CN108615052A - A kind of image-recognizing method without under similar training sample situation - Google Patents
A kind of image-recognizing method without under similar training sample situation Download PDFInfo
- Publication number
- CN108615052A CN108615052A CN201810335966.9A CN201810335966A CN108615052A CN 108615052 A CN108615052 A CN 108615052A CN 201810335966 A CN201810335966 A CN 201810335966A CN 108615052 A CN108615052 A CN 108615052A
- Authority
- CN
- China
- Prior art keywords
- attribute
- training sample
- sample
- known class
- tested
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
- G06F18/24133—Distances to prototypes
- G06F18/24143—Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image-recognizing method without under similar training sample situation, this method can effectively reduce the influence that semantic migration and known class training sample attribute noise problem are brought in without similar training sample situation hypograph identification process.We go out the real property of known class training sample to reduce the influence that its attribute noise problem is brought using just too distributing virtual, learn an attribute forecast device using encoder-decoder model secondly based on known class training sample feature and its virtual real property influences to achieve the purpose that effectively reduce semantic migration problem, and the label of unknown classification sample to be tested is obtained finally by nearest neighbor classifier.Compared with existing other methods, our method is obtained on discrimination and recognition rate and is significantly improved.
Description
Technical field
The present invention relates to field of image recognition in pattern-recognition more particularly to a kind of without under similar training sample situation
Image-recognizing method.
Background technology
In area of pattern recognition, image recognition is one important branch, and the purpose is to allow trained computer to exist
Information is extracted in testing image and is identified.Why image recognition is furtherd investigate, and quickly grows, be because its extensively and
Important application value.Image recognition technology is increasingly ripe at present, is widely used in face, the identification of number and other objects
In.In image recognition a large amount of similar images of training are just needed to obtain relatively good discrimination.It provides and largely carries label
Similar training sample be costly, and it is possible that collect the feelings of the similar training sample less than certain a kind of image
Condition, so how we solve to seem particularly important without the problem of image recognition under similar training sample situation, and this is asked
Topic is increasingly paid close attention to by industrial quarters.
Solution is built upon without most of the conventional method of image recognition under similar training sample situation on the basis of attribute
's.Attribute is artificially defined to describe category Properties information.Directly belong in image recognition under no similar training sample situation
Property prediction (DAP) and proxy attribute prediction (IAP) be most basic two kinds of methods based on attribute.DAP is to utilize supporting vector
Machine SVM learns an attribute forecast device from known class training sample, is then applied directly to this attribute forecast device to be measured
It tries unknown classification sample and obtains the expression of its attribute, classify finally by nearest neighbor classifier.Due to direct attribute forecast method
Based on attribute, when attribute changes, attributive classification device needs re -training, so directly attribute forecast method lacks spirit
Activity.IAP is to learn a category classifier on known class training sample using support vector machines, is then answered
It uses unknown classification and obtains the output of its class probability, bayes method is finally utilized to obtain the class label of unknown classification image.
These can all encounter semantic migration problem based on attribute without the image-recognizing method under similar training sample situation.It is specific next
Say, from known class training sample go to school acquistion to attribute forecast device may be used in known class training sample, however by
In known class training sample and unknown classification test sample and intersection is not present, directly answers the attribute forecast device that study is got
All unknown classifications can not be necessarily suitble to when using unknown classification to be tested, can not thus obtain all unknown classifications to be tested
The real property of sample just will produce semantic migration problem, to which final discrimination can be caused to decline.How one is learnt efficiently
Attribute forecast device is come to effectively reduce the influence that semantic migration problem is brought be an important project in image recognition research.However
Ignorance attribute this key element is often all easy while considering to learn an efficient attribute forecast device, attribute is all people
There is error certainly between definition, with its real property value, that is, there is noise, the attribute of given known class training sample is retouched
It states and deviates its real property certainly, how to effectively reduce the influence of training sample attribute noise is also worth it is contemplated that research.
Invention content
Goal of the invention：The present invention provides a kind of image-recognizing methods without under similar training sample situation effectively to subtract
The influence that light semantic migration and training sample attribute noise problem are brought.
