CN114997340B - Literature classification method based on graph neural network system - Google Patents
Literature classification method based on graph neural network system Download PDFInfo
- Publication number
- CN114997340B CN114997340B CN202210915495.5A CN202210915495A CN114997340B CN 114997340 B CN114997340 B CN 114997340B CN 202210915495 A CN202210915495 A CN 202210915495A CN 114997340 B CN114997340 B CN 114997340B
- Authority
- CN
- China
- Prior art keywords
- graph
- neural network
- document
- module
- target
- 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.)
- Active
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 80
- 238000000034 method Methods 0.000 title claims abstract description 30
- 238000001914 filtration Methods 0.000 claims abstract description 20
- 238000009499 grossing Methods 0.000 claims abstract description 18
- 238000013145 classification model Methods 0.000 claims abstract description 16
- 239000011159 matrix material Substances 0.000 claims description 23
- 238000005457 optimization Methods 0.000 claims description 17
- 238000010606 normalization Methods 0.000 claims description 6
- 230000001174 ascending effect Effects 0.000 claims description 4
- 238000010276 construction Methods 0.000 claims description 3
- 238000002945 steepest descent method Methods 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 238000012546 transfer Methods 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims 1
- 238000012545 processing Methods 0.000 abstract description 9
- 230000006870 function Effects 0.000 description 24
- 238000013461 design Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 238000010801 machine learning Methods 0.000 description 2
- 241000689227 Cora <basidiomycete fungus> Species 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000000737 periodic effect Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
Images
Classifications
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention relates to a literature classification method based on a graph neural network system, which is characterized in that the graph neural network system based on a graph filtering kernel and a generalized non-convex norm is designed and divided into a first multilayer perceptron module, a linear rectification function module, a second multilayer perceptron module, a target graph neural network module and a normalized exponential function module according to a processing flow, wherein the target graph neural network module is constructed based on the graph filtering kernel and the generalized non-convex norm and comprises a graph filtering kernel term used for extracting useful information of a graph signal, a graph Laplace regularization term used for carrying out global graph smoothing processing on the graph signal and a generalized non-convex norm term used for carrying out local graph smoothing processing on the graph signal, and in application, the target graph neural network module is solved by adopting a prediction check descent and ascent algorithm; therefore, the graph neural network system forms a network to be trained, and the network is trained to obtain a document classification model, so that the classification precision of documents and the working efficiency of practical document application can be effectively improved.
Description
Technical Field
The invention relates to a literature classification method based on a graph neural network system, and belongs to the technical field of machine learning of graph neural network classification.
Background
In recent years, machine learning techniques, such as neural networks, have attracted considerable attention because they have been able to successfully promote the development of fields such as pattern recognition and data mining. Classical neural network methods include convolutional neural networks, periodic neural networks, autoencoders and the like, but most of such neural network methods are used for processing regular data, and the effect of processing image data existing in non-Euclidean spaces is not good enough, and for such reasons, image neural network technology comes along. At present, the graph neural networks are roughly divided into two categories, one is a spectrum-based graph neural network such as a chebyshev model and the like, and the other is a space-based graph neural network such as a graph convolution neural network, a graph attention machine network and the like. However, the existing graph neural network has low classification precision in practical application and low practical application efficiency.
Disclosure of Invention
The invention aims to solve the technical problem of providing a document classification method based on a graph neural network system, which adopts a brand-new network structure design and can effectively improve the precision and efficiency of document classification in application.
The invention adopts the following technical scheme for solving the technical problems: the invention designs a literature classification method based on a graph neural network system, which is characterized in that a literature classification model is obtained through steps A to C based on sample literatures which respectively correspond to preset literature categories and graph signals of matrixes formed by characteristics of preset types respectively, and the literature to be classified is classified by applying the literature classification model according to step i;
a, constructing a target graph neural network module based on a graph filtering kernel and a generalized non-convex norm, wherein the target graph neural network module comprises a graph filtering kernel item used for extracting useful information of a graph signal, a graph Laplace regular item used for carrying out global graph smoothing on the graph signal, and a generalized non-convex norm item used for carrying out local graph smoothing on the graph signal, and then entering a step B;
b, constructing a network to be trained, wherein the input end of the network to be trained is formed by the input end of a first multilayer sensor module, the output end of the first multilayer sensor module is sequentially connected with a linear rectification function module, a second multilayer sensor module, a target graph neural network module and a normalization index function module in series, and the output end of the normalization index function module forms the output end of the network to be trained, and then entering the step C;
c, training a network to be trained according to the incidence relation among the sample documents based on the graph signals respectively corresponding to the sample documents and the document types respectively corresponding to the sample documents, and obtaining a document classification model taking the graph signals corresponding to the documents as input and the document types corresponding to the documents as output;
and i, obtaining a diagram signal formed by preset various types of characteristics corresponding to the document to be classified, and applying a document classification model to obtain the document category corresponding to the document to be classified.
As a preferred technical scheme of the invention: the target graph neural network module comprises a graph filtering kernel item for extracting useful information of graph signals
Graph Laplace regularization term for global graph smoothing of graph signals
And a generalized non-convex norm term for local graph smoothing of graph signals
And the optimization equation corresponding to the target graph neural network module is as follows:
wherein,
，
representing the map signal received by the target map neural network module,
representation map signal
A transfer function is performed that can be learned,
a representation of the parameters that can be learned,
representing target graph neural network modules for received graph signals
The output signal after a number of iterations,
represents the norm of Frobenius,
、
、
respectively, the super-parameters are the parameters,
indicating a balance factor greater than 0 and,
the trace function of the matrix is represented by,
，
a matrix of the laplacian of the graph is represented,
the unit matrix is represented by a matrix of units,
，
a graph adjacency matrix representing the regularization,
a matrix of degrees representing the graph is shown,
in order to map the adjacency matrix of the figure,
，
show each itemThe document corresponds to the first of the association relation graph between the vertexes
The number of the top points is equal to the number of the top points,
the second of which represents the association relationship graph between the corresponding vertexes of each document
The number of the top points is equal to the number of the top points,
the first in the graph representing the association relationship between the corresponding vertexes of each document
The degree of each of the vertices,
the first in the graph representing the association relationship between the corresponding vertexes of each document
The degree of each of the vertices,
and
are respectively
To (1) a
A sum of
The output signal of the first and second switching circuits is,
is broad, non-convex, conformable, closedAnd a lower semi-continuous function of the function,
，
。
as a preferred technical scheme of the invention: the optimization equation corresponding to the target graph neural network module is solved by adopting a predictive check descending and ascending algorithm in the following mode;
first according to
Is a conjugate function of
And matrix variables to be solved
Supremum function, infimum function
Construction of
；
Then, the optimization equation corresponding to the target graph neural network module is updated as follows:
and finally solving the optimization equation corresponding to the updated target graph neural network module.
As a preferred technical scheme of the invention: based on the initialization
、
、
Solving an optimization equation corresponding to the updated graph neural network module according to the following steps;
step 1) updating by adopting a steepest descent method according to the following formula
By a value of (i), i.e. using
Represents;
wherein,
represents the solution process
The number of sub-iterations is,
which represents the first step-size factor,
the gradient operator is represented by a gradient operator,
representation solving
Of a minor iteration
A value of (d); then entering step 2);
step 2) solving by adopting the steepest ascent method according to the following formula
；
Wherein,
is a proximity operator which is a function of the proximity operator,
represents a second step size factor; then entering step 3);
step 3) according to the following formula:
updating variable updates
Then step 4) is entered;
step 4) judgment
Whether or not the number of iterations is not less than a predetermined maximum number of iterations
If yes, ending the iteration; otherwise to
Is updated by adding 1 and returns to step 1).
Compared with the prior art, the document classification method based on the graph neural network system has the following technical effects by adopting the technical scheme:
the invention designs a literature classification method based on a graph neural network system, designs a graph neural network system based on a graph filtering kernel and a generalized non-convex norm, and divides the graph neural network system into a first multilayer perceptron module, a linear rectification function module, a second multilayer perceptron module, a target graph neural network module and a normalized exponential function module according to a processing flow, wherein the target graph neural network module is constructed based on the graph filtering kernel and the generalized non-convex norm and comprises a graph filtering kernel term used for extracting useful information of a graph signal, a graph Laplace regularization term used for carrying out global graph smoothing processing on the graph signal and a generalized non-convex norm term used for carrying out local graph smoothing processing on the graph signal, and in application, the target graph neural network module is solved by adopting a prediction check descending and ascending algorithm; therefore, the graph neural network system forms a network to be trained, and the network is trained to obtain a document classification model, so that the classification precision of documents and the working efficiency of practical document application can be effectively improved.
Drawings
FIG. 1 is a system block diagram of a neural network system of the present design;
FIG. 2 is a flow chart of the solving of the neural network module of the target graph in the design of the present invention;
FIG. 3 is a graph comparing the effect of the document classification method based on the graph neural network system compared with other classical graph neural networks for classifying graph nodes.
Detailed Description
The following description will explain embodiments of the present invention in further detail with reference to the accompanying drawings.
The invention designs a literature classification method based on a graph neural network system, and in practical application, a literature classification model is obtained through steps A to C based on sample literatures which respectively correspond to preset literature categories and graph signals of matrixes formed by characteristics of various types, which are respectively corresponding to preset sample literatures.
A, constructing a target graph neural network module based on a graph filtering kernel and a generalized non-convex norm, wherein the target graph neural network module comprises a graph filtering kernel item used for extracting useful information of a graph signal, a graph Laplace regular item used for carrying out global graph smoothing on the graph signal, and a generalized non-convex norm item used for carrying out local graph smoothing on the graph signal, and then entering a step B;
step B, as shown in fig. 1, constructing a network to be trained, wherein the input end of the network to be trained is formed by the input end of a first multilayer perceptron (MLP) module, the output end of the first multilayer perceptron (MLP) module is sequentially connected in series with a linear rectification function module (ReLU), a second multilayer perceptron (MLP) module, a target graph neural network module and a normalization index function module (Softmax), the output end of the normalization index function module (Softmax) forms the output end of the network to be trained, and then the step C is carried out;
c, training a network to be trained according to the incidence relation among the sample documents based on the graph signals respectively corresponding to the sample documents and the document types respectively corresponding to the sample documents, and obtaining a document classification model taking the graph signals corresponding to the documents as input and the document types corresponding to the documents as output;
and (3) after the document classification model is obtained through the steps A to C, further applying the document classification model, and executing the step i to classify the document to be classified.
And i, acquiring a graph signal formed by preset various types of characteristics corresponding to the document to be classified, and applying a document classification model to acquire the document category corresponding to the document to be classified.
In practical application, the optimization equation corresponding to the designed target graph neural network module is shown as formula (1):
wherein,
and
for the graph filtering kernel, the invention will
And
are all provided with
，
Is about
Is set as the generalized non-convex function in the present invention
Then, the optimization model corresponding to equation (1) becomes the following optimization equation:
wherein, the design target graph neural network module comprises a graph filtering kernel term for extracting useful information of the graph signal
Graph Laplace regularization term for global graph smoothing of graph signals
And generalized non-convex norm term for local graph smoothing of graph signals
。
Wherein,
，
representing the map signal received by the target map neural network module,
show to the picture letterNumber (C)
A transfer function is performed that can be learned,
a representation of the parameters that can be learned,
representing target graph neural network modules for received graph signals
The output signal after a number of iterations,
represents the norm of Frobenius,
、
、
respectively, the parameters are the super-parameters,
indicating a balance factor greater than 0 and,
the trace function of the matrix is represented by,
，
a matrix of laplacian of the graph is represented,
the matrix of the unit is expressed by,
，
a graph adjacency matrix representing the regularization,
a matrix of degrees representing the graph is shown,
in order to be a graph of the adjacency matrix,
，
the second of which represents the association relationship graph between the corresponding vertexes of each document
The number of the top points is equal to the number of the top points,
the second of which represents the association relationship graph between the corresponding vertexes of each document
The number of the vertexes is equal to that of the vertex,
the first in the graph representing the association relationship between the corresponding vertexes of each document
The degree of each of the vertices,
the first one in the graph representing the association relationship between the corresponding vertexes of each document
The degree of each of the vertices,
and
are respectively
To (1) a
A sum of
The output signal of the first and second switching circuits is,
is a generalized, non-convex, fitting, closed and lower semicontinuous function,
，
。
and in practical application, the optimization equation corresponding to the neural network module of the target graph is obtained because
Difficult to solve directly, the design of the invention converts it into
The conjugate function of the target graph neural network module is solved, namely, a prediction check descending and ascending algorithm is adopted, and the optimization equation corresponding to the target graph neural network module is solved in the following mode.
