquake fault detection results, the lack of continuity and the incompleteness of the earthquake fault structure; and that the training of the existing deep learning model requires the use of a high-performance graphics card, and the prediction process will greatly consume the GPU memory.

[0009] The technical solution of the present invention to solve the above technical problems is as follows: a method for detecting earthquake faults based on an N-type network, wherein the N-type network includes an input layer, an encoder, a first decoder, a second decoder and an output layer; earthquake fault detection using the N-type network specifically includes:

[0010] Acquire an initial earthquake fault image, preprocess the initial earthquake fault image to obtain a preprocessed image, input the preprocessed image into the input layer for downsampling processing, and obtain a downsampled fault feature map;

[0011] Inputting the downsampled fault feature map into the encoder, performing semantic extraction on the downsampled fault feature map using a cascade structure formed by stacking multiple first double convolutional layers built into the encoder, and obtaining multiple semantic extraction feature maps corresponding to the outputs of the multiple first double convolutional layers;

[0012] Inputting the downsampled fault feature map and the plurality of semantic extraction feature maps into the first decoder and the second decoder respectively, and decoding the downsampled fault feature map and the plurality of semantic extraction feature maps in reverse order by using the first decoder and the second decoder to obtain a first decoded feature map and a second decoded feature map respectively;

[0013] The first decoding feature map, the second decoding feature map and the semantic extraction feature map output by the last first double convolutional layer in the cascade structure are input into the output layer for splicing to obtain a fault splicing feature map, the fault splicing feature map is adjusted to an initial resolution, and fault detection is performed on the adjusted fault splicing feature map to obtain a fault detection result; wherein the initial resolution is the resolution of the initial seismic tomography image.

[0014] Furthermore, the method further includes: constructing an initial network, training the initial network, and obtaining an N-type network. The obtaining of the N-type network specifically includes:

[0015] Acquire original seismic images, and generate a corresponding fault label for each original seismic image to obtain an original seismic image set, and divide the original seismic image set into a training set and a validation set;

[0016] Inputting each original seismic image in the training set into the initial network in sequence to perform fault prediction, obtaining a plurality of first fault prediction results, calculating a first loss value between the plurality of first fault prediction results and the fault labels corresponding to the training set, and updating the network parameters of the initial network according to the first loss value;

[0017] The initial network after updating the network parameters is used to repeatedly traverse the training set for iterative training, and one cycle is defined as one time when the initial network traverses the entire training set;

[0018] When each cycle ends, the initial network after updating the network parameters using the verification set as input is used to perform fault prediction to obtain a second fault prediction result, and a second loss value between the second fault prediction result and the fault label corresponding to the verification set is calculated;

[0019] Set a patience period. At the end of each cycle, record the loss difference between the second loss value of this cycle and the second loss value of the previous cycle. If the loss difference obtained after multiple consecutive cycles reaches the stopping condition and the sum of multiple consecutive cycles is greater than the patience period, stop training the initial network and select the initial network corresponding to the minimum value of all second loss values ​​as the N-type network.

[0020] Further, the first loss value and the second loss value are calculated using a balanced binary cross entropy loss function; the formula of the balanced binary cross entropy loss function is shown as follows:

[0021]

[0022] Where L represents the first loss value or the second loss value, β represents the ratio of non-fault pixels to all pixels of the original seismic image, N represents all the pixels of the original seismic image, y i Indicates the fault label, p i Represents the probability that pixel i is predicted to be a fault label.

[0023] Further, the input layer includes a first downsampling module, a first convolutional layer and a second downsampling module, and inputting the preprocessed image into the input layer for downsampling processing specifically includes:

[0024] The preprocessed image is processed by the first downsampling module, and the output of the first downsampling module is input into the second downsampling module after being processed by the first convolution layer. The second downsampling module performs feature extraction processing and outputs a downsampled fault feature map.

[0025] Further, the encoder includes a cascade structure formed by stacking three first double convolutional layers, and the multiple semantic extraction feature maps corresponding to the outputs of the multiple first double convolutional layers specifically include:

[0026] Inputting the downsampled fault feature map into a first double convolutional layer to obtain a first semantic extraction feature map;

[0027] Inputting the first semantic extraction feature map into the second first double convolutional layer to obtain a second semantic extraction feature map;

[0028] Inputting the second semantic extraction feature map into the third first double convolutional layer to obtain a third semantic extraction feature map;

[0029] Among them, the first semantic extraction feature map, the second semantic extraction feature map and the third semantic extraction feature map are multiple semantic extraction feature maps corresponding to the outputs of multiple first dual convolutional layers.