Technical solution：A kind of image-recognizing method without under similar training sample situation, includes the following steps：
Step 1) utilizes just too distributing virtual for the attribute noise problem for including in given known class training sample
Go out the real property of known class training sample to reduce the influence that attribute noise problem is brought；
Wherein, include C in given known class training samplesClass amounts to Ns width images, the training of given known class
Sample set representations are as follows：
Wherein, the character representation of given known class training sample is as follows：
In above formula, d is sample characteristics dimension；
Wherein, the attribute of given known class training sample indicates as follows：
In above formula, k is sample attribute dimension；
Given known class training sample concentrates all known class prototype attributes for including to indicate as follows：
According to the attribute of given known class training sample, the known class training sample that just too distributing virtual goes out is utilized
Real property indicate it is as follows：
Step 2) is directed to without semantic migration problem present in the identification of similar training sample situation hypograph, according to step 1)
Calculated given known class training sample real property, one feature from known class training sample of study is to known
The encoder of classification training sample real property, then learn one and known class training sample real property is mapped to feature sky
Between decoder, the category of unknown classification sample to be tested is predicted using one attribute forecast device of coder-decoder model learning
Property；The detailed step of step 2) is as follows：
Step 2.1) learns one from known class training sample feature XsThe real property Y virtual to itssEncoder W
∈Rk×dThat is WXs=Ys, then learn one again by YsIt is mapped to the decoder W of feature space*, obtain one it is new about known
The character representation X ' of classification training sample is X '=W*Ys=W*WXs, in order to enable X ' and XsBetween error it is as small as possible, therefore
Write out following object function：
Wherein | | | |FInclude two variables for F- norms, in formula (6), it is convenient for abbreviation, it is assumed that W*W=WTFormula (6) because
This is write as：
Wherein λ is over-fitting control coefrficient, and in conjunction with step 1), we write out final object function：
The virtual reality attribute of the i-th width known class training sample, i ∈ 1,2 ... Ns,It is known to jth class
The prototype attribute of classification training sample, j ∈ 1,2......CsIfBelong to jth class known class then mI, j=1, otherwise mI, j
=0, λ1For over-fitting control coefrficient；Formula (8) is related with two variables, is broken down into two optimization problems to carry out respectively
It solves：
For formula (9), we carry out derivation zero setting to it to obtain：
W=sylvester (A, B, C) (11)
Formula (11) is a sylvester equation, wherein
It is same to be obtained using the method for derivation zero setting for formula (10)：
Wherein For NsThe unit matrix of rank；
Step 2.2) finds out Y using the method being separately optimizedsWith the expression formula of W, following we utilize following iteration
Formula finds out optimal Ys：
Wherein α=0.01, YiIt is the known class training sample virtual attribute expression of ith iteration, determines optimal Ys=
YiAfterwards, we can find out optimum code device i.e. attribute forecast device W using formula (11).
The attribute forecast phase that step 3) is obtained with step 2) predicts the attribute of unknown classification sample to be tested, and will predict
The attribute of unknown classification sample to be tested compared by the prototype attribute of nearest neighbor classifier and unknown classification sample to be tested
To obtain the label of unknown classification sample to be tested.The detailed step of step 3) is as follows：
Step 3.1) chooses CtClass unknown classification sample to be tested is as model measurement sample, and prototype attribute is known and table
Show as follows：
Step 3.2) utilizes the attribute forecast that step 2) obtainsTo predict CtThe attribute of class sample to be tested, finally pre-
The attribute for the sample to be tested surveyed and given CtThe prototype attribute of class sample to be tested compares to obtain by nearest neighbor classifier
The label of all unknown classification samples, wherein we are using following nearest neighbor classifier model：
Wherein CjFor the label j of k-th of unknown classification sample to be tested, wherein k ∈ 1,2,,, t；It is to be measured for k-th
The attribute of unknown classification sample is tried,For the prototype attribute of the unknown classification of jth class.
Advantageous effect：1. we go out the real property of known class training sample and effectively mitigate instruction using just too distributing virtual
Practice the influence that the attribute noise problem of collection database is brought.
