CN113887575A - Graph data set enhancing method based on self-adaptive graph convolution network - Google Patents
Graph data set enhancing method based on self-adaptive graph convolution network Download PDFInfo
- Publication number
- CN113887575A CN113887575A CN202111066154.7A CN202111066154A CN113887575A CN 113887575 A CN113887575 A CN 113887575A CN 202111066154 A CN202111066154 A CN 202111066154A CN 113887575 A CN113887575 A CN 113887575A
- Authority
- CN
- China
- Prior art keywords
- graph
- adaptive
- data set
- graph data
- convolution network
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 20
- 230000002708 enhancing effect Effects 0.000 title abstract description 4
- 230000003044 adaptive effect Effects 0.000 claims abstract description 70
- 239000011159 matrix material Substances 0.000 claims abstract description 36
- 238000012549 training Methods 0.000 claims abstract description 18
- 238000010586 diagram Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a graph data set enhancing method based on a self-adaptive graph convolution network, which comprises the following steps: 1) constructing an adaptive graph volume layer; 2) constructing an adaptive graph convolution network by using the adaptive graph convolution layer; 3) arranging a graph data set to be enhanced, and using the graph data set for self-adaptive graph convolution network training; 4) generating an enhancement matrix by using the trained self-adaptive graph convolution network; 5) and carrying out image data enhancement on the image data set to be enhanced by utilizing the enhancement matrix to obtain the enhanced image data set. The invention utilizes the adjacency matrix of the self-adaptive graph convolution network to expand the graph data set, and can well solve the problem of insufficient graph data in the computer vision task.
Description
Technical Field
The invention relates to the technical field of image pattern recognition, in particular to a graph data set enhancement method based on an adaptive graph convolution network.
Background
The deep learning technology is widely applied to the field of computer vision. However, the deep learning technique, as a data-driven algorithm, has high requirements on training data, and when the training data is insufficient, the performance of the algorithm is obviously reduced, so that data enhancement is applied to the extended training data.
In recent years, the development of deep learning based on graph data is receiving wide attention by virtue of excellent performance, however, the deep learning based on graph data also faces the trouble of insufficient training data, and because the structure of graph data is quite different from that of traditional data, the traditional data enhancement mode is difficult to be applied to the expanded graph data, so that the development of a data enhancement method capable of expanding the graph data has great prospect.
Disclosure of Invention
The invention aims to overcome the defects of the prior art and provides a graph data set enhancing method based on an adaptive graph convolution network, which utilizes an adjacent matrix of the adaptive graph convolution network to expand a graph data set and can well solve the problem of insufficient graph data in a computer vision task.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a graph data set enhancement method based on an adaptive graph convolution network is mainly used for amplifying graph data by using an adjacent matrix of the adaptive graph convolution network and comprises the following steps:
1) constructing an adaptive graph volume layer;
2) constructing an adaptive graph convolution network by using the adaptive graph convolution layer;
3) arranging a graph data set to be enhanced, and using the graph data set for self-adaptive graph convolution network training;
4) generating an enhancement matrix by using the trained self-adaptive graph convolution network;
5) and carrying out image data enhancement on the image data set to be enhanced by utilizing the enhancement matrix to obtain the enhanced image data set.
Further, in the step 1), the constructed adaptive graph convolution layer is formed by sequentially connecting a BN layer, a Relu layer, an adaptive graph convolution operation AGC, a BN layer and a Relu layer; the formula of the adaptive graph convolution operation AGC is as follows:
fout＝σ(W·fin·Ae)
in the formula (f)inAnd foutRespectively representing an input feature diagram and an output feature diagram of the adaptive graph convolution operation, wherein W is a graph convolution weight matrix, and sigma (-) represents an activation function; a. theeFor the adaptive adjacency matrix, the size is N × N, where N is the total number of nodes in a graph data, and the formula is as follows:
Ae＝αA+(1-α)B
where α denotes a connection coefficient adaptively adjusted according to training, B denotes a matrix adaptively adjusted according to training, a denotes an adjacent matrix of the input map data connection structure, and the calculation method is as follows: if the nth node and the mth node of the input graph data are connected, the element in the nth row and the mth column in the A is 1, otherwise, the element is 0, and when n is equal to m, the mth row and the mth column are 1.
