CN108984629A - A kind of model training method and system - Google Patents
A kind of model training method and system Download PDFInfo
- Publication number
- CN108984629A CN108984629A CN201810637462.2A CN201810637462A CN108984629A CN 108984629 A CN108984629 A CN 108984629A CN 201810637462 A CN201810637462 A CN 201810637462A CN 108984629 A CN108984629 A CN 108984629A
- Authority
- CN
- China
- Prior art keywords
- training
- image data
- training pattern
- crawler
- module
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- 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
- G06V20/68—Food, e.g. fruit or vegetables
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of model training method and systems, this method comprises: step S1 crawls the image data for meeting default crawler condition；Described image data are input to first training pattern by step S2, filter out training set data；Step S3 is trained the first training pattern using the training set data, obtains final training pattern.The present invention automatically cleans image data by first training pattern, and it is reliable to reduce labor workload, data；And it is voluntarily iterated update, model is optimized automatically, the accuracy of obtained final training pattern is higher, meets project demand.
Description
Technical field
The present invention relates to model training field more particularly to a kind of model training method and systems.
Background technique
When the AI (Artificial Intelligence, artificial intelligence) of picture recognition carries out model training at present, use
To a large amount of training data, these training datas are by manually being cleaned to obtain.
The process workload that artificial cleaning obtains training data is extremely great, and due to subjective impression when everyone screens
Difference, the training data quality disunity for causing artificial screening to come out.Such as: wax gourd rib soup, only rib soup on picture do not have
There is wax gourd, but picture name has write wax gourd rib soup, somebody thinks that this can leave, and somebody feels to stay
Under.
The quality of training data is related to the accuracy of entire training pattern, uses the training data training mould manually cleaned
Type can make the model accuracy trained lower, and training process larger workload.
Summary of the invention
The object of the present invention is to provide a kind of model training method and systems, improve the accuracy of trained model, reduce
Labor workload.
Technical solution provided by the invention is as follows:
A kind of model training method, comprising: step S1 crawls the image data for meeting default crawler condition；Step S2 is by institute
It states image data and is input to first training pattern, filter out training set data；Step S3 is using the training set data to described
First training pattern is trained, and obtains final training pattern.
In the above-mentioned technical solutions, training set is automatically and quickly filtered out from image data by first training pattern
Data reduce manual intervention, and judge index is consistent, improve the accuracy of training set data.
Further, further includes: step S4 when the accuracy of the final training pattern not up to default accuracy threshold value,
The first training pattern in the step S2, and the S1 that gos to step are updated using the final training pattern.
In the above-mentioned technical solutions, it is voluntarily iterated update, model is optimized automatically, obtained final trained mould
The accuracy of type is higher, meets project demand.
Further, in the step S2 not by the first training pattern that the final training pattern updates be using
Typical picture training is got.
In the above-mentioned technical solutions, the first training pattern used at the beginning is defined, lays base for subsequent loop iteration
Plinth.
Further, the step S2 specifically: described image data are input to first training pattern by step S21, are obtained
The confidence level of each picture in described image data；Step S22 filters out the confidence bit in the figure of default confidence range
Piece is as the training set data.
In the above-mentioned technical solutions, default confidence range can self-setting, make training set data reliability be in can
Control state.
Further, the step S1 specifically: step S11 is crawled and institute according to the principal name in default crawler condition
State the relevant described image data of principal name.
In the above-mentioned technical solutions, the difference of crawler condition is preset, the dependent image data crawled out is also different, can apply
In the crawler demand of different types of data.
The present invention also provides a kind of model training systems, comprising: crawler module meets default crawler condition for crawling
Image data；Screening module filters out training set data for described image data to be input to first training pattern；Training
Module obtains final training pattern for being trained using the training set data to the first training pattern.
In the above-mentioned technical solutions, training set is automatically and quickly filtered out from image data by first training pattern
Data reduce manual intervention, and judge index is consistent, improve the accuracy of training set data.
