CN115544309A - Improved nearest neighbor data interconnection method based on GCN - Google Patents
Improved nearest neighbor data interconnection method based on GCN Download PDFInfo
- Publication number
- CN115544309A CN115544309A CN202211157579.3A CN202211157579A CN115544309A CN 115544309 A CN115544309 A CN 115544309A CN 202211157579 A CN202211157579 A CN 202211157579A CN 115544309 A CN115544309 A CN 115544309A
- Authority
- CN
- China
- Prior art keywords
- gcn
- node
- nodes
- prediction model
- matrix
- 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 32
- 238000005259 measurement Methods 0.000 claims abstract description 53
- 238000011156 evaluation Methods 0.000 claims abstract description 7
- 239000011159 matrix material Substances 0.000 claims description 42
- 239000013598 vector Substances 0.000 claims description 19
- 238000012549 training Methods 0.000 claims description 7
- 230000004913 activation Effects 0.000 claims description 3
- 239000000126 substance Substances 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 230000000694 effects Effects 0.000 abstract description 6
- 238000000605 extraction Methods 0.000 abstract description 3
- 230000006870 function Effects 0.000 description 7
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000008447 perception Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/901—Indexing; Data structures therefor; Storage structures
- G06F16/9024—Graphs; Linked lists
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- 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
Abstract
The invention discloses an improved nearest neighbor data interconnection method based on GCN, which comprises the following steps: step 1, obtaining sensor measurement values, wherein each sensor measurement value corresponds to one node to establish a graph network structure, and a plurality of sensors can correspondingly obtain a plurality of graph networks; step 2, constructing a GCN prediction model, and extracting characteristic information of a graph network structure node by using the GCN prediction model; step 3, calculating the node distance based on the node characteristics output by the GCN prediction model; and 4, solving an interconnection result by taking the characteristic distance between the nodes output by the GCN prediction model as an interconnection evaluation standard. The invention mainly solves the problem of deep feature extraction of multi-sensor measurement, utilizes GCN to automatically learn deep feature information and invariant topology information between measurements from original measurement data, and improves the data interconnection effect of the traditional nearest neighbor method under the scenes of dense clutter, random errors and system deviation.
Description
Technical Field
The invention belongs to a data processing technology, and particularly relates to an improved nearest neighbor data interconnection method based on GCN.
Background
The characteristics of various, dense, maneuvering, stealth, fine guidance and the like of marine targets are increasingly highlighted, the diversification development of marine perception means and the continuous expansion of detection range are realized, the phenomena of redundancy, fuzziness, conflict and the like of multi-source perception data information are more serious, and the phenomena of missed tracking, wrong tracking, multiple tracking and the like of the targets are frequent. The sea target perception data is influenced by multiple factors such as sea clutter, atmospheric waveguides and platform errors, the traditional target tracking technology tries to accurately model by using ideas such as probability theory, fuzzy set or random set, and inevitable and non-negligible errors exist, and actual engineering requirements are difficult to meet.
In recent years, with the rapid development of deep neural networks, deep learning techniques are becoming research hotspots in the field of complex-environment marine target tracking. However, the existing target tracking deep neural network is simple in design, and can not overcome the deviation of the traditional target tracking model mechanism, so that the bottleneck problem of point-point/point-navigation data interconnection exists when the target tracking deep neural network is applied to actual engineering.
The nearest neighbor method is used as a basic method in the multi-sensor multi-target data interconnection method, is suitable for static target measurement data interconnection in a sparse clutter environment, but for a dense clutter scene, a large number of measurement errors and a system deviation scene, the correlation correctness is greatly reduced, and the application scene of the algorithm is severely limited.
Based on the above, an improved nearest neighbor data interconnection method based on a Graph Convolutional neural Network (GCN) is provided.
Disclosure of Invention
In view of the above technical problems, the present invention provides an improved nearest neighbor data interconnection method based on GCN.
The technical scheme for solving the technical problems is as follows:
a GCN-based improved nearest neighbor data interconnection method comprises the following steps:
step 1, obtaining sensor measurement values, wherein each sensor measurement value corresponds to one node to establish a graph network structure, and a plurality of sensors can correspondingly obtain a plurality of graph networks;
step 2, constructing a GCN prediction model, and extracting characteristic information of a graph network structure node by using the GCN prediction model;
step 3, calculating the node distance based on the node characteristics output by the GCN prediction model;
and 4, solving an interconnection result by taking the characteristic distance between the nodes output by the GCN prediction model as an interconnection evaluation standard.
