CN108074325A - A kind of note denomination detection method and device - Google Patents
A kind of note denomination detection method and device Download PDFInfo
- Publication number
- CN108074325A CN108074325A CN201611037851.9A CN201611037851A CN108074325A CN 108074325 A CN108074325 A CN 108074325A CN 201611037851 A CN201611037851 A CN 201611037851A CN 108074325 A CN108074325 A CN 108074325A
- Authority
- CN
- China
- Prior art keywords
- component
- neural network
- feature vector
- value
- network parameter
- 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
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of note denomination detection method and device, this method includes:The RGB figures of bank note are gathered by color sensor and the RGB image is converted into V component figure;The feature vector of the V component figure will be determined after the V component figure decile;The value of money of the bank note is determined according to the feature vector of the V component figure and neural network parameter.Based on the above method and device, V component image is converted by bank note RGB image to extract feature, this feature has very high otherness, a variety of values of money can easily be distinguished very much, optimize entire money-checking flow so that value of money detection is more accurate, simply.
Description
Technical field
The present embodiments relate to bill handling field more particularly to a kind of note denomination detection method and device.
Background technology
With economic prosperity and development, the circulation of bank note is increasing, so the identification of note denomination is in money-checking process
In be very important step.
Existing note denomination identification is distinguished by multiple features in gray-scale map, some gray feature information are more multiple
It is miscellaneous, very high calculation amount is had, it is high to be easy to cause identification error rate.Especially when image capture environment is severe, the figure of acquisition
As being present with many noises, feature can be allowed to become less obvious.
The content of the invention
The present invention provides a kind of note denomination detection method and device, to realize simplicity and identify note denomination exactly.
To reach this purpose, the embodiment of the present invention uses following technical scheme:
A kind of note denomination detection method, including:
The RGB figures of bank note are gathered by color sensor and the RGB image is converted into V component figure;
The feature vector of the V component figure will be determined after the V component figure decile;
The value of money of the bank note is determined according to the feature vector of the V component figure and neural network parameter.
Further, the method determines institute in the feature vector and neural network parameter according to the V component figure
Before the value of money for stating bank note, further include:
Initialize neural network parameter;
Neural network model is input to after the feature vector of V component figure is normalized to be trained, and obtains output result;
It according to neural network parameter described in the output results modification and comes back for training, until the neutral net is joined
Number is stablized.
Further, in the above method, the RGB figures of bank note is gathered by color sensor and convert the RGB image
Include into V component figure:
V component value is calculated according to equation below:
V=0.615*R-0.515*G-0.100*B;
Wherein, RGB is respectively the parameter value of pixel in RGB image.
Further, the method, the feature vector bag that the V component figure will be determined after the V component figure decile
It includes:
Extract the blocking characteristic of all subgraphs after the V component figure decile;
By the blocking characteristic of all subgraphs as the feature vector of the V component figure.
Further, in the above method, the blocking characteristic is the average after all pixels value of each subgraph adds up.
Correspondingly, a kind of note denomination detection device is also disclosed in the embodiment of the present invention, including:
Image Acquisition modular converter, for gathering the RGB figures of bank note by color sensor and converting the RGB image
Into V component figure;
Feature vector determining module, for normalizing the feature vector of V component figure as the neural network model
Input parameter is trained, and obtains output result;
Note denomination determining module, determine for the feature vector according to the V component figure and neural network parameter described in
The value of money of bank note.
Further, described device further includes:
Parameter setting module, for determining institute in the feature vector and neural network parameter according to the V component figure
Neural network parameter is initialized before stating the value of money of bank note;
As a result output module is instructed for will be input to neural network model after the normalization of the feature vector of V component figure
Practice, obtain output result;
Parameters revision module for the neural network parameter according to the output results modification and comes back for training,
Until the neural network parameter is stablized.
Further, in above device, described image modular converter is specifically used for:
V component value is calculated according to equation below:
V=0.615*R-0.515*G-0.100*B;
Wherein, RGB is respectively the parameter value of pixel in RGB image.
Further, described device, described eigenvector determining module include:
Blocking characteristic extraction unit, for extracting the blocking characteristic of all subgraphs after the V component figure decile;
Feature vector determination unit, for by the blocking characteristic of all subgraphs as the V component figure feature to
Amount.
Further, in above device, the blocking characteristic is the average after all pixels value of each subgraph adds up.
The technical solution that the embodiment of the present invention is provided in note denomination detects identification application, passes through bank note RGB image
V component image is converted into extract feature, using the high diversity of feature, can easily distinguish very much a variety of coin
Value.The real technical solution provided of the invention, optimizes entire money-checking flow, can ensure that value of money detection is more accurate, simply.
Description of the drawings
Fig. 1 is a kind of flow diagram for note denomination detection method that the embodiment of the present invention one provides;
Fig. 2 is the flow diagram that a kind of neural network parameter provided by Embodiment 2 of the present invention determines method;
Fig. 3 is a kind of structure diagram for note denomination detection device that the embodiment of the present invention three provides.
Specific embodiment
The present invention is described in further detail with reference to the accompanying drawings and examples.It is understood that this place is retouched
The specific embodiment stated is used only for explaining the present invention rather than limitation of the invention.It also should be noted that in order to just
Part related to the present invention rather than entire infrastructure are illustrated only in description, attached drawing.
Embodiment one
Attached drawing 1 is referred to, it, should for a kind of flow diagram for note denomination detection method that the embodiment of the present invention one provides
Method is suitable for the scene of note denomination detection identification during money-checking, and this method is performed by note denomination detection device, should
Device by software and/or hardware realization, can be integrated in the inside of currency examination device or financial transaction apparatus.This method specifically include as
Lower step:
The RGB image is simultaneously converted into V component figure by S110, the RGB figures that bank note is gathered by color sensor;
It should be noted that the colored RGB figures color of different note denominations is all different, and each pixel in RGB image
It is made of R (red), G (green) and 3 color components of B (indigo plant).RGB component is converted into YUV components, Y-component represents brightness, U
Representation in components tone, V component represent saturation degree.Since the difference of V component is most apparent, be conducive to note denomination identification, so only
V component need to be separated, can specifically calculate and convert according to equation below:V=0.615*R-0.515*G-0.100*B, wherein, RGB
The parameter value of pixel respectively in RGB image.
S120, the feature vector that the V component figure will be determined after the V component figure decile;
Wherein, it is described to determine that the feature vector of the V component figure includes after the V component figure decile:
Extract the blocking characteristic of all subgraphs after the V component figure decile;
By the blocking characteristic of all subgraphs as the feature vector of the V component figure.
Specifically, by every component map, often row is divided into 12 deciles, and each column is divided into 6 deciles, and whole figure is just divided into 12*
6 pieces of subgraphs.Every whole figure represents that every piece of subgraph can represent that i is from 1 to 72 with Vi with vector V.Every piece of subgraph width is w, a height of h,
It adds up to all pixels value of every piece of subgraph and averages again after summing, this average is the value of Vi, special for the piecemeal as subgraph
Sign.
Specifically, by the set of the blocking characteristic of all subgraphs, i.e., value from V1 to V72 as V component figure to
Measure feature, collection are combined into:V=V1, V2 ... V72 }.
S130, the value of money that the bank note is determined according to the feature vector and neural network parameter of the V component figure.
It should be noted that neural network parameter is not unique, there is different nerve nets corresponding to different note denominations
Network parameter, being equivalent to a kind of note denomination has a corresponding template, different between template, is independent of each other.
The V component eigen vector of a piece of paper coin is obtained specifically, working as, i.e., after the Vi values of all decile subgraphs, it will carry out
With template, the i.e. comparison with neural network parameter, the corresponding value of money of neural network parameter matched can be identified as the paper
The value of money of coin.
In conclusion the embodiment of the present invention one discloses a kind of note denomination detection method, detect and identify in note denomination
In, V component image is converted by bank note RGB image to extract feature, it, can be very simple using the high diversity of feature
Just a variety of values of money are distinguished.The real technical solution provided of the invention, optimizes entire money-checking flow, can ensure coin
Value detection is more accurate, simply.
Embodiment two
Fig. 2 determines the flow diagram of method, this implementation for a kind of neural network parameter disclosed in the embodiment of the present invention two
Example is also wrapped before embodiment one determines the value of money of the bank note according to the feature vector and neural network parameter of the V component figure
Include the method that neural network parameter is determined based on neural metwork training.This method specifically comprises the following steps:
S210, initialization neural network parameter;
S220, will V component figure feature vector normalize after be input to neural network model and be trained, obtain output knot
Fruit;
S230, neural network parameter and come back for training according to the output results modification, until the nerve
Network parameter is stablized.
Specifically, neural network model is divided into input layer, hidden layer and output layer.
Input layer is vector P (In), that is, needs every image V feature vector of input training.
The output function of each node of hidden layer is:Out (Hid)=F (W (Hid) * P (In)+B (Hid));
Out (Hid) exports for each hidden layer node, and W (Hid) is the weight vectors of hidden layer node, and B (Hid) is hidden
Offset vector containing node layer, P (In) be hidden layer node input, F be hidden layer node kernel function, hidden layer node kernel function
For logarithm S type functions F=1/ (1+exp (- n)).
Exporting node layer output function is:Out (Out)=F (W (Out) * Out (Hid)+B (Out));
Out (Out) exports node layer output valve, and W (Out) is output layer node weight vector, and B (Out) exports node layer
Offset vector, the input vector for exporting node layer are exactly the output vector Out (Hid) of hidden layer node;Output layer section simultaneously
Point kernel function F is linear function Y=X.
Specifically, neural network parameter initializes, and if to distinguish 50,100 and 500 yuan of 3 kinds of value of money types of Hongkong dollar, setting
Output layer is 1 node, exports the random vector of W (Out) 1 row 1 row, and B (Out) is also set to the random vector of 1 row 1 row.Setting
Hidden layer is 1 node, and W (Hid) is the random vector of 72 row, 10 row, and B (Hid) is the random vector of 1 row, 10 row.Learning rate is set
It is set to 0.1.
Specifically, W (Out), B (Out) and W (Hid), B (Hid) vector ginsengs are reversely changed according to each output resultant error
Numerical value, i.e. neural network parameter, until state vector parameter stability, note denomination discrimination will also be stablized.
It should be noted that the feature vector of training input data, i.e. V component figure, normalized main cause is to disappear
Except the difference between different dimensions data, the convergence rate of training algorithm may also speed up.It is inconsistent dimension can also to be removed
Defect, time series are unstable that problem, the big several influences to result of reduction are also unlikely to lose some critically important to result
Small Value Data.
In conclusion the embodiment of the present invention two, which discloses a kind of neural network parameter, determines method, it can be according to the V of bank note
The feature vector of component map determines neural network parameter, is used as the template of note denomination identification so that value of money detects more
Accurately.
Embodiment three
Attached drawing 3 is referred to, it, should for a kind of structure diagram for note denomination detection device that the embodiment of the present invention three provides
Device specifically includes following module:
Image Acquisition modular converter 310, for the RGB figures that bank note is gathered by color sensor and by the RGB image
It is converted into V component figure;
Feature vector determining module 320, for normalizing the feature vector of V component figure as the neural network model
Input parameter be trained, obtain output result;
Note denomination determining module 330 determines institute for the feature vector according to the V component figure and neural network parameter
State the value of money of bank note.
Preferably, described device further includes:
Parameter setting module, for determining institute in the feature vector and neural network parameter according to the V component figure
Neural network parameter is initialized before stating the value of money of bank note;
As a result output module is instructed for will be input to neural network model after the normalization of the feature vector of V component figure
Practice, obtain output result;
Parameters revision module for the neural network parameter according to the output results modification and comes back for training,
Until the neural network parameter is stablized.
Preferably, in above device, Image Acquisition modular converter 310 is specifically used for:
V component value is calculated according to equation below:
V=0.615*R-0.515*G-0.100*B;
Wherein, RGB is respectively the parameter value of pixel in RGB image.
Preferably, described device includes:
Blocking characteristic extraction unit, for extracting the blocking characteristic of all subgraphs after the V component figure decile;
Feature vector determination unit, for by the blocking characteristic of all subgraphs as the V component figure feature to
Amount.
The present embodiment gathers the RGB figures of bank note by color sensor and the RGB image is converted into V points according to formula
Spirogram;The feature vector of the V component figure will be determined after the V component figure decile;According to the feature vector of the V component figure and
Neural network parameter determines the value of money of the bank note.Based on the above method and device, V component is converted by bank note RGB image
Image extracts feature, and this feature has very high otherness, can easily distinguish very much a variety of values of money, optimize
Entire money-checking flow so that value of money detection is more accurate, simply.
The said goods can perform the method that any embodiment of the present invention is provided, and possess the corresponding function module of execution method
And advantageous effect.
Note that it above are only presently preferred embodiments of the present invention and institute's application technology principle.It will be appreciated by those skilled in the art that
The invention is not restricted to specific embodiment described here, can carry out for a person skilled in the art various apparent variations,
It readjusts and substitutes without departing from protection scope of the present invention.Therefore, although being carried out by above example to the present invention
It is described in further detail, but the present invention is not limited only to above example, without departing from the inventive concept, also
It can include other more equivalent embodiments, and the scope of the present invention is determined by scope of the appended claims.
Claims (10)
1. a kind of note denomination detection method, which is characterized in that including:
The RGB figures of bank note are gathered by color sensor and the RGB image is converted into V component figure;
The feature vector of the V component figure will be determined after the V component figure decile;
The value of money of the bank note is determined according to the feature vector of the V component figure and neural network parameter.
2. according to the method described in claim 1, it is characterized in that, in the feature vector according to the V component figure and god
Before the value of money that the bank note is determined through network parameter, further include:
Initialize neural network parameter;
Neural network model is input to after the feature vector of V component figure is normalized to be trained, and obtains output result;
It according to neural network parameter described in the output results modification and comes back for training, until the neural network parameter is steady
It is fixed.
3. according to the method described in claim 1, it is characterized in that, the RGB figures of bank note are gathered and by institute by color sensor
Stating RGB image and being converted into V component figure includes:
V component value is calculated according to equation below:
V=0.615*R-0.515*G-0.100*B;
Wherein, RGB is respectively the parameter value of pixel in RGB image.
4. according to the method described in claim 1, it is characterized in that, described will determine the V component after the V component figure decile
The feature vector of figure includes:
Extract the blocking characteristic of all subgraphs after the V component figure decile;
By the blocking characteristic of all subgraphs as the feature vector of the V component figure.
5. according to the method described in claim 4, it is characterized in that, the blocking characteristic tires out for all pixels value of each subgraph
Average after adding.
6. a kind of note denomination detection device, which is characterized in that including:
Image Acquisition modular converter, for gathering the RGB figures of bank note by color sensor and the RGB image being converted into V
Component map;
Feature vector determining module, for the input using the normalization of the feature vector of V component figure as the neural network model
Parameter is trained, and obtains output result;
Note denomination determining module determines the bank note for the feature vector according to the V component figure and neural network parameter
Value of money.
7. device according to claim 6, which is characterized in that further include:
Parameter setting module, for determining the paper in the feature vector and neural network parameter according to the V component figure
Neural network parameter is initialized before the value of money of coin;
As a result output module being trained for will be input to neural network model after the normalization of the feature vector of V component figure, obtaining
To output result;
Parameters revision module for the neural network parameter according to the output results modification and comes back for training, until
The neural network parameter is stablized.
8. device according to claim 6, which is characterized in that described image modular converter is specifically used for according to equation below
Calculate V component value:
V=0.615*R-0.515*G-0.100*B;
Wherein, RGB is respectively the parameter value of pixel in RGB image.
9. device according to claim 6, which is characterized in that described eigenvector determining module includes:
Blocking characteristic extraction unit, for extracting the blocking characteristic of all subgraphs after the V component figure decile;
Feature vector determination unit, for the feature vector by the blocking characteristic of all subgraphs as the V component figure.
10. device according to claim 9, which is characterized in that the blocking characteristic is all pixels value of each subgraph
Average after cumulative.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611037851.9A CN108074325A (en) | 2016-11-10 | 2016-11-10 | A kind of note denomination detection method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611037851.9A CN108074325A (en) | 2016-11-10 | 2016-11-10 | A kind of note denomination detection method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108074325A true CN108074325A (en) | 2018-05-25 |
Family
ID=62161299
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611037851.9A Pending CN108074325A (en) | 2016-11-10 | 2016-11-10 | A kind of note denomination detection method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108074325A (en) |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2340931A (en) * | 1998-08-21 | 2000-03-01 | Celestica Ltd | Object colour validation |
CN101583978A (en) * | 2007-05-15 | 2009-11-18 | Lgn-Sys株式会社 | Apparatus for media recognition and method for media kind distinction with the same |
CN102750771A (en) * | 2012-07-13 | 2012-10-24 | 中山大学 | Method for identifying denominations of fifth series of renminbi applied in smart phone |
CN105740774A (en) * | 2016-01-25 | 2016-07-06 | 浪潮软件股份有限公司 | Text region positioning method and apparatus for image |
CN105844789A (en) * | 2016-05-26 | 2016-08-10 | 北京化工大学 | Integrated banknote sorting machine and application thereof |
-
2016
- 2016-11-10 CN CN201611037851.9A patent/CN108074325A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB2340931A (en) * | 1998-08-21 | 2000-03-01 | Celestica Ltd | Object colour validation |
CN101583978A (en) * | 2007-05-15 | 2009-11-18 | Lgn-Sys株式会社 | Apparatus for media recognition and method for media kind distinction with the same |
CN102750771A (en) * | 2012-07-13 | 2012-10-24 | 中山大学 | Method for identifying denominations of fifth series of renminbi applied in smart phone |
CN105740774A (en) * | 2016-01-25 | 2016-07-06 | 浪潮软件股份有限公司 | Text region positioning method and apparatus for image |
CN105844789A (en) * | 2016-05-26 | 2016-08-10 | 北京化工大学 | Integrated banknote sorting machine and application thereof |
Non-Patent Citations (2)
Title |
---|
沈晋原编著: "《多媒体通信技术基础》", 31 May 2015, 北京:国防工业出版社 * |
程海玉: "人民币智能分捡器软件系统设计", 《中国优秀博硕士学位论文全文数据库 (硕士) 工程科技Ⅱ辑》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106407917B (en) | The retinal vessel extracting method and system distributed based on Dynamic Multi-scale | |
KR102102161B1 (en) | Method, apparatus and computer program for extracting representative feature of object in image | |
CN104834933B (en) | A kind of detection method and device in saliency region | |
CN109670452A (en) | Method for detecting human face, device, electronic equipment and Face datection model | |
CN104598908B (en) | A kind of crops leaf diseases recognition methods | |
Barron | Convolutional color constancy | |
CN104966085B (en) | A kind of remote sensing images region of interest area detecting method based on the fusion of more notable features | |
CN104182763B (en) | A kind of floristics identifying system based on flower feature | |
CN105518709B (en) | The method, system and computer program product of face for identification | |
CN106780485A (en) | SAR image change detection based on super-pixel segmentation and feature learning | |
CN109146856A (en) | Picture quality assessment method, device, computer equipment and storage medium | |
CN106504064A (en) | Clothes classification based on depth convolutional neural networks recommends method and system with collocation | |
CN107423701A (en) | The non-supervisory feature learning method and device of face based on production confrontation network | |
CN108280455B (en) | Human body key point detection method and apparatus, electronic device, program, and medium | |
Qu et al. | A pedestrian detection method based on yolov3 model and image enhanced by retinex | |
CN104103033B (en) | View synthesis method | |
CN106296638A (en) | Significance information acquisition device and significance information acquisition method | |
CN104658097B (en) | A kind of rmb paper currency denomination identifying method of Histogram Matching based on image | |
CN104217221A (en) | Method for detecting calligraphy and paintings based on textural features | |
CN111582359B (en) | Image identification method and device, electronic equipment and medium | |
CN106920211A (en) | U.S. face processing method, device and terminal device | |
Raffei et al. | A low lighting or contrast ratio visible iris recognition using iso-contrast limited adaptive histogram equalization | |
CN107657619A (en) | A kind of low-light (level) Forest fire image dividing method | |
CN108108753A (en) | A kind of recognition methods of check box selection state based on support vector machines and device | |
CN103984964A (en) | Image identification method and system for test strip |
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: 20180525 |
|
RJ01 | Rejection of invention patent application after publication |