CN109344823A - Based on the OCR deep learning method of block chain mechanism, storage medium - Google Patents
Based on the OCR deep learning method of block chain mechanism, storage medium Download PDFInfo
- Publication number
- CN109344823A CN109344823A CN201811054240.4A CN201811054240A CN109344823A CN 109344823 A CN109344823 A CN 109344823A CN 201811054240 A CN201811054240 A CN 201811054240A CN 109344823 A CN109344823 A CN 109344823A
- Authority
- CN
- China
- Prior art keywords
- model
- block chain
- state
- deep learning
- data
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Character Discrimination (AREA)
Abstract
The present invention provide based on the OCR deep learning method of block chain mechanism, storage medium, method include: according to learning data creation by each state and its it is corresponding return the model constituted instantly, each state corresponds to each sub- learning data in learning data；Privately owned storage model to block chain network each node；The state selected in model at random is trained, and obtains the corresponding new return of state；According to the new return Optimized model, the training pattern of the corresponding model is obtained；The corresponding training pattern collection of the model enough according to quantity carries out convolution deep learning；The result of shared storage convolution deep learning is to each node.The resource-sharing of learning data may be implemented in the present invention；Effective protection can be carried out to self-teaching achievement, improve learning value, while helping speed up research and development progress；It can also guarantee user data privacy.
Description
Technical field
The present invention relates to OCR deep learning methods, and in particular to OCR deep learning method based on block chain mechanism is deposited
Storage media.
Background technique
Common OCR scheme is top-down mode based on the scanning of sliding window full figure and based on bottom rule at present
First divide to zonule be combined into character area the bottom of from and on mode.And the deep learning of OCR is then special based on HOG
5 layers of CNN network of input are levied as OCR identification model.In identification network training, CNN will identify complicated text, face
Problem be how to obtain it is enough meet multifarious training sample, and only skilled sample meets diversity and could protect
Card trains the OCR identification network for meeting business demand.
Traditional Text region is to describe to feature more demanding by the way of based on template matching, is difficult to meet multiple
Identification under heterocycle border, deep learning is the study based on CNN network at present, is independently carried out.Study requires Pang every time
Big and various data sample, including font, color, background etc..From the point of view of existing product, for handwritten form discrimination simultaneously
The stability of block letter discrimination is not reached.Hinder handwriting recognition maximum problem be exactly identify that sample data is insufficient so that
The study of machine does not reach requirement in 5 layers of CNN network.And the fonts such as current characters in a fancy style, individualized signature are more and more, it is right
For OCR, need constantly to learn new font.And per mostly a kind of font, it is only just how up to ten thousand by taking Chinese character as an example
Kind text, it is necessary to a large amount of learning time.This problem so can solve by the data sharing feature of block chain.
Study for new data, each company is all in specific rate, and for OCR, CNN is not problem, and key exists
Identification learning is carried out to new literacy faster in for whom, and by block chain to the privately owned protection feature of data, it can be in state and shape
The corresponding return of state is chosen, the fixed model stage protects private data.The identification of OCR is using the financial row of slowly more multi-steering
Industry, whether bill, digital signature, all have certain private ownership, and after OCR identification, it only can determine whether word content, it can't
Text is protected.Data line can so be protected by the private data mechanism of block chain with really attributes.
Summary of the invention
The technical problems to be solved by the present invention are: providing the OCR deep learning method based on block chain mechanism, storage is situated between
Matter not only greatly reduces learning time；Training pattern privatization can also be protected simultaneously, improve learning value.
In order to solve the above-mentioned technical problem, the technical solution adopted by the present invention are as follows:
OCR deep learning method based on block chain mechanism, comprising:
According to learning data creation by each state and its it is corresponding return the model constituted instantly, each state is corresponding to be learned
Practise each sub- learning data in data；
Privately owned storage model to block chain network each node；
The state selected in model at random is trained, and obtains the corresponding new return of state；
According to the new return Optimized model, the training pattern of the corresponding model is obtained；
The corresponding training pattern collection of the model enough according to quantity carries out convolution deep learning；
The result of shared storage convolution deep learning is to each node.