Further, the step 1 of obtaining a sensor measurement value and constructing a graph network structure according to the sensor measurement value includes the following steps:
acquiring sensor measurement values, wherein each sensor measurement value corresponds to one node, calculating the relationship between any two sensor measurement values, namely the relationship between the nodes, and constructing a relationship edge according to the relationship of the nodes;
and constructing a graph network structure according to the nodes, the associated nodes and the relationship edges.
Further, calculating the relationship between any two sensor measurement values, namely the relationship between nodes, and constructing a relationship edge according to the relationship between the nodes, comprising the following steps:
calculating the distance d between any two measured values n,ij If the distance is largeIf the relation of the formula (1) is less satisfied, establishing an edge between two nodes corresponding to the two measurement values, otherwise, not establishing the edge;
d n,ij ＝(z n,i -z n,j ) T (z n,i -z n,j )≤γ,n＝1,2 (1)
in the formula (d) n,ij Represents a distance between the i-th and the j-th measurement values of the n-th sensor; z is a radical of n,i 、z n,j An i-th measurement value and a j-th measurement value respectively representing the n-th sensor; t represents transposition; and gamma is a threshold value for establishing the edge.
Further, before constructing the GCN prediction model, the nodes and the relation edges are in a matrix form;
given a total of N nodes in the graph network structure, the input characteristics of the nodes are
Wherein z is n,i ∈R D D is the input feature number of each node, and the feature vectors of all nodes in the layer I
Expressed in matrix form to obtain
The method comprises the following steps of (1) forming an N multiplied by D dimensional characteristic matrix, wherein each row in the matrix corresponds to the characteristic representation of each node;
using an N-dimensional adjacency matrix, A represents the edge relationship between the nodes, where A ij =1 indicates node i and node j are connected, there is an edge, otherwise A ij ＝0。
Furthermore, the GCN prediction model constructed in step 2 uses a two-layer GCN network structure, the inputs of the GCN prediction model are a node feature matrix and an adjacency matrix, and the propagation mode between layers is as follows:
wherein the content of the first and second substances,
i is an identity matrix;
is that
The degree matrix of (c) is,
is a normalized Laplace matrix;
for the characteristic matrix of the l layer, representing the characteristic vector of all nodes of the nth sensor on the l layer, l =0,1 for the input layer, there is
W (l) Represents the weight of the l-th layer, which is shared for different sensors n; σ is a nonlinear activation function.
Further, the step 3 of calculating the node distance based on the node features output by the GCN prediction model includes the following steps:
node characteristic matrix output based on GCN prediction model
Obtaining a feature vector h of each node n,i And calculating the characteristic distance between any two nodes in different graphs:
wherein h is n-1,i 、h n,j Respectively representing the ith node in the nth sensor based on GCN outputAnd a feature vector of a jth node in the (n-1) th sensor;
and the distance between the nodes i and j representing the output characteristic of the GCN prediction model.
Further, in step 4, the data interconnection result that minimizes the global feature distance is solved as follows:
wherein ρ ij Is a binary data interconnection variable, is represented by 0 or 1, if the ith node is interconnected with the jth node, then rho is ij =1; otherwise ρ ij ＝0；N n-1 、N n Respectively represent the total N of the N-1 th sensors n-1 The number of nodes and the nth sensor is N n And (4) each node.
Further, the training of the GCN prediction model comprises the following steps:
interconnecting the obtained output result rho with the target real data rho * And (4) performing difference, taking the matrix norm to establish a loss function, and training a GCN prediction model through back propagation.
Compared with the prior art, the invention has the following technical effects:
1. constructing a sensor measurement value into a graph network structure, so that the GCN can extract depth characteristic information and topology characteristic information between target measurements;
2. the distance between the features is used for measuring the similarity of measurement of the homologous targets, and the data interconnection effect under the scenes of dense clutter, random errors and system deviation is improved;
3. and constructing a loss function by using the output result of the optimal distribution model and the target real data interconnection result, and realizing network training.
Drawings
Fig. 1 is a schematic flow chart of the improved nearest neighbor data interconnection method based on GCN of the present invention.
Detailed Description
The principles and features of this invention are described below in conjunction with the following drawings, which are set forth to illustrate, but are not to be construed to limit the scope of the invention.
In order to improve the application range and the association effect of the nearest neighbor method in the multi-sensor multi-target data interconnection, the GCN method is introduced to improve the traditional nearest neighbor method, deep characteristic information and invariant topology information between measurements are automatically learned from original measurement data by the GCN, and the data interconnection effect under the scenes of dense clutter, random errors and system deviation is remarkably improved.
Fig. 1 is a schematic flow chart of an improved nearest neighbor data interconnection method based on GCN according to the present invention. Referring to fig. 1, an improved nearest neighbor data interconnection method based on GCN includes the following steps: step 1, obtaining sensor measurement values, wherein each sensor measurement value corresponds to one node to establish a graph network structure, and a plurality of sensors can correspondingly obtain a plurality of graph networks; step 2, constructing a GCN prediction model, and extracting characteristic information of a graph network structure node by using the GCN prediction model; step 3, calculating the node distance based on the node characteristics output by the GCN prediction model; and 4, solving an interconnection result by taking the characteristic distance between the nodes output by the GCN prediction model as an interconnection evaluation standard.
The following steps are detailed:
step 1, sensor measurement values are obtained, each sensor measurement value corresponds to one node to establish a graph network structure, and a plurality of sensors can correspondingly obtain a plurality of graph networks.
Acquiring sensor measurement values, wherein each sensor measurement value corresponds to one node, calculating the relationship between any two sensor measurement values, namely the relationship between the nodes, and constructing a relationship edge according to the relationship of the nodes;
calculating the relation between any two sensor measurement values, namely the relation between nodes, and constructing a relation edge according to the relation of the nodes, wherein the method comprises the following steps:
calculating the distance d between any two measured values n,ij If the distance satisfies the relation of formula (1), the two measured values are pairedEstablishing an edge between the corresponding two nodes, otherwise, not establishing the edge;
d n,ij ＝(z n,i -z n,j ) T (z n,i -z n,j )≤γ,n＝1,2 (1)
in the formula, d n,ij Represents a distance between the i-th and the j-th measurement values of the n-th sensor; z is a radical of n,i 、z n,j An i-th measurement value and a j-th measurement value respectively representing the n-th sensor; t represents transposition; and gamma is a threshold value for establishing edges.
And constructing a graph network structure according to the nodes, the associated nodes and the relationship edges. The Graph (Graph) structure is represented as G = (V, E), where V represents Node (Node) and E represents Edge (Edge), so as to obtain local topology information between the metrics.
And 2, constructing a GCN prediction model, and extracting the characteristic information of the graph network structure node by using the GCN prediction model.
The GCN is a basic convolutional graph neural network, which belongs to a model in deep learning, and is a feature extractor, which is a graph data object, in fact, the function of the convolutional neural CNN. The GCN directly acts on a neural network on a graph structure, obtains a feature vector representing the relationship between a node and a neighboring node thereof through feature extraction, and prepares for node interconnection analysis later.
Given a total of N nodes in a given Graph, the input characteristics of the nodes are
Wherein z is n,i ∈R D D is the input characteristic number (including multi-dimensional characteristics such as position, speed, course, attribute and type) of each node, and all the node characteristic vectors of the layer I are combined
Expressed in matrix form to obtain
The method comprises the following steps of (1) forming an N multiplied by D dimensional characteristic matrix, wherein each row in the matrix corresponds to the characteristic representation of each node; an Adjacency Matrix (Adjacency Matrix) a of N × N dimensions is used to represent an edge relation between respective nodes, where a ij =1 indicates node i and node j are connected, there is an edge, otherwise A ij ＝0。
In this embodiment, a two-layer GCN network structure is used to extract feature information of a graph structure, the input of the network is a feature matrix and an adjacency matrix, and the propagation mode between layers is:
wherein the content of the first and second substances,
i is an identity matrix;
is that
The degree matrix of (c) is,
is a normalized Laplace matrix;
for the characteristic matrix of the l layer, representing the characteristic vector of all nodes of the nth sensor on the l layer, l =0,1 for the input layer, there is
W (l) Representing the weight of the l layer, which is shared by different sensors n; σ is a nonlinear activation function.
Different from structured feature vectors input by the conventional prediction method, the prediction model is input into an unstructured graph, the traditional deep convolutional network can extract features of structured data (images, voices, sequences and the like), but the graph data has the characteristics of unstructured property, disorder and randomness, the constructed nodes and the relation number are not fixed, the expression form is more flexible, the feature matrixes which are aligned to be fixed scales cannot be aligned to extract the features by using the traditional convolutional network, therefore, the convolutional network for the graph is required to be adopted, the graph convolutional network is not limited by the dependency relationship on a two-dimensional structure when capturing the features, and richer associated node information can be aggregated.
