CN112130200A - Fault identification method based on grad-CAM attention guidance - Google Patents

Fault identification method based on grad-CAM attention guidance Download PDF

Info

Publication number
CN112130200A
CN112130200A CN202011005843.2A CN202011005843A CN112130200A CN 112130200 A CN112130200 A CN 112130200A CN 202011005843 A CN202011005843 A CN 202011005843A CN 112130200 A CN112130200 A CN 112130200A
Authority
CN
China
Prior art keywords
attention
neural network
fault
cam
grad
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202011005843.2A
Other languages
Chinese (zh)
Other versions
CN112130200B (en
Inventor
姚兴苗
李岱
周成
胡光岷
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202011005843.2A priority Critical patent/CN112130200B/en
Publication of CN112130200A publication Critical patent/CN112130200A/en
Application granted granted Critical
Publication of CN112130200B publication Critical patent/CN112130200B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
    • G01V1/282—Application of seismic models, synthetic seismograms
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V1/00—Seismology; Seismic or acoustic prospecting or detecting
    • G01V1/28—Processing seismic data, e.g. for interpretation or for event detection
    • G01V1/30—Analysis
    • G01V1/306—Analysis for determining physical properties of the subsurface, e.g. impedance, porosity or attenuation profiles
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V2210/00—Details of seismic processing or analysis
    • G01V2210/60—Analysis
    • G01V2210/61—Analysis by combining or comparing a seismic data set with other data
    • G01V2210/616—Data from specific type of measurement
    • G01V2210/6161—Seismic or acoustic, e.g. land or sea measurements
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01V—GEOPHYSICS; GRAVITATIONAL MEASUREMENTS; DETECTING MASSES OR OBJECTS; TAGS
    • G01V2210/00—Details of seismic processing or analysis
    • G01V2210/60—Analysis
    • G01V2210/62—Physical property of subsurface
    • G01V2210/624—Reservoir parameters

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Remote Sensing (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Biomedical Technology (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Acoustics & Sound (AREA)
  • Environmental & Geological Engineering (AREA)
  • Geology (AREA)
  • General Life Sciences & Earth Sciences (AREA)
  • Geophysics (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a fault identification method based on grad-CAM attention guidance, which comprises the following steps: s1, obtaining an attention map of the convolutional neural network through the grad-CAM; s2, adding a cross entropy loss function of the attention diagram and the attention diagram marked by a geoscience expert into the objective function of the convolutional neural network to obtain a new objective function of the convolutional neural network; s3, training the fault recognition model by using the objective function obtained in the step S2. According to the invention, on the basis of a typical deep learning framework, an attention guide mechanism is introduced, the attention of a network to a fault and adjacent pixels of the fault can be effectively increased, effective guide for fault classification judgment of a neural network can be realized, the fracture condition of a fault identification result can be effectively improved, and the identification result with better continuity is obtained.

