CN108920528B - Convolutional neural network-oriented space-time geographic dynamic attribute vectorization method and system - Google Patents
Convolutional neural network-oriented space-time geographic dynamic attribute vectorization method and system Download PDFInfo
- Publication number
- CN108920528B CN108920528B CN201810584141.0A CN201810584141A CN108920528B CN 108920528 B CN108920528 B CN 108920528B CN 201810584141 A CN201810584141 A CN 201810584141A CN 108920528 B CN108920528 B CN 108920528B
- Authority
- CN
- China
- Prior art keywords
- vertex
- vertexes
- central
- neural network
- sub
- 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
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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a convolution neural network-oriented space-time geographic dynamic attribute vectorization method and system, wherein an adjacent matrix of a research area is obtained, the research area is abstracted into a topological graph according to the adjacent matrix, then event point data of each sub-area in the research area is counted, preset w central vertexes are selected from the topological graph, a preset value k is obtained, and for each central vertex: selecting k-1 neighborhood vertexes from small to large according to the shortest path from the center vertex to the maximum; and finally, sequentially arranging the w groups of central vertexes and the selected corresponding k-1 domain vertexes, extracting corresponding columns from the C TS matrixes according to the numbers of the w-xk vertexes, reconstructing C MxV matrixes, and forming a C-xMxV three-dimensional tensor T by the C MxV matrixes as a final data processing result. The invention can embed the point data without accurate geographical position information into the tensor, thereby providing a foundation for learning and predicting through the deep convolutional network.
Description
Technical Field
The invention relates to the field of data processing, in particular to a convolution neural network-oriented space-time geographic dynamic attribute vectorization method and system.
Background
Deep Convolutional Neural Networks (CNNs) provide an efficient method for automatically extracting multilevel features from raw data. Although convolutional neural networks have proven to be successful in solving a large number of machine learning problems, conventional convolutional networks require the input to be a tensor. Such as speech and pictures, are modeled as 1D and 2D tensors, respectively. If the convolutional neural network is to extract the spatio-temporal regularity, the spatio-temporal dynamic attribute data must be embedded into the tensor.
For point data with accurate geographical location information, we can make the spatial unit of the data a grid, similar to the pixels of an image, by rasterization. For point data without accurate geographical location information, regular rasterization cannot be performed. For example, if the relevant public administration records an event in spatial units of administrative divisions, such data is difficult to embed in the tensor because the data is spatially irregular due to administrative divisions. The irregularity of the region where the data is located makes it difficult to embed the tensor, and thus learning and prediction cannot be performed through the deep convolutional network.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a convolution neural network-oriented space-time geographic dynamic attribute vectorization method and system aiming at the technical defects that in the prior art, no regular rasterization can be performed on point data without accurate geographic position information, tensor embedding is difficult due to irregularity of an area where the data is located, and further learning and prediction can not be performed through a deep convolution network.