And 3, calculating the node distance based on the node characteristics output by the GCN prediction model.
Node characteristic matrix output based on GCN prediction model
Obtaining a feature vector h of each node n,i Calculating the characteristic distance between any two nodes in different graphs:
wherein h is n-1,i 、h n,j Respectively representing the feature vector of the ith node in the nth sensor and the feature vector of the jth node in the n-1 th sensor based on the GCN output;
and the distance between the nodes i and j representing the output characteristic of the GCN prediction model.
Specifically, taking two sensors as an example, according to the constructed graph network structure, the node feature matrix based on GCN output
The feature vector h of each node can be obtained n,i And calculating the characteristic distance between any two nodes in different graphs:
wherein h is 1,i 、h 2,j Respectively representing the feature vector of the ith node in the 1 st sensor and the feature vector of the jth node in the 2 nd sensor based on the GCN output;
and representing the GCN prediction model output characteristic distance between the node i and the node j.
And 4, solving an interconnection result by taking the characteristic distance between the nodes output by the GCN prediction model as an interconnection evaluation standard.
In order to obtain the data interconnection relation among different sensor measurement sets, the characteristic distance among the nodes output by the GCN prediction model is used as an interconnection evaluation standard, a data interconnection result which enables the global characteristic distance to be minimum is solved, and an optimal distribution model is established as follows:
where ρ is ij Is a binary data interconnection variable, is represented by 0 or 1, and if the ith node is interconnected with the jth node, rho is ij =1; otherwise ρ ij ＝0；N n-1 、N n Respectively represent the total N of the N-1 th sensors n-1 The number of nodes and the N sensor is N n And (4) each node.
Specifically, taking two sensors as an example, in order to obtain a data interconnection relationship between two sensor measurement sets, the feature distance between nodes output by the GCN prediction model is used as an interconnection evaluation standard, a data interconnection result which enables the global feature distance to be minimum is solved, and an optimal distribution model is established as follows:
where ρ is ij Is a binary data interconnection variable, is represented by 0 or 1, if the ith node is interconnected with the jth nodeρ ij =1; otherwise ρ ij ＝0。N 1 、N 2 Respectively representing the total number N of the 1 st sensors 1 The number of nodes and the number 2 of sensors is N 2 And (4) each node.
The training of the GCN prediction model comprises the following steps: interconnecting the obtained output result rho with the target real data rho * And (4) performing difference, taking the matrix norm to establish a loss function, and training a GCN prediction model through back propagation.
Specifically, the obtained output result ρ of the optimal distribution model and the target real data interconnection result ρ are interconnected · And (3) taking the matrix norm to establish a loss function as follows:
Loss＝||ρ-ρ * || (5)
after the Loss is calculated, the GCN model is trained through back propagation.
The invention mainly solves the problem of deep characteristic extraction of multi-sensor measurement, utilizes GCN to automatically learn deep characteristic information and invariant topology information between measurements from original measurement data, and improves the data interconnection effect of the traditional nearest neighbor method under the scenes of dense clutter, random errors and system deviation.
The above description is only for the purpose of illustrating the preferred embodiments of the present invention and should not be taken as limiting the scope of the present invention, which is intended to cover any modifications, equivalents, improvements, etc. within the spirit and scope of the present invention.
Claims (8)
1. An improved nearest neighbor data interconnection method based on GCN is characterized by comprising the following steps:
step 1, acquiring sensor measurement values, wherein each sensor measurement value corresponds to one node to establish a graph network structure, and a plurality of sensors correspond to obtain a plurality of graph network structures;
step 2, constructing a GCN prediction model, and extracting characteristic information of a graph network structure node by using the GCN prediction model;
step 3, calculating the node distance based on the node characteristics output by the GCN prediction model;
and 4, solving an interconnection result by taking the characteristic distance between the nodes output by the GCN prediction model as an interconnection evaluation standard.
2. The GCN-based improved nearest neighbor data interconnection method of claim 1, wherein sensor measurement values are obtained in the step 1, each sensor measurement value corresponds to a node to establish a graph network structure, and the method comprises the following steps:
acquiring sensor measurement values, wherein each sensor measurement value corresponds to one node, calculating the relationship between any two sensor measurement values, namely the relationship between the nodes, and constructing a relationship edge according to the relationship of the nodes;
and constructing a graph network structure according to the nodes, the associated nodes and the relationship edges.
3. The GCN-based improved nearest neighbor data interconnection method according to claim 2, wherein a relationship between any two sensor measurement values, namely a relationship between nodes, is calculated, and a relationship edge is constructed according to the relationship between the nodes, comprising the following steps:
calculating the distance d between any two measured values n,ij If the distance satisfies the relation of formula (1), establishing an edge between two nodes corresponding to the two measurement values, otherwise, not establishing an edge;
d n,ij ＝(z n,i -z n,j ) T (z n,i -z n,j )≤γ,n＝1,2 (1)
in the formula (d) n,ij Representing a distance between an i-th measurement value and a j-th measurement value of an n-th sensor; z is a radical of n,i 、z n,j An i-th measurement value and a j-th measurement value respectively representing the n-th sensor; t represents transposition; and gamma is a threshold value for establishing edges.
4. The GCN-based improved nearest neighbor data interconnection method according to claim 1, wherein before the GCN prediction model is constructed, the nodes and the relational edges are in a matrix form;
given a total of N nodes in the graph network structure, the input characteristics of the nodes are
Wherein z is n,i ∈R D D is the input feature number of each node, and the feature vectors of all nodes in the layer I
Expressed in matrix form to obtain
The feature matrix is an N multiplied by D dimension feature matrix, and each row in the matrix corresponds to the feature representation of each node;
using an N-dimensional adjacency matrix, A represents the edge relationship between the nodes, where A ij =1 indicates node i and node j are connected, there is an edge, otherwise A ij ＝0。
5. The GCN-based improved nearest neighbor data interconnection method according to claim 4, wherein the GCN prediction model constructed in the step 2 uses a two-layer GCN network structure, the GCN prediction model inputs are a node feature matrix and an adjacency matrix, and the propagation modes between layers are as follows:
wherein the content of the first and second substances,
i is an identity matrix;
is that
The degree matrix of (c) is,
is a normalized Laplace matrix;
for the characteristic matrix of the l layer, representing the characteristic vector of all nodes of the nth sensor on the l layer, l =0,1 for the input layer, there is
W (l) Represents the weight of the l-th layer, which is shared for different sensors n; σ is a nonlinear activation function.
6. The GCN-based improved nearest neighbor data interconnection method according to claim 5, wherein the step 3 of calculating the node distance based on the node characteristics output by the GCN prediction model comprises the steps of:
node characteristic matrix output based on GCN prediction model
Obtaining a feature vector h of each node n,i And calculating the characteristic distance between any two nodes in different graphs:
wherein h is n-1,i 、h n,j Respectively representing the feature vector of the ith node in the nth sensor and the feature vector of the jth node in the n-1 th sensor based on the GCN output;
representing the GCN prediction model between node i and node jThe distance of the feature is determined.
7. The GCN-based improved nearest neighbor data interconnection method according to claim 6, wherein in the step 4, the data interconnection result with the minimum global feature distance is solved as follows:
wherein ρ ij Is a binary data interconnection variable, is represented by 0 or 1, and if the ith node is interconnected with the jth node, rho is ij =1; otherwise ρ ij ＝0；N n-1 、N n Respectively represent the total N of the N-1 th sensors n-1 The number of nodes and the N sensor is N n And (4) each node.