Description

Fault identification method based on grad-CAM attention guidance
Technical Field
The invention belongs to the technical field of seismic data identification, and particularly relates to a fault identification method based on grad-CAM attention guidance.
Background
Seismic data interpretation is an important step in the work of oil and gas exploration, while the identification of faults is an important component of seismic data interpretation. The fault is a structure with obvious relative displacement of the rock blocks on two sides of the fracture surface when the crust is fractured under stress. Faults break the continuity of the formation. The properties, the breaking and tightening degrees of the fault, the contact relationship between lithological combinations on two sides of the fault surface and the like have close relations to oil and gas migration, aggregation and damage. The same fault, with different roles in deep and shallow parts; in the course of historical development, it is also possible to play two opposite roles of blocking or destroying in different periods. The fault plays an important control role in movement and accumulation of oil and gas, so the fault identification method has high practical value.
Fault identification is one of the most important tasks in seismic data interpretation, as faults play an important role in controlling the movement and accumulation of hydrocarbons. Because geoscience data is very large, methods for automatically interpreting faults have been studied in the industry. Faults are one of seismic attributes, and common attributes include coherence, variance, chaos, and the like. Relatively speaking, the method has high use and popularization rate and good identification effect, and the stable and mature algorithm is the attribute of the coherent body, which mainly uses the amplitude discontinuity between adjacent tracks to highlight the fault. The idea of coherent body implementation is to estimate the waveform similarity in the inline and crossline directions. Generally speaking, the coherence characteristics of coherent bodies will exhibit differences in coherence between the faulted seismic trace waveforms, especially where faults running parallel to the earth are more pronounced in coherence properties.
At present, the fault interpretation methods mainly include: (1) fault identification based on a seismic attribute method; (2) fault identification based on image analysis; (3) fault identification based on deep learning. The effectiveness of fault identification is often empirically evaluated by geological experts. The fault on the seismic section is observed and can be divided into a large fault, a medium fault and a small fault according to the size of the fault. Large faults are mostly fracture zones formed by a group of faults with similar properties, and the fracture zones control the construction direction of main faults. Generally, the fracture has a certain fracture thickness, the fault spreading direction is not greatly changed, and the adjacent fault zone is often accompanied by strong anticline and syncline of folds and fracture branches. The plane of the interrupted layer extends generally farther and runs generally parallel or oblique to the direction of the structure. The small fault is mostly a fault crack, the fracture direction is uncertain, and the plane transversely extends not far. Besides the observation of whether the fault distribution and the form accord with the geological rule, the fault identification effect is judged, and the detailed part of the fault, such as whether the fault edge has a large amount of burrs, the noise resistance of the slice background, the continuity in the fault extending direction and the like, is also considered.
Deep learning has been used successfully in computer vision, natural language processing, and other fields. The development of the deep learning technology is combined with the characteristic of huge volume of seismic data, so that a new method is provided for fault identification. An advantage of the deep learning approach is that the neural network model has the ability to generate higher order functions to fit the model. The neural network can automatically extract fault features under supervised training, the trouble of manually selecting the fault features is avoided, and the complexity of the whole fault identification process is greatly reduced. On the other hand, a good deep learning model should have good generalization capability, when the neural network model learns the fault morphological characteristics and fault distribution rules behind a large amount of seismic data, good fault recognition effect can be given even if the seismic data of other work areas are input, and the deep neural network model has excellent effect in the seismic interpretation neighborhood and has infinite potential. However, a large number of parameters exist in the end-to-end 'black box' model of deep learning, and people can hardly understand what influence the hidden layer, the neuron, the activation function and other key parts in the model can have on the final learning result. Current methods of identifying faults using deep learning may produce significant fractures when predicting partial fault samples. The key pixels of the neural network for fault classification may have some deviation, which is not in accordance with the objective knowledge of human beings.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provide a fault identification method based on grad-CAM attention guidance, which introduces an attention guidance mechanism on the basis of a typical deep learning framework, can effectively increase the attention of a network to a fault and adjacent pixels thereof, can realize effective guidance for carrying out fault classification judgment on a neural network, can effectively improve the fracture condition of a fault identification result and obtains a better continuous identification result.
The purpose of the invention is realized by the following technical scheme: a fault identification method based on grad-CAM attention guidance comprises the following steps:
s1, inputting the seismic data marked with the faults as labels and the seismic data not marked with the faults as training sets into a convolutional neural network for training;
s2, obtaining an attention map of the convolutional neural network through the grad-CAM;
s3, adding a cross entropy loss function of the attention diagram and the attention diagram marked by a geoscience expert into the objective function of the convolutional neural network to obtain a new objective function of the convolutional neural network;
s4, training the fault recognition model by using the objective function obtained in the step S3.
Furthermore, the convolutional neural network sequentially comprises two convolutional layers, a pooling layer, three convolutional layers and two full-connection layers.
Furthermore, the attention map is obtained by assuming that the k-th characteristic map output by the last convolutional layer is Ak，
Figure BDA0002695862830000021
Showing a characteristic diagram AkPoint (i, j) above; defining the weight of the kth feature map to the class c as
Figure BDA0002695862830000022
Figure BDA0002695862830000023
Wherein Z represents the number of pixels in the kth feature map; scA classification score representing class c;
the attention map is calculated by:
