CN104484684B - A kind of Manuscripted Characters Identification Method and system - Google Patents
A kind of Manuscripted Characters Identification Method and system Download PDFInfo
- Publication number
- CN104484684B CN104484684B CN201510001954.9A CN201510001954A CN104484684B CN 104484684 B CN104484684 B CN 104484684B CN 201510001954 A CN201510001954 A CN 201510001954A CN 104484684 B CN104484684 B CN 104484684B
- Authority
- CN
- China
- Prior art keywords
- sample
- training sample
- self
- target
- encoding encoder
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
- G06V30/19—Recognition using electronic means
- G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
- G06V30/194—References adjustable by an adaptive method, e.g. learning
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of Manuscripted Characters Identification Method and system, method is:Using with smooth norm L1Self-encoding encoder each training sample that training sample is concentrated is handled, obtain corresponding target training sample, the target training sample forms target training sample set, the smooth norm L of band with the sample label that the training sample is concentrated1Self-encoding encoder object function in be equipped with sparse penalty term, the sparse penalty term be smooth L1Then norm utilizes target training sample to train grader, object classifiers is obtained, using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target sample to be predicted, target sample to be predicted be finally input to the object classifiers, with the classification of determination sample to be predicted.The scheme of the application is by smooth norm L1It is introduced into self-encoding encoder, the feature of more identification can be obtained as new sparse penalty term instead of common KL divergences so that final handwriting recongnition rate higher.
Description
Technical field
This application involves mode identification technologies, more specifically to a kind of Manuscripted Characters Identification Method and system.
Background technology
The identification of handwriting digital (such as postal, bank and e-commerce field) in real life has more far-reaching
Application demand.It always is the research hotspot of area of pattern recognition.In recent years, with computer technology and image procossing skill
The rapid development of art, it has been proposed that much for realizing the method for Handwritten Digital Recognition, such as the calculation based on stroke feature
Method, the algorithm based on k nearest neighbor, the algorithm based on support vector machines and the algorithm etc. based on neural network.But due to hand-written
Number varies with each individual and changes very much, causes the recognition effect of all kinds of algorithms still not ideal enough.Therefore, research is efficient hand-written
The identification of body number is still an important direction.
The method of artificial neural network provides a kind of be good for for approaching the object function of real number value, centrifugal pump or vector value
The very strong method of strong property.Self-encoding encoder is a three-layer neural network, including input layer, hidden layer and output layer.Self-encoding encoder
By the reconstructed error of minimum input data come the statistical framework inside acquistion input data, to obtain more discriminating power
Feature.Andrew professors Ng of Stanford University in the object function of self-encoding encoder by adding KL divergence regularization terms
The sparse coding to data is successfully realized punishing larger feature, and learns to have arrived good feature.But KL dissipates
Degree is limited to the ability of data sparse coding, therefore there are still certain for the identification of handwriting digital for finally obtained feature
Limitation.
Invention content
In view of this, this application provides a kind of Manuscripted Characters Identification Method and system, for solving existing handwriting recongnition
The low problem of method recognition effect.
To achieve the goals above, it is proposed that scheme it is as follows:
A kind of Manuscripted Characters Identification Method, including:
Using with smooth norm L1Self-encoding encoder each training sample that training sample is concentrated is handled, obtain pair
The target training sample answered, the target training sample form target training sample with the sample label that the training sample is concentrated
Collection, the smooth norm L of band1Self-encoding encoder object function in be equipped with sparse penalty term, the sparse penalty term be smooth L1Model
Number;
Grader is trained using the target training sample set, obtains object classifiers;
Using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target sample to be predicted;
Target sample to be predicted is input to the object classifiers, with the classification of determination sample to be predicted.