[0030] Furthermore, the encoder is jump-connected to the first decoder and the second decoder respectively, the first decoder includes three second dual convolutional layers, and the first decoding feature map obtained specifically includes:

[0031] Inputting the downsampled fault feature map and the first semantic extraction feature map into the first decoder for feature splicing to obtain a first spliced ​​feature map;

[0032] Inputting the first concatenated feature map into the first second double convolutional layer to obtain a third decoding feature map, and concatenating the third decoding feature map with the second semantic extraction feature map to obtain a second concatenated feature map;

[0033] Inputting the second concatenated feature map into the second second dual convolutional layer to obtain a fourth decoding feature map, and concatenating the fourth decoding feature map with the third semantic extraction feature map to obtain a third concatenated feature map;

[0034] Inputting the third concatenated feature map into the third second double convolutional layer to obtain a first decoding feature map;

[0035] The second decoder includes three third dual convolutional layers, and obtaining the second decoding feature map specifically includes:

[0036] Inputting the third semantic extraction feature map and the second semantic extraction feature map into the second decoder for feature splicing to obtain a fourth splicing feature map;

[0037] Inputting the fourth concatenated feature map into the first third dual convolutional layer to obtain a fifth decoding feature map, and concatenating the fifth decoding feature map with the first semantic extraction feature map to obtain a fifth concatenated feature map;

[0038] Inputting the fifth spliced ​​feature map into the second third dual convolutional layer to obtain a sixth decoded feature map, and splicing the sixth decoded feature map with the down-sampled fault feature map to obtain a sixth spliced ​​feature map;

[0039] The sixth concatenated feature map is input into the third double convolutional layer to obtain a second decoding feature map.

[0040] Further, the output layer includes a first upsampling module, a second upsampling module and a fifth convolutional layer, and the fault detection result obtained specifically includes:

[0041] Inputting the third semantic extraction feature map, the first decoding feature map and the second decoding feature map into the output layer for splicing to obtain a fault splicing feature map;

[0042] Processing the fault splicing feature map using the first upsampling module and the second upsampling module to restore the fault splicing feature map to an initial resolution, thereby obtaining a fault upsampling feature map;

[0043] The upsampled feature map of the fault is input into the fifth convolutional layer to perform fault detection to obtain a fault detection result.

[0044] In a second aspect, in order to solve the above technical problems, the present invention also provides an N-type network-based earthquake fault detection system, comprising:

[0045] An input layer module, used for acquiring an initial earthquake fault image, preprocessing the initial earthquake fault image to obtain a preprocessed image, and downsampling the preprocessed image to obtain a downsampled fault feature map;

[0046] An encoder, configured to perform semantic extraction on the downsampled fault feature map by utilizing a cascade structure formed by stacking a plurality of first double convolutional layers built into the encoder, and obtain a plurality of semantic extraction feature maps corresponding to outputs of the plurality of first double convolutional layers;

[0047] A first decoder is used to receive the downsampled fault feature map and the multiple semantic extraction feature maps input by the encoder, and decode the downsampled fault feature map and the multiple semantic extraction feature maps according to the input order to obtain a first decoded feature map;

[0048] A second decoder is used to receive the downsampled fault feature map and the multiple semantic extraction feature maps input by the encoder, and decode the downsampled fault feature map and the multiple semantic extraction feature maps in a decoding order opposite to that of the first decoder to obtain a second decoded feature map;

[0049] The output layer module is used to input the first decoding feature map, the second decoding feature map and the semantic extraction feature map output by the last first double convolutional layer in the cascade structure into the output layer for splicing to obtain a fault splicing feature map, adjust the fault splicing feature map to an initial resolution, and perform fault detection on the adjusted fault splicing feature map to obtain a fault detection result; wherein the initial resolution is the resolution of the initial seismic fault image.

[0050] In the third aspect, in order to solve the above-mentioned technical problems, the present invention also provides an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, an N-type network-based earthquake fault detection method of the present application is implemented.

[0051] In a fourth aspect, in order to solve the above-mentioned technical problem, the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, an N-type network-based earthquake fault detection method of the present application is implemented.

[0052] Compared with the prior art, the present invention has the following advantages:

[0053] 1. The N-type network designed by the present invention can realize the communication and fusion of low-level semantic information and high-level semantic information through the jump connection structure between the encoder and the decoder, and obtain better fault detection results. The image resolution of the encoder and the decoder at different stages in the N-type network of the present invention remains unchanged, and feature fusion and decoding in different directions can be realized, while better preserving the image edge information.