Figure BDA0002695862830000031
n represents the total number of signatures.
Further, the cross entropy loss function of the attention map and the geo-expert labeled attention map is:
Figure BDA0002695862830000032
wherein q isiFor indicating whether or not to use the attention mechanism for point (i, j), if so, qi1, otherwise qi0; n represents the total number of points to be classified,
Figure BDA0002695862830000033
an attention map representing points (i, j) of a geo expert mark,
Figure BDA0002695862830000034
an attention map representing the (i, j) points generated by the grad-CAM;
the new objective function is obtained as:
Figure BDA0002695862830000035
wherein,
Figure BDA0002695862830000036
the original objective function of the convolutional neural network is beta, and the beta is a hyper-parameter defined by a user.
The invention has the beneficial effects that: according to the invention, on the basis of a typical deep learning framework, an attention guide mechanism is introduced, the attention of a network to a fault and adjacent pixels of the fault can be effectively increased, effective guide for fault classification judgment of a neural network can be realized, the fracture condition of a fault identification result can be effectively improved, and the identification result with better continuity is obtained. The fault identification method based on the grad-CAM attention guidance provided by the invention introduces the important experience of geological experts on fault identification, realizes the explanation of the convolutional neural network identification fault, effectively guides the attention of the neural network to a key area, finally obtains a fault identification result with better continuity, and lays a foundation for subsequent work.
Drawings
FIG. 1 is a flow chart of a method of fault identification based on grad-CAM attention-directed according to the present invention;
FIG. 2 is a schematic diagram of the structure of a convolutional neural network of the present invention;
FIG. 3 is a 401 x 101 two-dimensional seismic data image;
FIG. 4 is a graph of the results of two-dimensional seismic data identification 401 x 101 using a conventional convolutional network;
FIG. 5 is a seismic recognition image and an attention map thereof;
fig. 6 is a diagram of the result of two-dimensional image recognition using the method of the present invention when β is 10 and β is 100, respectively;
figure 7 is a graph of the results of identifying three-dimensional seismic data containing faults 401 x 101 using the method of the invention.
Detailed Description
Prior art relating to the invention
(1) Convolutional Neural Network (CNN)
A convolutional neural network is a neural network that is used to process data having a grid-like structure. Such as time series data (which may be considered as a one-dimensional grid formed by regular sampling on a time axis, such as text speech) and image data (which may be considered as a two-dimensional grid of pixels). Convolutional networks are excellent in many application fields such as computer vision and natural language processing. Convolutional neural networks use the mathematical operation of convolution (convolution). Convolution is a special linear operation. The convolution network refers to a neural network in which at least one layer in the network uses convolution operation to replace general matrix multiplication operation. Typically, a convolutional network also contains a pooling layer.
In the two-dimensional image data, the convolution operation specifically includes the following steps:
and starting sliding on the upper left corner of the image by utilizing the convolution kernel, multiplying the gray value of the pixel on the image by the corresponding numerical value on the convolution kernel, then adding all multiplied values to obtain the gray value of the pixel on the image corresponding to the convolution result, and obtaining the final result after sliding all pixels.
The image pooling operation comprises the following specific steps:
similar to the convolution operation, the sliding is started from the upper left corner of the image, within a window with a specified size, a value is obtained according to different types of pooling (such as maximum value, average value) and is used as a gray value at a corresponding position of the image of the pooling result, and the final result is obtained after all pixels are processed by the sliding window.
(2) Class activation mapping CAM
The class activation mapping CAM is able to locate areas in the image where the convolutional neural network plays an important role in its classification. The class activation map generates a thermodynamic map that is superimposed on the original image, and it can be seen that the parts that play an important role in image classification. The CAM algorithm mainly uses the Global Average Pooling (GAP) layer proposed in NIN. In convolutional neural networks, convolutional layers of pooling are always followed by one or n fully-connected layers, and finally passed through a classifier and sorted with softmax. The convolutional neural network is characterized in that parameters of a full connection layer are excessive, so that the model per se becomes very large. And the GAP layer performs mean pooling on all the feature maps of the last convolution layer, each feature map forms a feature value, and the feature values form a final feature vector, so that the parameters of the network are greatly reduced, and the actual class meaning of each feature map can be directly given. The CAM is realized on the basis of classifying GAP, the output of GAP layer is connected with a full-connection layer to realize classification, and the weight of each full-connection classifier of a certain class is multiplied by the corresponding original feature graph and added to obtain the CAM result. Intuitively, the feature map with a larger average feature value has a larger effect on obtaining the corresponding classification result.
Let a convolutional neural network have n characteristic maps in the last layer, and mark A1,A2,…AnIn the classification layer, one neuron has n weights, each neuron corresponds to one class, and the weight of the ith neuron is set as w1 i,w2 i,…wn iCAM of class c (denoted as M)c) Comprises the following steps:
Figure BDA0002695862830000041
analysis of why the computed CAM can then derive the category-related regions. The above symbols are used along with the notation fi(x, y) represents the pixel value of the ith feature map at position (x, y), then the GAP output of the feature map is:
Figure BDA0002695862830000051
where Z represents the number of feature map pixels. Class c classification score:
Figure BDA0002695862830000052
it can be seen that the score for class c is equal to the sum of all the pixels of its CAM, from which it can be seen that CAM map McThe larger the middle pixel value, the larger the influence on the result of classification.
(3)grad-CAM
The CAM generated attention tries to explain the neural network well, but the CAM requires the structure of the original model to be modified, and the existing model needs to be retrained when the CAM is used, so that the use scene of the model is greatly limited. The limitations of the CAM algorithm are very large if the model is already online or trained at a very high cost. Thus, grad-CAM has been proposed to solve this problem.
