CN107679510A - A kind of bank client identifying system and method based on deep learning - Google Patents
A kind of bank client identifying system and method based on deep learning Download PDFInfo
- Publication number
- CN107679510A CN107679510A CN201710979947.5A CN201710979947A CN107679510A CN 107679510 A CN107679510 A CN 107679510A CN 201710979947 A CN201710979947 A CN 201710979947A CN 107679510 A CN107679510 A CN 107679510A
- Authority
- CN
- China
- Prior art keywords
- face
- characteristic information
- deep learning
- vip
- client
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Collating Specific Patterns (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of bank client identifying system and method based on deep learning, the system includes face acquisition module, face normalization module, characteristic information extracting module and VIP identification modules, face acquisition module automatically extracts the facial image of all personnel by the assembled classifier of a reverse pyramid from hall of bank scene photo, plurality of top-level categories device is rough sort device, and gradually fine and number gradually decreases each sub-classification device；Face normalization module is used to be corrected the facial image got；Characteristic information extracting module is used to extract the characteristic information in facial image；VIP identification modules are contrasted the characteristic information of extraction and VIP client's face characteristic information data storehouse, determine whether VIP client.Identity of the invention by automatic identification business handling crowd, improve the intelligence degree of service robot so that service robot can provide the intelligent, Site Service of hommization, improve customer experience.
Description
Technical field
The present invention relates to intellect service robot field, more particularly to a kind of bank client identification system based on deep learning
System and method.
Background technology
Bank welcome services humanoid robot as banking industries upgrading and the breach transformed, progressively in drug in some provinces
Bank outlets carry out trial operation, it is contemplated that tomorrow requirement will further enhance.Bank's welcome's service robot is different from simple
Meal delivery robot, in addition to welcome amuses function, more emphasize the functions such as business consultation, guiding shunting, sales publicity, therefore technology
Complexity and composite request are of a relatively high.
Traditional service robot only possesses simple business handling function, can not be directed to different clients and provide specific aim clothes
Business, is even more the service that can not be provided with differentiation for different levels client, and this just needs " wisdom " that improves service robot, made
It can be directed to different levels, different types of client provides the service customized.
The content of the invention
The technical problems to be solved by the invention are that existing service robot only possesses simple business handling function, no
The problem of customizing service can be provided for different levels, different types of client.
In order to solve the above-mentioned technical problem, the technical solution adopted in the present invention is to provide a kind of silver based on deep learning
Row user identification system, including：
Face acquisition module, it is described for automatically extracting the facial image of all personnel from hall of bank scene photo
Face acquisition module is extracted by the assembled classifier of a reverse pyramid, multiple top layers in the assembled classifier
Grader is rough sort device, and gradually fine and number gradually decreases each sub-classification device：
Face normalization module, for being corrected to the facial image got；
Characteristic information extracting module, for extracting the characteristic information in facial image；
VIP identification modules, the characteristic information of extraction and VIP client's face characteristic information data storehouse are contrasted, judged
Whether it is VIP client.
In said system, the face normalization module uses multiple dimensioned face normalization method, based on by slightly to essence
Architectural framework carries out face normalization, and each yardstick carries out school using general deep learning network to low-resolution image
Just, then correction result is inputted as upper strata, be corrected step by step.
In said system, the characteristic information extracting module is using the face characteristic extraction mould based on deep learning framework
Type extracts face characteristic information, and the face characteristic extraction model based on deep learning framework includes 7 layers of convolutional layer and 3 layers of full connection
Layer, the convolution kernel size of initial convolutional layer is 9 × 9, and the convolution kernel size of second layer convolutional layer is two 3 × 3, each layer of convolution
Feature quantity is 2048 in layer.
In said system, in addition to frequent customer's identification module, when the VIP identification modules are judged to be not present in hall
During VIP client, face characteristic information within the vision is preserved a period of time by frequent customer's identification module, if at this moment
Between client's transacting business again in the data set in section, then be considered as frequent customer, service be preferentially provided；Otherwise corresponding face is removed
Data.