Further, further includes: update module, for when the not up to default accuracy of the accuracy of the final training pattern
When threshold value, the first training pattern in the screening module is updated using the final training pattern；The crawler module, when making
After updating the first training pattern in the screening module with the final training pattern, the crawler module for crawling again
Meet the image data of default crawler condition.
Further, the first training pattern not updated by the final training pattern in the screening module is to make
It is got with typical picture training.
Further, screening module specifically includes: confidence level submodule, for described image data to be input to first training
Model obtains the confidence level of each picture in described image data；Submodule is screened, for filtering out the confidence bit in pre-
If the picture of confidence range is as the training set data.
Further, the crawler module, for crawling the image data for meeting default crawler condition specifically: the crawler
Module, for crawling described image data relevant to the principal name according to the principal name preset in crawler condition.
Compared with prior art, model training method of the invention and system beneficial effect are:
The present invention automatically cleans image data by first training pattern, and it is reliable to reduce labor workload, data；
And it is voluntarily iterated update, model is optimized automatically, the accuracy of obtained final training pattern is higher, meets project
It is required that.
Detailed description of the invention
Below by clearly understandable mode, preferred embodiment is described with reference to the drawings, to a kind of model training method and
Above-mentioned characteristic, technical characteristic, advantage and its implementation of system are further described.
Fig. 1 is the flow chart of model training method one embodiment of the present invention；
Fig. 2 is the flow chart of another embodiment of model training method of the present invention；
Fig. 3 is the structural schematic diagram of model training systems one embodiment of the present invention；
Fig. 4 is the structural schematic diagram of another embodiment of model training systems of the present invention.
Drawing reference numeral explanation:
10. crawler module, 20. screening modules, 21. confidence level submodules, 22. screening submodules, 30. training modules, 40.
Update module.
Specific embodiment
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, Detailed description of the invention will be compareed below
A specific embodiment of the invention.It should be evident that drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing, and obtain other embodiments.
To make simplified form, part related to the present invention is only schematically shown in each figure, they are not represented
Its practical structures as product.In addition, there is identical structure or function in some figures so that simplified form is easy to understand
Component only symbolically depicts one of those, or has only marked one of those.Herein, "one" is not only indicated
" only this ", can also indicate the situation of " more than one ".
Model training method and system of the invention, can be applied to text, also can be applied to picture, mainly adopts below
It is laid down a definition with volume nerual network technique to the model training of picture recognition.
In one embodiment of the invention, as shown in Figure 1, a kind of model training method, comprising:
Step S1 crawls the image data for meeting default crawler condition；
Image data is input to first training pattern by step S2, filters out training set data；
Step S3 is trained first training pattern using training set data, obtains final training pattern.
Firstly, picture concerned is crawled by crawler module, as image data.Picture concerned is according to default crawler condition
Crawled, such as: if default crawler condition be it is related to vegetable, crawl picture relevant to vegetable as picture number
According to.
Step S1 specifically: step S11 is crawled relevant to principal name according to the principal name in default crawler condition
Image data.
Specifically, principal name is set in default crawler condition, and such as: tomato, egg, green vegetables etc. also can be set
" vegetable " is used as principal name.Crawler module crawls relevant picture as image data according to the principal name of setting, such as:
It crawls title and contains the picture of tomato, egg or green vegetables, or, crawling all pictures under related web site under vegetable list.
Crawler module can go to crawl to related web site data relevant to default crawler condition, related web site by engineer from
Row setting, the quantity of related web site can be it is multiple, when each crawler module is run, only crawled from one or more related web sites
Image data relevant to default crawler condition, again run crawler module when, from another or other multiple websites crawl with
The default relevant image data of crawler condition.Data may be crawled for sequential selection related web site, correlation also can be randomly selected
Network crawls data, as long as can be crawled from different places every time to the image data for meeting default crawler condition.
Image data is cleaned by first training pattern, filters out training set data.
The image data that crawler module crawls has very big inaccuracy, and existing technology is by manually carrying out to figure
As data are screened, comparison between projects are great, and the standard disunity of training set data.And the present invention uses first training pattern
Automatically screening is carried out, reduces manual labor, and judgment criteria is unitized, obtained training set data manual intervention degree drop
It is low.