The basic idea of the grad-CAM is consistent with that of the CAM, and each piece of the card is required to be obtainedAnd finally, obtaining the weighted sum of the weights corresponding to the classification by the characteristic graph. The main difference between the grad-CAM and the CAM is that the weight w is determinedi cThe process of (1). CAM gets weights by replacing the fully-connected layer with the GAP layer, retraining, and grad-CAM explores a path to compute weights with a global average of gradients. In fact, the grad-CAM is equivalent to the weights computed by the CAM, through rigorous mathematical derivations. Defining the weight of the kth characteristic graph in the grad-CAM to the class c as
Figure BDA0002695862830000053
Can be calculated by the following formula:
Figure BDA0002695862830000054
after the weight corresponding to each feature map is obtained, the attention map can be calculated by the formula (1).
Grad-CAM is a generalization of CAM, as briefly demonstrated below.
Suppose that the Kth feature map of the final convolutional layer output is Ak,Aij kThe (i, j) position of the feature map is shown. CAM requires calculation of GAP values:
Figure BDA0002695862830000061
the CAM calculates the final classification score from the following equation:
Figure BDA0002695862830000062
the gradient of the classification score to the feature map is found:
Figure BDA0002695862830000063
(5) formula (I) isij kPartial derivatives of (A) having
Figure BDA0002695862830000064
Carry over (7) to get:
Figure BDA0002695862830000065
the composite material is prepared from (6),
Figure BDA0002695862830000066
therefore, there are:
Figure BDA0002695862830000067
(9) both sides sum all pixels:
Figure BDA0002695862830000068
Figure BDA0002695862830000069
Figure BDA00026958628300000610
the invention provides a convolutional neural network fault identification method based on grad-CAM attention guidance. The method adopts a grad-CAM method to deeply deepen a class activation diagram of a neural network, and can explain the obtained classification result basis. And combining regular terms marked by geological experts and adding the original loss function of the neural network as a target function of the convolutional neural network, so that the problem of fault identification result fracture caused by the fact that all pixels in geological data play equal roles in the convolutional neural network is solved, and the continuity of the fault identification result is improved. The technical scheme of the invention is further explained by combining the drawings and the specific embodiment.
As shown in FIG. 1, the fault identification method based on the grad-CAM attention guidance of the invention comprises the following steps:
s1, inputting the seismic data marked with the faults as labels and the seismic data not marked with the faults as training sets into a convolutional neural network for training;
s2, obtaining an attention map of the convolutional neural network through the grad-CAM;
the present invention is based on a convolutional neural network with a simpler structure, which comprises two convolutional layers, a pooling layer, three convolutional layers and two fully-connected layers in sequence from the input, as shown in fig. 2. In this embodiment, a convolutional neural network is used to identify 401 × 101 two-dimensional seismic data shown in fig. 3, and the obtained identification result is shown in fig. 4.
For faults, the geological body surfaces staggered on two sides of the faults are generally considered to play a guiding role when the faults are identified manually, however, the identification of a neural network can have deviation. As shown in fig. 5, the attention map derived from the fact that the center point of fig. 5(a) is determined as a fault and the type of the point is identified as a fault is (b), and it can be seen that the position having a large effect on the determination of the point as a fault is located at the upper left of the point, which is inconsistent with our visual understanding, and thus, the model obtained by the neural network training has certain limitations. Therefore, the invention combines the fault label of a geological expert with the Grad-CAM attention map, provides a method for guiding the attention of the neural network, modifies the conventional objective function of the neural network, and adds a cross entropy loss function to make the neural network pay more attention to the area around the fault in the training process.
The attention map is obtained by assuming that the k-th characteristic map output by the last convolutional layer is Ak，
Figure BDA0002695862830000071
Showing a characteristic diagram AkPoint (i, j) above; defining the weight of the kth feature map to the class c as
Figure BDA0002695862830000072
Figure BDA0002695862830000073
Wherein Z represents the number of pixels in the kth feature map; scA classification score representing class c;
the attention map is calculated by:
Figure BDA0002695862830000074
n represents the total number of signatures.
S3, adding a cross entropy loss function of the attention diagram and the attention diagram marked by a geoscience expert into the objective function of the convolutional neural network to obtain a new objective function of the convolutional neural network;
the cross entropy loss function of the attention map and the geo-expert labeled attention map is:
Figure BDA0002695862830000075
wherein q isiFor indicating whether or not to use the attention mechanism for point (i, j), if so, qi1, otherwise qi0; n represents the total number of points to be classified,
Figure BDA0002695862830000081
an attention map representing points (i, j) of a geo expert mark,
Figure BDA0002695862830000082
an attention map representing the (i, j) points generated by the grad-CAM;
the new objective function is obtained as:
Figure BDA0002695862830000083
wherein,
Figure BDA0002695862830000084
beta is a user-defined hyper-parameter, which is the original objective function of the convolutional neural network, and is used for balancing the predicted loss function and the attention guidance term.
S4, training the fault recognition model by using the objective function obtained in the step S3.
In fig. 6(a) and (b), when β is 10 and β is 100, respectively, the recognition results obtained by the method of the present invention are obtained. As can be seen from the figure, the attention map of the fault of fig. 5(a) focuses on the vicinity of the fault classification point of fig. 6. The parameter beta is increased, the obtained effect is better, and the method is consistent with the subjective understanding that the reading area should be focused on the fault identification result.
Figure 7 is a graph of the results of identifying three-dimensional seismic data containing faults 401 x 101 using the method of the invention. In the figure, (a) is a seismic data fault; (b) the fault identification result is obtained when the guiding is not done with attention; (c) the fault identification result when the beta is 10 by using the method of the invention. As can be seen from the figure, when the neural network without attention guidance is used for fault identification, part of fault identification results are broken, and after an attention guidance mechanism is added, the broken condition is improved, and the continuity of the identification results is improved.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims (4)