8. The GCN-based improved nearest neighbor data interconnection method according to claim 7, wherein the training of the GCN prediction model comprises the following steps:
interconnecting the obtained output result rho with the target real data rho * And (4) performing difference, taking the matrix norm to establish a loss function, and training a GCN prediction model through back propagation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211157579.3A CN115544309B (en) | 2022-09-22 | 2022-09-22 | Improved nearest neighbor data interconnection method based on GCN |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211157579.3A CN115544309B (en) | 2022-09-22 | 2022-09-22 | Improved nearest neighbor data interconnection method based on GCN |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115544309A true CN115544309A (en) | 2022-12-30 |
| CN115544309B CN115544309B (en) | 2024-03-19 |
Family
ID=84728790
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211157579.3A Active CN115544309B (en) | 2022-09-22 | 2022-09-22 | Improved nearest neighbor data interconnection method based on GCN |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115544309B (en) |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113468288A (en) * | 2021-07-23 | 2021-10-01 | 平安国际智慧城市科技股份有限公司 | Content extraction method of text courseware based on artificial intelligence and related equipment |
| CN113591380A (en) * | 2021-07-28 | 2021-11-02 | 浙江大学 | Traffic flow prediction method, medium and equipment based on graph Gaussian process |
| CN114418174A (en) * | 2021-12-13 | 2022-04-29 | 国网陕西省电力公司电力科学研究院 | Electric vehicle charging load prediction method |
| WO2022134576A1 (en) * | 2020-12-23 | 2022-06-30 | 深圳壹账通智能科技有限公司 | Infrared video timing behavior positioning method, apparatus and device, and storage medium |
-
2022
- 2022-09-22 CN CN202211157579.3A patent/CN115544309B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2022134576A1 (en) * | 2020-12-23 | 2022-06-30 | 深圳壹账通智能科技有限公司 | Infrared video timing behavior positioning method, apparatus and device, and storage medium |
| CN113468288A (en) * | 2021-07-23 | 2021-10-01 | 平安国际智慧城市科技股份有限公司 | Content extraction method of text courseware based on artificial intelligence and related equipment |
| CN113591380A (en) * | 2021-07-28 | 2021-11-02 | 浙江大学 | Traffic flow prediction method, medium and equipment based on graph Gaussian process |
| CN114418174A (en) * | 2021-12-13 | 2022-04-29 | 国网陕西省电力公司电力科学研究院 | Electric vehicle charging load prediction method |
Non-Patent Citations (2)
| Title |
|---|
| NAIXUE KUANG 等: "" Network Link Connectivity Prediction Based on GCN and Differentiable Pooling Model"", 《2022 IEEE 14TH INTERNATIONAL CONFERENCE ON ADVANCED INFOCOMM TECHNOLOGY (ICAIT)》, 11 July 2022 (2022-07-11), pages 1 - 6, XP034185354, DOI: 10.1109/ICAIT56197.2022.9862715 * |
| 党吉圣;杨军;: "多特征融合的三维模型识别与分割", 西安电子科技大学学报, no. 04, pages 153 - 161 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115544309B (en) | 2024-03-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109919108B (en) | Remote sensing image rapid target detection method based on deep hash auxiliary network | |
| CN111489358B (en) | Three-dimensional point cloud semantic segmentation method based on deep learning | |
| Wang et al. | Hybrid inference network for few-shot SAR automatic target recognition | |
| CN112308019B (en) | SAR ship target detection method based on network pruning and knowledge distillation | |
| CN110245709B (en) | 3D point cloud data semantic segmentation method based on deep learning and self-attention | |
| CN112818969B (en) | Knowledge distillation-based face pose estimation method and system | |
| CN111126134B (en) | Radar radiation source deep learning identification method based on non-fingerprint signal eliminator | |
| CN113033520A (en) | Tree nematode disease wood identification method and system based on deep learning | |
| CN112364747B (en) | Target detection method under limited sample | |
| WO2020086176A1 (en) | Artificial neural network and method of training an artificial neural network with epigenetic neurogenesis | |
| JP6892606B2 (en) | Positioning device, position identification method and computer program | |
| CN116580322A (en) | Unmanned aerial vehicle infrared small target detection method under ground background | |
| CN111523586B (en) | Noise-aware-based full-network supervision target detection method | |
| Cai et al. | Underwater distortion target recognition network (UDTRNet) via enhanced image features | |
| CN114973031A (en) | Visible light-thermal infrared image target detection method under view angle of unmanned aerial vehicle | |
| CN112668662B (en) | Outdoor mountain forest environment target detection method based on improved YOLOv3 network | |
| CN114492755A (en) | Target detection model compression method based on knowledge distillation | |
| CN116486089A (en) | Point cloud segmentation network light-weight method, device and equipment based on knowledge distillation | |
| CN116958809A (en) | Remote sensing small sample target detection method for feature library migration | |
| CN115544309A (en) | Improved nearest neighbor data interconnection method based on GCN | |
| CN115423847A (en) | Twin multi-modal target tracking method based on Transformer | |
| CN116030292A (en) | Concrete surface roughness detection method based on improved ResNext | |
| CN114819325A (en) | Score prediction method, system, device and storage medium based on graph neural network | |
| CN111652246B (en) | Image self-adaptive sparsization representation method and device based on deep learning | |
| CN115019053A (en) | Dynamic graph semantic feature extraction method for point cloud classification and segmentation |
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 |