[0054] 2. The N-type network proposed in the present invention adopts a "bidirectional decoding" strategy, which can learn richer and more detailed fault features. Through the "bidirectional decoding" strategy, low-level semantic information and high-level semantic information can be fully interactively integrated to obtain more accurate and continuous fault detection results.

[0055] 3. After the seismic data of the present invention is input into the N-type network, it is first downsampled twice, and the image size becomes one-fourth of the original size, and then the encoding and decoding operations are performed. Therefore, the computational consumption of the N-type network is smaller and the computational speed is faster. At the same time, the maximum number of convolution kernel channels in the N-type network is 16. Thanks to the targeted network structure design, the N-type network can use a smaller network size to achieve excellent prediction performance. The compression of the number of channels greatly saves computational consumption.

[0056] 4. The N-type network of the present invention uses synthetic seismic data during training and testing. Compared with the model using actual seismic data with manual annotation, it can ensure more sufficient training samples; secondly, there will be no false negative label problem in the training data, which ensures the correctness of model training and improves the model prediction effect.

[0057] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0059] Figure 1 A schematic diagram of an N-type network structure of an N-type network-based earthquake fault detection method according to an embodiment of the present invention is shown;

[0060] FIG2( a ) shows a schematic diagram of an original seismic image of a training set of a seismic fault detection method based on an N-type network according to an embodiment of the present invention;

[0061] FIG2( b ) shows a schematic diagram of the fault label corresponding to FIG2( a );

[0062] FIG3( a ) shows a schematic diagram of an original seismic image of a validation set of an earthquake fault detection method based on an N-type network according to an embodiment of the present invention;

[0063] FIG3( b ) shows a schematic diagram of a fault prediction result image corresponding to FIG3( a );

[0064] FIG3( c ) shows a schematic diagram of the fault label corresponding to FIG3( a );

[0065] FIG4( a ) shows a schematic diagram of a seismic image to be detected;

[0066] FIG4( b ) is a schematic diagram showing the fault detection result based on FIG4( a ) of the present invention;

[0067] FIG4(c) is a schematic diagram showing the fault detection result based on FIG4(a) in the prior art;

[0068] FIG5( a ) shows a three-dimensional schematic diagram of a seismic image to be detected;

[0069] FIG5(b) shows a schematic diagram of the fault detection result based on FIG5(a) of the present invention;

[0070] FIG5(c) is a schematic diagram showing the fault detection result of the prior art based on FIG5(a);

[0071] FIG6( a ) shows a schematic diagram of a time slice of a seismic image to be detected;

[0072] FIG6( b ) is a schematic diagram showing the fault detection result of the present invention based on FIG6( a );

[0073] FIG6(c) is a schematic diagram showing the fault detection result based on FIG6(a) in the prior art. DETAILED DESCRIPTION

[0074] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0075] Figure 1 FIG. 4 shows a schematic diagram of an N-type network structure of an earthquake fault detection method based on an N-type network according to an embodiment of the present invention. Figure 1 As shown, an earthquake fault detection method based on an N-type network according to an embodiment of the present invention, wherein the N-type network includes an input layer, an encoder, a first decoder, a second decoder and an output layer; earthquake fault detection using the N-type network specifically includes:

[0076] Acquire an initial earthquake fault image, preprocess the initial earthquake fault image to obtain a preprocessed image, input the preprocessed image into the input layer for downsampling processing, and obtain a downsampled fault feature map;

[0077] Inputting the downsampled fault feature map into the encoder, performing semantic extraction on the downsampled fault feature map using a cascade structure formed by stacking multiple first double convolutional layers built into the encoder, and obtaining multiple semantic extraction feature maps corresponding to the outputs of the multiple first double convolutional layers;

[0078] Inputting the downsampled fault feature map and the plurality of semantic extraction feature maps into the first decoder and the second decoder respectively, and decoding the downsampled fault feature map and the plurality of semantic extraction feature maps in reverse order by using the first decoder and the second decoder to obtain a first decoded feature map and a second decoded feature map respectively;

[0079] The first decoding feature map, the second decoding feature map and the semantic extraction feature map output by the last first double convolutional layer in the cascade structure are input into the output layer for splicing to obtain a fault splicing feature map, the fault splicing feature map is adjusted to an initial resolution, and fault detection is performed on the adjusted fault splicing feature map to obtain a fault detection result; wherein the initial resolution is the resolution of the initial seismic tomography image.