1. A fault identification method based on grad-CAM attention guidance is characterized by comprising the following steps:
s1, inputting the seismic data marked with the faults as labels and the seismic data not marked with the faults as training sets into a convolutional neural network for training;
s2, obtaining an attention map of the convolutional neural network through the grad-CAM;
s3, adding a cross entropy loss function of the attention diagram and the attention diagram marked by a geoscience expert into the objective function of the convolutional neural network to obtain a new objective function of the convolutional neural network;
s4, training the fault recognition model by using the objective function obtained in the step S3.
2. The method of claim 1, wherein the convolutional neural network comprises two convolutional layers, one pooling layer, three convolutional layers, and two fully-connected layers in sequence.
3. The method according to claim 1, wherein the attention map is obtained by assuming that the kth feature map output by the last convolutional layer is Ak，
Figure FDA0002695862820000011
Showing a characteristic diagram AkPoint (i, j) above; defining the weight of the kth feature map to the class c as
Figure FDA0002695862820000012
Figure FDA0002695862820000013
Wherein Z represents the number of pixels in the kth feature map; scA classification score representing class c;
the attention map is calculated by:
Figure FDA0002695862820000014
n represents the total number of signatures.
4. The method for fault identification based on grad-CAM attention guidance according to claim 1, wherein the cross entropy loss function of the attention map and the geo-expert labeled attention map is as follows:
Figure FDA0002695862820000015
wherein q isiFor indicating whether or not to use the attention mechanism for point (i, j), if so, qi1, otherwise qi0; n represents the total number of points to be classified,
Figure FDA0002695862820000016
an attention map representing points (i, j) of a geo expert mark,
Figure FDA0002695862820000017
an attention map representing the (i, j) points generated by the grad-CAM;
the new objective function is obtained as:
Figure FDA0002695862820000018
wherein,
Figure FDA0002695862820000021
the original objective function of the convolutional neural network is beta, and the beta is a hyper-parameter defined by a user.
CN202011005843.2A 2020-09-23 2020-09-23 Fault identification method based on grad-CAM attention guidance Active CN112130200B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011005843.2A CN112130200B (en) 2020-09-23 2020-09-23 Fault identification method based on grad-CAM attention guidance