Effectively mitigate without similar trained sample 2. we are reached by one attribute forecast device of coder-decoder model learning
The purpose that semantic migration problem influences in image recognition under this situation.
3. being compared with existing method, the method that we are proposed is significantly increased on discrimination and recognition rate.
Description of the drawings
Fig. 1 is the work flow diagram of the present invention.
Specific implementation mode
A kind of image-recognizing method without under similar training sample situation, includes the following steps：
Step 1) utilizes just too distributing virtual for the attribute noise problem for including in given known class training sample
Go out the real property of known class training sample to reduce the influence that attribute noise problem is brought；
Wherein, include C in given known class training samplesClass amounts to Ns width images, the training of given known class
Sample set representations are as follows：
Wherein, the character representation of given known class training sample is as follows：
In above formula, d is sample characteristics dimension；
Wherein, the attribute of given known class training sample indicates as follows：
In above formula, k is sample attribute dimension；
Given known class training sample concentrates all known class prototype attributes for including to indicate as follows：
In traditional mode identification, belong to same category of several samples in attribute space in being just distributed very much, therefore I
Using just too distributing virtual goes out the real property of known class training sample；According to the known class training given in formula (3)
The attribute of sample indicates as follows using the real property for the known class training sample that just too distributing virtual goes out：
Step 2) is directed to without semantic migration problem present in the identification of similar training sample situation hypograph, according to step 1)
Calculated given known class training sample real property, one feature from known class training sample of study is to known
The encoder of classification training sample real property, then learn one and known class training sample real property is mapped to feature sky
Between decoder, the category of unknown classification sample to be tested is predicted using one attribute forecast device of coder-decoder model learning
Property；The detailed step of step 2) is as follows：
Step 2.1) learns one from known class training sample feature XsThe real property Y virtual to itssEncoder W
∈Rk×dThat is WXs=Ys, then learn one again by YsIt is mapped to the decoder W of feature space*, obtain one it is new about known
The character representation X ' of classification training sample is x '=W*Ys=W*WXs, in order to enable X ' and XsBetween error it is as small as possible, therefore
Write out following object function：
Wherein | | | |FInclude two variables for F- norms, in formula (6), it is convenient for abbreviation, it is assumed that W*W=WTFormula (6) because
This is write as：
Wherein λ is over-fitting control coefrficient, and in conjunction with step 1), we write out final object function：
The virtual reality attribute of the i-th width known class training sample, i ∈ 1,2 ... Ns,It is known to jth class
The prototype attribute of classification training sample, j ∈ 1,2......Cs：IfBelong to jth class known class then mI, j=1, otherwise mI, j
=0, λ1For over-fitting control coefrficient；Formula (8) is related with two variables, is broken down into two optimization problems to carry out respectively
It solves：
For formula (9), we carry out derivation zero setting to it to obtain：
W=sylvester (A, B, C) (11)
Formula (11) is a sylvester equation, wherein
It is same to be obtained using the method for derivation zero setting for formula (10)：
Wherein For NsThe unit matrix of rank；
Step 2.2) finds out Y using the method being separately optimizedsWith the expression formula of W, following we utilize following iteration
Formula finds out optimal Ys：
Wherein α=0.01, YiIt is the known class training sample virtual attribute expression of ith iteration, determines optimal Ys=
YiAfterwards, we can find out optimum code device i.e. attribute forecast device W using formula (11).