[0080] Among them, the present invention regards fault detection as an image segmentation problem, that is, dividing the original three-dimensional image of seismic data into two parts, the fault part and the non-fault part. The proportion of fault pixels in the three-dimensional seismic image is very small, and the whole presents a continuous planar structure, which requires the network to accurately learn the detailed information in the image. According to the characteristics of seismic data, starting from the difficulties and existing problems of fault detection tasks, the present invention designs a lightweight and efficient network, the N-type network, for fault detection.

[0081] Optionally, the method further includes: constructing an initial network, training the initial network to obtain an N-type network, and obtaining the N-type network specifically includes:

[0082] Acquire original seismic images, and generate a corresponding fault label for each original seismic image to obtain an original seismic image set, and divide the original seismic image set into a training set and a validation set;

[0083] Inputting each original seismic image in the training set into the initial network in sequence to perform fault prediction, obtaining a plurality of first fault prediction results, calculating a first loss value between the plurality of first fault prediction results and the fault labels corresponding to the training set, and updating the network parameters of the initial network according to the first loss value;

[0084] The initial network after updating the network parameters is used to repeatedly traverse the training set for iterative training, and one cycle is defined as one time when the initial network traverses the entire training set;

[0085] When each cycle ends, the initial network after updating the network parameters using the verification set as input is used to perform fault prediction to obtain a second fault prediction result, and a second loss value between the second fault prediction result and the fault label corresponding to the verification set is calculated;

[0086] Set a patience period. At the end of each cycle, record the loss difference between the second loss value of this cycle and the second loss value of the previous cycle. If the loss difference obtained after multiple consecutive cycles reaches the stopping condition and the sum of multiple consecutive cycles is greater than the patience period, stop training the initial network and select the initial network corresponding to the minimum value of all second loss values ​​as the N-type network.

[0087] Among them, the present invention is based on the principle of seismic image generation, and uses mathematical methods to artificially synthesize three-dimensional original seismic images and corresponding fault labels. A total of 200 pairs of training set data and 20 pairs of validation set data are generated, and the data size is 128×128×128. First, all original seismic images are standardized by subtracting the mean and dividing the standard deviation, and then data augmentation is performed. Specifically, each data body is rotated around the timeLine axis and flipped up and down. This operation can effectively avoid overfitting during network training, and the trained model is more robust, while significantly improving data quality. Finally, the processed data is input into the network for training. Figure 2 (a) and Figure 2 (b) show the original seismic images and corresponding fault labels of the training set; Figure 3 (a), Figure 3 (b) and Figure 3 (c) show the original seismic images, fault prediction results and corresponding fault labels of the validation set.

[0088] Optionally, a balanced binary cross entropy loss function is used to calculate the first loss value and the second loss value; the formula of the balanced binary cross entropy loss function is shown as follows:

[0089]

[0090] Where L represents the first loss value or the second loss value, β represents the ratio of non-fault pixels to all pixels of the original seismic image, N represents all the pixels of the original seismic image, y i Indicates the fault label, p i Represents the probability that pixel i is predicted to be a fault label.

[0091] in,

[0092] Optionally, the input layer includes a first downsampling module, a first convolutional layer, and a second downsampling module, and inputting the preprocessed image into the input layer for downsampling processing specifically includes:

[0093] The preprocessed image is processed by the first downsampling module, and the output of the first downsampling module is input into the second downsampling module after being processed by the first convolution layer. The second downsampling module performs feature extraction processing and outputs a downsampled fault feature map.

[0094] The input layer of the present invention includes two convolution-based downsampling stages, and each downsampling reduces the resolution of the preprocessed image by half. The first downsampling layer uses a convolution module with a convolution kernel size of 5 and a step size of 2; the number of channels is 8 and the size is 64×64×64. The convolution kernel size of the second downsampling is 3, and other parameters remain unchanged; the number of channels of the output feature map is 16 and the size is 32×32×32. Between the two downsampling, a first convolution layer with a convolution kernel size of 3×3×3 is included.

[0095] Optionally, the encoder includes a cascade structure formed by stacking three first double convolutional layers, and obtaining multiple semantic extraction feature maps corresponding to the outputs of the multiple first double convolutional layers specifically includes:

[0096] Inputting the downsampled fault feature map into a first double convolutional layer to obtain a first semantic extraction feature map;

[0097] Inputting the first semantic extraction feature map into the second first double convolutional layer to obtain a second semantic extraction feature map;

[0098] Inputting the second semantic extraction feature map into the third first double convolutional layer to obtain a third semantic extraction feature map;

[0099] Among them, the first semantic extraction feature map, the second semantic extraction feature map and the third semantic extraction feature map are multiple semantic extraction feature maps corresponding to the outputs of multiple first dual convolutional layers.