Preferably, described using with smooth norm L1Self-encoding encoder each training sample that training sample is concentrated carry out
Processing, obtains corresponding target training sample, including:
Defining training sample set is:
Wherein, y(i)It is and training sample x(i)Corresponding sample label, m are the numbers of training sample, and d is training sample dimension
Degree;
Define self-encoding encoder hypothesis function be:
hW,b(x(i))
Wherein, W and b indicates weight and the biasing of self-encoding encoder respectively;
The output for defining j-th of hidden unit of i-th of training sample is expressed asAnd the number of hidden unit is n;
It determines with smooth norm L1The object function of self-encoding encoder be:
Wherein, first item is reconstruct item, and Section 2 is weight attenuation term, and λ is weight attenuation coefficient, and Section 3 is sparse punishes
It is the weight of coefficient penalty factor to penalize item, β, and S () indicates smooth L1Norm, it is specific as follows:
Wherein, μ>0 is parameter preset;
Solve the parameter W so that the object function minimumoptAnd bopt;
By WoptAnd boptIt brings into the hypothesis function of self-encoding encoder, obtains goal hypothesis function;
The training sample x that training sample is concentrated(i)It brings the goal hypothesis function into, obtains target training sample a(i)。
Preferably, the parameter W so that the object function minimum is being solvedoptAnd boptWhen, using back-propagation algorithm into
Row calculates.
Preferably, described using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target and wait for
Forecast sample, including:
It brings the sample to be predicted into the goal hypothesis function, obtains target sample to be predicted.
Preferably, the grader is Softmax graders.
A kind of handwriting recongnition system, including:
Training sample processing unit, for using with smooth norm L1Self-encoding encoder each instruction that training sample is concentrated
Practice sample to be handled, obtains corresponding target training sample, the sample that the target training sample is concentrated with the training sample
This label forms target training sample set, the smooth norm L of band1Self-encoding encoder object function in be equipped with sparse punishment
, which is smooth L1Norm;
Classifier training unit obtains object classifiers for training grader using the target training sample set;
Sample to be tested processing unit, for using with smooth norm L1Self-encoding encoder treat forecast sample and handled,
Obtain target sample to be predicted;
Classification determination unit is waited for pre- for target sample to be predicted to be input to the object classifiers with determination
The classification of test sample sheet.
Preferably, the training sample processing unit includes:
Parameter definition unit is for defining training sample set:
Wherein, y(i)It is and training sample x(i)Corresponding sample label, m are the numbers of training sample, and d is training sample dimension
Degree;
Define self-encoding encoder hypothesis function be:
hW,b(x(i))
Wherein, W and b indicates weight and the biasing of self-encoding encoder respectively;
The output for defining j-th of hidden unit of i-th of training sample is expressed asAnd the number of hidden unit is n;
Object function determination unit, for determining with smooth norm L1The object function of self-encoding encoder be:
Wherein, first item is reconstruct item, and Section 2 is weight attenuation term, and λ is weight attenuation coefficient, and Section 3 is sparse punishes
It is the weight of coefficient penalty factor to penalize item, β, and S () indicates smooth L1Norm, it is specific as follows:
Wherein, μ>0 is parameter preset;
Object function solves unit, for solving the parameter W so that the object function minimumoptAnd bopt;
Assuming that function determination unit, is used for WoptAnd boptIt brings into the hypothesis function of self-encoding encoder, obtains goal hypothesis
Function;
Target training sample acquiring unit, the training sample x for concentrating training sample(i)Bring the goal hypothesis into
Function obtains target training sample a(i)。
Preferably, the parameter W so that the object function minimum is being solvedoptAnd boptWhen, using back-propagation algorithm into
Row calculates.
Preferably, the sample to be tested processing unit includes:
First sample to be tested processing subelement is obtained for bringing the sample to be predicted into the goal hypothesis function
Target sample to be predicted.
Preferably, the grader is Softmax graders.
It can be seen from the above technical scheme that Manuscripted Characters Identification Method provided by the embodiments of the present application, first with band
Smooth norm L1Self-encoding encoder each training sample that training sample is concentrated is handled, obtain corresponding target training sample
This, the target training sample forms target training sample set with the sample label that the training sample is concentrated, and the band is smooth
Norm L1Self-encoding encoder object function in be equipped with sparse penalty term, the sparse penalty term be smooth L1Then norm utilizes mesh
It marks training sample and trains grader, object classifiers are obtained, using with smooth norm L1Self-encoding encoder treat forecast sample into
Row processing, obtains target sample to be predicted, target sample to be predicted is finally input to the object classifiers, with determination
The classification of sample to be predicted.The scheme of the application is by smooth norm L1It is introduced into self-encoding encoder, instead of common KL divergences, as
New sparse penalty term, can obtain the feature of more identification so that final handwriting recongnition rate higher.