Step 3) chooses CtClass unknown classification sample to be tested as test sample, prototype attribute and classification it is known that
Feature can directly be extracted from sample；The attribute forecast device W obtained using step 2) passes through CtClass unknown classification sample to be tested
Feature to predict the attribute of unknown classification sample to be tested, and the attribute of the unknown classification sample to be tested predicted is passed through
The prototype attribute of nearest neighbor classifier and unknown classification sample to be tested is compared to obtain the classification of unknown classification sample to be tested
Thus label judges the accuracy of attribute forecast device W；The detailed step of step 3) is as follows：
Step 3.1) chooses CtClass unknown classification sample to be tested is as model measurement sample, and prototype attribute is known and table
Show as follows：
Step 3.2) utilizes the attribute forecast that step 2) obtainsTo predict CtThe attribute of class sample to be tested, finally pre-
The attribute for the sample to be tested surveyed and given CtThe prototype attribute of class sample to be tested compares to obtain by nearest neighbor classifier
The label of all unknown classification samples, wherein we are using following nearest neighbor classifier model：
Wherein CjFor the label j of k-th of unknown classification sample to be tested, wherein k ∈ 1,2,,, t；It is to be measured for k-th
The attribute of unknown classification sample is tried,For the prototype attribute of the unknown classification of jth class.
By a kind of image-recognizing method without under similar training sample situation of the present invention in AWA and CUB databases
On tested.
Amount to 30475 photos comprising 50 class animals in AWA databases, the attribute of a row 85 dimension is corresponded to per one kind image
Vector.In AWA databases, we take out 40 classes as known class training sample, remaining 10 class is as unknown class to be tested
Very originally.Containing a total of 11788 pictures of 200 kinds of different birds in CUB databases, corresponding per pictures one
The attribute vector of 312 dimensions.We take out 150 classes as known class training sample in CUB databases, and remaining 50 classes, which are used as, to be waited for
Test unknown classification sample.The discrimination that we test various methods in two databases is as follows：
Table one：The discrimination of various methods compares on AWA and CUB databases
| Method | AWA databases | CUB databases |
| SAE | 81.4% | 57.4% |
| RKT | 66.2% | 38.4% |
| SMS | 74.8% | 43.6% |
| RZSL | 65.6% | 31.4% |
| Our Method | 82.6% | 58.1% |
Table two：Each method recognition rate situation compares on AWA databases
| Method | Training stage | Test phase |
| RKT | 49.71s | 6.65s |
| RZSL | 71.13s | 15.85s |
| SMS | 59.63s | 11.65s |
| Our Method | 13.78s | 0.12s |
Method proposed by the present invention is can be seen that in same database than its other party from the result in table one and table two
Method all significantly improves on discrimination and recognition rate.
Those skilled in the art of the present technique are it is understood that unless otherwise defined, all terms used herein (including skill
Art term and scientific terminology) there is meaning identical with the general understanding of the those of ordinary skill in fields of the present invention.Also
It should be understood that those terms such as defined in the general dictionary should be understood that with in the context of the prior art
The consistent meaning of meaning, and unless defined as here, will not be explained with the meaning of idealization or too formal.
Above-described specific implementation mode has carried out further the purpose of the present invention, technical solution and advantageous effect
It is described in detail, it should be understood that the foregoing is merely the specific implementation mode of the present invention, is not limited to this hair
Bright, all within the spirits and principles of the present invention, any modification, equivalent substitution, improvement and etc. done should be included in the present invention
Protection domain within.
Claims (4)
1. a kind of image-recognizing method without under similar training sample situation, it is characterised in that：Include the following steps：
Step 1) is for the attribute noise problem for including in given known class training sample, using just too distributing virtual goes out
Know the real property of classification training sample；
Step 2) is directed to without semantic migration problem present in the identification of similar training sample situation hypograph, is calculated according to step 1)
The given known class training sample real property gone out, feature of the study one from known class training sample to known class
The encoder of training sample real property, then learn one and known class training sample real property is mapped to feature space
Decoder predicts the attribute of unknown classification sample to be tested using one attribute forecast device of coder-decoder model learning；
The attribute forecast device that step 3) is obtained with step 2) predicts the attribute of unknown classification sample to be tested, and is waited for what is predicted
The attribute for testing unknown classification sample is compared by the prototype attribute of nearest neighbor classifier and unknown classification sample to be tested, with
To the label of unknown classification sample to be tested.