[0100] The encoder of the present application receives the sampled fault feature map processed by the input layer and learns the fault information therefrom. The encoder includes three first double convolutional layers, each of which is composed of two convolutional layers with a convolution kernel of 3×3×3. The number of channels of the feature map output by each first double convolutional layer is 16, and the size is 32×32×32.

[0101] Among them, two 3×3×3 convolutional layers are used to extract features in each stage. No pooling layer is used for downsampling between the two stages, so the three first double convolutional layers of the encoder extract features at the same image resolution, which can effectively preserve the edge information of the image.

[0102] Optionally, the encoder is jump-connected to the first decoder and the second decoder respectively, the first decoder includes three second dual convolutional layers, and obtaining the first decoding feature map specifically includes:

[0103] Inputting the downsampled fault feature map and the first semantic extraction feature map into the first decoder for feature splicing to obtain a first spliced ​​feature map;

[0104] Inputting the first concatenated feature map into the first second double convolutional layer to obtain a third decoding feature map, and concatenating the third decoding feature map with the second semantic extraction feature map to obtain a second concatenated feature map;

[0105] Inputting the second concatenated feature map into the second second dual convolutional layer to obtain a fourth decoding feature map, and concatenating the fourth decoding feature map with the third semantic extraction feature map to obtain a third concatenated feature map;

[0106] Inputting the third concatenated feature map into the third second double convolutional layer to obtain a first decoding feature map;

[0107] The second decoder includes three third dual convolutional layers, and obtaining the second decoding feature map specifically includes:

[0108] Inputting the third semantic extraction feature map and the second semantic extraction feature map into the second decoder for feature splicing to obtain a fourth splicing feature map;

[0109] Inputting the fourth concatenated feature map into the first third dual convolutional layer to obtain a fifth decoding feature map, and concatenating the fifth decoding feature map with the first semantic extraction feature map to obtain a fifth concatenated feature map;

[0110] Inputting the fifth spliced ​​feature map into the second third dual convolutional layer to obtain a sixth decoded feature map, and splicing the sixth decoded feature map with the down-sampled fault feature map to obtain a sixth spliced ​​feature map;

[0111] The sixth concatenated feature map is input into the third double convolutional layer to obtain a second decoding feature map.

[0112] The decoder receives four feature maps obtained by the encoder, namely, the downsampled fault feature map, the first semantic extraction feature map, the second semantic extraction feature map, and the third semantic extraction feature map. The decoder based on "bidirectional decoding" includes a first decoder and a second decoder, and the first decoder and the second decoder decode independently. The first decoder includes three second dual convolutional layers, and the second decoder includes three third dual convolutional layers. The second dual convolutional layer and the third dual convolutional layer are both composed of two 3×3×3 convolutional layers. The two decoders decode in opposite directions.

[0113] Among them, the first decoder and the second decoder based on the "bidirectional decoding" strategy can decode the output feature map of the encoder from low-level semantics to high-level semantics and from high-level semantics to low-level semantics at the same time, and can extract richer and more detailed fault features. The second double convolution layer and the third double convolution layer both use two 3×3×3 convolution layers to extract features, and the number of channels is 16. There is no upsampling operation between the two feature extraction operations, and features are extracted at the same image resolution. The number of channels of the output feature map is 16, and the size is 32×32×32.

[0114] Optionally, the output layer includes a first upsampling module, a second upsampling module and a fifth convolutional layer, and obtaining the fault detection result specifically includes:

[0115] Inputting the third semantic extraction feature map, the first decoding feature map and the second decoding feature map into the output layer for splicing to obtain a fault splicing feature map;

[0116] Processing the fault splicing feature map using the first upsampling module and the second upsampling module to restore the fault splicing feature map to an initial resolution, thereby obtaining a fault upsampling feature map;

[0117] The upsampled feature map of the fault is input into the fifth convolutional layer to perform fault detection to obtain a fault detection result.

[0118] Among them, the output layer first concatenates the three feature maps obtained from the encoder and decoder branches to obtain the fault splicing feature map, and then uses two upsampling + 3×3×3 convolutions to restore the fault splicing feature map to the original resolution. Finally, a 1×1×1 convolution module is used to output the fault detection result. The number of channels of the spliced ​​feature map is 48 and the size is 32×32×32. After the first upsampling, the number of channels of the feature map is reduced to 16, and the size is restored to 64×64×64. After the second upsampling, the number of channels is still 16, and the size is restored to 128×128×128. Finally, a 1×1×1 convolution layer is used to reduce the channel dimension to 1 and output the fault detection result.

[0119] Among them, the output layer fuses the three feature maps and outputs the fault detection results after two upsamplings. The fusion of multiple features can enable the network to learn richer fault information.

[0120] Optionally, the nearest neighbor interpolation is used for upsampling. This method does not contain learnable parameters, is simple to calculate, and can effectively reduce computational overhead.

[0121] The existing earthquake fault detection method and the earthquake fault detection method based on the N-type network of the present application are verified by using two real data, namely the F3 earthquake data of the North Sea of ​​the Netherlands and the Kerry3D data of New Zealand.

[0122] Verification Example 1

[0123] As shown in Figures 4(a), 4(b) and 4(c), Figure 4(a) is a three-dimensional view of the F3 earthquake data volume in the North Sea of ​​the Netherlands, Figure 4(b) is the fault detection result obtained by Figure 4(a) based on the N-type network of the present invention, and Figure 4(c) is the fault detection result obtained by Figure 4(a) based on the existing earthquake fault detection method. As shown in the black boxes in Figures 4(b) and 4(c), the fault results predicted by the present invention are richer. The fault continuity is greatly improved and the fault structure is more complete.

[0124] Verification Example 2

[0125] As shown in Figures 5(a), 5(b) and 5(c), Figure 5(a) is the Kerry3D data of New Zealand, Figure 5(b) is the fault detection result of Figure 5(a) based on the N-type network of the present invention, and Figure 5(c) is the fault detection result of Figure 5(a) based on the existing earthquake fault detection method. As shown by the black arrows in Figures 5(b) and 5(c), it can be seen that the prediction results of the N-type network have better fault continuity, while the existing network is more likely to predict broken fault structures.

[0126] Verification Example 3

[0127] As shown in Figures 6(a), 6(b) and 6(c), Figure 6(a) is the Kerry3D data of New Zealand, Figure 6(b) is the fault detection result of Figure 6(a) based on the N-type network of the present invention, and Figure 6(c) is the fault detection result of Figure 6(a) based on the existing earthquake fault detection method. As shown in the black boxes in Figures 6(b) and 6(c), the faults in the boxes predicted by the existing network have poor continuity and incomplete structure, while the prediction results of the N-type network proposed in the present invention are more accurate, and a clear and continuous fault structure can be seen.

[0128] Compared with the existing network, in terms of GPU memory consumption and time consumption of the network, during the training phase, the N-type network consumes the same GPU memory, and the amount of training data input at one time is about 4 times that of the existing network, and the training time is only half of the existing network; in the prediction phase, using the same GPU memory, the size of data predicted at one time by the N-type network is about 6 times that of the existing network. The resolution of the Dutch North Sea F3 earthquake data used in the test is 512*384*128. When tested on a 2080Ti graphics card, the N-type network can input the entire data body at one time, and the network prediction takes 0.33s. However, due to the limitation of GPU memory, the existing network can only input a 128*128*128 data body at a time, and it takes 4.09s to predict the complete Dutch North Sea F3 earthquake data body.

[0129] Based on the above method, an embodiment of the present invention further provides an N-type network-based earthquake fault detection system corresponding to the above method, including:

[0130] An input layer module, used for acquiring an initial earthquake fault image, preprocessing the initial earthquake fault image to obtain a preprocessed image, and downsampling the preprocessed image to obtain a downsampled fault feature map;

[0131] An encoder, configured to perform semantic extraction on the downsampled fault feature map by utilizing a cascade structure formed by stacking a plurality of first double convolutional layers built into the encoder, and obtain a plurality of semantic extraction feature maps corresponding to outputs of the plurality of first double convolutional layers;

[0132] A first decoder is used to receive the downsampled fault feature map and the multiple semantic extraction feature maps input by the encoder, and decode the downsampled fault feature map and the multiple semantic extraction feature maps according to the input order to obtain a first decoded feature map;

[0133] A second decoder is used to receive the downsampled fault feature map and the multiple semantic extraction feature maps input by the encoder, and decode the downsampled fault feature map and the multiple semantic extraction feature maps in a decoding order opposite to that of the first decoder to obtain a second decoded feature map;

[0134] The output layer module is used to input the first decoding feature map, the second decoding feature map and the semantic extraction feature map output by the last first double convolutional layer in the cascade structure into the output layer for splicing to obtain a fault splicing feature map, adjust the fault splicing feature map to an initial resolution, and perform fault detection on the adjusted fault splicing feature map to obtain a fault detection result; wherein the initial resolution is the resolution of the initial seismic fault image.

