CN109754016A - Image data intelligent identifying system - Google Patents
Image data intelligent identifying system Download PDFInfo
- Publication number
- CN109754016A CN109754016A CN201910007002.6A CN201910007002A CN109754016A CN 109754016 A CN109754016 A CN 109754016A CN 201910007002 A CN201910007002 A CN 201910007002A CN 109754016 A CN109754016 A CN 109754016A
- Authority
- CN
- China
- Prior art keywords
- image data
- training
- picture
- data
- trained
- 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
Abstract
The present invention relates to a kind of image data intelligent identifying systems comprising image data memory module for storing image data, and obtains the store path and picture identification of the image data；Image data labeling module for being standardized the image data is also used to that the image data is marked；Training module obtains a variety of training patterns for being trained to the multiple network model prestored；Identification module, for being classified according to the training pattern to the image data.Image data intelligent identifying system provided by the invention can be improved recognition efficiency, reduce cost.
Description
Technical field
The present invention relates to field of computer technology, more particularly to a kind of image data intelligent identifying system.
Background technique
Along with the continuous development of artificial intelligence, in civilian computer vision, video intelligent process field, in military affairs
Target decomposition identification, Intellisense field etc. has outstanding performance.
But the bottom frame being related at present using deep learning as the artificial intelligence approach of representative is multifarious, each research group
Team wants to carry out the intelligent algorithm for being suitable for respective application demand, to realize training, the identification of respective vision neural network
And verifying, it must just be constructed from data set, marking software exploitation, the deployment of deep learning bottom frame, computing resource pool scheduling
Etc. spend a large amount of energy material resources, it is at high cost.
Summary of the invention
The purpose of the present invention is to provide a kind of image data intelligent identifying systems, can calculate to avoid numerous and complicated intelligence
The deployment of method bottom, efficiently completes the research and development of intelligent algorithm, improves recognition efficiency, reduces cost.
The purpose of the present invention is achieved through the following technical solutions:
A kind of image data intelligent identifying system, the system comprises:
Image data memory module for storing image data, and obtains the store path and picture of the image data
Mark；
Image data labeling module is also used to for being standardized to the image data to the picture number
According to being marked；
Training module obtains a variety of training patterns for being trained to the multiple network model prestored；
Identification module, for being classified according to the training pattern to the image data.
The image data memory module includes: in one of the embodiments,
Image data storage unit, for storing the image data；
Image data processing unit, for handling functions using a variety of default pictures, to the image data of storage into
Row processing.
The image data memory module in one of the embodiments, further include:
Network protocol processing unit, for obtaining and the picture according to the uniform resource locator for having picture identification
Identify the corresponding data picture.
The image data labeling module includes: in one of the embodiments,
Picture query unit is looked into for receiving the picture query instruction for carrying the picture identification according to the picture
It askes instructions query image data corresponding with the picture identification and shows.
The block diagram mark module in one of the embodiments, further include:
Standardisation Cell, for receiving picture standardized instruction, according to the picture standardized instruction, to described in displaying
Image data is standardized.
The image data labeling module in one of the embodiments, further include:
Semantic segmentation marking unit carries out the image data shown for receiving semantic segmentation mark instructions
Semantic segmentation label；
Irregular marking unit does not advise the image data shown for receiving irregular mark instructions
Then mark.
The training module includes: in one of the embodiments,
Data capture unit is updated, for obtaining the image data updated and caching；
Data volume acquiring unit, the number of the image data for obtaining the update cached in the update data capture unit
According to amount.
The training module in one of the embodiments, further include:
Model storage unit, for storing multiple network structure, wherein the network structure includes housebroken training
Network structure and the network structure to be trained not being trained also.
The training module in one of the embodiments, further include:
Training unit, the quantity of the image data of the update for obtaining when the data volume acquiring unit are more than pre-
If when threshold value, using the image data of the update cached as training input data, inputting training network structure, being instructed
Practice classification results.
The training unit is also used to according to the trained input data and the training point in one of the embodiments,
Class is as a result, be trained the network structure to be trained.
Image data intelligent identifying system provided by the invention, image data memory module, for storing image data,
And obtain the store path and picture identification of the image data；Image data labeling module, for the image data into
Row standardization is also used to that the image data is marked；Training module, for the multiple network model prestored into
Row training, obtains a variety of training patterns；Identification module, for being classified according to the training pattern to the image data；
It can be disposed to avoid numerous and complicated intelligent algorithm bottom, efficiently complete the research and development of intelligent algorithm, improve efficiency, dropped
Low cost.
Detailed description of the invention
Fig. 1 is the structural block diagram of image data intelligent identifying system in one embodiment；
Fig. 2 is the structural block diagram of military products data acquisition module in one embodiment；
Fig. 3 is the structural block diagram of military products data memory module in one embodiment；
Fig. 4 is the structural block diagram of military products data memory module in one embodiment.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention more comprehensible, with reference to the accompanying drawings and embodiments, to this
Invention is described in further detail.It should be appreciated that the specific embodiments described herein are only used to explain the present invention,
And the scope of protection of the present invention is not limited.
In one embodiment, as shown in Figure 1, providing a kind of image data intelligent identifying system, comprising:
Image data memory module 101 for storing image data, and obtains the store path and figure of the image data
Piece mark.
In the specific implementation process, realize that image data dispersion is stored in more by distributed storage using multiple servers
In independent equipment, storage load is shared jointly.
Image data labeling module 102 is also used to for being standardized to the image data to the picture
Data are marked.
In the specific implementation process, image data labeling module 102 can realize the cloud production of data set, have very strong
Configurability, versatility and flexibility have friendly man-machine interaction and label use value.Block diagram label realizes online
The production of data set cloud, supports cloud data set to check and provide picture standardization.It breaks through semantic segmentation label and does not advise
Creation is then marked, supports multiple person cooperational label, improves the pretreatment efficiency such as research team's data mark.
Training module 103 obtains a variety of training patterns for being trained to the multiple network model prestored.
In the specific implementation process, the compatible a variety of isomery intelligent recognition algorithm frames of training module 103 are, it can be achieved that training
Data classification, the flow setting of algorithm towed, fining arameter optimization, data increment expand training, training parameter and network layer
The intelligent boot flow of grade visualization output.By optimization design, increase isomery intelligent framework type, realizes and be based on increment
The expansion training of habit, and carry out the more resource parallel computations acceleration of multitask etc..By optimization high-performance calculation training, use can be formed
In the high network quality model of intelligent recognition.
Identification module 104, for being classified according to the training pattern to the image data.
Image data intelligent identifying system provided by the invention, image data memory module, for storing image data,
And obtain the store path and picture identification of the image data；Image data labeling module, for the image data into
Row standardization is also used to that the image data is marked；Training module, for the multiple network model prestored into
Row training, obtains a variety of training patterns；Identification module, for being classified according to the training pattern to the image data；
It can be disposed to avoid numerous and complicated intelligent algorithm bottom, efficiently complete the research and development of intelligent algorithm, improve efficiency, dropped
Low cost.
In one of the embodiments, as shown in Fig. 2, the image data memory module 101 includes:
Image data storage unit 110, for storing the image data.
In the specific implementation process, realize that image data dispersion is stored in more by distributed storage using multiple servers
In independent equipment, storage load is shared jointly；Storage is using memcached caching plus the scheme of SSDB storage.
Wherein, Memcached is a high performance distributed memory target cache system, for dynamic network apply with
Mitigate database loads；It reduces the number of reading database by data cached in memory and object, to improve dynamic
State, the speed of database-driven website.
Wherein, SSDB is a set of non-relational database, is more suitable for the storage of mass data.
In the specific implementation process, it also needs to save mark information while image data storage unit 110 saves picture,
By using the scheme of the corresponding tables of data of a data set, mark information and the distribution task dispatching information of picture are stored
Into traditional SQL database, can by way of tables of data and image identification rapidly locating concentrate picture.
Wherein, SQL is the abbreviation of Structured Query Language (structured query language).SQL is to aim at number
The operation commands set established according to library is a kind of multiple functional database language.
Image data processing unit 120, for handling function using a variety of default pictures, to the image data of storage
It is handled.
In the specific implementation process, picture processing uses the library imagemagick, and imagemagick is to generally acknowledge function now
Most strong, the best picture of performance handles function library.
In one of the embodiments, as shown in Fig. 2, the image data memory module further include:
Network protocol processing unit 130, for obtaining and the figure according to the uniform resource locator for having picture identification
Piece identifies the corresponding data picture.
In the specific implementation process, network protocol processing unit 130 is for handling http, wherein http is network protocol.
Wherein, uniform resource locator is one kind of the position and access method to the resource that can be obtained from internet
Succinct expression is the address of standard resource on internet.Each file on internet has a unique unified resource
Finger URL, the information that it includes point out how the position of file and browser should handle it.
In one of the embodiments, as shown in figure 3, the image data labeling module 102 includes:
Picture query unit 210, for receiving the picture query instruction for carrying the picture identification, according to the picture
Inquiry instruction is inquired image data corresponding with the picture identification and is shown.
Wherein, due to also preserving mark information while image data storage unit 110 saves picture, by using one
The scheme of the corresponding tables of data of a data set, mark information and the distribution task dispatching information of picture are stored to traditional
In SQL database, therefore, picture query unit 210 can be concentrated rapidly locating by way of tables of data and image identification
Picture.
In one of the embodiments, as shown in figure 3, the block diagram mark module 102 further include:
Standardisation Cell 220, for receiving picture standardized instruction, according to the picture standardized instruction, to displaying
The image data is standardized.
Wherein, picture standardization is by image data by going mean value to realize the processing of centralization, according to convex optimum theory
With data probability distributions relevant knowledge, data center meets data distribution rule, it is easier to obtain extensive effect after training
Fruit, picture standardization is one of pretreated common methods of image data.
In one of the embodiments, as shown in figure 3, the image data labeling module 102 further include:
Semantic segmentation marking unit 230, for receiving semantic segmentation mark instructions, to the image data shown into
Row semantic segmentation label.
Wherein, image data is made of many pixels, and semantic segmentation exactly contains pixel according to expression semanteme in image
The difference of justice is grouped or divides；Machine is divided automatically and identifies the content in image, for example provides a people and ride motor
The photo of vehicle should can generate right part of flg after machine judgement, red mark is people, and green is vehicle.
In the specific implementation process, one or more users send semantic segmentation mark instructions to server, server root
According to semantic segmentation mark instructions, pixel is grouped or is divided according to the difference for expressing semantic meaning in image data, is identified
Content in image out.
Irregular marking unit 240 carries out not the image data shown for receiving irregular mark instructions
Regular marks.
In the specific implementation process, multiple user collaborations can be supported to mark, improve the pre- places such as research team's data mark
Manage efficiency.
In one of the embodiments, as shown in figure 4, the training module 103 includes:
Data capture unit 310 is updated, for obtaining the image data updated and caching.
In the specific implementation process, expansion training method sustainable utilization of the training module 103 based on incremental learning is new
On-board data is trained, and the target signature learnt automatically is added in training process, therefore, updates data acquisition list
Member 310 constantly obtains the image data updated and caching.
Data volume acquiring unit 320, for obtaining the image data of the update cached in the update data capture unit
Data volume.
In implementation process, when the data volume that image data is updated is more than preset threshold, so that it may more by caching
New image data introducing is trained；When the data volume that image data is updated is no more than preset threshold, continue buffer update
Image data.
In one of the embodiments, as shown in figure 4, the training module 103 further include:
Model storage unit 330, for storing multiple network structure, wherein the network structure includes housebroken
Training network structure and the network structure to be trained not being trained also.
In the specific implementation process, housebroken network structure is obtained for classifying to the image data of update
Criteria classification as a result, then by the image data of update and criteria classification as a result, to also untrained network structure to be trained into
Row training.
In the specific implementation process, it during housebroken network structure is to image data classification, can also obtain
The label manually added carries out manual sort to uncertain classification results.
In one of the embodiments, as shown in figure 4, the training module further include:
The quantity of training unit 340, the image data of the update for obtaining when the data volume acquiring unit is super
When crossing preset threshold, using the image data of the update cached as training input data, training network structure is inputted, is obtained
To training classification results.
In the specific implementation process, training unit 340 is inputted the image data of the update cached as training
Data input training network structure, obtain training classification results；The process that housebroken network structure classifies to image data
In, the label manually added can also be obtained, manual sort is carried out to uncertain classification results.
The training unit 340 is also used to according to the trained input data and the instruction in one of the embodiments,
Practice classification results, the network structure to be trained is trained.
In the specific implementation process, training unit 340 is by the image data of update and the classification results exported, to also
Untrained network structure to be trained is trained；The label manually added can also be obtained, uncertain classification is tied
Fruit carries out manual sort, and the label that then will acquire inputs network to be trained and is trained as a result, treating trained network structure.
In the specific implementation process, in classification the most determine a few class data will be directly inputted in neural network continue into
Row training, and uncertain data are then input in network by manually adding label and are trained, to reduce artificial addition to the greatest extent
For the purpose of label.Continuous loop iteration aforesaid operations, to achieve the purpose that really intelligent continuous learning.
Each technical characteristic of above embodiments can be combined arbitrarily, for simplicity of description, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield all should be considered as described in this specification.
The embodiments described above only express several embodiments of the present invention, and the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, without departing from the inventive concept of the premise, various modifications and improvements can be made, these belong to protection of the invention
Range.Therefore, the scope of protection of the patent of the invention shall be subject to the appended claims.
Claims (10)
1. a kind of image data intelligent identifying system, which is characterized in that the system comprises:
Image data memory module for storing image data, and obtains the store path and picture identification of the image data；
Image data labeling module, for being standardized to the image data, be also used to the image data into
Line flag；
Training module obtains a variety of training patterns for being trained to the multiple network model prestored；
Identification module, for being classified according to the training pattern to the image data.
2. system according to claim 1, which is characterized in that the image data memory module includes:
Image data storage unit, for storing the image data；
Image data processing unit, for handling function using a variety of default pictures, at the image data of storage
Reason.
3. system according to claim 2, which is characterized in that the image data memory module further include:
Network protocol processing unit, for obtaining and the picture identification according to the uniform resource locator for having picture identification
The corresponding data picture.
4. system according to claim 1, which is characterized in that the image data labeling module includes:
Picture query unit refers to for receiving the picture query instruction for carrying the picture identification according to the picture query
It enables and inquires image data corresponding with the picture identification and show.
5. system according to claim 4, which is characterized in that the block diagram mark module further include:
Standardisation Cell, for receiving picture standardized instruction, according to the picture standardized instruction, to the picture of displaying
Data are standardized.
6. system according to claim 4, which is characterized in that the image data labeling module further include:
Semantic segmentation marking unit carries out the image data shown semantic for receiving semantic segmentation mark instructions
Dividing mark；
Irregular marking unit irregularly marks the image data shown for receiving irregular mark instructions
Note.
7. system according to claim 1, which is characterized in that the training module includes:
Data capture unit is updated, for obtaining the image data updated and caching；
Data volume acquiring unit, the data of the image data for obtaining the update cached in the update data capture unit
Amount.
8. system according to claim 7, which is characterized in that the training module further include:
Model storage unit, for storing multiple network structure, wherein the network structure includes housebroken trained network
Structure and the network structure to be trained not being trained also.
9. system according to claim 8, which is characterized in that the training module further include:
Training unit, the quantity of the image data of the update for obtaining when the data volume acquiring unit are more than default threshold
When value, using the image data of the update cached as training input data, training network structure is inputted, training point is obtained
Class result.
10. system according to claim 9, which is characterized in that the training unit is also used to according to the training input
Data and the trained classification results, are trained the network structure to be trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910007002.6A CN109754016A (en) | 2019-01-04 | 2019-01-04 | Image data intelligent identifying system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910007002.6A CN109754016A (en) | 2019-01-04 | 2019-01-04 | Image data intelligent identifying system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109754016A true CN109754016A (en) | 2019-05-14 |
Family
ID=66404419
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910007002.6A Pending CN109754016A (en) | 2019-01-04 | 2019-01-04 | Image data intelligent identifying system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109754016A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112836714A (en) * | 2019-11-22 | 2021-05-25 | 杭州海康威视数字技术股份有限公司 | Intelligent model training method and device |
| CN114330714A (en) * | 2022-03-08 | 2022-04-12 | 北京环境特性研究所 | Convolutional neural network pruning optimization method and device, electronic equipment and storage medium |
| CN112836714B (en) * | 2019-11-22 | 2024-05-10 | 杭州海康威视数字技术股份有限公司 | Training method and device for intelligent model |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103069415A (en) * | 2010-07-02 | 2013-04-24 | 埃森哲环球服务有限公司 | A computer-implemented method, a computer program product and a computer system for image processing |
| US20170177943A1 (en) * | 2015-12-21 | 2017-06-22 | Canon Kabushiki Kaisha | Imaging system and method for classifying a concept type in video |
| US20170228616A1 (en) * | 2015-02-05 | 2017-08-10 | Tolga Tasdizen | Systems and methods for image classification |
| CN108734211A (en) * | 2018-05-17 | 2018-11-02 | 腾讯科技（深圳）有限公司 | The method and apparatus of image procossing |
-
2019
- 2019-01-04 CN CN201910007002.6A patent/CN109754016A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103069415A (en) * | 2010-07-02 | 2013-04-24 | 埃森哲环球服务有限公司 | A computer-implemented method, a computer program product and a computer system for image processing |
| US20170228616A1 (en) * | 2015-02-05 | 2017-08-10 | Tolga Tasdizen | Systems and methods for image classification |
| US20170177943A1 (en) * | 2015-12-21 | 2017-06-22 | Canon Kabushiki Kaisha | Imaging system and method for classifying a concept type in video |
| CN108734211A (en) * | 2018-05-17 | 2018-11-02 | 腾讯科技（深圳）有限公司 | The method and apparatus of image procossing |
Non-Patent Citations (1)
| Title |
|---|
| 企鹅号小编: "TensorFlow图像分类教程", 《云+社区》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112836714A (en) * | 2019-11-22 | 2021-05-25 | 杭州海康威视数字技术股份有限公司 | Intelligent model training method and device |
| CN112836714B (en) * | 2019-11-22 | 2024-05-10 | 杭州海康威视数字技术股份有限公司 | Training method and device for intelligent model |
| CN114330714A (en) * | 2022-03-08 | 2022-04-12 | 北京环境特性研究所 | Convolutional neural network pruning optimization method and device, electronic equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Wang et al. | Learning to combine: Knowledge aggregation for multi-source domain adaptation | |
| Leroux et al. | The cascading neural network: building the internet of smart things | |
| CN112559007B (en) | Parameter updating method and device of multitask model and electronic equipment | |
| CN110059807A (en) | Image processing method, device and storage medium | |
| CN110413786B (en) | Data processing method based on webpage text classification, intelligent terminal and storage medium | |
| CN106095966B (en) | User extensible label labeling method and system | |
| CN112016633A (en) | Model training method and device, electronic equipment and storage medium | |
| Fang et al. | Deep learning for automatic recognition of magnetic type in sunspot groups | |
| CN108241867B (en) | Classification method and device | |
| CN108197656A (en) | A kind of attribute reduction method based on CUDA | |
| CN109816001A (en) | A kind of more attribute recognition approaches of vehicle based on deep learning, device and equipment | |
| CN109754016A (en) | Image data intelligent identifying system | |
| CN108304253A (en) | Map method for scheduling task based on cache perception and data locality | |
| CN105069133B (en) | A kind of digital picture sorting technique based on Unlabeled data | |
| Olorunnimbe et al. | Intelligent adaptive ensembles for data stream mining: a high return on investment approach | |
| DE102022126287A1 (en) | STORAGE AND RETRIEVAL OF UNSTRUCTURED DATA IN CONVERSATIONAL ARTIFICIAL INTELLIGENCE APPLICATIONS | |
| CN109523031A (en) | A kind of big data intelligence machine learning system for depth analysis | |
| Jiao et al. | Plant leaf recognition based on conditional generative adversarial nets | |
| CN109977881B (en) | Character action feature extraction and identification optimization method based on radio frequency technology | |
| CN108960291A (en) | A kind of image processing method and system based on parallelization Softmax classification | |
| Rudorfer | Towards robust object detection and pose estimation as a service for manufacturing industries | |
| CN111290855A (en) | GPU card management method, system and storage medium for multiple GPU servers in distributed environment | |
| CN113344097B (en) | Image processing method and device based on multiple models | |
| Liu et al. | Learn a deep convolutional neural network for image smoke detection | |
| Arsenovic et al. | Deep learning powered automated tool for generating image based datasets |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190514 |
|
| RJ01 | Rejection of invention patent application after publication |