2. a kind of image-recognizing method without under similar training sample situation according to claim 1, it is characterised in that：Step
It is rapid 1) described in given known class training sample in include CsClass amounts to Ns width images, and given known class trains sample
This set representations is as follows：
Wherein, the character representation of given known class training sample is as follows：
In above formula, d is sample characteristics dimension；
Wherein, the attribute of given known class training sample indicates as follows：
In above formula, k is sample attribute dimension；
Given known class training sample concentrates all known class prototype attributes for including to indicate as follows：
According to the attribute of given known class training sample, the true of the known class training sample that just too distributing virtual goes out is utilized
Real attribute indicates as follows：
3. a kind of image-recognizing method without under similar training sample situation according to claim 2, it is characterised in that：Institute
The detailed step for stating step 2) is as follows：
Step 2.1) learns one from known class training sample feature XsThe real property Y virtual to itssEncoder W ∈ Rk×d
That is WXs=Ys, then learn one again by YsIt is mapped to the decoder W of feature space*, obtain one it is new about known class
The character representation X ' of training sample is X '=W*Ys=W*WXs, in order to enable X ' and XsBetween error it is as small as possible, therefore write out
Following object function：
Wherein | | | |FInclude two variables for F- norms, in formula (6), it is convenient for abbreviation, it is assumed that W*W=WTIt formula (6) therefore writes
At：
Wherein λ is over-fitting control coefrficient, and in conjunction with step 1), we write out final object function：
The virtual reality attribute of the i-th width known class training sample, i ∈ 1,2 ... Ns,It is jth class known class
The prototype attribute of training sample, j ∈ 1,2......Cs：IfBelong to jth class known class then mI, j=1, otherwise mI, j=0,
λ1For over-fitting control coefrficient；Formula (8) is related with two variables, is broken down into two optimization problems to be asked respectively
Solution：
For formula (9), we carry out derivation zero setting to it to obtain：
W=sylvester (A, B, C) (11)
Formula (11) is a sylvester equation, wherein
It is same to be obtained using the method for derivation zero setting for formula (10)：
Wherein For NsThe unit matrix of rank；
Step 2.2) finds out Y using the method being separately optimizedsWith the expression formula of W, following we are asked using following iterative formula
Go out optimal Ys：
Wherein α=0.01, YiIt is the known class training sample virtual attribute expression of ith iteration, determines optimal Ys=Yi
Afterwards, we can find out optimum code device i.e. attribute forecast device W using formula (11).
4. a kind of image-recognizing method without under similar training sample situation according to claim 3, it is characterised in that：Institute
The detailed step for stating step 3) is as follows：
Step 3.1) chooses CtFor class unknown classification sample to be tested as model measurement sample, prototype attribute is known and indicates such as
Under：
Attribute forecast device W that step 3.2) is obtained using step 2) predicts CtThe attribute of class sample to be tested, finally prediction
The attribute of sample to be tested and given CtThe prototype attribute of class sample to be tested is compared by nearest neighbor classifier to be owned
The label of unknown classification sample, wherein we are using following nearest neighbor classifier model：
Wherein CjFor the label j of k-th of unknown classification sample to be tested, wherein k ∈ 1,2,,, t；For k-th it is to be tested not
Know the attribute of classification sample,For the prototype attribute of the unknown classification of jth class.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810335966.9A CN108615052A (en) | 2018-04-13 | 2018-04-13 | A kind of image-recognizing method without under similar training sample situation |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810335966.9A CN108615052A (en) | 2018-04-13 | 2018-04-13 | A kind of image-recognizing method without under similar training sample situation |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108615052A true CN108615052A (en) | 2018-10-02 |
Family
ID=63660071
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810335966.9A Pending CN108615052A (en) | 2018-04-13 | 2018-04-13 | A kind of image-recognizing method without under similar training sample situation |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108615052A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909643A (en) * | 2019-11-14 | 2020-03-24 | 北京航空航天大学 | Remote sensing ship image small sample classification method based on nearest neighbor prototype representation |
| CN113409821A (en) * | 2021-05-27 | 2021-09-17 | 南京邮电大学 | Method for recognizing unknown emotional state of voice signal |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160292538A1 (en) * | 2015-03-31 | 2016-10-06 | Disney Enterprises, Inc. | Object Classification Through Semantic Mapping |
| CN106980875A (en) * | 2017-03-13 | 2017-07-25 | 南京邮电大学 | The zero sample image recognition methods based on attribute low-rank representation |
| CN107563444A (en) * | 2017-09-05 | 2018-01-09 | 浙江大学 | A kind of zero sample image sorting technique and system |
-
2018
- 2018-04-13 CN CN201810335966.9A patent/CN108615052A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160292538A1 (en) * | 2015-03-31 | 2016-10-06 | Disney Enterprises, Inc. | Object Classification Through Semantic Mapping |
| CN106980875A (en) * | 2017-03-13 | 2017-07-25 | 南京邮电大学 | The zero sample image recognition methods based on attribute low-rank representation |
| CN107563444A (en) * | 2017-09-05 | 2018-01-09 | 浙江大学 | A kind of zero sample image sorting technique and system |
Non-Patent Citations (3)
| Title |
|---|
| ELYOR KODIROV 等: "Semantic Autoencoder for Zero-Shot Learning", 《CVPR》 * |
| KUN WANG 等: "Learning Autoencoder of Attribute Constraint for Zero-Shot Classification", 《2017 4TH IAPR ASIAN CONFERENCE ON PATTERN RECOGNITION》 * |
| 冀中 等: "基于典型相关分析和距离度量学习的零样本学习", 《天津大学学报(自然科学与工程技术版)》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110909643A (en) * | 2019-11-14 | 2020-03-24 | 北京航空航天大学 | Remote sensing ship image small sample classification method based on nearest neighbor prototype representation |
| CN110909643B (en) * | 2019-11-14 | 2022-10-28 | 北京航空航天大学 | Remote sensing ship image small sample classification method based on nearest neighbor prototype representation |
| CN113409821A (en) * | 2021-05-27 | 2021-09-17 | 南京邮电大学 | Method for recognizing unknown emotional state of voice signal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Khalil et al. | Energy efficiency prediction using artificial neural network | |
| US12061966B2 (en) | Relevance score assignment for artificial neural networks | |
| CN109766557B (en) | Emotion analysis method and device, storage medium and terminal equipment | |
| CN114912612A (en) | Bird identification method and device, computer equipment and storage medium | |
| CN106980876A (en) | A kind of zero sample image recognition methods learnt based on distinctive sample attribute | |
| Wang et al. | Research on maize disease recognition method based on improved resnet50 | |
| CN116310519A (en) | Surface defect classification method for semi-supervised deep learning | |
| CN111274396B (en) | Visual angle level text emotion classification method and system based on external knowledge | |
| CN116432184A (en) | Malicious software detection method based on semantic analysis and bidirectional coding characterization | |
| CN116186250A (en) | Multi-mode learning level mining method, system and medium under small sample condition | |
| Rodzin et al. | Deep learning techniques for natural language processing | |
| CN108615052A (en) | A kind of image-recognizing method without under similar training sample situation | |
| CN113779249B (en) | Cross-domain text emotion classification method and device, storage medium and electronic equipment | |
| CN109034182A (en) | A kind of zero sample image identification new method based on attribute constraint | |
| Vargas et al. | Relu-based activations: Analysis and experimental study for deep learning | |
| Makwe et al. | An empirical study of neural network hyperparameters | |
| US20230121404A1 (en) | Searching for normalization-activation layer architectures | |
| CN112489689B (en) | Cross-database voice emotion recognition method and device based on multi-scale difference countermeasure | |
| CN110222737A (en) | A kind of search engine user satisfaction assessment method based on long memory network in short-term | |
| CN116975743A (en) | Industry information classification method, device, computer equipment and storage medium | |
| CN111598580A (en) | XGboost algorithm-based block chain product detection method, system and device | |
| CN106095811A (en) | A kind of image search method of the discrete Hash of supervision based on optimum code | |
| CN110851600A (en) | Text data processing method and device based on deep learning | |
| Horrace et al. | Lasso for stochastic frontier models with many efficient firms | |
| CN117009621A (en) | Information searching method, device, electronic equipment, storage medium and program product |
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: 20181002 |
|
| RJ01 | Rejection of invention patent application after publication |