According to one aspect of the present invention, the technical solution adopted by the present invention to solve the technical problem is: a convolutional neural network-oriented space-time geographic dynamic attribute vectorization method is constructed, and comprises the following steps:
s1, obtaining an adjacent matrix A of the research areaN×N(ii) a Wherein the adjacency matrix AN×NAny element A in (1)i,jObtained by the following formula:
in the formula, Ai,jRepresents the adjacency matrix AN×NIn the ith row and the jth column, i ≠ j, i ≠ 1,2, …, N, j ≠ 1,2, …, N, where N is the number of subregions included in the study region, Z is the number of subregions included in the study region, and Z is the number of subregions included in the study regioniAnd ZjRespectively an ith sub-area and a jth sub-area;
s2, according to the adjacency matrix AN×NAbstracting a research area into a topological graph, wherein each sub-area is a vertex, and an edge is formed between every two adjacent sub-areas;
s3, counting the event point data of each sub-area in the research area: for each type of event point data, counting the event number in each preset time period of each sub-region according to the time mark of the event point and the number of the sub-region to which the event point belongs, and sequentially arranging the event number according to the region number and the time period; c TS matrixes are formed, wherein C is the total number of event types;
s4, selecting w central vertexes from the topological graph, wherein w is a preset value, and the selection rule of the central vertexes is as follows:
s41, respectively calculating the sum of the shortest distances from each vertex to other vertices in the topology, and sorting all the vertices from small to large according to the sum of the shortest distances;
s42, sorting vertexes with the equal sum of the shortest distances according to the degrees of the vertexes from large to small;
s43, intercepting the first w vertexes after finishing sequencing as central vertexes;
s5, acquiring a preset value k, and for each central vertex: selecting k-1 neighborhood vertexes from small to large according to the shortest path from the center vertex to the maximum;
s6, sequentially arranging w sets of central vertices and the selected corresponding k-1 domain vertices, extracting corresponding columns from the C TS matrices according to the numbers of the w × k vertices, and reconstructing C M × V matrices to form a C × M × V three-dimensional tensor T as a final data processing result; wherein, arranging the w group central vertex and the selected corresponding k-1 field vertex in sequence specifically means that: for the w central vertices, each central vertex and the corresponding field vertex selected according to step S5 are sequentially arranged in one row or one column, for w rows or w columns.
Further, in the convolutional neural network-oriented space-time geographic dynamic attribute vectorization method of the present invention, in step S1, when i is j, a isi,j＝0。
Further, in the convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method of the present invention, in step S4, if the purpose of unique sorting cannot be achieved after step S42, an additional sorting rule is invoked for sorting.
Further, in the convolutional neural network-oriented space-time geographic dynamic attribute vectorization method of the present invention, in step S5, when the shortest path from the center vertex is selected from small to large, if distances from several neighborhood vertices to the center vertex are equal, the additional sorting rule is invoked to sort, and the neighborhood vertex with the top sorting result is selected.
Further, in the convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method of the present invention, the additional ranking rule refers to a ranking rule set according to one or more of the total event number, population and economy corresponding to the vertex.
Further, in the convolutional neural network-oriented space-time geographic dynamic attribute vectorization method of the present invention, in step S5, when the shortest path from the central vertex is selected from small to large and the number of neighborhood vertices of the central vertex is less than k-1, all the domain vertices of the central vertex are selected as corresponding selection results; in step S6, when the area vertex of the center vertex is less than k-1, zero padding is performed in the row or column in which the center vertex is arranged, and the zero padding position is at the end of the row or column.
Further, in the convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method of the present invention, in step S1, for any two sub-regions, their common edges or common points are considered as being contiguous, otherwise, they are considered as not being contiguous.
Further, in the convolution neural network-oriented space-time geographic dynamic attribute vectorization method, the adjacent matrix A is usedN×NAbstracting a research area into a topological graph specifically as follows:
sub-region ZiAbstract as a vertex v in a topological graphiN vertexes form a vertex set V; will adjoin the matrix AN×NElement A of 1i,jCorresponding to vertex v in the topological graphiAnd vertex viEdge (v) in betweeni，vj) All the edges form an edge set E; the set V and the set E of edges form a topology G (V, E).
According to another aspect of the present invention, to solve the technical problem, the present invention further provides a convolutional neural network-oriented space-time geographic dynamic attribute vectorization system, and the convolutional neural network-oriented space-time geographic dynamic attribute vectorization is performed by using any one of the above mentioned convolutional neural network-oriented space-time geographic dynamic attribute vectorization methods.
The implementation of the convolution neural network-oriented space-time geographic dynamic attribute vectorization method and the convolution neural network-oriented space-time geographic dynamic attribute vectorization system has the following beneficial effects: the invention can embed the point data without accurate geographical position information into the tensor, thereby providing a foundation for learning and predicting through the deep convolutional network.
Drawings
The invention will be further described with reference to the accompanying drawings and examples, in which:
FIG. 1 is a flow chart of an embodiment of the convolutional neural network-oriented spatiotemporal geographic dynamic attribute vectorization method of the present invention;
FIG. 2 is a topological diagram of a study area constructed according to a adjacency matrix in the present invention;
FIG. 3 is a schematic diagram of a TS matrix in the present invention;
FIG. 4 is a schematic diagram of the summation of shortest paths in the present invention;
fig. 5 is a schematic diagram of space division not starved for one channel as a picture for one type of time slot in the present invention.
Detailed Description
For a more clear understanding of the technical features, objects and effects of the present invention, embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
Referring to fig. 1, it is a flowchart of an embodiment of the convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method of the present invention. In this embodiment, taking the example that the original input data of the crime prediction model is the sub-region data of the research region and the attribute data thereof at different times as an example, the main flow of the method of this embodiment may be summarized as follows:
s1, regarding the sub-region data of the study region, the set Z of sub-regions is assumed to have N sub-regions in total. Two sub-regions ZiAnd Zj(i, j ≠ j) common edge orThe common points are considered contiguous, otherwise they are considered non-contiguous. Then, the adjacent matrix A of Z is obtained by the following definitionN×N.
Wherein i, j is 1,2, …, N. Preferably, in the adjacency matrix AN×NWhen i is j, Ai,jAlthough not limited thereto, in other embodiments, when i is j, ai,jOther values are also possible.
S2, according to the adjacency matrix AN×NThe research area is abstracted into a topological graph, each sub-area is a vertex, and an edge is formed between every two adjacent sub-areas. Referring to fig. 2, it is a topological graph constructed by the study region according to the adjacency matrix in the present invention, where the left side is the study region, each small block represents a sub-region, and the right side is the topological graph: sub-region ZiAbstract as a vertex v in a topological graphiN vertexes form a vertex set V; will adjoin the matrix AN×NElement A of 1i,jCorresponding to vertex v in the topological graphiAnd vertex viEdge (v) in betweeni，vj) All the edges form an edge set E; the set V and the set E of edges form a topology G (V, E).
S3, counting the point data of the criminal event: c event types are set, and for each type of event point data, the event number of each time period of each area is counted according to the time mark of the event point and the number of the area to which the event point belongs. The event numbers are arranged according to the area numbers and the time periods to form C TS matrixes. As shown in fig. 3, a TS matrix of crime counts of a type is counted.
S4, selecting w central vertexes with important structural roles in the topological graph. The importance of a structural character is determined by attributes such as the center distance and the degree of a vertex (the number of edges of a vertex becomes the degree of the vertex). The selection rule is as follows: the first step, calculating the sum of the shortest paths from the top point to other top points in the graph, and sequencing all top points from small to large according to the sum of the shortest paths; secondly, sorting according to the degrees of the vertexes from big to small; the third step: this step is skipped if the first two steps achieve the goal of unique ordering. If the first two steps can not be uniquely sorted, the invention can achieve the purpose of unique sorting by adopting additional sorting rules for the attributes of the vertexes, such as setting the sorting rules for the total number of crime cases, population, economy and the like. And fourthly, intercepting the first w vertexes as central vertexes.
To understand the first step, please refer to fig. 4, where the vertices of the graph have the shortest path between each two vertices. The following diagram is a schematic diagram of a shortest path matrix of a diagram having 12 vertices, and elements in an ith row (column) of the matrix are the shortest paths between a vertex i and a vertex j (j is 1,2,3, … 12).
Assuming that there are vertices D1, D2, …, Dn, and taking D1 as an example, the shortest path from vertex D1 to other vertices can be found, calculating D1_1+ D1_2+ … + D1_ n (where D1_ x represents the shortest path from vertex D1 to Dx) is a simple addition. Like the SP matrix, the sum of the shortest paths of D1 to other vertices can be obtained by summing the first row elements of the SP matrix.