Description of the drawings
In order to illustrate the technical solutions in the embodiments of the present application or in the prior art more clearly, to embodiment or will show below
There is attached drawing needed in technology description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
The embodiment of application for those of ordinary skill in the art without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 is a kind of Manuscripted Characters Identification Method flow chart disclosed in the embodiment of the present application;
Fig. 2 is a kind of handwriting recongnition system structure diagram disclosed in the embodiment of the present application.
Specific implementation mode
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation describes, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on
Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other
Embodiment shall fall in the protection scope of this application.
The purport of the application is to be used to approach L by what Beck and Teboulle were proposed1The smooth function of norm is introduced into
In self-encoding encoder, instead of common KL divergences, as new sparse penalty term, and by the feature with sparsity of acquistion, make
Grader is trained for new sample.
Referring to Fig. 1, Fig. 1 is a kind of Manuscripted Characters Identification Method flow chart disclosed in the embodiment of the present application.
As shown in Figure 1, this method includes:
Step S100, using with smooth norm L1Self-encoding encoder each training sample that training sample is concentrated at
Reason, obtains corresponding target training sample;
Specifically, the smooth norm L of the band1Self-encoding encoder object function in be equipped with sparse penalty term, this is sparse to punish
It is smooth L to penalize item1Norm.The existing sparse penalty term of KL divergences is replaced in the present embodiment, substitution generation refers to smooth model
Number L1。
Wherein, the target training sample forms target training sample set with the sample label that the training sample is concentrated.
That is, by being trained to former training sample, the sample label of the target training sample obtained after training as before, by
Target training sample collectively forms target training sample set with sample label.
Step S110, grader is trained using the target training sample set, obtains object classifiers;
Step S120, using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target wait for it is pre-
Test sample sheet;
Specifically, it also needs to treat forecast sample and carries out identical processing, the target sample to be predicted that obtains that treated.
Step S130, target sample to be predicted is input to the object classifiers, with determination sample to be predicted
Classification.
The classification namely its sample label of target sample to be predicted are predicted using object classifiers.
Manuscripted Characters Identification Method provided by the embodiments of the present application, first with smooth norm L1Self-encoding encoder to training
Each training sample in sample set is handled, and obtains corresponding target training sample, the target training sample with it is described
The sample label that training sample is concentrated forms target training sample set, the smooth norm L of band1Self-encoding encoder object function
In be equipped with sparse penalty term, the sparse penalty term be smooth L1Then norm utilizes target training sample to train grader, obtains
Object classifiers, using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target sample to be predicted,
Target sample to be predicted is finally input to the object classifiers, with the classification of determination sample to be predicted.The application's
Scheme is by smooth norm L1It is introduced into self-encoding encoder, can be obtained more as new sparse penalty term instead of common KL divergences
Has the feature of identification so that final handwriting recongnition rate higher.
Training sample set is handled next, we introduce, obtains the process of target training sample set.
First, defining training sample set is:
Wherein, y(i)It is and training sample x(i)Corresponding sample label, m are the numbers of training sample, and d is training sample dimension
Degree.
The hypothesis function of self-encoding encoder is:hW,b(x(i))
Wherein, W and b indicates weight and the biasing of self-encoding encoder respectively.
The output of j-th of hidden unit of i-th of training sample is expressed asAnd the number of hidden unit is n.
It determines with smooth norm L1The object function of self-encoding encoder be:
Wherein, first item is reconstruct item, and Section 2 is weight attenuation term, and λ is weight attenuation coefficient, and Section 3 is sparse punishes
It is the weight of coefficient penalty factor to penalize item, β, and S () indicates smooth L1Norm, it is specific as follows:
Wherein, μ>0 is parameter preset, and μ control S (x) approach norm L1Degree.
Then, the parameter W so that the object function minimum is solvedoptAnd bopt.Then by WoptAnd boptBring own coding into
The hypothesis function h of deviceW,b(x(i)) in, obtain goal hypothesis function.Finally, training sample x training sample concentrated(i)It brings into
The goal hypothesis function obtains target training sample a(i).Target training sample a(i)With training sample x(i)Sample label one
Sample, it is thus determined that target training sample set is
It is to be understood that in the solution procedure for carrying out object function optimal solution, can be asked using back-propagation algorithm
Solution.
Further, when the above-mentioned training grader using target training sample set, Softmax graders can be used.
It after training finishes grader, needs to treat forecast sample x processing, namely carries it into above-mentioned target vacation
If in function, it may be determined that corresponding target sample a to be predicted.Then target sample a to be predicted is input to and has been trained
Object classifiers in, to obtain the prediction classification of sample x.
In order to further confirm the superiority of the application method, now illustrated by a specific example.
We are for MNIST Handwritten Digital Recognitions.The data set shares 60000 training samples and 10000
Test sample.Each size is that 28*28 namely d values are 784.In this experiment, it is instructed using all training samples
Practice with smooth L1The self-encoding encoder of norm, i.e. m=60000, and tested on entire test set.
Parameter setting procedure:
With smooth L1The number of the hidden unit of the self-encoding encoder of norm is 14 × 14, i.e. n=196, weight attenuation coefficient λ=
1e-3, weight beta=1 of sparsity penalty factor, sparsity parameter μ=0.9.In experiment, optimized using back-propagation algorithm
Self-encoding encoder model and Softmax sorter models.
Prediction of result:
As comparison, on same training set and test set, we to without sparse item self-encoding encoder and with KL dissipate
The self-encoding encoder of degree penalty term is trained and has been tested.Finally, k nearest neighbor algorithm is also used on same training set to surveying
Examination collection is classified, and the recognition effect of experiment is shown in Table 1, and table 1 is the comparison of Handwritten Digit Classification performance (discrimination %).
K nearest neighbor | Without sparse item | The band sparse item of KL divergences | The application |
(97.08 K=3) | 93.03 | 96.96 | 97.21 |
Table 1
Shown after introducing sparse item by upper table 1, the feature to more identification can be learnt, and the application is made
Smooth L1Norm penalty term is more preferable than common KL divergences penalty term performance.
Handwriting recongnition system provided by the embodiments of the present application is described below, handwriting recongnition system described below
System can correspond reference with above-described Manuscripted Characters Identification Method.
Referring to Fig. 2, Fig. 2 is a kind of handwriting recongnition system structure diagram disclosed in the embodiment of the present application.
As shown in Fig. 2, the system includes:
Training sample processing unit 21, for using with smooth norm L1Self-encoding encoder training sample is concentrated it is each
Training sample is handled, and corresponding target training sample is obtained, and the target training sample is concentrated with the training sample
Sample label forms target training sample set, the smooth norm L of band1Self-encoding encoder object function in be equipped with sparse punishment
, which is smooth L1Norm;
Classifier training unit 22 obtains object classifiers for training grader using the target training sample set;
Specifically, grader used herein can be Softmax graders.
Sample to be tested processing unit 23, for using with smooth norm L1Self-encoding encoder treat at forecast sample
Reason, obtains target sample to be predicted;
Classification determination unit 24 is waited for for target sample to be predicted to be input to the object classifiers with determination
The classification of forecast sample.
Optionally, the training sample processing unit 21 may include:
Parameter definition unit is for defining training sample set:
Wherein, y(i)It is and training sample x(i)Corresponding sample label, m are the numbers of training sample, and d is training sample dimension
Degree;
Define self-encoding encoder hypothesis function be:
hW,b(x(i))
Wherein, W and b indicates weight and the biasing of self-encoding encoder respectively;
The output for defining j-th of hidden unit of i-th of training sample is expressed asAnd the number of hidden unit is n;
Object function determination unit, for determining with smooth norm L1The object function of self-encoding encoder be:
Wherein, first item is reconstruct item, and Section 2 is weight attenuation term, and λ is weight attenuation coefficient, and Section 3 is sparse punishes
It is the weight of coefficient penalty factor to penalize item, β, and S () indicates smooth L1Norm, it is specific as follows:
Wherein, μ>0 is parameter preset;
Object function solves unit, for solving the parameter W so that the object function minimumoptAnd bopt;
Assuming that function determination unit, is used for WoptAnd boptIt brings into the hypothesis function of self-encoding encoder, obtains goal hypothesis
Function;
Target training sample acquiring unit, the training sample x for concentrating training sample(i)Bring the goal hypothesis into
Function obtains target training sample a(i)。
Optionally, the sample to be tested processing unit 23 may include:
First sample to be tested processing subelement is obtained for bringing the sample to be predicted into the goal hypothesis function
Target sample to be predicted.
Optionally, the parameter W so that the object function minimum is being solvedoptAnd boptWhen, it can be calculated using backpropagation
Method is calculated.
Handwriting recongnition system provided by the embodiments of the present application, first with smooth norm L1Self-encoding encoder to training
Each training sample in sample set is handled, and obtains corresponding target training sample, the target training sample with it is described
The sample label that training sample is concentrated forms target training sample set, the smooth norm L of band1Self-encoding encoder object function
In be equipped with sparse penalty term, the sparse penalty term be smooth L1Then norm utilizes target training sample to train grader, obtains
Object classifiers, using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target sample to be predicted,
Target sample to be predicted is finally input to the object classifiers, with the classification of determination sample to be predicted.The application's
Scheme is by smooth norm L1It is introduced into self-encoding encoder, can be obtained more as new sparse penalty term instead of common KL divergences
Has the feature of identification so that final handwriting recongnition rate higher.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning
Covering non-exclusive inclusion, so that the process, method, article or equipment including a series of elements includes not only that
A little elements, but also include other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged
Except there is also other identical elements in the process, method, article or apparatus that includes the element.
Each embodiment is described by the way of progressive in this specification, the highlights of each of the examples are with other
The difference of embodiment, just to refer each other for identical similar portion between each embodiment.
The foregoing description of the disclosed embodiments enables professional and technical personnel in the field to realize or use the application.
Various modifications to these embodiments will be apparent to those skilled in the art, as defined herein
General Principle can in other embodiments be realized in the case where not departing from spirit herein or range.Therefore, the application
It is not intended to be limited to the embodiments shown herein, and is to fit to and the principles and novel features disclosed herein phase one
The widest range caused.
Claims (8)
1. a kind of Manuscripted Characters Identification Method, which is characterized in that including:
Using with smooth norm L1Self-encoding encoder each training sample that training sample is concentrated is handled, obtain corresponding
Target training sample, the target training sample form target training sample set with the sample label that the training sample is concentrated,
The smooth norm L of band1Self-encoding encoder object function in be equipped with sparse penalty term, the sparse penalty term be smooth L1Norm;
Grader is trained using the target training sample set, obtains object classifiers;
Using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain target sample to be predicted;
Target sample to be predicted is input to the object classifiers, with the classification of determination sample to be predicted;
It is described to utilize with smooth norm L1Self-encoding encoder each training sample that training sample is concentrated is handled, obtain pair
The target training sample answered, including:
Defining training sample set is:
Wherein, y(i)It is and training sample x(i)Corresponding sample label, m are the numbers of training sample, and d is training sample dimension;
Define self-encoding encoder hypothesis function be:
hW,b(x(i))
Wherein, W and b indicates weight and the biasing of self-encoding encoder respectively;
The output for defining j-th of hidden unit of i-th of training sample is expressed asAnd the number of hidden unit is n;
It determines with smooth norm L1The object function of self-encoding encoder be:
Wherein, first item is reconstruct item, and Section 2 is weight attenuation term, and λ is weight attenuation coefficient, and Section 3 is sparse punishment
, β is the weight of coefficient penalty factor, and S () indicates smooth L1Norm, it is specific as follows:
Wherein, μ > 0 are parameter preset;
Solve the parameter W so that the object function minimumoptAnd bopt;
By WoptAnd boptIt brings into the hypothesis function of self-encoding encoder, obtains goal hypothesis function;
The training sample x that training sample is concentrated(i)It brings the goal hypothesis function into, obtains target training sample a(i)。
2. according to the method described in claim 1, it is characterized in that, solving the parameter W so that the object function minimumopt
And boptWhen, it is calculated using back-propagation algorithm.
3. according to the method described in claim 1, it is characterized in that, described using with smooth norm L1Self-encoding encoder treat it is pre-
Test sample is originally handled, and target sample to be predicted is obtained, including:
It brings the sample to be predicted into the goal hypothesis function, obtains target sample to be predicted.
4. according to the method described in claim 1, it is characterized in that, the grader is Softmax graders.
5. a kind of handwriting recongnition system, which is characterized in that including:
Training sample processing unit, for using with smooth norm L1Self-encoding encoder each trained sample that training sample is concentrated
This is handled, and corresponding target training sample, the sample mark that the target training sample is concentrated with the training sample are obtained
Label composition target training sample set, the smooth norm L of band1Self-encoding encoder object function in be equipped with sparse penalty term, should
Sparse penalty term is smooth L1Norm;
Classifier training unit obtains object classifiers for training grader using the target training sample set;
Sample to be tested processing unit, for using with smooth norm L1Self-encoding encoder treat forecast sample and handled, obtain mesh
Mark sample to be predicted;
Classification determination unit, for target sample to be predicted to be input to the object classifiers, with determination sample to be predicted
This classification;
The training sample processing unit includes:
Parameter definition unit is for defining training sample set:
Wherein, y(i)It is and training sample x(i)Corresponding sample label, m are the numbers of training sample, and d is training sample dimension;
Define self-encoding encoder hypothesis function be:
hW,b(x(i))
Wherein, W and b indicates weight and the biasing of self-encoding encoder respectively;
The output for defining j-th of hidden unit of i-th of training sample is expressed asAnd the number of hidden unit is n;
Object function determination unit, for determining with smooth norm L1The object function of self-encoding encoder be:
Wherein, first item is reconstruct item, and Section 2 is weight attenuation term, and λ is weight attenuation coefficient, and Section 3 is sparse punishment
, β is the weight of coefficient penalty factor, and S () indicates smooth L1Norm, it is specific as follows:
Wherein, μ > 0 are parameter preset;
Object function solves unit, for solving the parameter W so that the object function minimumoptAnd bopt;
Assuming that function determination unit, is used for WoptAnd boptIt brings into the hypothesis function of self-encoding encoder, obtains goal hypothesis function;
Target training sample acquiring unit, the training sample x for concentrating training sample(i)Bring the goal hypothesis function into,
Obtain target training sample a(i)。
6. system according to claim 5, which is characterized in that solving the parameter W so that the object function minimumopt
And boptWhen, it is calculated using back-propagation algorithm.
7. system according to claim 5, which is characterized in that the sample to be tested processing unit includes:
First sample to be tested processing subelement obtains target for bringing the sample to be predicted into the goal hypothesis function
Sample to be predicted.