Here first training pattern is got by typical picture training.Typical picture collection is by several by artificial
The picture composition filtered out.Such as: it goes to train by the typical picture collection that 8 typical pictures form, obtains first training module.
The one first training module of picture training first gone out with a small amount of artificial screening, then thus first training module is gone automatically
It screens the image data crawled and obtains the process of training set data instead of the process of existing artificial screening image data
More rapidly.
After Automatic sieve selects training set data, first training pattern is trained using this training set data, is obtained
Final training pattern.The accuracy of training set data is much higher compared to the accuracy of image data, and schemes in training set data
The quantity of piece is also more much higher than the picture number that typical picture is concentrated, and is instructed using training set data to first training pattern
Practice, the output accuracy rate of obtained final training pattern can be higher than the accuracy rate of first training pattern.
In the present embodiment, training set data is automatically and quickly filtered out from image data by first training pattern, is subtracted
Lack manual intervention, and judge index is consistent, improves the accuracy of training set data.
In another embodiment of the present invention, as shown in Fig. 2, a kind of model training method, comprising:
Step S1 crawls the image data for meeting default crawler condition；
Image data is input to first training pattern by step S2, filters out training set data；
Step S3 is trained first training pattern using training set data, obtains final training pattern；
Step S4 is when the accuracy of final training pattern not up to default accuracy threshold value, more using final training pattern
First training pattern in new step S2, and the S1 that gos to step.
Specifically, having testing tool when obtaining final training pattern and being carried out to the accuracy of this final training pattern
Test, if having reached default accuracy threshold value, the model that final training pattern specifically obtained is just intended to, no longer need into
Row training；If not up to default accuracy threshold value, illustrates that the accuracy of current final training pattern also not up to requires, need
Continue to train.Default accuracy threshold value be arranged according to actual project demand, such as: 70%, or, 75%, or, 80%
Deng.
Final training pattern is updated to first training pattern, re-execute the steps S1-3, to updated first training
Model continues to train, if the final training pattern specifically obtained reaches default accuracy threshold value, terminates；If be still not up to
Default accuracy threshold value, then be updated to first training pattern with current final training pattern again, re-execute the steps S1-
3 ... are recycled with this, until obtained final training pattern meets default accuracy threshold value.
Step S1 specifically: step S11 is crawled relevant to principal name according to the principal name in default crawler condition
Image data.
Principal name is set in default crawler condition, such as: " vegetable " work also can be set in tomato, egg, green vegetables etc.
Based on title.Crawler module crawls relevant picture as image data according to the principal name of setting, such as: crawl title
Picture containing tomato, egg or green vegetables, or, crawling all pictures under related web site under vegetable list.
Crawler module can go to crawl to related web site data relevant to default crawler condition, related web site by engineer from
Row setting, the quantity of related web site can be it is multiple, when each crawler module is run, only crawl and preset from a related web site
Crawler condition relevant image data crawls related to default crawler condition when running crawler module again from another website
Image data.Data may be crawled for sequential selection related web site, network of relation also can be randomly selected and crawl data, as long as
It can be crawled from different places every time to the image data for meeting default crawler condition.
It should be noted that the first training pattern not updated by final training pattern in step S2 is using typical figure
Piece training is got.
The first training pattern of first time is that the pre- typical picture collection that first passes through is trained, and typical picture collection is by several
It is made of the picture that artificial screening goes out.Such as: it goes to train by the typical picture collection that 10 typical electric appliance pictures form, obtains one
The first training module started.
Step S2 specifically:
Image data is input to first training pattern by step S21, obtains the confidence level of each picture in image data；
Step S22 filter out confidence bit in default confidence range picture as training set data.
Specifically, first training pattern can be in image data when image data is input to first training pattern
Every picture provides confidence level, that is, illustrates the degree of correlation of this picture.
Confidence range is preset to be determined by the accuracy of project, such as: it is set as 80%-100%.According to every picture
Confidence level can filter out the higher picture of the degree of correlation as training set data, quickly, accurately at once.
The specific example model for wanting one identification vegetable of training is as follows:
1, presetting crawler condition is " vegetable ", and the first time image data (800 under vegetable catalogue is crawled from first website
Picture), the picture number crawled be it is random, determined according to the result that actually crawls.
2, first time image data (800 picture) is input in the first training pattern of first time, filters out confidence
Spend 80%-100% first time training set data (200 picture), wherein the first training pattern of first time be using
The typical vegetable picture training that 8 artificial screenings go out forms.
3, the first training pattern of first time is trained using the training set data of first time (200 picture), is obtained
To the final training pattern of first time.
If 4, the not up to default accuracy threshold value of the accuracy of the final training pattern of first time, i.e., 70%, then just
First training pattern is updated using the final training pattern of first time, obtains secondary first training pattern.
5, second of image data (900 picture) under vegetable catalogue is crawled from second website again.
6, second of image data (900 picture) is input in secondary first training pattern, filters out confidence
Spend secondary training set data (250 picture) in 80%-100%.
7, secondary first training pattern is trained using secondary training set data (250 picture), is obtained
To secondary final training pattern.
If 8, the accuracy of secondary final training pattern reaches 70%, terminate, this secondary final instruction
Practice the training pattern that model is just intended to；If the accuracy of secondary final training pattern is not up to 70%, the is used
Secondary final training pattern updates first training pattern, obtains the first training pattern ... of third time until obtaining accuracy
Reach 70% final training pattern.
The present embodiment automatically cleans image data by first training pattern, and reduction labor workload, data can
It leans on；And it is voluntarily iterated update, model is optimized automatically, the accuracy of obtained final training pattern is higher, meets
Project demand.
In one embodiment of the invention, as shown in figure 3, a kind of model training systems, comprising:
Crawler module 10, for crawling the image data for meeting default crawler condition；
Screening module 20 is electrically connected with crawler module 10, for image data to be input to first training pattern, is filtered out
Training set data；
Training module 30 is electrically connected with screening module 20, for being instructed using training set data to first training pattern
Practice, obtains final training pattern.
Firstly, picture concerned is crawled by crawler module 10, as image data.Picture concerned is according to default crawler item
What part was crawled, such as: if default crawler condition be it is related to vegetable, crawl picture relevant to vegetable as picture number
According to.
Crawler module 10, for crawling the image data for meeting default crawler condition specifically:
Crawler module 10, for crawling image relevant to principal name according to the principal name preset in crawler condition
Data.
Specifically, principal name is set in default crawler condition, and such as: tomato, egg, green vegetables etc. also can be set
" vegetable " is used as principal name.Crawler module 10 crawls relevant picture as image data, example according to the principal name of setting
Such as: crawling title and contain the picture of tomato, egg or green vegetables, or, crawling all pictures under related web site under vegetable list.
Crawler module 10 can go to crawl data relevant to default crawler condition to related web site, and related web site is by engineer
Self-setting, the quantity of related web site can be it is multiple, when each crawler module is run, only climbed from one or more related web sites
Image data relevant to default crawler condition is taken, when running crawler module again, is crawled from another or other multiple websites
Image data relevant to default crawler condition.Data may be crawled for sequential selection related web site, phase also can be randomly selected
It closes network and crawls data, as long as can be crawled from different places every time to the image data for meeting default crawler condition.
The image data that crawler module 10 crawls has very big inaccuracy, and existing technology is by manually carrying out pair
Image data is screened, and comparison between projects are great, and the standard disunity of training set data.And the present invention is using first training mould
Type carries out automatically screening, reduces manual labor, and judgment criteria is unitized, obtained training set data manual intervention degree drop
It is low, data reliability is high.
Here first training pattern is got by typical picture training.Typical picture collection is by several by artificial
The picture composition filtered out.Such as: it goes to train by the typical picture collection that 8 typical pictures form, obtains first training module.
The one first training module of picture training first gone out with a small amount of artificial screening, then thus first training module is gone automatically
It screens the image data crawled and obtains the process of training set data instead of the process of existing artificial screening image data
More rapidly.
After Automatic sieve selects training set data, first training pattern is trained using this training set data, is obtained
Final training pattern.The accuracy of training set data is much higher compared to the accuracy of image data, and schemes in training set data
The quantity of piece is also more much higher than the picture number that typical picture is concentrated, and is instructed using training set data to first training pattern
Practice, the output accuracy rate of obtained final training pattern can be higher than the accuracy rate of first training pattern.
In the present embodiment, training set data is automatically and quickly filtered out from image data by first training pattern, is subtracted
Lack manual intervention, and judge index is consistent, improves the accuracy of training set data.
In another embodiment of the present invention, as shown in figure 4, a kind of model training systems, comprising:
Crawler module 10, for crawling the image data for meeting default crawler condition；
Screening module 20 is electrically connected with crawler module 10, for image data to be input to first training pattern, is filtered out
Training set data；
Training module 30 is electrically connected with screening module 20, for being instructed using training set data to first training pattern
Practice, obtains final training pattern；
Update module 40 is electrically connected with training module 30 and crawler module 10, for working as the accuracy of final training pattern
When not up to presetting accuracy threshold value, final training pattern is used to update the first training pattern in screening module；
Crawler module 10, after updating the first training pattern in screening module using final training pattern, crawler module
For crawling the image data for meeting default crawler condition again.
Specifically, having testing tool when obtaining final training pattern and being carried out to the accuracy of this final training pattern
Test, if having reached default accuracy threshold value, the model that final training pattern specifically obtained is just intended to, no longer need into
Row training；If not up to default accuracy threshold value, illustrates that the accuracy of current final training pattern also not up to requires, need
Continue to train.Default accuracy threshold value be arranged according to actual project demand, such as: 70%, or, 75%, or, 80%
Deng.
Final training pattern is updated to first training pattern, re-executes crawler module, screening module, training module,
Updated first training pattern is continued to train, if the final training pattern specifically obtained reaches default accuracy threshold value,
Then terminate；If still not up to default accuracy threshold value, is updated to train mould for the first time with current final training pattern again
Type is re-executed crawler module, screening module, training module ... and is recycled with this, until obtained final training pattern meets
Default accuracy threshold value.
Crawler module 10, for crawling the image data for meeting default crawler condition specifically:
Crawler module 10, for crawling image relevant to principal name according to the principal name preset in crawler condition
Data.
Specifically, principal name is set in default crawler condition, and such as: tomato, egg, green vegetables etc. also can be set
" vegetable " is used as principal name.Crawler module 10 crawls relevant picture as image data, example according to the principal name of setting
Such as: crawling title and contain the picture of tomato, egg or green vegetables, or, crawling all pictures under related web site under vegetable list.
Crawler module can go to crawl to related web site data relevant to default crawler condition, related web site by engineer from
Row setting, the quantity of related web site can be it is multiple, when each crawler module is run, only crawled from one or more related web sites
Image data relevant to default crawler condition, again run crawler module when, from another or other multiple websites crawl with
The default relevant image data of crawler condition.Data may be crawled for sequential selection related web site, correlation also can be randomly selected
Network crawls data, as long as can be crawled from different places every time to the image data for meeting default crawler condition.
It should be noted that the first training pattern not updated by final training pattern in screening module is using typical case
Pictures training obtains.
The first training pattern of first time is that the pre- typical picture collection that first passes through is trained, and typical picture collection is by several
It is made of the picture that artificial screening goes out.Such as: it goes to train by the typical picture collection that 10 typical electric appliance pictures form, obtains one
The first training module started.
Screening module 20 specifically includes:
Confidence level submodule 21 obtains each figure in image data for image data to be input to first training pattern
The confidence level of piece；
Screen submodule 22, for filter out confidence bit in default confidence range picture as training set data.
Specifically, first training pattern can be in image data when image data is input to first training pattern
Every picture provides confidence level, that is, illustrates the degree of correlation of this picture.
Confidence range is preset to be determined by the accuracy of project, such as: it is set as 80%-100%.According to every picture
Confidence level can filter out the higher picture of the degree of correlation as training set data, quickly, accurately at once.
The specific implementation process of this system embodiment is identical as the specific implementation process in above method embodiment, herein not
It is described in detail again.
The present embodiment automatically cleans image data by first training pattern, and reduction labor workload, data can
It leans on；And it is voluntarily iterated update, model is optimized automatically, the accuracy of obtained final training pattern is higher, meets
Project demand.
It should be noted that above-described embodiment can be freely combined as needed.The above is only preferred implementations of the invention
Mode, it is noted that for those skilled in the art, without departing from the principle of the present invention, also
Several improvements and modifications can be made, these modifications and embellishments should also be considered as the scope of protection of the present invention.
Claims (10)
1. a kind of model training method characterized by comprising
Step S1 crawls the image data for meeting default crawler condition；
Described image data are input to first training pattern by step S2, filter out training set data；
Step S3 is trained the first training pattern using the training set data, obtains final training pattern.
2. model training method as described in claim 1, which is characterized in that further include:
Step S4 uses the final trained mould when the accuracy of the final training pattern not up to default accuracy threshold value
Type updates the first training pattern in the step S2, and the S1 that gos to step.
3. model training method as claimed in claim 2, which is characterized in that in the step S2 not by the final training
The first training pattern of model modification is got using typical picture training.
4. model training method as described in claim 1, which is characterized in that the step S2 specifically:
Described image data are input to first training pattern by step S21, obtain the confidence of each picture in described image data
Degree；
Step S22 filter out the confidence bit in default confidence range picture as the training set data.
5. the model training method as described in claim 1-4 any one, which is characterized in that the step S1 specifically: step
Rapid S11 crawls described image data relevant to the principal name according to the principal name in default crawler condition.
6. a kind of model training systems characterized by comprising
Crawler module, for crawling the image data for meeting default crawler condition；
Screening module filters out training set data for described image data to be input to first training pattern；
Training module obtains finally training mould for being trained the first training pattern using the training set data
Type.
7. model training systems as claimed in claim 6, which is characterized in that further include:
Update module, for when the accuracy of the final training pattern not up to default accuracy threshold value, using it is described most
Whole training pattern updates the first training pattern in the screening module；
The crawler module, after updating the first training pattern in the screening module using the final training pattern, institute
Crawler module is stated for crawling the image data for meeting default crawler condition again.
8. model training systems as claimed in claim 7, which is characterized in that in the screening module not by the final instruction
The first training pattern for practicing model modification is got using typical picture training.
9. model training systems as claimed in claim 6, which is characterized in that screening module specifically includes:
Confidence level submodule obtains each in described image data for described image data to be input to first training pattern
The confidence level of picture；
Screen submodule, for filter out the confidence bit in default confidence range picture as the training set number
According to.
10. the model training method as described in claim 6-9 any one, which is characterized in that the crawler module, for climbing
Take the image data for meeting default crawler condition specifically:
The crawler module, for crawling institute relevant to the principal name according to the principal name preset in crawler condition
State image data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810637462.2A CN108984629A (en) | 2018-06-20 | 2018-06-20 | A kind of model training method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810637462.2A CN108984629A (en) | 2018-06-20 | 2018-06-20 | A kind of model training method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108984629A true CN108984629A (en) | 2018-12-11 |
Family
ID=64541541
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810637462.2A Pending CN108984629A (en) | 2018-06-20 | 2018-06-20 | A kind of model training method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108984629A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872333A (en) * | 2019-02-20 | 2019-06-11 | 腾讯科技（深圳）有限公司 | Medical image dividing method, device, computer equipment and storage medium |
| CN109993079A (en) * | 2019-03-19 | 2019-07-09 | 嘉兴智设信息科技有限公司 | A kind of indoor design picture recognition and screening technique and its device |
| CN112541544A (en) * | 2020-12-09 | 2021-03-23 | 福州大学 | Garbage classification method based on deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080159622A1 (en) * | 2006-12-08 | 2008-07-03 | The Nexus Holdings Group, Llc | Target object recognition in images and video |
| CN106529564A (en) * | 2016-09-26 | 2017-03-22 | 浙江工业大学 | Food image automatic classification method based on convolutional neural networks |
| CN107958263A (en) * | 2017-11-13 | 2018-04-24 | 浙江工业大学 | A kind of semi-supervised Image Classifier training method |
-
2018
- 2018-06-20 CN CN201810637462.2A patent/CN108984629A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080159622A1 (en) * | 2006-12-08 | 2008-07-03 | The Nexus Holdings Group, Llc | Target object recognition in images and video |
| CN106529564A (en) * | 2016-09-26 | 2017-03-22 | 浙江工业大学 | Food image automatic classification method based on convolutional neural networks |
| CN107958263A (en) * | 2017-11-13 | 2018-04-24 | 浙江工业大学 | A kind of semi-supervised Image Classifier training method |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109872333A (en) * | 2019-02-20 | 2019-06-11 | 腾讯科技（深圳）有限公司 | Medical image dividing method, device, computer equipment and storage medium |
| CN109872333B (en) * | 2019-02-20 | 2021-07-06 | 腾讯科技（深圳）有限公司 | Medical image segmentation method, medical image segmentation device, computer equipment and storage medium |
| US11854205B2 (en) | 2019-02-20 | 2023-12-26 | Tencent Technology (Shenzhen) Company Limited | Medical image segmentation method and apparatus, computer device, and storage medium |
| CN109993079A (en) * | 2019-03-19 | 2019-07-09 | 嘉兴智设信息科技有限公司 | A kind of indoor design picture recognition and screening technique and its device |
| CN112541544A (en) * | 2020-12-09 | 2021-03-23 | 福州大学 | Garbage classification method based on deep learning |
| CN112541544B (en) * | 2020-12-09 | 2022-05-13 | 福州大学 | Garbage classification method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108984629A (en) | A kind of model training method and system | |
| DE10197407B4 (en) | Adaptive feedback / feedforward PID controller | |
| DE112012005419B4 (en) | System and method for efficient service-oriented energy management in the Internet of Things | |
| CN103605738B (en) | Web page access data statistical method and device | |
| CN112842149B (en) | Control method of intelligent cleaning equipment and intelligent cleaning equipment | |
| CN108111860B (en) | Video sequence lost frame prediction recovery method based on depth residual error network | |
| CN106650228B (en) | It improves the noise data minimizing technology of k-means algorithm and implements system | |
| DE102004057021A1 (en) | Computer system test method involves designating load and applying that load to resources corresponding to load specification | |
| Kristensen et al. | Phenology of two interdependent traits in migratory birds in response to climate change | |
| CN107894956A (en) | A kind of long-range BIOS promotion and demotion refresh automated testing method | |
| DE102010034546A1 (en) | Image processing apparatus and image processing method | |
| CN110058756A (en) | A kind of mask method and device of image pattern | |
| CN110087110A (en) | Using the method and device of deep search dynamic regulation video playing | |
| CN106383832B (en) | A kind of generation method of data mining training pattern | |
| CN104714718B (en) | The method and apparatus that operable icon is marked in interactive application | |
| CN108536793A (en) | A kind of method and system for preventing ajax requests from repeating to submit | |
| CN106303724B (en) | The method and apparatus that smart television adds dynamic expression automatically | |
| CN110490319A (en) | Distributed deeply study based on fused neural network parameter | |
| EP2804342A2 (en) | Method and device for clearing configuration command in communication equipment | |
| DE102011085301A1 (en) | Image processing device, image processing method, and image editing program | |
| CN109783728A (en) | Page crawler rule update method and system | |
| CN106776290A (en) | A kind of intelligent continuous integrating method of testing of incremental learning | |
| CN105912706A (en) | Method and device for improving rank of search engine | |
| EP3113574A1 (en) | Device for a cooking hob and method for operating a device for a cooking hob | |
| CN107391724A (en) | A kind of screening technique of big data |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181211 |