[0135] The earthquake fault detection system of the embodiment of the present invention can execute the earthquake fault detection method provided by the embodiment of the present invention, and the implementation principle is similar. The actions performed by each module and unit in the earthquake fault detection system in each embodiment of the present invention correspond to the steps in the earthquake fault detection method in each embodiment of the present invention. For the detailed functional description of each module of the earthquake fault detection system, please refer to the description in the corresponding earthquake fault detection method shown in the previous text, which will not be repeated here.

[0136] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting earthquake faults based on an N-type network, characterized in that: The N-type network includes an input layer, an encoder, a first decoder, a second decoder and an output layer; using the N-type network to perform earthquake fault detection specifically includes: Acquire an initial earthquake fault image, preprocess the initial earthquake fault image to obtain a preprocessed image, input the preprocessed image into the input layer for downsampling processing, and obtain a downsampled fault feature map; Inputting the downsampled fault feature map into the encoder, performing semantic extraction on the downsampled fault feature map using a cascade structure formed by stacking multiple first double convolutional layers built into the encoder, and obtaining multiple semantic extraction feature maps corresponding to the outputs of the multiple first double convolutional layers; Inputting the downsampled fault feature map and the plurality of semantic extraction feature maps into the first decoder and the second decoder respectively, and decoding the downsampled fault feature map and the plurality of semantic extraction feature maps in reverse order by using the first decoder and the second decoder to obtain a first decoded feature map and a second decoded feature map respectively; The first decoding feature map, the second decoding feature map and the semantic extraction feature map output by the last first double convolutional layer in the cascade structure are input into the output layer for splicing to obtain a fault splicing feature map, the fault splicing feature map is adjusted to an initial resolution, and fault detection is performed on the adjusted fault splicing feature map to obtain a fault detection result; wherein the initial resolution is the resolution of the initial seismic tomography image.

2. The earthquake fault detection method based on N-type network according to claim 1, characterized in that: Also includes: Constructing an initial network, training the initial network, and obtaining an N-type network, wherein obtaining the N-type network specifically includes: Acquire original seismic images, and generate a corresponding fault label for each original seismic image to obtain an original seismic image set, and divide the original seismic image set into a training set and a verification set; Inputting each original seismic image in the training set into the initial network in sequence to perform fault prediction, obtaining a plurality of first fault prediction results, calculating a first loss value between the plurality of first fault prediction results and the fault labels corresponding to the training set, and updating the network parameters of the initial network according to the first loss value; The initial network after updating the network parameters is used to repeatedly traverse the training set for iterative training, and one cycle is defined as one time when the initial network traverses the entire training set; When each cycle ends, the initial network after updating the network parameters using the verification set as input is used to perform fault prediction to obtain a second fault prediction result, and a second loss value between the second fault prediction result and the fault label corresponding to the verification set is calculated; Set a patience period. At the end of each cycle, record the loss difference between the second loss value of this cycle and the second loss value of the previous cycle. If the loss difference obtained after multiple consecutive cycles reaches the stopping condition and the sum of multiple consecutive cycles is greater than the patience period, stop training the initial network and select the initial network corresponding to the minimum value of all second loss values ​​as the N-type network.

3. The earthquake fault detection method based on N-type network according to claim 2, characterized in that: The first loss value and the second loss value are calculated using a balanced binary cross entropy loss function; the formula of the balanced binary cross entropy loss function is as follows: Where L represents the first loss value or the second loss value, β represents the ratio of non-fault pixels to all pixels of the original seismic image, N represents all the pixels of the original seismic image, y i Indicates the fault label, p i Represents the probability that pixel i is predicted to be a fault label.

4. The earthquake fault detection method based on N-type network according to claim 1, characterized in that: The input layer includes a first downsampling module, a first convolutional layer, and a second downsampling module, and inputting the preprocessed image into the input layer for downsampling processing specifically includes: The preprocessed image is processed by the first downsampling module, and the output of the first downsampling module is input into the second downsampling module after being processed by the first convolution layer. The second downsampling module performs feature extraction processing and outputs a downsampled fault feature map.

5. The earthquake fault detection method based on N-type network according to claim 1, characterized in that: The encoder includes a cascade structure formed by stacking three first double convolutional layers, and the multiple semantic extraction feature maps corresponding to the outputs of the multiple first double convolutional layers specifically include: Inputting the downsampled fault feature map into a first double convolutional layer to obtain a first semantic extraction feature map; Inputting the first semantic extraction feature map into the second first double convolutional layer to obtain a second semantic extraction feature map; Inputting the second semantic extraction feature map into the third first double convolutional layer to obtain a third semantic extraction feature map; Among them, the first semantic extraction feature map, the second semantic extraction feature map and the third semantic extraction feature map are multiple semantic extraction feature maps corresponding to the outputs of multiple first dual convolutional layers.

6. The earthquake fault detection method based on N-type network according to claim 5, characterized in that: The encoder is jump-connected with the first decoder and the second decoder respectively, the first decoder includes three second double convolutional layers, and the second decoder includes three third double convolutional layers, and obtaining the first decoding feature map and the second decoding feature map specifically includes: inputting the down-sampled fault feature map and the first semantic extraction feature map into the first decoder for feature splicing to obtain a first spliced ​​feature map; Inputting the first concatenated feature map into the first second double convolutional layer to obtain a third decoding feature map, and concatenating the third decoding feature map with the second semantic extraction feature map to obtain a second concatenated feature map; Inputting the second concatenated feature map into the second second dual convolutional layer to obtain a fourth decoding feature map, and concatenating the fourth decoding feature map with the third semantic extraction feature map to obtain a third concatenated feature map; Inputting the third concatenated feature map into the third second double convolutional layer to obtain a first decoding feature map; Inputting the third semantic extraction feature map and the second semantic extraction feature map into the second decoder for feature splicing to obtain a fourth splicing feature map; Inputting the fourth concatenated feature map into the first third dual convolutional layer to obtain a fifth decoding feature map, and concatenating the fifth decoding feature map with the first semantic extraction feature map to obtain a fifth concatenated feature map; Inputting the fifth spliced ​​feature map into the second third dual convolutional layer to obtain a sixth decoded feature map, and splicing the sixth decoded feature map with the down-sampled fault feature map to obtain a sixth spliced ​​feature map; The sixth concatenated feature map is input into the third double convolutional layer to obtain a second decoding feature map.

7. The earthquake fault detection method based on N-type network according to claim 6, characterized in that: The output layer includes a first upsampling module, a second upsampling module and a fifth convolutional layer, and the fault detection result obtained specifically includes: Inputting the third semantic extraction feature map, the first decoding feature map and the second decoding feature map into the output layer for splicing to obtain a fault splicing feature map; Processing the fault splicing feature map by using the first upsampling module and the second upsampling module to restore the fault splicing feature map to an initial resolution, thereby obtaining a fault upsampling feature map; The upsampled feature map of the fault is input into the fifth convolutional layer to perform fault detection to obtain a fault detection result.

8. An earthquake fault detection system based on an N-type network, characterized in that: include: An input layer module, used for acquiring an initial earthquake fault image, preprocessing the initial earthquake fault image to obtain a preprocessed image, and downsampling the preprocessed image to obtain a downsampled fault feature map; An encoder, configured to perform semantic extraction on the downsampled fault feature map by utilizing a cascade structure formed by stacking a plurality of first double convolutional layers built into the encoder, and obtain a plurality of semantic extraction feature maps corresponding to outputs of the plurality of first double convolutional layers; A first decoder is used to receive the downsampled fault feature map and the multiple semantic extraction feature maps input by the encoder, and decode the downsampled fault feature map and the multiple semantic extraction feature maps according to the input order to obtain a first decoded feature map; A second decoder is used to receive the downsampled fault feature map and the multiple semantic extraction feature maps input by the encoder, and decode the downsampled fault feature map and the multiple semantic extraction feature maps in a decoding order opposite to that of the first decoder to obtain a second decoded feature map; The output layer module is used to input the first decoding feature map, the second decoding feature map and the semantic extraction feature map output by the last first double convolutional layer in the cascade structure into the output layer for splicing to obtain a fault splicing feature map, adjust the fault splicing feature map to an initial resolution, and perform fault detection on the adjusted fault splicing feature map to obtain a fault detection result; wherein the initial resolution is the resolution of the initial seismic fault image.

9. An electronic device, characterized in that: The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images