8. system according to claim 5, which is characterized in that the grader is Softmax graders.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510001954.9A CN104484684B (en) | 2015-01-05 | 2015-01-05 | A kind of Manuscripted Characters Identification Method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510001954.9A CN104484684B (en) | 2015-01-05 | 2015-01-05 | A kind of Manuscripted Characters Identification Method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104484684A CN104484684A (en) | 2015-04-01 |
CN104484684B true CN104484684B (en) | 2018-11-02 |
Family
ID=52759225
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510001954.9A Active CN104484684B (en) | 2015-01-05 | 2015-01-05 | A kind of Manuscripted Characters Identification Method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104484684B (en) |
Families Citing this family (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107391996B (en) * | 2017-08-02 | 2021-01-26 | 广东工业大学 | Identity verification method and device based on L1 norm neural network |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103544392A (en) * | 2013-10-23 | 2014-01-29 | 电子科技大学 | Deep learning based medical gas identifying method |
CN103778432A (en) * | 2014-01-08 | 2014-05-07 | 南京邮电大学 | Human being and vehicle classification method based on deep belief net |
CN104077580A (en) * | 2014-07-15 | 2014-10-01 | 中国科学院合肥物质科学研究院 | Pest image automatic recognition method based on high-reliability network |
-
2015
- 2015-01-05 CN CN201510001954.9A patent/CN104484684B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103544392A (en) * | 2013-10-23 | 2014-01-29 | 电子科技大学 | Deep learning based medical gas identifying method |
CN103778432A (en) * | 2014-01-08 | 2014-05-07 | 南京邮电大学 | Human being and vehicle classification method based on deep belief net |
CN104077580A (en) * | 2014-07-15 | 2014-10-01 | 中国科学院合肥物质科学研究院 | Pest image automatic recognition method based on high-reliability network |
Non-Patent Citations (1)
Title |
---|
"基于稀疏自编码深度神经网络的林火图像分类";王勇 等;《计算机工程与应用》;20141215;174页左边栏倒数第1段-右边栏第1段,175页左边栏,176页左边栏倒数第11行-倒数第8行,176页右边栏第2段 * |
Also Published As
Publication number | Publication date |
---|---|
CN104484684A (en) | 2015-04-01 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Lin et al. | Forecasting concentrations of air pollutants by logarithm support vector regression with immune algorithms | |
CN100380396C (en) | Object detection apparatus, learning apparatus, object detection system, object detection method | |
CN103258214A (en) | Remote sensing image classification method based on image block active learning | |
CN1307579C (en) | Methods and apparatus for classifying text and for building a text classifier | |
CN105184298B (en) | A kind of image classification method of quick local restriction low-rank coding | |
Anil et al. | Convolutional neural networks for the recognition of Malayalam characters | |
CN110288030A (en) | Image-recognizing method, device and equipment based on lightweight network model | |
CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
CN104200224A (en) | Valueless image removing method based on deep convolutional neural networks | |
CN103699523A (en) | Product classification method and device | |
Katiyar et al. | A hybrid recognition system for off-line handwritten characters | |
CN104834940A (en) | Medical image inspection disease classification method based on support vector machine (SVM) | |
CN106778863A (en) | The warehouse kinds of goods recognition methods of dictionary learning is differentiated based on Fisher | |
CN103426004B (en) | Model recognizing method based on error correcting output codes | |
CN103020122A (en) | Transfer learning method based on semi-supervised clustering | |
CN109635946A (en) | A kind of combined depth neural network and the clustering method constrained in pairs | |
CN110069959A (en) | A kind of method for detecting human face, device and user equipment | |
CN113159171B (en) | Plant leaf image fine classification method based on counterstudy | |
CN110472417A (en) | Malware operation code analysis method based on convolutional neural networks | |
CN106991049A (en) | A kind of Software Defects Predict Methods and forecasting system | |
CN105976390A (en) | Steel tube counting method by combining support vector machine threshold statistics and spot detection | |
Li et al. | Dating ancient paintings of Mogao Grottoes using deeply learnt visual codes | |
CN110414587A (en) | Depth convolutional neural networks training method and system based on progressive learning | |
CN105893941B (en) | A kind of facial expression recognizing method based on area image | |
Chen et al. | A semisupervised deep learning framework for tropical cyclone intensity estimation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20220310 Address after: 200000 Room 502, floor 5, No. 250, JIANGCHANG Third Road, Jing'an District, Shanghai Patentee after: Ren Tuo data technology (Shanghai) Co.,Ltd. Address before: 215123 No. 199 benevolence Road, Suzhou Industrial Park, Jiangsu, China Patentee before: SOOCHOW University |
|
TR01 | Transfer of patent right |