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011005843.2A CN112130200B (en) 2020-09-23 2020-09-23 Fault identification method based on grad-CAM attention guidance

Publications (2)

Publication Number Publication Date
CN112130200A true CN112130200A (en) 2020-12-25
CN112130200B CN112130200B (en) 2021-07-20

Family

ID=73842538

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011005843.2A Active CN112130200B (en) 2020-09-23 2020-09-23 Fault identification method based on grad-CAM attention guidance

Country Status (1)

Country Link
CN (1) CN112130200B (en)

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112613574A (en) * 2020-12-30 2021-04-06 清华大学 Training method of image classification model, image classification method and device
CN112700434A (en) * 2021-01-12 2021-04-23 苏州斯玛维科技有限公司 Medical image classification method and classification device thereof
CN112734739A (en) * 2021-01-18 2021-04-30 福州大学 Visual building crack identification method based on attention mechanism and ResNet fusion
CN112749667A (en) * 2021-01-15 2021-05-04 中国科学院宁波材料技术与工程研究所 Deep learning-based nematode classification and identification method
CN113156513A (en) * 2021-04-14 2021-07-23 吉林大学 Convolutional neural network seismic signal denoising method based on attention guidance
CN113298084A (en) * 2021-04-01 2021-08-24 山东师范大学 Feature map extraction method and system for semantic segmentation
CN113688901A (en) * 2021-08-23 2021-11-23 西南石油大学 Reservoir discontinuous boundary identification method based on expansion convolution neural network
CN116660982A (en) * 2023-08-02 2023-08-29 东北石油大学三亚海洋油气研究院 Full waveform inversion method based on attention convolution neural network
CN118015329A (en) * 2023-12-28 2024-05-10 兰州大学 Symbolized execution method for explaining convolutional neural network model classification basis

Citations (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104166163A (en) * 2014-08-27 2014-11-26 电子科技大学 Method for automatically extracting fault curved surface based on three-dimensional large-data-volume seismic data cube
CN107356968A (en) * 2017-08-25 2017-11-17 四川易诚智讯科技有限公司 Three-dimensional level set tomography curved surface extraction method based on crease
CN107621655A (en) * 2017-08-29 2018-01-23 电子科技大学 Three-dimensional data tomography Enhancement Method based on DoS filtering
CN109086773A (en) * 2018-08-29 2018-12-25 电子科技大学 Fault plane recognition methods based on full convolutional neural networks
CN109271898A (en) * 2018-08-31 2019-01-25 电子科技大学 Solution cavity body recognizer based on optimization convolutional neural networks
CN109709603A (en) * 2018-11-23 2019-05-03 中国石油天然气股份有限公司 Seismic horizon identification and method for tracing, system
CN110441820A (en) * 2019-08-21 2019-11-12 中国矿业大学(北京) Architectonic intelligent interpretation method
KR20190128292A (en) * 2018-05-08 2019-11-18 서울대학교산학협력단 Method and System for Early Diagnosis of Glaucoma and Displaying suspicious Area
CN110516740A (en) * 2019-08-28 2019-11-29 电子科技大学 A kind of fault recognizing method based on Unet++ convolutional neural networks
CN110579354A (en) * 2019-10-16 2019-12-17 西安交通大学 Bearing detection method based on convolutional neural network
CN110728183A (en) * 2019-09-09 2020-01-24 天津大学 Human body action recognition method based on attention mechanism neural network
CN110737021A (en) * 2019-11-06 2020-01-31 中国矿业大学(北京) Fault recognition method and model training method and device thereof
EP3627391A1 (en) * 2018-09-24 2020-03-25 Siemens Aktiengesellschaft Deep neural net for localising objects in images, methods for preparing such a neural net and for localising objects in images, corresponding computer program product, and corresponding computer-readable medium
CN111046939A (en) * 2019-12-06 2020-04-21 中国人民解放军战略支援部队信息工程大学 CNN (CNN) class activation graph generation method based on attention
CN111046962A (en) * 2019-12-16 2020-04-21 中国人民解放军战略支援部队信息工程大学 Sparse attention-based feature visualization method and system for convolutional neural network model
US20200183031A1 (en) * 2018-12-11 2020-06-11 Exxonmobil Upstream Research Company Automated seismic interpretation-guided inversion
US20200258223A1 (en) * 2018-05-14 2020-08-13 Tempus Labs, Inc. Determining biomarkers from histopathology slide images
CN111562612A (en) * 2020-05-20 2020-08-21 大连理工大学 Deep learning microseismic event identification method and system based on attention mechanism
CN111580161A (en) * 2020-05-21 2020-08-25 长江大学 Earthquake random noise suppression method based on multi-scale convolution self-coding neural network