Another technical solution provided by the invention are as follows:
A kind of computer readable storage medium is stored thereon with computer program, described program when being processed by the processor,
It is able to achieve the step of above-mentioned OCR deep learning method based on block chain mechanism is included.
The beneficial effects of the present invention are: firstly, passing through the data sharing feature of block chain, it can not only guarantee that OCR knows
Other data can be multiplexed by all machines, and can also be obviously improved learning efficiency according to shared data, to contract significantly
Subtract learning time；Secondly, can not only be guaranteed obtaining the training pattern stage by the privately owned protection of block chain and really attributes
The private ownership of training pattern helps quickly to research and develop, and improves enterprise value；And in the identification storing data stage, moreover it is possible to guarantee
User data privacy.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow diagrams of the OCR deep learning method of block chain mechanism；
Fig. 2 is the flow diagram of the embodiment of the present invention one.
Specific embodiment
To explain the technical content, the achieved purpose and the effect of the present invention in detail, below in conjunction with embodiment and cooperate attached
Figure is explained.
The most critical design of the present invention is: by the data sharing feature of block chain, guaranteeing can be by by learning data
All machine multiplexings；By the privately owned protection of block chain and really attributes, in the training pattern stage, guarantees product private ownership, have
Help quickly research and develop, improves enterprise value；And in the identification storing data stage, moreover it is possible to guarantee user data privacy.
Explanation of technical terms of the present invention:
| Technical term | It explains |
| OCR deep learning | The ability of self-teaching of handwriting recognition |
| Block chain mechanism | Storage characteristics based on block chain network |
| State | The action data transmitted when deep learning |
| The corresponding return of state | The result data that movement generates after occurring |
| Model | Wish the result reached |
| Training pattern | Model after optimization, i.e. deep learning data |
| Convolution deep learning | Self-teaching is carried out by convolutional neural networks |
| Fixed model | Fixed target, prevents learning process from shaking |
| CNN | Convolutional neural networks |
Fig. 1 is please referred to, the present invention provides the OCR deep learning method based on block chain mechanism, comprising:
According to learning data creation by each state and its it is corresponding return the model constituted instantly, each state is corresponding to be learned
Practise each sub- learning data in data；
Privately owned storage model to block chain network each node；
The state selected in model at random is trained, and obtains the corresponding new return of state；
According to the new return Optimized model, the training pattern of the corresponding model is obtained, and its privately owned is stored to each
Node；
The corresponding training pattern collection of the model enough according to quantity carries out convolution deep learning；
The result of shared storage convolution deep learning is to each node.
As can be seen from the above description, the beneficial effects of the present invention are: the OCR deep learning of new font is carried out by the application
And identification, the privately owned attribute stored by block chain, it can guarantee that the training pattern of selected learning data is able to private first
Having saves, and prevents corporate espionage；Secondly learning data is stored in block chain network, as long as meeting all of single font
Stroke learning, can by the sharing of block chain network platform by stroke combination at font, without all knowing to text
Not, greatly reduce learnt duration.
Further, the training pattern for obtaining the corresponding model, specifically:
Model after fixed optimization, using the model after fixation as the training pattern of the model.
Seen from the above description, by fixing model, occurs the problems such as such as shaking when preventing from training, to improve instruction
Experienced stability and accuracy.
Further, further includes: shared storage OCR identifies each node of data to block chain network.
Seen from the above description, using the shared mechanism of block chain, the resource-sharing of OCR identification data is realized, to reduce
The learning time of OCR identification.
Further, described corresponding to return by each state and its instantly the model constituted, institute according to learning data creation
It states each state and corresponds to each sub- learning data in learning data, specifically:
The learning data of an individual character is chosen from OCR identification data；
The sub- learning data of predetermined number is randomly selected from the learning data as state, learning outcome is as state
Corresponding to return instantly, creation is by multiple states and its corresponding mould corresponding with an individual character that return is constituted instantly
Type.