First according to
Is a conjugate function of
And the matrix variables to be solved for
Supremum function
Construction of
。
Then, the optimization equation corresponding to the neural network module of the target graph is updated as follows:
finally, as shown in FIG. 2, based on the initial
、
、
And solving the optimization equation corresponding to the updated graph neural network module according to the following steps 1) to 4).
Step 1) updating by adopting a steepest descent method according to the following formula
By a value of (i), i.e. using
Representing;
wherein,
represents the solution process
The number of sub-iterations is,
which represents the first step-size factor,
the gradient operator is represented by a gradient operator,
representation solution
Of a minor iteration
A value of (d); then step 2) is entered.
Step 2) solving by adopting the steepest ascent method according to the following formula
；
Wherein,
is a proximity operator, which is a function of the proximity operator,
represents a second step size factor; then step 3) is entered.
Step 3) according to the following formula:
updating variable updates
Then step 4) is entered.
Step 4) judgment
Whether or not it is not less than a preset maximum number of iterations
If yes, ending the iteration; otherwise to
Is updated by adding 1 and returns to step 1).
The technical scheme includes that a graph neural network system based on a graph filtering kernel and a generalized non-convex norm is designed, and the graph neural network system based on the graph filtering kernel and the generalized non-convex norm is divided into a first multilayer perceptron module, a linear rectification function module (ReLU), a second multilayer perceptron module, a target graph neural network module and a normalized exponential function module (Softmax) according to a processing flow, wherein the target graph neural network module is constructed based on the graph filtering kernel and the generalized non-convex norm and comprises a graph filtering regular term used for extracting useful information of a graph signal, a graph Laplace term used for carrying out global graph smoothing on the graph signal and a generalized non-convex norm term used for carrying out local graph smoothing on the graph signal, and in application, the target graph neural network module is solved by adopting a prediction check descent and ascent algorithm; in this way, the graph neural network system forms a network to be trained, and trains the network to obtain a document classification model, and in practical application, as shown in fig. 3, the OurGNN is a new graph neural network system proposed by the present invention, and compared with graph node classification performed by other classical graph neural networks GCN (graph convolution neural network), GAT (graph attention neural network), SGC (simplified graph convolution network), APPNP (accelerated neural prediction personalized propagation), graphSAGE (graph sample aggregation neural network), the classification method provided by the present invention can effectively improve the classification accuracy of documents and the work efficiency of practical document application from the experimental effects of three document cited graph data, namely Cora, citeseer and Pubmed. Meanwhile, the application of the design method is expanded to the node classification problem of a document author relationship diagram (comprising a relationship diagram between two document authors, namely CS and Physics) and a purchase relationship diagram (comprising two purchase relationship diagrams, namely Computers and Photo), and the simulation effect of two groups of experiments expanded by the design method is further verified by the invention, so that the effectiveness of the score classification method in practical application is further verified.
The embodiments of the present invention have been described in detail with reference to the drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge of those skilled in the art without departing from the gist of the present invention.
Claims (4)
1. A literature classification method based on a graph neural network system is characterized in that: obtaining a document classification model through steps A to C based on sample documents which respectively correspond to preset document types and graph signals of matrixes formed by characteristics of the sample documents which respectively correspond to preset types of the documents, and classifying the documents to be classified by applying the document classification model according to the step i;
a, constructing a target graph neural network module based on a graph filtering kernel and a generalized non-convex norm, wherein the target graph neural network module comprises a graph filtering kernel item used for extracting useful information of a graph signal, a graph Laplace regular item used for carrying out global graph smoothing on the graph signal, and a generalized non-convex norm item used for carrying out local graph smoothing on the graph signal, and then entering a step B;
b, constructing a network to be trained, wherein the input end of the network to be trained is formed by the input end of a first multilayer sensor module, the output end of the first multilayer sensor module is sequentially connected with a linear rectification function module, a second multilayer sensor module, a target graph neural network module and a normalization index function module in series, and the output end of the normalization index function module forms the output end of the network to be trained, and then entering the step C;
c, training a network to be trained according to the incidence relation among the sample documents based on the graph signals respectively corresponding to the sample documents and the document types respectively corresponding to the sample documents, and obtaining a document classification model taking the graph signals corresponding to the documents as input and the document types corresponding to the documents as output;
and i, acquiring a graph signal formed by preset various types of characteristics corresponding to the document to be classified, and applying a document classification model to acquire the document category corresponding to the document to be classified.
2. The method of claim 1, wherein the method comprises: the target graph neural network module comprises a graph filtering kernel item for extracting useful information of a graph signal
Graph Laplace regularization term for global graph smoothing of graph signals
And a generalized non-convex norm term for local graph smoothing of graph signals
And the optimization equation corresponding to the target graph neural network module is as follows:
wherein,
，
representing the map signal received by the target map neural network module,
representation map signal
A transfer function is performed that can be learned,
the representation of the learnable parameter is,
representing target graph neural network modules against received graph signals
The output signal after a number of iterations,
represents the Frobenius norm,
、
、
respectively, the super-parameters are the parameters,
indicating a balance factor greater than 0 and,
the trace function of the matrix is represented by,
，
a matrix of the laplacian of the graph is represented,
the unit matrix is represented by a matrix of units,
，
a graph adjacency matrix representing the regularization,
a matrix of degrees representing the graph is shown,
in order to map the adjacency matrix of the figure,
，
the second of which represents the association relationship graph between the corresponding vertexes of each document
The number of the top points is equal to the number of the top points,
the first to the second of the association graph
The number of the top points is equal to the number of the top points,
the first in the graph representing the association relationship between the corresponding vertexes of each document
The degree of each of the vertices is,
the first in the graph representing the association relationship between the corresponding vertexes of each document
The degree of each of the vertices,
and
are respectively
To (1) a
An
The output signal of the first and second switching circuits is output,
is a generalized, non-convex, fitting, closed and lower semicontinuous function,
，
。
3. the method of claim 2, wherein the method comprises: the optimization equation corresponding to the target graph neural network module is solved by adopting a predictive check descending and ascending algorithm in the following mode;
first according to
Conjugate function of (2)
And matrix variables to be solved
Supremum function
Construction of
；
Then, the optimization equation corresponding to the neural network module of the target graph is updated as follows:
and finally solving the optimization equation corresponding to the updated target graph neural network module.
4. The method of claim 3, wherein the method comprises: based on the initial
、
、
Aiming at the neural network model of the updated graph according to the following stepsSolving an optimization equation corresponding to the block;
step 1) updating by adopting a steepest descent method according to the following formula
By a value of (i), i.e. using
Represents;
wherein,
represents the solution process
The number of sub-iterations is,
a first step-size factor is represented by,
a gradient operator is represented by a gradient operator, which,
representation solving
Of a minor iteration
A value of (d); then entering step 2);
step 2) solving by adopting the steepest ascent method according to the following formula
；
Wherein,
is a proximity operator, which is a function of the proximity operator,
represents a second step size factor; then entering step 3);
step 3) according to the following formula:
updating variable updates
Then go to step 4);
step 4) judgment
Whether or not it is not less than a preset maximum number of iterations
If yes, ending the iteration; otherwise to
Is updated by adding 1 and returns to step 1).
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210915495.5A CN114997340B (en) | 2022-08-01 | 2022-08-01 | Literature classification method based on graph neural network system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210915495.5A CN114997340B (en) | 2022-08-01 | 2022-08-01 | Literature classification method based on graph neural network system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114997340A CN114997340A (en) | 2022-09-02 |
| CN114997340B true CN114997340B (en) | 2022-10-14 |
Family
ID=83022511
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210915495.5A Active CN114997340B (en) | 2022-08-01 | 2022-08-01 | Literature classification method based on graph neural network system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114997340B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019084571A1 (en) * | 2017-10-23 | 2019-05-02 | Spangenberg Erich Lawson | Ico and crowdfunding and presale payment system using alternative currency |
| CN112733933A (en) * | 2021-01-08 | 2021-04-30 | 北京邮电大学 | Data classification method and device based on unified optimization target frame graph neural network |
| CN113553440A (en) * | 2021-06-25 | 2021-10-26 | 武汉理工大学 | Medical entity relationship extraction method based on hierarchical reasoning |
-
2022
- 2022-08-01 CN CN202210915495.5A patent/CN114997340B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2019084571A1 (en) * | 2017-10-23 | 2019-05-02 | Spangenberg Erich Lawson | Ico and crowdfunding and presale payment system using alternative currency |
| CN112733933A (en) * | 2021-01-08 | 2021-04-30 | 北京邮电大学 | Data classification method and device based on unified optimization target frame graph neural network |
| CN113553440A (en) * | 2021-06-25 | 2021-10-26 | 武汉理工大学 | Medical entity relationship extraction method based on hierarchical reasoning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114997340A (en) | 2022-09-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Huixian | The analysis of plants image recognition based on deep learning and artificial neural network | |
| CN109002845B (en) | Fine-grained image classification method based on deep convolutional neural network | |
| CN112836672B (en) | Unsupervised data dimension reduction method based on self-adaptive neighbor graph embedding | |
| CN108596053B (en) | Vehicle detection method and system based on SSD and vehicle posture classification | |
| CN110929029A (en) | Text classification method and system based on graph convolution neural network | |
| CN105701507A (en) | Image classification method based on dynamic random pooling convolution neural network | |
| CN110866530A (en) | Character image recognition method and device and electronic equipment | |
| CN108805222A (en) | A kind of deep learning digital handwriting body recognition methods based on ARM platforms | |
| CN111126127A (en) | High-resolution remote sensing image classification method guided by multi-level spatial context characteristics | |
| CN112686242B (en) | Fine-grained image classification method based on multilayer focusing attention network | |
| CN112364747B (en) | Target detection method under limited sample | |
| Chen et al. | Application of improved convolutional neural network in image classification | |
| CN112115993A (en) | Zero sample and small sample evidence photo anomaly detection method based on meta-learning | |
| CN113010683B (en) | Entity relationship identification method and system based on improved graph attention network | |
| CN112149612A (en) | Marine organism recognition system and recognition method based on deep neural network | |
| CN115393631A (en) | Hyperspectral image classification method based on Bayesian layer graph convolution neural network | |
| Wang et al. | Learning to find reliable correspondences with local neighborhood consensus | |
| CN105809200B (en) | Method and device for autonomously extracting image semantic information in bioauthentication mode | |
| CN114399642A (en) | Convolutional neural network fluorescence spectrum feature extraction method | |
| CN114997340B (en) | Literature classification method based on graph neural network system | |
| CN116912576A (en) | Self-adaptive graph convolution brain disease classification method based on brain network higher-order structure | |
| Qiao et al. | Urban scene recognition via deep network integration | |
| CN112307914B (en) | Open domain image content identification method based on text information guidance | |
| CN114170457A (en) | Rock classification method, terminal device and storage medium | |
| Yang et al. | A neural-network-based framework for cigarette laser code identification |
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 |