Patent Citations (19)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104166163A (en) * 2014-08-27 2014-11-26 电子科技大学 Method for automatically extracting fault curved surface based on three-dimensional large-data-volume seismic data cube
CN107356968A (en) * 2017-08-25 2017-11-17 四川易诚智讯科技有限公司 Three-dimensional level set tomography curved surface extraction method based on crease
CN107621655A (en) * 2017-08-29 2018-01-23 电子科技大学 Three-dimensional data tomography Enhancement Method based on DoS filtering
KR20190128292A (en) * 2018-05-08 2019-11-18 서울대학교산학협력단 Method and System for Early Diagnosis of Glaucoma and Displaying suspicious Area
US20200258223A1 (en) * 2018-05-14 2020-08-13 Tempus Labs, Inc. Determining biomarkers from histopathology slide images
CN109086773A (en) * 2018-08-29 2018-12-25 电子科技大学 Fault plane recognition methods based on full convolutional neural networks
CN109271898A (en) * 2018-08-31 2019-01-25 电子科技大学 Solution cavity body recognizer based on optimization convolutional neural networks
EP3627391A1 (en) * 2018-09-24 2020-03-25 Siemens Aktiengesellschaft Deep neural net for localising objects in images, methods for preparing such a neural net and for localising objects in images, corresponding computer program product, and corresponding computer-readable medium
CN109709603A (en) * 2018-11-23 2019-05-03 中国石油天然气股份有限公司 Seismic horizon identification and method for tracing, system
US20200183031A1 (en) * 2018-12-11 2020-06-11 Exxonmobil Upstream Research Company Automated seismic interpretation-guided inversion
CN110441820A (en) * 2019-08-21 2019-11-12 中国矿业大学(北京) Architectonic intelligent interpretation method
CN110516740A (en) * 2019-08-28 2019-11-29 电子科技大学 A kind of fault recognizing method based on Unet++ convolutional neural networks
CN110728183A (en) * 2019-09-09 2020-01-24 天津大学 Human body action recognition method based on attention mechanism neural network
CN110579354A (en) * 2019-10-16 2019-12-17 西安交通大学 Bearing detection method based on convolutional neural network
CN110737021A (en) * 2019-11-06 2020-01-31 中国矿业大学(北京) Fault recognition method and model training method and device thereof
CN111046939A (en) * 2019-12-06 2020-04-21 中国人民解放军战略支援部队信息工程大学 CNN (CNN) class activation graph generation method based on attention
CN111046962A (en) * 2019-12-16 2020-04-21 中国人民解放军战略支援部队信息工程大学 Sparse attention-based feature visualization method and system for convolutional neural network model
CN111562612A (en) * 2020-05-20 2020-08-21 大连理工大学 Deep learning microseismic event identification method and system based on attention mechanism
CN111580161A (en) * 2020-05-21 2020-08-25 长江大学 Earthquake random noise suppression method based on multi-scale convolution self-coding neural network

Non-Patent Citations (5)