Present invention also offers a kind of bank client recognition methods based on deep learning, comprise the following steps:
Using VIP client's photo, the human face data for meeting deep learning normal size is obtained；
The multidimensional face characteristic information of VIP client is obtained using the face characteristic extraction model based on deep learning framework,
It is stored in VIP customer databases；
The facial image of all personnel is automatically extracted from hall of bank scene photo, the step passes through an inverted pyramid
The assembled classifier of shape is extracted, and multiple top-level categories devices therein are rough sort device, each sub-classification device it is gradually fine and
Number gradually decreases；
Position the characteristic point of every face respectively according to every facial image, and face is corrected；
All normal size faces are input in the face characteristic extraction model based on deep learning framework, extract people
Face characteristic information；
Face characteristic information and VIP customer databases are contrasted, determine whether VIP client.
In the above-mentioned methods, using multiple dimensioned face normalization method based on by the thick architectural framework progress face school to essence
Just, low-resolution image is corrected using general deep learning network under each yardstick, then will corrects result
Input as upper strata, be corrected step by step.
In the above-mentioned methods, the face characteristic extraction model based on deep learning framework enters pedestrian using 10 layer network models
Face feature information extraction, 10 layer network models include 7 layers of convolutional layer and 3 layers of full articulamentum, the convolution kernel size of initial convolutional layer
For 9 × 9, the convolution kernel size of second layer convolutional layer is two 3 × 3, and feature quantity is 2048 in each layer of convolutional layer.
The present invention, by the identity of automatic identification business handling crowd, the intelligence degree of service robot is improved, is assigned
The ability of Yu Liao robots " understanding " people so that service robot can provide the intelligent, Site Service of hommization, improve
Customer experience.Face acquisition algorithm and feature information extraction algorithm are particularly improved, substantially increases speed and efficiency.
Brief description of the drawings
Fig. 1 is the bank client identifying system schematic diagram based on deep learning in the present invention；
Fig. 2 is the bank client recognition methods flow chart based on deep learning in the present invention.
Embodiment
The invention provides a kind of bank client identifying system based on deep learning, pass through automatic identification business handling people
The identity of group, the intelligence degree of service robot is improved, impart the ability of robot " understanding " people so that service-delivery machine
People can provide the intelligent, Site Service of hommization, improve customer experience.With reference to Figure of description and specific implementation
Mode is described in detail to the present invention.
As shown in figure 1, bank VIP client's recognition methods provided by the invention based on deep learning, comprises the following steps：
Step 1, using VIP client's photo face in photo is automatically extracted, face is cut and alignment operation, gone forward side by side
Row normalized, finally give the human face data for meeting deep learning normal size.
Step 2, the face characteristic extraction model based on improved AlexNet deep learnings framework is established, by normal size
Human face data input the model, finally give multidimensional face characteristic information, in this, as the unique mark of VIP client, preserve
In VIP customer databases.
Step 3, hall of bank scene photo is obtained using the camera installed on hall of bank service robot, and therefrom
Automatically extract the facial image of all personnel.
Present invention improves over traditional pyramid various visual angles face recognition algorithms, the group of a reverse pyramid is employed
Grader is closed to be extracted, top-level categories device therein be multiple rough sort devices, each sub-classification device gradually finely and number by
It is decrescence few.Such as：Assembled classifier can be three layers, and top layer is 3 rough sort devices, and the second layer is 2 middle classification devices, and bottom is
1 sophisticated category device, grader is gradually fine from top to bottom.Top-level categories device can quickly reject a large amount of non-face windows,
So as to which face is limited in smaller range, the speed of identification is improved, meanwhile, keep very high by lower floor's sophisticated category device
Discrimination, the result of unique recognition of face is finally acquired in bottom.
Step 4, the characteristic point for positioning according to every facial image every face respectively, and face is corrected.
The present invention uses multiple dimensioned face normalization method based on face normalization is carried out by the thick architectural framework to essence, every
Low-resolution image is corrected using general deep learning network under one yardstick, then will correct result as upper strata
Input, is corrected step by step.
Step 5, all normal size faces are input in the face characteristic extraction model based on deep learning framework, carried
Take out face characteristic information.
The present invention is improved traditional AlexNet deep learning networks, 10 layer network models is established, wherein wrapping
Include 7 layers of convolutional layer and 3 layers of full articulamentum.Compared to AlexNet, the initial convolutional layer of the present invention is substituted using 9 × 9 sizes
AlexNet 11 × 11 size convolutional layers, so as to reduce computation complexity.By the volume of the size of the second layer in AlexNet 5 × 5
Lamination is changed to the convolutional layer of two 3 × 3, while removes all local acknowledgement's normalization layers.In addition, present invention reduces each layer
The quantity of middle feature, 2048 are dropped to from 4096, and adds additional one layer of convolutional layer.
Step 6, face characteristic information and VIP client's face database contrasted, if VIP client, then know
Other system can be by feedback of the information to voice system, and service robot can be preferably VIP customer services；If it is not, then perform step
Rapid 7；
Step 7, by face characteristic information within the vision preserve a period of time, if retrieved within the time period
The client of identical face characteristic information transacting business again, then be considered as frequent customer, preferentially provides service, service robot can with it is old
Client is greeted, and inquiry facility is handled；Otherwise corresponding human face data is removed.
System can automatically save the business that user handles on service robot, and itself and client's human face data are tied up
It is fixed, form client traffic and handle database.Taxonomic revision and statistics are carried out to business in database automatically every half a year, and provided
Service robot business handling Optimizing Suggestions.
As shown in Fig. 2 the bank client identifying system provided by the invention based on deep learning, including with lower module：
Face acquisition module 10：By the assembled classifier of a reverse pyramid from hall of bank scene photo it is automatic
The facial image of all personnel is extracted, the top-level categories device in assembled classifier be multiple more coarse graders, and lower floor divides
Gradually fine but number also gradually decreases class device.Top-level categories device can quickly reject a large amount of non-face windows, so as to by people
Face is limited in smaller range, while improving recognition speed, keeps very high discrimination by lower floor's sophisticated category device, most
The result of unique recognition of face will be obtained in bottom eventually.
Face normalization module 20：Using multiple dimensioned face normalization method, equally based on by slightly entering to the architectural framework of essence
Row correction, under each yardstick, all low-resolution image is corrected using general deep learning network, then high-ranking officers
Positive result inputs as upper strata, is corrected step by step.
Characteristic information extracting module 30：Traditional AlexNet deep learning networks are improved, establish 10 layers of net
Network model, including 7 layers of convolutional layer and 3 layers of full articulamentum.Compared to AlexNet, the initial convolutional layer of the present invention using 9 ×
9 sizes substitute AlexNet 11 × 11 size convolutional layers, so as to reduce computation complexity.By the second layer 5 × 5 in AlexNet
The convolutional layer of size is changed to the convolutional layer of two 3 × 3, while removes all local acknowledgement's normalization layers.In addition, the present invention reduces
The quantity (dropping to 2048 from 4096) of feature in each layer, and adds additional one layer of convolutional layer.
VIP identification modules 40：The face characteristic information of extraction and VIP client's face database are contrasted, judgement is
No is VIP client.If it is, extraction VIP customer informations, and data exchange is carried out with service robot voice system, preferentially
For VIP customer services.
Frequent customer's identification module 50：When VIP client is not present in hall, face characteristic information within the vision is protected
A period of time (time can freely set) is deposited, if retrieving the client of identical face characteristic information in this time again
Secondary transacting business, this client will be classified as frequent customer by identifying system, and preferentially provide service；And do not have in imitating when stored
The client's human face data come again can be eliminated.
VIP client's face database 60, preserve the face characteristic information of VIP client.
The present invention, by the identity of automatic identification business handling crowd, the intelligence degree of service robot is improved, is assigned
The ability of Yu Liao robots " understanding " people so that service robot can provide the intelligent, Site Service of hommization, improve
Customer experience.Face acquisition algorithm and feature information extraction algorithm are particularly improved, substantially increases speed and efficiency.
The present invention is not limited to above-mentioned preferred forms, and anyone should learn that the knot made under the enlightenment of the present invention
Structure changes, and the technical schemes that are same or similar to the present invention, each falls within protection scope of the present invention.
Claims (7)
- A kind of 1. bank client identifying system based on deep learning, it is characterised in that including：Face acquisition module, for automatically extracting the facial image of all personnel, the face from hall of bank scene photo Acquisition module is extracted by the assembled classifier of a reverse pyramid, multiple top-level categories in the assembled classifier Device is rough sort device, and gradually fine and number gradually decreases each sub-classification device：Face normalization module, for being corrected to the facial image got；Characteristic information extracting module, for extracting the characteristic information in facial image；VIP identification modules, the characteristic information of extraction and VIP client's face characteristic information data storehouse are contrasted, judged whether For VIP client.
- 2. the system as claimed in claim 1, it is characterised in that the face normalization module uses multiple dimensioned face normalization side Method, based on face normalization is carried out by the thick architectural framework to essence, each yardstick is using general deep learning network to low Image in different resolution is corrected, and then correction result is inputted as upper strata, is corrected step by step.
- 3. the system as claimed in claim 1, it is characterised in that the characteristic information extracting module uses and is based on deep learning frame The face characteristic extraction model extraction face characteristic information of frame, the face characteristic extraction model based on deep learning framework include 7 Layer convolutional layer and 3 layers of full articulamentum, the convolution kernel size of initial convolutional layer is 9 × 9, and the convolution kernel size of second layer convolutional layer is Two 3 × 3, feature quantity is 2048 in each layer of convolutional layer.
- 4. the system as claimed in claim 1, it is characterised in that also including frequent customer's identification module, when the VIP identification modules When judging to be not present VIP client in hall, frequent customer's identification module face characteristic information will preserve one within sweep of the eye The section time, if retrieving the client of identical face characteristic information transacting business again within the time period, it is considered as old visitor Family, preferentially provide service；Otherwise corresponding human face data is removed.
- 5. a kind of bank client recognition methods based on deep learning, it is characterised in that comprise the following steps:Using VIP client's photo, the human face data for meeting deep learning normal size is obtained；The multidimensional face characteristic information of VIP client is obtained using the face characteristic extraction model based on deep learning framework, is preserved In VIP customer databases；Automatically extract the facial image of all personnel from hall of bank scene photo, the step passes through reverse pyramid Assembled classifier is extracted, and multiple top-level categories devices therein are rough sort device, the gradual fine and number of each sub-classification device Gradually decrease；Position the characteristic point of every face respectively according to every facial image, and face is corrected；The face of all normal sizes is input in the face characteristic extraction model based on deep learning framework, extracts face Characteristic information；Face characteristic information and VIP customer databases are contrasted, determine whether VIP client.
- 6. method as claimed in claim 5, it is characterised in that using multiple dimensioned face normalization method based on by slightly to essence Architectural framework carries out face normalization, and school is carried out to low-resolution image using general deep learning network under each yardstick Just, then correction result is inputted as upper strata, be corrected step by step.
- 7. method as claimed in claim 5, it is characterised in that the face characteristic extraction model based on deep learning framework uses 10 layer network models carry out face characteristic information extraction, and 10 layer network models include 7 layers of convolutional layer and 3 layers of full articulamentum, initially The convolution kernel size of convolutional layer is 9 × 9, and the convolution kernel size of second layer convolutional layer is two 3 × 3, special in each layer of convolutional layer It is 2048 to levy quantity.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710979947.5A CN107679510A (en) | 2017-10-19 | 2017-10-19 | A kind of bank client identifying system and method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710979947.5A CN107679510A (en) | 2017-10-19 | 2017-10-19 | A kind of bank client identifying system and method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107679510A true CN107679510A (en) | 2018-02-09 |
Family
ID=61141051
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710979947.5A Pending CN107679510A (en) | 2017-10-19 | 2017-10-19 | A kind of bank client identifying system and method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107679510A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491773A (en) * | 2018-03-12 | 2018-09-04 | 中国工商银行股份有限公司 | A kind of recognition methods and system |
| CN111814670A (en) * | 2020-07-08 | 2020-10-23 | 中国工商银行股份有限公司 | Special customer identification method and device for bank outlets |
| CN112215183A (en) * | 2020-10-21 | 2021-01-12 | 中国银行股份有限公司 | Bank customer identification method and device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063607A (en) * | 2009-11-16 | 2011-05-18 | 日电（中国）有限公司 | Method and system for acquiring human face image |
| US20150347820A1 (en) * | 2014-05-27 | 2015-12-03 | Beijing Kuangshi Technology Co., Ltd. | Learning Deep Face Representation |
| CN105426875A (en) * | 2015-12-18 | 2016-03-23 | 武汉科技大学 | Face identification method and attendance system based on deep convolution neural network |
| CN105718873A (en) * | 2016-01-18 | 2016-06-29 | 北京联合大学 | People stream analysis method based on binocular vision |
| CN106558282A (en) * | 2017-02-09 | 2017-04-05 | 广州动创信息科技有限公司 | A kind of market advertisement machine with pageview statistics |
| CN106826855A (en) * | 2017-03-14 | 2017-06-13 | 山东大学 | A kind of banking assistant robot based on cloud data identification |
-
2017
- 2017-10-19 CN CN201710979947.5A patent/CN107679510A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102063607A (en) * | 2009-11-16 | 2011-05-18 | 日电（中国）有限公司 | Method and system for acquiring human face image |
| US20150347820A1 (en) * | 2014-05-27 | 2015-12-03 | Beijing Kuangshi Technology Co., Ltd. | Learning Deep Face Representation |
| CN105426875A (en) * | 2015-12-18 | 2016-03-23 | 武汉科技大学 | Face identification method and attendance system based on deep convolution neural network |
| CN105718873A (en) * | 2016-01-18 | 2016-06-29 | 北京联合大学 | People stream analysis method based on binocular vision |
| CN106558282A (en) * | 2017-02-09 | 2017-04-05 | 广州动创信息科技有限公司 | A kind of market advertisement machine with pageview statistics |
| CN106826855A (en) * | 2017-03-14 | 2017-06-13 | 山东大学 | A kind of banking assistant robot based on cloud data identification |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108491773A (en) * | 2018-03-12 | 2018-09-04 | 中国工商银行股份有限公司 | A kind of recognition methods and system |
| CN111814670A (en) * | 2020-07-08 | 2020-10-23 | 中国工商银行股份有限公司 | Special customer identification method and device for bank outlets |
| CN112215183A (en) * | 2020-10-21 | 2021-01-12 | 中国银行股份有限公司 | Bank customer identification method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107067020B (en) | Image identification method and device | |
| WO2021088300A1 (en) | Rgb-d multi-mode fusion personnel detection method based on asymmetric double-stream network | |
| CN106997629B (en) | Access control method, apparatus and system | |
| CN104537336B (en) | A kind of face identification method and system for possessing self-learning function | |
| CN108009482A (en) | One kind improves recognition of face efficiency method | |
| CN107657249A (en) | Method, apparatus, storage medium and the processor that Analysis On Multi-scale Features pedestrian identifies again | |
| CN112926506B (en) | Non-controlled face detection method and system based on convolutional neural network | |
| CN106372648A (en) | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method | |
| CN105488517A (en) | Vehicle brand model identification method based on deep learning | |
| CN106845510A (en) | Chinese tradition visual culture Symbol Recognition based on depth level Fusion Features | |
| CN109711366A (en) | A kind of recognition methods again of the pedestrian based on group information loss function | |
| CN107220604A (en) | A kind of fall detection method based on video | |
| CN106529499A (en) | Fourier descriptor and gait energy image fusion feature-based gait identification method | |
| CN106599925A (en) | Plant leaf identification system and method based on deep learning | |
| CN107832835A (en) | The light weight method and device of a kind of convolutional neural networks | |
| CN102682302A (en) | Human body posture identification method based on multi-characteristic fusion of key frame | |
| CN104063686B (en) | Crop leaf diseases image interactive diagnostic system and method | |
| CN107679510A (en) | A kind of bank client identifying system and method based on deep learning | |
| CN109635694A (en) | A kind of pedestrian detection method, device, equipment and computer readable storage medium | |
| CN107145889A (en) | Target identification method based on double CNN networks with RoI ponds | |
| CN107122796A (en) | A kind of remote sensing image sorting technique based on multiple-limb network integration model | |
| CN106778653A (en) | Towards the exchange method and device based on recognition of face Sample Storehouse of intelligent robot | |
| CN101253535A (en) | Image retrieval device and image retrieval method | |
| CN106682628B (en) | Face attribute classification method based on multilayer depth feature information | |
| CN107169106A (en) | Video retrieval method, device, storage medium and processor |
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: 20180209 |