Further, in step 2), an adaptive graph convolution network is constructed by using the adaptive graph convolution layer and is used for classifying the graph data set, and the adaptive graph convolution network is formed by sequentially connecting a BN layer, 9 adaptive graph convolution layers L1, L2, L3, L4, L5, L6, L7, L8, L9, a global average pooling layer GAP and a Softmax classifier.
Further, in step 3), the graph data set is composed of a plurality of graph data, each graph data includes N node information, the graph data set F to be enhanced is arranged into a tensor with a size of C × T × N, where C represents the number of channels of an image, T represents the total number of data in the graph data set F, and N represents the total number of nodes in each graph data, and the arranged graph data set F and the corresponding sample label thereof are input to the adaptive graph convolution network for X-round iterative training.
Further, in step 4), recording the adaptive adjacency matrix A in the trained adaptive graph convolution networkeGenerating an enhancement matrix AaugThe formula is as follows:
in the formula, i represents adaptationThe number of the layer to be wrapped up in the figure,
adaptive adjacency matrix A representing ith adaptive graph convolution layere。
Further, in step 5), an enhancement matrix a is utilizedaugCarrying out graph data enhancement on the graph data set F to be enhanced to obtain an enhanced graph data set FαThe formula is as follows:
Fα＝F·Aaug。
compared with the prior art, the invention has the following advantages and beneficial effects:
the conventional data enhancement method cannot be applied to graph data, and the invention constructs an innovative graph data set enhancement method based on the adaptive graph convolution network, expands the graph data set by using the adjacent matrix in the adaptive graph convolution network and fills the vacancy for the deficiency of the graph data enhancement method. In a word, the invention performs experiments on a plurality of graph data sets, the expanded graph data is practical and effective, the problem of insufficient graph data in the application occasion of partial graph convolution is solved, and the method has very high practical value and is worthy of popularization.
Drawings
FIG. 1 is a block diagram of an adaptive map convolutional layer.
Fig. 2 is a block diagram of an adaptive graph convolution network.
FIG. 3 is a graph data diagram.
Detailed Description
The present invention will be described in further detail with reference to examples and drawings, but the present invention is not limited thereto.
The embodiment discloses a graph data set enhancement method based on an adaptive graph convolution network, which is mainly used for amplifying graph data by using an adjacency matrix of the adaptive graph convolution network and comprises the following steps:
1) constructing an adaptive graph volume layer;
as shown in fig. 1, the constructed adaptive graph convolution layer is formed by sequentially connecting a BN layer, a Relu layer, an adaptive graph convolution operation AGC, a BN layer, and a Relu layer. The formula of the adaptive graph convolution operation AGC is as follows:
fout＝σ(W·fin·Ae)
in the formula (f)inAnd foutRespectively representing an input feature diagram and an output feature diagram of the adaptive graph convolution operation, wherein W is a graph convolution weight matrix, and sigma (-) represents an activation function; a. theeFor the adaptive adjacency matrix, the size is N × N, where N is the total number of nodes in a graph data, and the formula is as follows:
Ae＝αA+(1-α)B
where α denotes a connection coefficient adaptively adjusted according to training, B denotes a matrix adaptively adjusted according to training, a denotes an adjacent matrix of the input map data connection structure, and the calculation method is as follows: if the nth node and the mth node of the input graph data are connected, the element in the nth row and the mth column in the A is 1, otherwise, the element is 0, and when n is equal to m, the mth row and the mth column are 1.
In this embodiment, the initialization of α to 0.5 can achieve a good effect through trial and error.
2) An adaptive graph convolution network is constructed by using an adaptive graph convolution layer and is used for classifying graph data sets, and as shown in fig. 2, the adaptive graph convolution network is formed by sequentially connecting a BN layer, 9 adaptive graph convolution layers L1, L2, L3, L4, L5, L6, L7, L8, L9, a global average pooling layer GAP and a Softmax classifier.
3) Arranging a graph data set to be enhanced, and using the graph data set for self-adaptive graph convolution network training;
the graph data set is composed of a plurality of graph data, the graph data is illustrated in FIG. 3, each graph data contains N pieces of node information, the graph data set F to be enhanced is arranged into a tensor with the size of C X T X N, wherein C represents the number of channels of an image, T represents the total number of data in the graph data set F, N represents the total number of nodes in each graph data, and the arranged graph data set F and a sample Label Label corresponding to the arranged graph data set F are input into an adaptive graph convolution network for X-round iterative training.