S5, acquiring a preset value k, and for each central vertex: and selecting the k-1 neighborhood vertex from small to large according to the shortest path from the center vertex. The k value set in this step is the number of elements of the local perception field of the convolution kernel. When the shortest path from the center vertex is selected from small to large, if the distances from the neighbor vertices to the center vertex are equal, an additional sorting rule is called to sort to achieve the purpose of unique sorting, and the neighbor vertex with the top sorting result is selected. The additional sort rules may or may not be the same as those in step S4, but the present invention preferably does so.
S6, sequentially arranging w sets of central vertices and the selected corresponding k-1 domain vertices, extracting corresponding columns from the C TS matrices according to the numbers of the w × k vertices, and reconstructing C M × V matrices to form a C × M × V three-dimensional tensor T as a final data processing result; wherein, arranging the w group central vertex and the selected corresponding k-1 field vertex in sequence specifically means that: for the w central vertices, each central vertex and the corresponding field vertex selected according to step S5 are sequentially arranged in one row or one column, for w rows or w columns. The tensor T _ (C × M × V) is the final result of data preprocessing, and is composed of C M × V matrices, and the M × V matrices reflect the space-time distribution of a type of event and can be used as a channel of an image, as shown in FIG. 5. T is similar to the image of C channels.
In step S5, when the shortest path from the center vertex is selected from small to large and the number of neighborhood vertices of the center vertex is less than k-1, all the neighborhood vertices of the center vertex are selected as corresponding selection results; in step S6, when the area vertex of the center vertex is less than k-1, zero padding is performed in the row or column in which the center vertex is arranged, and the zero padding position is at the end of the row or column. That is, the relevant process in steps S5, S6 can be understood as: for each central vertex, k-1 domain vertices are found, the central vertex and the k-1 domain vertices form a vector with the length of k, and when the domain vertices of the central vertex are less than k-1, the last few missing elements of the vector are set to be 0, so that w vectors are formed, and the w vectors form a matrix with the dimension of w multiplied by k.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (6)
1. A convolution neural network-oriented space-time geographic dynamic attribute vectorization method is characterized by comprising the following steps:
s1, obtaining an adjacent matrix A of the research areaN×N(ii) a Wherein the adjacency matrix AN×NAny element A in (1)i，jObtained by the following formula:
in the formula, Ai，jRepresents the adjacency matrix AN×NIn the ith row and the jth column, i ≠ j, i ≠ 1,2, …, N, j ≠ 1,2, …, N, where N is the number of subregions included in the study region, Z is the number of subregions included in the study region, and Z is the number of subregions included in the study regioniAnd ZjRespectively an ith sub-area and a jth sub-area;
s2, according to the adjacency matrix AN×NAbstracting a research area into a topological graph, wherein each sub-area is a vertex, and an edge is formed between every two adjacent sub-areas;
s3, counting the event point data of each sub-area in the research area: for each type of event point data, counting the event number in each preset time period of each sub-region according to the time mark of the event point and the number of the sub-region to which the event point belongs, and sequentially arranging the event number according to the region number and the time period; c TS matrixes are formed, wherein C is the total number of event types;
s4, selecting w central vertexes from the topological graph, wherein w is a preset value, and the selection rule of the central vertexes is as follows:
s41, respectively calculating the sum of the shortest distances from each vertex to other vertices in the topology, and sorting all the vertices from small to large according to the sum of the shortest distances;
s42, sorting vertexes with the equal sum of the shortest distances according to the degrees of the vertexes from large to small;
s43, intercepting the first w vertexes after finishing sequencing as central vertexes;
if the unique sorting purpose cannot be achieved after the step S42, calling an additional sorting rule for sorting;
s5, acquiring a preset value k, and for each central vertex: selecting k-1 neighborhood vertexes from small to large according to the shortest path from the center vertex, wherein if the distances from the neighborhood vertexes to the center vertex are equal, calling the additional sorting rule for sorting, and selecting the neighborhood vertex with the top sorting result;
selecting all adjacent domain vertexes of the central vertex as corresponding selection results when the shortest path from the central vertex is selected from small to large and the number of the adjacent domain vertexes of the central vertex is less than k-1;
s6, sequentially arranging w sets of central vertices and the selected corresponding k-1 neighborhood vertices, extracting corresponding columns from the C TS matrices according to the numbers of the w × k vertices, and reconstructing C M × V matrices to form a C × M × V three-dimensional tensor T as a final data processing result; wherein, arranging the w group central vertexes and the selected corresponding k-1 adjacent domain vertexes in sequence specifically means that: for the w central vertexes, each central vertex and the adjacent domain vertex corresponding to the central vertex are selected according to the step S5 and are sequentially arranged in a row or a column, wherein the row or the column is w;
and if the vertex of the adjacent domain of the central vertex is less than k-1, performing zero padding in the row or the column corresponding to the central vertex, wherein the zero padding position is at the tail of the row or the column.
2. The convolutional neural network-oriented space-time geographic dynamic attribute vectorization method as claimed in claim 1, wherein in step S1, when i is j, a isi，j＝0。
3. The convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method as claimed in claim 1, wherein the additional ranking rules refer to the ranking rules set according to one or more of total event number, population and economy corresponding to the vertex.
4. The convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method as claimed in claim 1, wherein in step S1, for any two sub-regions, their common edge or common point is considered as adjacent, otherwise it is considered as not adjacent.
5. The convolutional neural network-oriented spatio-temporal geographic dynamic attribute vectorization method as claimed in claim 1, wherein the neighboring matrix is based onAN×NAbstracting a research area into a topological graph specifically as follows:
sub-region ZiAbstract as a vertex v in a topological graphiN vertexes form a vertex set V; will adjoin the matrix AN×NElement A of 1i，jCorresponding to vertex v in the topological graphiAnd vertex viEdge (v) in betweeni，vj) All the edges form an edge set E; the set V and the set E of edges form a topology G (V, E).
6. A convolution neural network-oriented space-time geographic dynamic attribute vectorization system is characterized in that the convolution neural network-oriented space-time geographic dynamic attribute vectorization is carried out by adopting the convolution neural network-oriented space-time geographic dynamic attribute vectorization method of any one of claims 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810584141.0A CN108920528B (en) | 2018-06-08 | 2018-06-08 | Convolutional neural network-oriented space-time geographic dynamic attribute vectorization method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810584141.0A CN108920528B (en) | 2018-06-08 | 2018-06-08 | Convolutional neural network-oriented space-time geographic dynamic attribute vectorization method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108920528A CN108920528A (en) | 2018-11-30 |
| CN108920528B true CN108920528B (en) | 2021-07-20 |
Family
ID=64418968
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810584141.0A Active CN108920528B (en) | 2018-06-08 | 2018-06-08 | Convolutional neural network-oriented space-time geographic dynamic attribute vectorization method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108920528B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390259A (en) * | 2019-06-11 | 2019-10-29 | 中国科学院自动化研究所南京人工智能芯片创新研究院 | Recognition methods, device, computer equipment and the storage medium of diagram data |
| CN113298302B (en) * | 2021-05-18 | 2022-06-28 | 昆明理工大学 | Irregular shape space-time scanning method aiming at disease prediction |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1191459A1 (en) * | 2000-09-22 | 2002-03-27 | Nightingale Technologies Ltd. | Data clustering methods and applications |
| CN102968804A (en) * | 2012-11-23 | 2013-03-13 | 西安工程大学 | Method for carrying out compression storage on adjacent matrixes of sparse directed graph |
| CN107564288A (en) * | 2017-10-10 | 2018-01-09 | 福州大学 | A kind of urban traffic flow Forecasting Methodology based on tensor filling |
| CN107957993A (en) * | 2017-12-13 | 2018-04-24 | 北京邮电大学 | The computational methods and device of english sentence similarity |
-
2018
- 2018-06-08 CN CN201810584141.0A patent/CN108920528B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1191459A1 (en) * | 2000-09-22 | 2002-03-27 | Nightingale Technologies Ltd. | Data clustering methods and applications |
| CN102968804A (en) * | 2012-11-23 | 2013-03-13 | 西安工程大学 | Method for carrying out compression storage on adjacent matrixes of sparse directed graph |
| CN107564288A (en) * | 2017-10-10 | 2018-01-09 | 福州大学 | A kind of urban traffic flow Forecasting Methodology based on tensor filling |
| CN107957993A (en) * | 2017-12-13 | 2018-04-24 | 北京邮电大学 | The computational methods and device of english sentence similarity |
Non-Patent Citations (4)
| Title |
|---|
| 地理学第一定律之争及其对地理学理论建设的启示;孙俊等;《地理研究》;20121015;第31卷(第10期);全文 * |
| 基于传输效率矩阵的复杂网络节点重要度排序方法;范文礼等;《西南交通大学学报》;20140415;第49卷(第2期);全文 * |
| 盗窃犯罪时空分布特征研究;黄智函;《中国优秀硕士学位论文全文数据库 社会科学I辑》;20161015(第10期);正文第1页-第41页 * |
| 网络重要节点排序方法综述;任晓龙等;《科学通报》;20140510;第59卷(第13期);正文第1175页-第1197页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108920528A (en) | 2018-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111199214B (en) | Residual network multispectral image ground object classification method | |
| CN109858461B (en) | Method, device, equipment and storage medium for counting dense population | |
| JP5463415B2 (en) | Method and system for quasi-duplicate image retrieval | |
| TWI700599B (en) | Method and device for embedding relationship network diagram, computer readable storage medium and computing equipment | |
| CN108804551B (en) | Spatial interest point recommendation method considering diversity and individuation | |
| CN111368943B (en) | Method and device for identifying object in image, storage medium and electronic device | |
| CN108920528B (en) | Convolutional neural network-oriented space-time geographic dynamic attribute vectorization method and system | |
| CN110765320B (en) | Data processing method, device, storage medium and computer equipment | |
| Kumar et al. | Adaptive cluster tendency visualization and anomaly detection for streaming data | |
| US9665773B2 (en) | Searching for events by attendants | |
| CN113793284B (en) | Image restoration method for nonuniform blocking of aerodynamic optical effect space-variant blurred image | |
| CN111291768A (en) | Image feature matching method and device, equipment and storage medium | |
| KR102297195B1 (en) | Population dispersal prediction device based on population dispersal simulation model and population dispersal prediction method using the same | |
| CN111681165A (en) | Image processing method, image processing device, computer equipment and computer readable storage medium | |
| CN110705340A (en) | Crowd counting method based on attention neural network field | |
| CN111709415B (en) | Target detection method, device, computer equipment and storage medium | |
| CN114067041A (en) | Material generation method and device of three-dimensional model, computer equipment and storage medium | |
| CN115828011A (en) | Data analysis method and platform based on big data | |
| CN115391561A (en) | Method and device for processing graph network data set, electronic equipment, program and medium | |
| CN115810149A (en) | High-resolution remote sensing image building extraction method based on superpixel and image convolution | |
| CN112862023B (en) | Object density determination method and device, computer equipment and storage medium | |
| CN110209863A (en) | Method and apparatus for similar pictures retrieval | |
| Borkowski et al. | Matching 2D image segments with genetic algorithms and approximation spaces | |
| CN111223046A (en) | Image super-resolution reconstruction method and device | |
| CN115730217A (en) | Model training method, material recalling method and device |
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 |