* Cited by examiner, † Cited by third party
Title
BING ZHAO 等: ""Visualization of Railway Scene Classification Model via Grad-CAM"", 《2018 IEEE 3RD INTERNATIONAL CONFERENCE ON SIGNAL AND IMAGE PROCESSING（ICSIP）》 *
SHAOWEI ZHENG: ""Fine-Grained Image Classification Algorithm Based on Grad-CAM and B-CNN"", 《COMPUTER SCIENCE AND APPLICATION》 *
赵迪 等: ""基于Grad-CAM的探地雷达公路地下目标检测算法"", 《电子测量技术》 *
车向前 等: ""利用改进组合交叉熵实现煤层气储层地震属性约简"", 《煤田地质与勘探》 *
邓博文 等: ""基于多属性的地质目标体表面重建方法研究"", 《现代计算机》 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112613574A (en) * 2020-12-30 2021-04-06 清华大学 Training method of image classification model, image classification method and device
CN112613574B (en) * 2020-12-30 2022-07-19 清华大学 Training method of image classification model, image classification method and device
CN112700434A (en) * 2021-01-12 2021-04-23 苏州斯玛维科技有限公司 Medical image classification method and classification device thereof
CN112700434B (en) * 2021-01-12 2024-07-19 广州一达健康管理有限公司 Medical image classification method and classification device thereof
CN112749667A (en) * 2021-01-15 2021-05-04 中国科学院宁波材料技术与工程研究所 Deep learning-based nematode classification and identification method
CN112734739A (en) * 2021-01-18 2021-04-30 福州大学 Visual building crack identification method based on attention mechanism and ResNet fusion
CN112734739B (en) * 2021-01-18 2022-07-08 福州大学 Visual building crack identification method based on attention mechanism and ResNet fusion
CN113298084B (en) * 2021-04-01 2023-04-07 山东师范大学 Feature map extraction method and system for semantic segmentation
CN113298084A (en) * 2021-04-01 2021-08-24 山东师范大学 Feature map extraction method and system for semantic segmentation
CN113156513A (en) * 2021-04-14 2021-07-23 吉林大学 Convolutional neural network seismic signal denoising method based on attention guidance
CN113156513B (en) * 2021-04-14 2024-01-30 吉林大学 Convolutional neural network seismic signal denoising method based on attention guidance
CN113688901B (en) * 2021-08-23 2024-03-01 西南石油大学 Reservoir discontinuous boundary line identification method based on expansion convolutional neural network
CN113688901A (en) * 2021-08-23 2021-11-23 西南石油大学 Reservoir discontinuous boundary identification method based on expansion convolution neural network
CN116660982A (en) * 2023-08-02 2023-08-29 东北石油大学三亚海洋油气研究院 Full waveform inversion method based on attention convolution neural network
CN116660982B (en) * 2023-08-02 2023-09-29 东北石油大学三亚海洋油气研究院 Full waveform inversion method based on attention convolution neural network
CN118015329A (en) * 2023-12-28 2024-05-10 兰州大学 Symbolized execution method for explaining convolutional neural network model classification basis

Also Published As

Publication number Publication date
CN112130200B (en) 2021-07-20

Similar Documents

Publication Publication Date Title
CN112130200B (en) Fault identification method based on grad-CAM attention guidance
Liu et al. Seismic facies classification using supervised convolutional neural networks and semisupervised generative adversarial networks
Arras et al. Explaining predictions of non-linear classifiers in NLP
Bu et al. Detection of fabric defects by auto-regressive spectral analysis and support vector data description
CN107545577B (en) Sedimentary facies image segmentation method based on neural network
Bai et al. Hyperspectral image classification based on multibranch attention transformer networks
CN105740823A (en) Dynamic gesture trace recognition method based on depth convolution neural network
CN107689052A (en) Visual target tracking method based on multi-model fusion and structuring depth characteristic
CN109001801B (en) Fault variable-scale identification method based on multiple iteration ant colony algorithm
CN113269228B (en) Method, device and system for training graph network classification model and electronic equipment
CN109271546A (en) The foundation of image retrieval Feature Selection Model, Database and search method
CN104634265A (en) Soft measurement method for thickness of mineral floating foam layer based on multivariate image feature fusion
Dixit et al. Texture feature based satellite image classification scheme using SVM
CN114913378A (en) Image classification interpretable method based on comprehensive class activation mapping
CN106056577A (en) Hybrid cascaded SAR image change detection method based on MDS-SRM
Jafrasteh et al. Generative adversarial networks as a novel approach for tectonic fault and fracture extraction in high resolution satellite and airborne optical images
Maloney et al. The use of probabilistic neural networks to improve solution times for hull-to-emitter correlation problems
Bi et al. 3D relative geologic time estimation with deep learning
Orozco-Del-Castillo et al. Seismic data interpretation using the Hough transform and principal component analysis
Jung et al. A metric to measure contribution of nodes in neural networks
Daribayev et al. Implementation of the solution to the oil displacement problem using machine learning classifiers and neural networks
US20220188647A1 (en) Model learning apparatus, data analysis apparatus, model learning method and program
Amin et al. Feature selection using multivariate adaptive regression splines in telecommunication fraud detection
Singh et al. Choquet fuzzy integral based verification of handwritten signatures
Shi et al. Permeability estimation of rock reservoir based on pca and elman neural networks

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant