CN108229578B - Image data target identification method based on three layers of data, information and knowledge map framework - Google Patents
Image data target identification method based on three layers of data, information and knowledge map framework Download PDFInfo
- Publication number
- CN108229578B CN108229578B CN201810074539.XA CN201810074539A CN108229578B CN 108229578 B CN108229578 B CN 108229578B CN 201810074539 A CN201810074539 A CN 201810074539A CN 108229578 B CN108229578 B CN 108229578B
- Authority
- CN
- China
- Prior art keywords
- image
- entity
- data
- identified
- information
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/22—Matching criteria, e.g. proximity measures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/30—Information retrieval; Database structures therefor; File system structures therefor of unstructured textual data
- G06F16/36—Creation of semantic tools, e.g. ontology or thesauri
- G06F16/367—Ontology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/51—Indexing; Data structures therefor; Storage structures
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Databases & Information Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Animal Behavior & Ethology (AREA)
- Computational Linguistics (AREA)
- Image Analysis (AREA)
Abstract
The present invention is a kind of image data target identification method based on three layers of data, information and knowledge map framework, belongs to distributed computing and Software Engineering technology crossing domain.The present invention is primarily introduced into data map, Information Atlas, knowledge mapping framework and carries out knowledge reasoning to image is not identified, thus realize automated intelligent identify in picture or shot by camera to image in physical object.Specific implementation step is to have identified target by analysis, past path can be connected to unidentified target by finding in data map, Information Atlas or knowledge mapping, entity in traverse path carries out characteristic matching simultaneously, is eventually found and the unidentified highest recognition result of object matching degree.
Description
Technical field
The present invention is a kind of image data target identification method based on three layers of data, information and knowledge map framework.It is main
Be used for machine automated intelligent identify in picture or shot by camera to image in physical object.Belong to distribution
Formula calculates and Software Engineering technology crossing domain.
Background technique
Most traditional image recognition technology is manual identified, and having passed through can be identified by the method for deep learning and marked
The image of knowledge, machine gradually simulate an equation in identifying same class object repeatedly, this equation, which approaches, to be identified pair
The common trait of elephant finally reaches the purpose of identification, but the method for deep learning not can solve the identification for not identifying image
Problem.Knowledge mapping, which has become, indicates the strong tools of knowledge with the digraph form of label, and provides the language of text information
Justice.Knowledge mapping is by the way that each project, entity or user are done node expression, and by edge by interaction between each other
Those nodes are chained up the figure of construction.Side between node can indicate any relationship.Compared with uml class, knowledge mapping
With richer natural semantic, expression mechanism is closer and natural language, contains more more complete semantic informations.Now very
Carry out auto-modeling application program using UML less, one reason for this is that UML lacks the formal key needed to application program
The semanteme that part is modeled, by the three-decker proposed by the present invention based on data map, Information Atlas and knowledge mapping
Relationship frame carries out image information identification.The present invention be mainly by by data map, Information Atlas, knowledge mapping frame
Structure goes correctly to identify the object as much as possible not being identified.
Summary of the invention
Technical problem: the present invention be primarily introduced into data map, Information Atlas, knowledge mapping framework to do not identify image into
Row knowledge reasoning, thus automated intelligent identify in picture or shot by camera to image in physical object.This
Invention is analyzed identified target first, and finding in data map, Information Atlas or knowledge mapping can connect with unidentified target
Lead to past path, the entity in traverse path carries out characteristic matching simultaneously, is eventually found and unidentified object matching degree highest
Recognition result.
Technical solution: the present invention is a kind of tactic method, and the image object that can be applied to picture or video camera is known
Not, facilitate the identification for solving the problems, such as not identifying physical object in current machine study.The present invention be primarily introduced into data map,
Information Atlas, knowledge mapping framework carry out knowledge reasoning to image is not identified, to realize the identifying in picture of automated intelligent
Or shot by camera to image in physical object.Embodiment be by analysis identified target, data map,
Past path can be connected to unidentified target by finding in Information Atlas or knowledge mapping, the entity in traverse path simultaneously into
Row characteristic matching is eventually found and the unidentified highest recognition result of object matching degree.We define in knowledge mapping
Semantic relation between where, when and entity node is as shown in Figure 1:
1. abundant necessity relationship: node A is certainly existed under conditions of when and where are determined, if while node A deposit
In the where or when of available determination;
2. abundant unnecessary relationship: node A is certainly existed under conditions of when and where is determined, if but node A deposit
Where and when can not be determined;
3. necessary insufficient condition: node A is not necessarily present under conditions of when and where is determined, if but node A deposit
Where or when can be uniquely determined;
4. both insufficient or unnecessary condition: node A is not necessarily present under conditions of when and where is determined, simultaneously
If node A exists, where or when can not also be determined.
Architecture: the present invention be primarily introduced into data map, Information Atlas, knowledge mapping framework to do not identify image into
Row knowledge reasoning, thus realize automated intelligent identify in picture or shot by camera to image in entity mesh
Mark.The present invention sets three kinds of situations first: having identified that image and unidentified image are in same data structure, identification figure
As with unidentified image have directly or indirectly interactive relation, identified image and unidentified image is in same environment, needle
To these three situations respectively to having identified that image traverses on data map, Information Atlas and knowledge mapping, and at three layers
Past path can be connected to unidentified target by finding in map, found the entity occurred on path and carried out characteristic matching, most
After find with the unidentified highest entity of object matching degree as recognition result.Currently, we can use deep learning
Identify identified image, the present invention on this basis, introduces data map, Information Atlas, knowledge mapping framework to not marking
Know image and carry out knowledge reasoning, to realize the image entities that in picture or shot by camera arrives of identifying of automated intelligent
Target.Building data map, Information Atlas, knowledge mapping is given below to illustrate:
Data map: data map can record the frequency of entity appearance, the frequency including three structure, time and space levels
Degree.Our definition structure frequency are that entity appears in the number in different data structure, and time frequency is the time locus of entity,
Spatial frequency is defined as the space tracking of entity.Data map can describe the entity category that can be observed directly in the picture
Property, including color, shape etc., can also describe the static relation between entity, including direction relations and topological relation and each
Associated tightness degree between node, we term it density, it is close to can reflect out which entity relationship, which entity relationship
It is sparse.But the accuracy of entity is not analyzed on data map, in fact it could happen that the entity but the same thing of expression of different names
Object, this generates data redundancies.To sum up, data map can only handle static relation image recognition (identified image with not
Identification image be in the same data structure, such as apple is grown on apple tree), it is unpredictable and analyze have interaction pass
The image of system;
Information Atlas: information is conveyed by the context after data and data combination, by concept mapping and
The information of suitable analysis and explanation after correlativity combination.Information Atlas can be expressed according to relational database.Information
Map can express the interactive relation (such as shock of apple and ground) between entity, and the frequency on Information Atlas refers to reality
The frequency of interaction between body and entity can carry out data cleansing on Information Atlas, eliminate redundant data.According to algorithm 1
All interaction paths between available two entities:
Knowledge mapping: knowledge mapping is the overall understanding and consciousness obtained from the information of accumulation, and information is carried out into one
The abstract and classification of step can form knowledge.Knowledge mapping can carry out table by the inclusion of the digraph of relationship between node and node
It reaches.Knowledge mapping can express various semantic relations, improve knowledge graph by information inference and entity link on knowledge mapping
The side density and node density of spectrum, making without architectural characteristic for knowledge mapping itself can be with seamless link.Information inference needs
The support of dependency rule, these rules can be time-consuming and laborious by people's manual construction, but often, and obtain in complex relationship all pushes away
Reason rule is more difficult.Currently, information inference depends on the co-occurrence of relationship, and is searched and pushed away automatically using association mining technology
Reason rule.Use each different relation path as one-dimensional characteristic, by constructing a large amount of relation path in knowledge mapping
Relationship is extracted to construct feature vector and the relationship extractor of relationship classification, and it is to think that the correctness of relationship, which is more than a certain threshold value,
New relation is set up.The correctness Cr of the new relation obtained by reasoning can be obtained by formula 1.Between entity E1 and entity E2
New relation can be expressed as E1 → E2, and z indicates all relationships, | z | it is the corresponding quantity of relationship,Indicate new relation weight,A relationship between presentation-entity 1 and entity 2 is to think that the relationship is set up when correctness is more than a threshold value:
The utility model has the advantages that the present invention is a kind of based on data, the knowledge of the image data target of three layers of map framework of information and knowledge
Other method has the following remarkable advantage:
(1) present invention constructs data map, Information Atlas and knowledge mapping three-tier architecture, progressive, realizes efficient
Identification;
(2) image unidentified in deep learning method can be correctly identified;
(3) present invention is for having identified that the relationship of image and unidentified image in data map, Information Atlas and knows respectively
Know and traversed on map, keeps recognition result more accurate.
Detailed description of the invention
Fig. 1 is to data map, the formal definitions of Information Atlas and knowledge mapping;
Wherein (1) data map
Circle represents entity, and box represents entity attributes, wherein A1, A2, B1, C1, C2, C3, in same data structure
In.A3, B2, C4 are in same data structure.
(2) Information Atlas
B1 and D1 interaction frequency is that 2 and D2 interaction frequency is 3;B1 and E2, E1 indirect interaction;B2 and D4, D5 are directly interactive
(3) knowledge mapping
1. sufficiently necessity relationship 2. sufficiently necessary insufficient relationship 4. of unnecessary relationship 3. is both insufficient or unnecessary relationship
Fig. 2 is the specific of the image entities target identification process based on data map, Information Atlas and knowledge mapping framework
Flow chart.