In this embodiment, through repeated experiments, the iterative training times X of 50 can achieve better effect.
4) Recording the adaptive adjacent matrix A in the trained adaptive graph convolution networkeGenerating an enhancement matrix AaugThe formula is as follows:
wherein i represents the number of the adaptive map convolution layer,
adaptive adjacency matrix A representing ith adaptive graph convolution layere。
In this embodiment, through trial and error, a better effect can be obtained when i ═ {1,3,5 }.
5) Using an enhancement matrix AaugCarrying out graph data enhancement on the graph data set F to be enhanced to obtain an enhanced graph data set FαThe formula is as follows:
Fα＝F·Aaug。
the above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (6)
1. A graph data set enhancement method based on an adaptive graph convolution network is characterized in that graph data are amplified mainly by using an adjacency matrix of the adaptive graph convolution network, and the method comprises the following steps:
1) constructing an adaptive graph volume layer;
2) constructing an adaptive graph convolution network by using the adaptive graph convolution layer;
3) arranging a graph data set to be enhanced, and using the graph data set for self-adaptive graph convolution network training;
4) generating an enhancement matrix by using the trained self-adaptive graph convolution network;
5) and carrying out image data enhancement on the image data set to be enhanced by utilizing the enhancement matrix to obtain the enhanced image data set.
2. The graph data set enhancement method based on the adaptive graph convolution network according to claim 1, characterized in that: in the step 1), the constructed self-adaptive graph volume layer is formed by sequentially connecting a BN layer, a Relu layer, a self-adaptive graph volume operation AGC, a BN layer and a Relu layer; the formula of the adaptive graph convolution operation AGC is as follows:
fout＝σ(W·fin·Ae)
in the formula (f)inAnd foutRespectively representing an input feature diagram and an output feature diagram of the adaptive graph convolution operation, wherein W is a graph convolution weight matrix, and sigma (-) represents an activation function; a. theeFor the adaptive adjacency matrix, the size is N × N, where N is the total number of nodes in a graph data, and the formula is as follows:
Ae＝αA+(1-α)B
where α denotes a connection coefficient adaptively adjusted according to training, B denotes a matrix adaptively adjusted according to training, a denotes an adjacent matrix of the input map data connection structure, and the calculation method is as follows: if the nth node and the mth node of the input graph data are connected, the element in the nth row and the mth column in the A is 1, otherwise, the element is 0, and when n is equal to m, the mth row and the mth column are 1.
3. The graph data set enhancement method based on the adaptive graph convolution network according to claim 1, characterized in that: in step 2), an adaptive graph convolution network is constructed by using the adaptive graph convolution layer and is used for classifying the graph data set, wherein the adaptive graph convolution network is formed by sequentially connecting a BN layer, 9 adaptive graph convolution layers L1, L2, L3, L4, L5, L6, L7, L8, L9, a global average pooling layer GAP and a Softmax classifier.
4. The graph data set enhancement method based on the adaptive graph convolution network according to claim 1, characterized in that: in step 3), the graph data set is composed of a plurality of graph data, each graph data contains N pieces of node information, the graph data set F to be enhanced is arranged into a tensor with the size of C × T × N, wherein C represents the number of channels of an image, T represents the total number of data in the graph data set F, and N represents the total number of nodes in each graph data, and the arranged graph data set F and a sample label corresponding to the graph data set F are input into the adaptive graph convolution network for X-round iterative training.
5. The graph data set enhancement method based on the adaptive graph convolution network according to claim 1, characterized in that: in step 4), recording the adaptive adjacent matrix A in the trained adaptive graph convolution networkeGenerating an enhancement matrix AaugThe formula is as follows:
wherein i represents the number of the adaptive map convolution layer,
adaptive adjacency matrix A representing ith adaptive graph convolution layere。
6. The graph data set enhancement method based on the adaptive graph convolution network according to claim 1, characterized in that: in step 5), an enhancement matrix A is utilizedaugCarrying out graph data enhancement on the graph data set F to be enhanced to obtain an enhanced graph data set FαThe formula is as follows:
Fα＝F·Aaug。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111066154.7A CN113887575B (en) | 2021-09-13 | 2021-09-13 | Graph dataset enhancement method based on self-adaptive graph rolling network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111066154.7A CN113887575B (en) | 2021-09-13 | 2021-09-13 | Graph dataset enhancement method based on self-adaptive graph rolling network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113887575A true CN113887575A (en) | 2022-01-04 |
| CN113887575B CN113887575B (en) | 2024-04-05 |
Family
ID=79009006
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111066154.7A Active CN113887575B (en) | 2021-09-13 | 2021-09-13 | Graph dataset enhancement method based on self-adaptive graph rolling network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113887575B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023216269A1 (en) * | 2022-05-13 | 2023-11-16 | 北京小米移动软件有限公司 | Data enhancement method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020248581A1 (en) * | 2019-06-11 | 2020-12-17 | 中国科学院自动化研究所 | Graph data identification method and apparatus, computer device, and storage medium |
| CN112216108A (en) * | 2020-10-12 | 2021-01-12 | 中南大学 | Traffic prediction method based on attribute-enhanced space-time graph convolution model |
| CN112464042A (en) * | 2020-11-25 | 2021-03-09 | 平安科技（深圳）有限公司 | Task label generation method according to relation graph convolution network and related device |
-
2021
- 2021-09-13 CN CN202111066154.7A patent/CN113887575B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020248581A1 (en) * | 2019-06-11 | 2020-12-17 | 中国科学院自动化研究所 | Graph data identification method and apparatus, computer device, and storage medium |
| CN112216108A (en) * | 2020-10-12 | 2021-01-12 | 中南大学 | Traffic prediction method based on attribute-enhanced space-time graph convolution model |
| CN112464042A (en) * | 2020-11-25 | 2021-03-09 | 平安科技（深圳）有限公司 | Task label generation method according to relation graph convolution network and related device |
Non-Patent Citations (1)
| Title |
|---|
| 董荣胜 等: "大规模图数据的k~2-MDD表示方法与操作研究", 计算机研究与发展, no. 12, 15 December 2016 (2016-12-15), pages 120 - 129 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2023216269A1 (en) * | 2022-05-13 | 2023-11-16 | 北京小米移动软件有限公司 | Data enhancement method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113887575B (en) | 2024-04-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11301727B2 (en) | Efficient image classification method based on structured pruning | |
| CN110782462B (en) | Semantic segmentation method based on double-flow feature fusion | |
| Chen et al. | ReGAN: A pipelined ReRAM-based accelerator for generative adversarial networks | |
| CN110232394B (en) | Multi-scale image semantic segmentation method | |
| CN112560432B (en) | Text emotion analysis method based on graph attention network | |
| JP6838161B2 (en) | End-to-end modeling methods and systems | |
| CN109508784B (en) | Design method of neural network activation function | |
| CN110197217B (en) | Image classification method based on deep interleaving fusion packet convolution network | |
| CN110648292A (en) | High-noise image denoising method based on deep convolutional network | |
| CN111916144B (en) | Protein classification method based on self-attention neural network and coarsening algorithm | |
| CN111276187B (en) | Gene expression profile feature learning method based on self-encoder | |
| CN112685504A (en) | Production process-oriented distributed migration chart learning method | |
| CN109886389B (en) | Novel bidirectional LSTM neural network construction method based on Highway and DC | |
| CN113887575A (en) | Graph data set enhancing method based on self-adaptive graph convolution network | |
| CN113297427A (en) | Graph classification method based on U-shaped nested network | |
| CN114064627A (en) | Knowledge graph link completion method and system for multiple relations | |
| CN116740527A (en) | Remote sensing image change detection method combining U-shaped network and self-attention mechanism | |
| CN111199237A (en) | Attention-based convolutional neural network frequency division feature extraction method | |
| CN113962262B (en) | Continuous learning-based intelligent radar signal sorting method | |
| CN116434039B (en) | Target detection method based on multiscale split attention mechanism | |
| CN111882028B (en) | Convolution operation device for convolution neural network | |
| CN112131403A (en) | Knowledge graph representation learning method in dynamic environment | |
| CN115601744B (en) | License plate detection method for vehicle body and license plate with similar colors | |
| CN116883746A (en) | Graph node classification method based on partition pooling hypergraph neural network | |
| CN116152570A (en) | Commodity image classification method based on hierarchical convolutional neural network |
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 |