Seen from the above description, corresponding individual character creates model, and optimizes and obtain individualized training model.It is total in model (individual character)
Quantity is enough, it is sufficient in the case where covering all stroke features of such font, need to can only pass through the depth to these models
Study, can easily, quickly and accurately grasp the recognition methods of this kind of font, without all carrying out to all identification data
Study, to greatly reduce learning time.
Further, the corresponding training pattern collection of the model enough according to quantity carries out convolution deep learning, tool
Body are as follows:
According to the corresponding training pattern collection of multiple models of covering all stroke features of font, pass through 5 layers of convolutional Neural net
Network carries out OCR deep learning.
Seen from the above description, it need to only ensure all stroke features of training pattern collection energy coverage goal font, Bian Ketong
The identification method for crossing such font of deep learning and mastering is not only significantly mentioned without all learning to all identification data
Learning efficiency has been risen, has reduced learning time, and can guarantee the accuracy of study again.Meanwhile in a specific embodiment, lead to
5 layers of CNN are crossed to guarantee the accuracy of study.
It is further, described according to the new return Optimized model, specifically:
By bellman equation dynamic optimization model, obtain by closest to model state and its corresponding return constitute
Training pattern.
Seen from the above description, in a specific embodiment, Optimized model through the above way, can efficient automatic screening
Optimal policy out obtains training pattern.
Another technical solution provided by the invention are as follows:
A kind of computer readable storage medium is stored thereon with computer program, described program when being processed by the processor,
It is able to achieve the step of above-mentioned OCR deep learning method based on block chain mechanism is included.
It corresponds to those of ordinary skill in the art will appreciate that realizing all or part of the process in above-mentioned technical proposal, being can
It is realized with instructing relevant hardware by computer program, the program can be stored in one and computer-readable deposit
In storage media, the program is when being executed, it may include such as the process of above-mentioned each method.
Wherein, the storage medium can be disk, optical disc, read-only memory (Read-Only Memory,
) or random access memory (Random Access Memory, RAM) etc. ROM.
Embodiment one
Referring to figure 2., the present embodiment is to further limit to Fig. 2, provides a kind of OCR depth based on block chain mechanism
Learning method, comprising the following steps:
S1: block chain network platform is built.
Build a block chain network platform, i.e. block chain network, each node storage of the block chain network platform
Identical data, and default open use.
S2: shared storage OCR identifies each node of data to block chain network.
Based on the data sharing spy of block chain to, allow OCR identification data obtained by any machine, realization can answer
With.For example, a new personalized fonts emerge, if it is desired to can identify the font, then need to its OCR identify data into
Row deep learning.
S2: corresponding the model constituted, each state pair are returned by each state and its instantly according to learning data creation
Answer each sub- learning data in learning data.
Specifically, may include following sub-step:
S21: the learning data of an individual character is chosen from OCR identification data；
S22: a part of data (i.e. sub- learning data) is randomly selected out from the corresponding learning data of the individual character as shape
State, the learning outcome obtained after learning to selected sub- learning data are returned instantly as the state is corresponding；Creation
It is corresponding instantly by the state (carrying out randomly selecting for corresponding number according to predetermined number) of predetermined number and each state
Return constituted model.
The corresponding individual character (learning data) of one model.During OCR deep learning, the state is convolution mind
When working through network, the state transmitted；Corresponding return of the state is the result provided after state is transmitted.That is,
When machine starts deep learning, some status datas can be transmitted, for example tell network to be scanned, window translation；State pair
The return answered are as follows: after state occurs, generated reward data, i.e., as a result, such as pixel value.In the present embodiment, described
State is a state given at random from learning data, i.e., a randomly selected sub- learning data；The state is corresponding
Return be this state behavior return obtained instantly, that is, the result returned when learning.
S3: each node of privately owned storage model to block chain network.
I.e. by each state and state it is corresponding return instantly it is privately owned be stored in block chain network, i.e., privatization, really
Quan Hua gets up model protection, and only oneself can be looked into, on the one hand for property information protection and quick utility strategies, need
It will be to state and the corresponding return privatization of state；On the other hand the model (enterprise itself product) by constructing enterprise itself
Private ownership, help quickly to research and develop.
S4: the state selected in model at random is trained, and obtains the corresponding new return of state；
S5: the model according to the new return Automatic Optimal model, after obtaining the optimization of the corresponding model.
Some states are picked out from model at random to be trained, and automatic to model progress excellent using these states
Change.Optionally, optimal data are gone out by bellman equation Dynamic Programming.
This is because state and the corresponding return of state are to be randomly generated, more tend to correct state, obtained return
It is more accurate, therefore, most suitable state can be found out by exhaustion analysis.The Automatic Optimal is system corresponding to state
When return analysis, the corresponding return of state and state closest to model is selected automatically, is preferably advised by bellman equation dynamic
Draw theory optimize, no matter original state and initial decision, using first piece of decision be formed by stage and state as
When primary condition considers, remaining decision also constitutes optimal policy for remaining problem.By taking the present embodiment as an example, when the side A
When formula is optimal, next step decision can be gone out by ballman equation calculation, Automatic sieve selects subsequent all optimal policy, i.e.,
Select the state and the corresponding return of state closest to set model, the model after being optimized.The purpose for the arrangement is that making
Time Dependent between training sample disappears.
S6: the model after fixed optimization, using the model after fixation as the training pattern of the model, and by training pattern
It is privately owned to store to each node.
Model after will optimizing is fixed, and the training pattern for corresponding to the individual character is just obtained, and privately owned is stored in block chain
In network.The training pattern refers to corresponding selected learning data, and what is obtained after the model of optimum option is optimal
Model；It needs to carry out privately owned protection to training pattern；It is fixed in this way so that the obtained optimal policy of model (trains mould
Type) it will not change, it is not in the problems such as such as shake, to improve study accuracy when training.
S7: the corresponding training pattern collection of the model enough according to quantity carries out convolution deep learning.
Firstly, the model that the quantity is enough, corresponding is the model that can cover all stroke features of such font
Set.Individualized training model is corresponding is the target learning model of such one individual character of font, therefore individualized training result is directed to
It is the accurate learning outcome that some individual character of such font is made.The deep learning of individualized training model is as a result, i.e. single instruction
Practice result to be also stored in privately owned in block chain network.Due to the number of fonts of a kind of font (such as Chinese character) be usually it is huge,
Single word can not be used as sample data, it is therefore desirable to preserve individualized training result is privately owned, when acquisition can cover such
Training pattern set corresponding to the word set of font whole stroke, and all after learning outcome, i.e., all strokes are all known
Not Cheng Gong after, can be realized and all words (all Chinese characters) of such font are carried out and its deep learning of self, and herein it
Preceding training result is only to find optimal policy, so needing to the privately owned storage of individualized training result.
Secondly, convolution deep learning (CNN) refers to a kind of deep learning mode, Chinese is convolutional neural networks, usually
In OCR study, 5 layers of CNN is needed to guarantee study accuracy.
S8: the result of shared storage convolution deep learning to each node.
It is the learning outcome of all training patterns, (training of corresponding diagram 2 of as final learning outcome that this result is corresponding
As a result).The self-teaching to such font is had been completed at this time, therefore " OCR achievement " i.e. learning outcome can be shown
Come, is used in each block chain user.
Particularly, for financial industry, such as the OCR identification of bill, digital signature is nowadays more applied to, due to this reality
The training pattern set private ownership for applying example is stored in each node, therefore its identification process is secrecy.Thus it just may be implemented in
When identifying to the above-mentioned content to be identified with security requirement, secret protection can be carried out to user data simultaneously.
Embodiment two
The present embodiment specifically uses scene for the one of corresponding embodiment one:
OCR deep learning mode based on block chain mechanism can be directed to personalized Chinese character, the English increasingly changed at present
The unconventional text such as text and Mars word, Chinese-traditional, foreign language.
Such as after a new personalized text emerges, its font is unconventional, the possibility in known character library
It can not search out, if that machine, which wants the such Text region of study, needs to carry out deep learning, and use embodiment one
The method can reach safe and quick purpose.
Specifically, first choosing the less word of stroke when identifying new individual character font.First according to deep learning mode,
A corresponding word give a template (i.e. set model, deep learning be by providing set objective after, machine is learnt,
Here template refers to set target)；Font randomly chooses some states in identification, obtains the corresponding return of these states,
Optimal strategy, i.e. discrimination height, accurate numerical value are selected in return again.We are corresponding by such state and state
Return (i.e. above-mentioned training template) is stored in the private room of block chain network platform.It is inconsistent that other strokes are chosen again
Font identified according to aforesaid way, when choosing take fully more than enough stroke can cover font when (it is few preferably first to choose stroke,
Learn stroke therein by these words, finally, if the word that study is enough, can cover the study of all strokes, such as
Study has the word of Philosophy, then learns the word of the strokes such as cross break, point, hook again, can cover after learning enough strokes
Cover all Chinese characters), these states and the corresponding return (i.e. corresponding trained template set) of corresponding state just meet this time
The learning data of deep learning, these data for being stored in block chain network private room can not be seen forever by unauthorized people
It arrives, it is ensured that the safety of our core datas.
These training patterns have been fixed on the private room of block chain by we, are carried out followed by 5 layers of CNN network
The identification of OCR, so in the process since data are fixed, not will lead to us trains the problems such as shaking.And we
Fixed data and model data be it is privately owned, obtain optimal protection in block chain network platform.
After training result comes out, we are stored the result in block chain network, it is noted that current storage is open
, because we, which have been directed to the individual character font, completes self-teaching, our OCR achievement is shown in time.In view of area
The public character of block chain network and high degree of spreading, our result can be obtained by each block chain user.
Example above we using certain personalized Chinese character as experiment body, and when for English or other foreign languages, due to language
Problem, it may be necessary to be identified for each letter, by taking individual character english font as an example, it would be desirable to 26 English alphabets,
Optimum state and the corresponding return of state are chosen respectively, by optimal policy store in block chain private room, are then directed to word
Mother's combination is made correctly sequence and is identified.
Mode carries out the OCR deep learning and identification of new font through this embodiment, can use the private of block chain storage
There is attribute, can guarantee that the learning data selected by us is able to privatization preservation first, prevent corporate espionage；Secondly by data
It is stored in block chain network, it, being total to by block chain network platform as long as meeting all stroke learnings of single font
Stroke combination is greatly reduced learnt duration without all identifying to text at font by enjoying property.
Embodiment three
The present embodiment corresponding embodiment one or embodiment two, provide a kind of computer readable storage medium, are stored thereon with
Computer program, described program are able to achieve described in above-described embodiment one or embodiment two when being processed by the processor based on area
The step of OCR deep learning method of block chain mechanism is included.Specific step content without repeating, for further details, please refer to herein
The record of embodiment one or embodiment two.
It corresponds to those of ordinary skill in the art will appreciate that realizing in the technical solution of above-described embodiment one or embodiment two
All or part of the process, relevant hardware can be instructed to realize by computer program, the program can deposit
It is stored in a computer-readable storage medium, the program is when being executed, it may include such as the process of above-mentioned each method.
Wherein, the storage medium can be disk, optical disc, read-only memory (Read-Only Memory,
) or random access memory (Random Access Memory, RAM) etc. ROM.
In conclusion provided by the invention based on the OCR deep learning method of block chain mechanism, storage medium, Ke Yishi
The resource-sharing of existing learning data；Effective protection can be carried out to self-teaching achievement, improve learning value, while help to add
Fast research and development progress；It can also guarantee user data privacy.Thus has wide utilization prospect.
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalents made by bright specification and accompanying drawing content are applied directly or indirectly in relevant technical field, similarly include
In scope of patent protection of the invention.
Claims (7)
1. the OCR deep learning method based on block chain mechanism characterized by comprising
According to learning data creation by each state and its it is corresponding return the model constituted instantly, the corresponding study number of each state
Each sub- learning data in；
Privately owned storage model to block chain network each node；
The state selected in model at random is trained, and obtains the corresponding new return of state；
According to the new return Optimized model, the training pattern of the corresponding model is obtained, and its privately owned is stored to each node；
The corresponding training pattern collection of the model enough according to quantity carries out convolution deep learning；
The result of shared storage convolution deep learning is to each node.
2. as described in claim 1 based on the OCR deep learning method of block chain mechanism, which is characterized in that the acquisition pair
The training pattern of the model is answered, specifically:
Model after fixed optimization, using the model after fixation as the training pattern of the model.
3. as described in claim 1 based on the OCR deep learning method of block chain mechanism, which is characterized in that further include: it is shared
Store each node that OCR identifies data to block chain network.
4. as described in claim 1 based on the OCR deep learning method of block chain mechanism, which is characterized in that described according to
Practise data creation by each state and its it is corresponding return the model constituted instantly, each state corresponds to each sub- in learning data
Data are practised, specifically:
The learning data of an individual character is chosen from OCR identification data；
The sub- learning data of predetermined number is randomly selected from the learning data as state, learning outcome is corresponding as state
Return instantly, creation is by multiple states and its corresponding returns the model corresponding with an individual character that constitutes instantly.
5. as claimed in claim 4 based on the OCR deep learning method of block chain mechanism, which is characterized in that described according to number
It measures the corresponding training pattern collection of enough models and carries out convolution deep learning, specifically:
According to covering all stroke features of font the corresponding training pattern collection of multiple models, by 5 layers of convolutional neural networks into
Row OCR deep learning.
6. as described in claim 1 based on the OCR deep learning method of block chain mechanism, which is characterized in that described according to institute
New return Optimized model is stated, specifically:
By bellman equation dynamic optimization model, obtains by the state closest to model and its corresponding return the instruction constituted
Practice model.
7. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that described program is processed
When device processing, it is able to achieve the OCR deep learning method institute described in the claims 1-6 any one based on block chain mechanism
The step of including.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811054240.4A CN109344823B (en) | 2018-09-11 | 2018-09-11 | OCR deep learning method based on block chain mechanism and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811054240.4A CN109344823B (en) | 2018-09-11 | 2018-09-11 | OCR deep learning method based on block chain mechanism and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109344823A true CN109344823A (en) | 2019-02-15 |
| CN109344823B CN109344823B (en) | 2022-06-07 |
Family
ID=65305212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811054240.4A Active CN109344823B (en) | 2018-09-11 | 2018-09-11 | OCR deep learning method based on block chain mechanism and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109344823B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197285A (en) * | 2019-05-07 | 2019-09-03 | 清华大学 | Security cooperation deep learning method and device based on block chain |
| CN110503202A (en) * | 2019-08-22 | 2019-11-26 | 联想(北京)有限公司 | A kind of information processing method and electronic equipment |
| CN112612641A (en) * | 2020-12-16 | 2021-04-06 | 苏州浪潮智能科技有限公司 | Protection method and device for model training, electronic equipment and storage medium |
| WO2022105297A1 (en) * | 2020-11-17 | 2022-05-27 | 深圳壹账通智能科技有限公司 | Table structure recovery method and system, computer device, and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| BE1024386B1 (en) * | 2016-12-22 | 2018-02-05 | Itext Group Nv | Distributed blockchain-based method for jointly signing a PDF-based document by multiple parties |
| CN108090443A (en) * | 2017-12-15 | 2018-05-29 | 华南理工大学 | Scene text detection method and system based on deeply study |
-
2018
- 2018-09-11 CN CN201811054240.4A patent/CN109344823B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| BE1024386B1 (en) * | 2016-12-22 | 2018-02-05 | Itext Group Nv | Distributed blockchain-based method for jointly signing a PDF-based document by multiple parties |
| CN108090443A (en) * | 2017-12-15 | 2018-05-29 | 华南理工大学 | Scene text detection method and system based on deeply study |
Non-Patent Citations (2)
| Title |
|---|
| HYESUNG KIM: "《On-Device Federated Learning via Blockchain and its Latency Analysis》", 《ARXIV》 * |
| 王一扬: "为区块链作画", 《新理财》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110197285A (en) * | 2019-05-07 | 2019-09-03 | 清华大学 | Security cooperation deep learning method and device based on block chain |
| CN110197285B (en) * | 2019-05-07 | 2021-03-23 | 清华大学 | Block chain-based safe cooperation deep learning method and device |
| CN110503202A (en) * | 2019-08-22 | 2019-11-26 | 联想(北京)有限公司 | A kind of information processing method and electronic equipment |
| WO2022105297A1 (en) * | 2020-11-17 | 2022-05-27 | 深圳壹账通智能科技有限公司 | Table structure recovery method and system, computer device, and storage medium |
| CN112612641A (en) * | 2020-12-16 | 2021-04-06 | 苏州浪潮智能科技有限公司 | Protection method and device for model training, electronic equipment and storage medium |
| CN112612641B (en) * | 2020-12-16 | 2022-12-02 | 苏州浪潮智能科技有限公司 | Protection method and device for model training, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109344823B (en) | 2022-06-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109344823A (en) | Based on the OCR deep learning method of block chain mechanism, storage medium | |
| Leifsen et al. | New mechanisms of participation in extractive governance: between technologies of governance and resistance work | |
| Carothers | Democracy aid at 25: Time to choose | |
| Pyle et al. | An executive’s guide to machine learning | |
| Kao et al. | Hierarchical aesthetic quality assessment using deep convolutional neural networks | |
| WO2019100723A1 (en) | Method and device for training multi-label classification model | |
| WO2019100724A1 (en) | Method and device for training multi-label classification model | |
| Cox | How artificial intelligence might change academic library work: Applying the competencies literature and the theory of the professions | |
| Salehi-Abari et al. | On purely automated attacks and click-based graphical passwords | |
| CN110096641A (en) | Picture and text matching process, device, equipment and storage medium based on image analysis | |
| CN104915972A (en) | Image processing apparatus, image processing method and program | |
| Vasilescu | Effective strategic decision making | |
| CN111951154B (en) | Picture generation method and device containing background and medium | |
| Datta et al. | Exploiting the human–machine gap in image recognition for designing CAPTCHAs | |
| Chan et al. | Metamorphic relation based adversarial attacks on differentiable neural computer | |
| CN108038484A (en) | Hollow identifying code method for quickly identifying | |
| Liu et al. | Pixel Privacy 2019: Protecting Sensitive Scene Information in Images. | |
| CN114549698A (en) | Text synthesis method and device and electronic equipment | |
| Powell et al. | Convergent minds: the evolution of cognitive complexity in nature | |
| CN107832271A (en) | Functional image method for drafting, device, equipment and computer-readable storage medium | |
| CN111242114A (en) | Character recognition method and device | |
| Cao et al. | Incremental learning for fine-grained image recognition | |
| Malhotra et al. | The contrasting shape representations that support object recognition in humans and CNNs | |
| CN114612989A (en) | Method and device for generating face recognition data set, electronic equipment and storage medium | |
| CN110721471B (en) | Virtual application object output method and device and computer storage medium |
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 |