Specific embodiment
It is a kind of based on data, the detailed process of the image data target identification method of three layers of map framework of information and knowledge such as
Under:
Step 1) according to have image resource, establish based on data map, Information Atlas, knowledge mapping frame;
Step 2) obtains images to be recognized;
Images to be recognized is divided by step 3) has identified two modules of image { Bi } and unidentified image X;
If step 4) A and B are present among a data structure: the image of identification { Bi } traversal obtained based on step 3
Data map, thus it is speculated that go out it and be likely to be present among the data structure of which entity, these entities are denoted as { Ai };To { Ai } into
Row analysis, acquires all entities being likely to occur in { Ai }, is denoted as { Ci };Note;{ Ai } ∪ { Ci } is { Xi };
The unidentified image X obtained based on step 3 and entity { Xi } are carried out characteristic matching by step 5), are obtained by formula 1
Matching degree R, when R is greater than some threshold value, using R as recognition result, it is assumed that entity attribute feature and Ai in data map
With value be μ (α), α indicate Xi can with an attribute of unidentified images match, x indicate it is all can with matched attribute, including
Shape, color, topological relation, direction relations etc.,
R (Xi)=∑ α ∈ x (X → Xi) μ (α) (2);
If step 6) A and B have direct interactive relation: the sequence time in Information Atlas according to interaction frequency from big to small
The neighbouring node { Di } for having identified image { Bi } is gone through, note { Di } is { Xi }, executes step 5;If A and B has indirect interaction relationship:
The neighbouring node { Ei } for having identified image { Di } according to the order traversal of interaction frequency from big to small in Information Atlas, remembers { Ei }
For { Xi }, executes step 5 and continually look for the neighbouring node of { Ei } if not finding recognition result, repeat the above process, directly
Until finding;
If step 7) has identified that image and unidentified image are in same environment: the identification figure obtained based on step 3
As { Bi } traverses knowledge mapping, find " Where " corresponding in image and " When ", and find at corresponding " Where " and
The entity type being likely to occur in " When " is denoted as { Xi }, executes step 5;
Step 8) exports recognition result.
Claims (1)
1. a kind of image data target identification method based on three layers of data, information and knowledge map framework, it is characterised in that draw
Enter data map, Information Atlas and knowledge mapping framework and carry out knowledge reasoning to image is not identified, to realize automated intelligent
Identify in picture or shot by camera to image in physical object, detailed process is as follows:
Step 1) establishes the frame based on data map, Information Atlas and knowledge mapping according to existing image resource;
Step 2 obtains images to be recognized;
Images to be recognized is divided by step 3) has identified two modules of image B and unidentified image X;
If step 4) has identified that the entity in image in B and unidentified image X is present among a data structure, wherein data
Structure includes the set of array, chained list, stack, queue, tree and figure, the image B ergodic data figure of identification obtained based on step 3)
Spectrum, thus it is speculated that go out it and be likely to be present among the data structure of which entity, these entities are denoted as { Ai };{ Ai } is divided
Analysis acquires all entities being likely to occur in { Ai }, is denoted as { Ci };Remember that { Ai } ∪ { Ci } is { Xi };
The unidentified image X obtained based on step 3) and entity { Xi } are carried out characteristic matching by step 5), are matched by formula 2
R is spent, when R is greater than some threshold value, XiAs recognition result, it is assumed that entity attribute feature and Ai matching value in data map
Indicate that Xi can indicate that matched attribute, including shape, color are opened up with an attribute of unidentified images match, x for μ (α), α
Flutter relationship and direction relations;
(2)
If entity in step 6) A and B has direct interactive relation: according to frequency of interaction from big to small suitable in Information Atlas
Sequence traversal has identified the adjacent node { Di } of the entity node { Bi } in image B, and note { Di } is the subset of { Xi }, executes step
5);If the entity in A and B has indirect interaction relationship: the order traversal in Information Atlas according to interaction frequency from big to small
The adjacent node { Ei } of the adjacent node { Di } of the entity node { Bi } in image is identified, note { Ei } is the subset of { Xi }, is held
Row step 5) continually looks for the adjacent node of { Ei } if not finding recognition result, step 6) is repeated, until finding;
If step 7) has identified that image and unidentified image are in same environment: the image of the identification B obtained based on step 3)
Knowledge mapping is traversed, finds " Where " corresponding in image and " When ", and find in corresponding " Where " and " When "
The entity type being likely to occur is denoted as { Xi }, executes step 5);
Step 8) exports recognition result.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810074539.XA CN108229578B (en) | 2018-01-25 | 2018-01-25 | Image data target identification method based on three layers of data, information and knowledge map framework |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810074539.XA CN108229578B (en) | 2018-01-25 | 2018-01-25 | Image data target identification method based on three layers of data, information and knowledge map framework |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108229578A CN108229578A (en) | 2018-06-29 |
CN108229578B true CN108229578B (en) | 2019-10-11 |
Family
ID=62667370
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810074539.XA Active CN108229578B (en) | 2018-01-25 | 2018-01-25 | Image data target identification method based on three layers of data, information and knowledge map framework |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108229578B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110210387B (en) * | 2019-05-31 | 2021-08-31 | 华北电力大学(保定) | Method, system and device for detecting insulator target based on knowledge graph |
CN111143626B (en) * | 2019-12-25 | 2023-06-09 | 深圳云天励飞技术有限公司 | Method, apparatus, device and computer readable storage medium for identifying group |
CN111309827A (en) * | 2020-03-23 | 2020-06-19 | 平安医疗健康管理股份有限公司 | Knowledge graph construction method and device, computer system and readable storage medium |
CN111460206B (en) * | 2020-04-03 | 2023-06-23 | 百度在线网络技术(北京)有限公司 | Image processing method, apparatus, electronic device, and computer-readable storage medium |
CN112528993A (en) * | 2020-12-17 | 2021-03-19 | 济南浪潮高新科技投资发展有限公司 | Method for improving image recognition accuracy rate based on knowledge graph |
CN115050085B (en) * | 2022-08-15 | 2022-11-01 | 珠海翔翼航空技术有限公司 | Method, system and equipment for recognizing objects of analog machine management system based on map |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10235423B2 (en) * | 2012-12-12 | 2019-03-19 | Google Llc | Ranking search results based on entity metrics |
CN106776564B (en) * | 2016-12-21 | 2020-04-24 | 张永成 | Semantic recognition method and system based on knowledge graph |
CN107085618A (en) * | 2017-06-01 | 2017-08-22 | 海南大学 | Study point and learning path towards the 5W target drives based on data collection of illustrative plates, Information Atlas and knowledge mapping are recommended |
CN107038508A (en) * | 2017-06-06 | 2017-08-11 | 海南大学 | The study point tissue and execution route of the learning ability modeling of knowledge based collection of illustrative plates and the target drives of dynamic self-adapting recommend method |
-
2018
- 2018-01-25 CN CN201810074539.XA patent/CN108229578B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN108229578A (en) | 2018-06-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108229578B (en) | Image data target identification method based on three layers of data, information and knowledge map framework | |
WO2021103492A1 (en) | Risk prediction method and system for business operations | |
Rhee et al. | Active and semi-supervised learning for object detection with imperfect data | |
Cai et al. | Discrete particle swarm optimization for identifying community structures in signed social networks | |
Alzahrani et al. | Community detection in bipartite networks: Algorithms and case studies | |
Leng et al. | Mining and matching relationships from interaction contexts in a social manufacturing paradigm | |
CN105824802A (en) | Method and device for acquiring knowledge graph vectoring expression | |
Le Bodic et al. | An integer linear program for substitution-tolerant subgraph isomorphism and its use for symbol spotting in technical drawings | |
CN107451210B (en) | Graph matching query method based on query relaxation result enhancement | |
CN103198228A (en) | Relational network link predicting method based on generalized relation hidden topic model | |
CN108052680B (en) | Image data target identification Enhancement Method based on data map, Information Atlas and knowledge mapping | |
Zare et al. | Detection of community structures in networks with nodal features based on generative probabilistic approach | |
CN107818328A (en) | With reference to the deficiency of data similitude depicting method of local message | |
Lonij et al. | Open-world visual recognition using knowledge graphs | |
CN110502669A (en) | The unsupervised chart dendrography learning method of lightweight and device based on the side N DFS subgraph | |
Wu et al. | Multi-graph-view learning for complicated object classification | |
Sun et al. | Heterogeneous network representation learning based on role feature extraction | |
CN113537292B (en) | Multi-source domain adaptation method based on tensor high-order mutual attention mechanism | |
CN111062484B (en) | Data set selection method and device based on multi-task learning | |
Yang et al. | A few-shot inductive link prediction model in knowledge graphs | |
Kalyonova et al. | Revealing of entities interconnections in system dynamics modelling process by applying multimodal data analysis paradigm | |
Wijayanto et al. | Predicting future potential flight routes via inductive graph representation learning | |
Sharma et al. | Review Of Data Mining Techniques: An Empirical Study | |
El Abri et al. | Learning probabilistic relational models with (partially structured) graph databases | |
Lander | Multi-objective graph mining algorithms for detecting and predicting communities in complex dynamic networks |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |