CN108564132A - A method of classified to depth characteristic based on integrated supporting vector machine - Google Patents
A method of classified to depth characteristic based on integrated supporting vector machine Download PDFInfo
- Publication number
- CN108564132A CN108564132A CN201810378739.4A CN201810378739A CN108564132A CN 108564132 A CN108564132 A CN 108564132A CN 201810378739 A CN201810378739 A CN 201810378739A CN 108564132 A CN108564132 A CN 108564132A
- Authority
- CN
- China
- Prior art keywords
- depth characteristic
- vector machine
- supporting vector
- integrated supporting
- characteristic based
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Abstract
The present invention provides a kind of method classified to depth characteristic based on integrated supporting vector machine, is related to digital signal processing technique field, a method of classified to depth characteristic based on integrated supporting vector machine, is included the following steps：S1：Input depth characteristic file to be sorted；S2：Establish first predetermined value kind dimensionality reduction dimension；S3：Establish second predetermined value kind kernel function；S4：The grader output information under any of the above-described dimensionality reduction dimension and the combination of any kernel function is obtained successively；S5：Each grader output information is voted by combining, obtains the classification results of depth characteristic file to be sorted.A kind of depth characteristic information extraction accuracy height for the method classified to depth characteristic based on integrated supporting vector machine of the present invention, multiple dimensionality reduction dimensions obtain information, and the result accuracy for prediction of classifying is high, can be effectively to the classification of depth characteristic and classification.
Description
Technical field
The present invention relates to digital signal processing technique field,
Especially, the present invention relates to a kind of methods classified to depth characteristic based on integrated supporting vector machine.
Background technology
Science and technology development picture pick-up device is popularized, the digital signal datas such as video data of enormous amount also with
Generation.Meanwhile it also coming into being for the application of digital signal：Intelligent video monitoring, video data classification, advanced man-machine friendship
Mutually etc..In such applications, for the action of people understand to be in the core of most crucial focus and people's research
Hold.
Since human action identification has prodigious potential value, so this project continue for as a research hotspot
At least ten years benefited from the huge progress acquired by image recognition in recent years, and video identification similarly achieves prodigious
Progress.Wherein most of action identification methods are suggested, many traditional classification of motion sides both for classification of motion task
The feature of method extraction is mostly the feature that can be directly extracted from image using manual feature extraction.Such as color characteristic, texture
Feature and histogram feature, these features are directed to the simple image of content, can there is good effect, but encounter content complexity
When image, manual feature is unable to Efficient Characterization image property, can not effectively be classified to these depth characteristics.
Such as Chinese patent patent application CN107844795A discloses a kind of convolutional Neural net based on principal component analysis
Network feature extracting method, this method have chosen on Imagenet data sets trained convolutional neural networks first, will
Then feature extractor of the network as image extracts feature from the output of each pond layer of the convolutional neural networks and reflects
Penetrate figure, principal component analysis finally be utilized as the depth characteristic of image in every layer of extraction all Feature Mapping figures, to its into
It has gone dimensionality reduction, and has utilized bilinear interpolation, last result Feature Mapping figure is reset to original image size, obtained efficiently
Picture depth feature.The depth characteristic that the present invention obtains, containing the semantic information that image is abundant, and characteristic dimension is low, data
It measures small, can be used for various identifications and the classification task of image.
However, the above method still remains following disadvantage：1, the low dimensional characteristic reading after dimensionality reduction is only carried out,
Can not various dimensions obtain information jointly, it is inadequate that feature reads accuracy；2, depth characteristic extraction identification process is complicated, after reading
Without classification analysis process, effective classification of depth characteristic can not be carried out.
Invention content
The purpose of the present invention is to provide a kind of depth characteristic information extraction accuracy height, multiple dimensionality reduction dimensions obtain letter
The result accuracy of breath, prediction of classifying is high, can be effectively to the classification of depth characteristic and classification based on integrated supporting vector machine
The method classified to depth characteristic.
To solve the above problems, the present invention is adopted the following technical scheme that and is achieved：
A method of classified to depth characteristic based on integrated supporting vector machine, is included the following steps：
S1：Input depth characteristic file to be sorted；
S2：Establish first predetermined value kind dimensionality reduction dimension；
S3：Establish second predetermined value kind kernel function；
S4：The grader output information under any of the above-described dimensionality reduction dimension and the combination of any kernel function is obtained successively；
S5：Each grader output information is voted by combining, obtains the classification results of depth characteristic file to be sorted.
Preferably, when executing step S1, depth characteristic file to be sorted includes text, picture, animation and video.
Preferably, before executing step S2, according to depth characteristic file type to be sorted, first predetermined value is generated.
Preferably, when executing step S2, the dimensionality reduction of different dimensions is carried out using principal component analytical method.
Preferably, after executing step S2, by established dimensionality reduction dimension, the eigenmatrix of various dimensions depth is formed.
Preferably, before executing step S3, according to depth characteristic file type to be sorted, second predetermined value is generated.
Preferably, when executing step S4, the quantity of output information is the product of first predetermined value and second predetermined value.
Preferably, when executing step S4, the quantity of grader is identical as the quantity of output information.
Preferably, step S5 is specially：
S51：The output of each grader is input to combination vote module；
S52：Carry out weight combination ballot；
S53：The result won of voting is final classification result.
Preferably, when executing step S53, the high result of ballot scoring is the result that ballot is won.
The present invention is a kind of to be the method advantageous effect that depth characteristic is classified based on integrated supporting vector machine：Depth
Feature information extraction accuracy is high, and multiple dimensionality reduction dimensions obtain information, and the result accuracy for prediction of classifying is high, can be effectively to depth
Spend the classification and classification of feature.
Description of the drawings
Fig. 1 is the flow chart of one embodiment of the invention.
Specific implementation mode
The following is specific embodiments of the present invention, and technical scheme of the present invention will be further described, but the present invention is simultaneously
It is not limited to these embodiments.
Carry out the various exemplary embodiments of detailed description of the present invention now with reference to attached drawing.It should be noted that：Unless in addition having
Body illustrates that the positioned opposite and step of the module and step that otherwise illustrate in these embodiments does not limit the scope of the invention.
The development of science and technology makes picture pick-up device be popularized, and the digital signal datas such as video data of enormous amount also produce therewith
It is raw.Meanwhile it also coming into being for the application of digital signal：Intelligent video monitoring, video data classification, advanced human-computer interaction
Deng.In such applications, the core content that understanding is most crucial focus and people's research is carried out for the action of people.
It is described since it cannot be inferred in the depth characteristic of the movement properties or pedestrian by pedestrian in the picture
One real pedestrian or other and pedestrian have the non-pedestrian target of the similar characteristics of motion, therefore using based on depth letter
Support vector machines (the SVM of the feature description of breath：Support Vector Machine) it is handled, depth characteristic is carried
It takes.
Embodiment one：As shown in Figure 1, only one of present invention embodiment, to solve the above problems, the present invention adopts
It is achieved with following technical solution：A method of classified to depth characteristic based on integrated supporting vector machine, including
Following steps：
S1：Input depth characteristic file to be sorted；
To classify to depth characteristic, it is necessary first to obtain depth characteristic file to be sorted, and to be sorted according to this
File is identified and classifies, and could have and targetedly classify.
S2：Establish first predetermined value kind dimensionality reduction dimension；
Multiple dimensionality reduction is carried out, establishes multiple and different dimensionality reduction dimensions, the number of dimensionality reduction dimension is first predetermined value, and described first is pre-
The size of definite value is different according to the difference of the type of the file of depth characteristic to be sorted, also for convenient for different
The file of depth characteristic to be sorted carries out in various degree and the identification of depth identifies accurate road in order to improve.
S3：Establish second predetermined value kind kernel function；
When support vector machines works, multigroup kernel function is generated, the number of kernel function is second predetermined value, likewise, described the
The size of two predetermined values is also different according to the difference of the type of the file of depth characteristic to be sorted, is waited for point for different
The file of the depth characteristic of class, is identified and work of classifying using different kernel functions.
S4：The grader output information under any of the above-described dimensionality reduction dimension and the combination of any kernel function is obtained successively；
The kernel function of the dimensionality reduction dimension of first predetermined value number and second predetermined value number is combined, it is special to depth to be sorted
Part of soliciting articles is identified, and is then exported in the output end of grader, and obtain all output informations.
It is readily apparent that the quantity of grader output information here is multiplying for first predetermined value and second predetermined value
Product.Ensure that each dimensionality reduction dimension and each kernel function have combination, the result for identifying and exporting more comprehensively, facilitates in next step
Accurate classification.
S5：Each grader output information is voted by combining, obtains the classification of depth characteristic file to be sorted
As a result.
Specifically, step S5 is specially：
S51：The output of each grader is input to combination vote module；
S52：Carry out weight combination ballot；
S53：The result won of voting is final classification result.
It should be noted that when executing step S53, ballot score high result be vote win as a result, as most
The result foundation classified eventually.
First predetermined value is input to the output of the product kind grader of second predetermined value and combines in vote module in total,
The SVM of each dimensionality reduction feature and each kernel function may be constructed a grader, and the output of each grader carries out weight
Combination ballot, obtains final classification results, and more comprehensively, final classification result is more acurrate for such dimensional characteristics, to realization pair
The classification and classification of depth characteristic file to be sorted.
Embodiment two：Still as shown in Figure 1, only one of present invention embodiment, to further understand the present invention, the present invention
Also following explanation.
First, when executing step S1, depth characteristic file to be sorted includes text, picture, animation and video.
When in view of step S2 and S3, can according to depth characteristic file type to be sorted, come determine first predetermined value and
The size of second predetermined value, so after step S1, it is thus necessary to determine that the type of depth characteristic file to be sorted, if it is figure
Piece is then directed to picture extraction depth characteristic and then first takes frame if it is video, extract depth characteristic in each frame, generates different
First predetermined value and different second predetermined values, the step of then execution below again.
Secondly, before executing step S2, according to depth characteristic file type to be sorted, first predetermined value is generated.
Here it is first to generate dimensionality reduction dimension herein, then counts the number of the dimensionality reduction dimension of generation again, and not according to these
Same dimensionality reduction dimension carries out analysis and identification to depth characteristic.
Depth characteristic file to be sorted is read target exists in the otherness of target characteristic dimension on different scale
The local feature in partial structurtes feature and multiple dimensions on multiple scales carries out integration statistics, obtains all issuable
The synthesis of dimensionality reduction dimension.
Moreover, when executing step S2, the dimensionality reduction of different dimensions is carried out using principal component analytical method.
Theoretically, reduction process can carry out dimensionality reduction again and again, obtain different dimensionality reduction dimensions, can also be to wait for
Dimension layering is directly carried out in the depth characteristic file of classification, is carried out at the same time the dimensionality reduction of multiple dimensions, it might even be possible to which the two is simultaneously
It carries out, obtains more possible dimensionality reduction dimensions so that more comprehensively, classification results are more acurrate for the dimension got.
It is readily apparent that after executing step S2, by established dimensionality reduction dimension, the feature square of various dimensions depth is formed
Battle array.
Multigroup kernel function in this eigenmatrix and step S3 is combined, and obtains comprehensive grader branch rule, side
Just inspection-classification device output information.Also it is that last combination ballot and obtaining for classification results provide guarantee.
Then, before executing step S3, according to depth characteristic file type to be sorted, second predetermined value is generated.
According to the difference of depth characteristic file type to be sorted, support vector machines screens it, selects all tools
There is the non-linear supporting vector of kernel function, generates a variety of different kernel functions, ensure diversification and the standard of last output information
Trueization.
Finally, when executing step S4, the quantity of grader is identical as the quantity of output information.One grader generates one
Output information corresponds, that is, the combination of a dimensionality reduction dimension and a kernel function corresponds to a unique grader,
Generate unique output information.
So when executing combination ballot, all dimensionality reduction dimensions that can be executed and kernel function situation can be known
Not, final classification results are obtained.
The present invention has following gain effect：
A kind of depth characteristic information extraction essence for the method classified to depth characteristic based on integrated supporting vector machine of the present invention
Exactness is high, and multiple dimensionality reduction dimensions obtain information, and the result accuracy for prediction of classifying is high, can effectively to the classification of depth characteristic and
Classification.
Although some specific embodiments of the present invention are described in detail by example, the skill of this field
Art personnel it should be understood that above example merely to illustrate, the range being not intended to be limiting of the invention, belonging to the present invention
Those skilled in the art can make various modifications or additions to described specific embodiment or use class
As mode substitute, but without departing from the direction or beyond the scope of the appended claims of the present invention.Ability
Domain it is to be understood by the skilled artisans that it is every according to the technical essence of the invention to made by embodiment of above it is any modification, etc.
With replacement, improvement etc., protection scope of the present invention should be included in.
Claims (10)
1. a kind of method classified to depth characteristic based on integrated supporting vector machine, which is characterized in that include the following steps：
S1：Input depth characteristic file to be sorted；
S2：Establish first predetermined value kind dimensionality reduction dimension；
S3：Establish second predetermined value kind kernel function；
S4：The grader output information under any of the above-described dimensionality reduction dimension and the combination of any kernel function is obtained successively；
S5：Each grader output information is voted by combining, obtains the classification results of depth characteristic file to be sorted.
2. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
When executing step S1, depth characteristic file to be sorted includes text, picture, animation and video.
3. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
Before executing step S2, according to depth characteristic file type to be sorted, first predetermined value is generated.
4. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
When executing step S2, the dimensionality reduction of different dimensions is carried out using principal component analytical method.
5. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
After executing step S2, by established dimensionality reduction dimension, the eigenmatrix of various dimensions depth is formed.
6. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
Before executing step S3, according to depth characteristic file type to be sorted, second predetermined value is generated.
7. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
When executing step S4, the quantity of output information is the product of first predetermined value and second predetermined value.
8. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is：
When executing step S4, the quantity of grader is identical as the quantity of output information.
9. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 1, special
Sign is that step S5 is specially：
S51：The output of each grader is input to combination vote module；
S52：Carry out weight combination ballot；
S53：The result won of voting is final classification result.
10. a kind of method classified to depth characteristic based on integrated supporting vector machine according to claim 9, special
Sign is：
When executing step S53, the high result of ballot scoring is the result that ballot is won.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810378739.4A CN108564132A (en) | 2018-04-25 | 2018-04-25 | A method of classified to depth characteristic based on integrated supporting vector machine |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810378739.4A CN108564132A (en) | 2018-04-25 | 2018-04-25 | A method of classified to depth characteristic based on integrated supporting vector machine |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108564132A true CN108564132A (en) | 2018-09-21 |
Family
ID=63536436
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810378739.4A Pending CN108564132A (en) | 2018-04-25 | 2018-04-25 | A method of classified to depth characteristic based on integrated supporting vector machine |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108564132A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220076003A1 (en) * | 2020-09-04 | 2022-03-10 | Hitachi, Ltd. | Action recognition apparatus, learning apparatus, and action recognition method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855488A (en) * | 2011-06-30 | 2013-01-02 | 北京三星通信技术研究有限公司 | Three-dimensional gesture recognition method and system |
| CN103345645A (en) * | 2013-06-27 | 2013-10-09 | 复旦大学 | Commodity image category forecasting method based on online shopping platform |
| CN103544963A (en) * | 2013-11-07 | 2014-01-29 | 东南大学 | Voice emotion recognition method based on core semi-supervised discrimination and analysis |
| CN106503746A (en) * | 2016-11-03 | 2017-03-15 | 哈尔滨工业大学 | A kind of Fault Diagnosis of Aeroengines method based on offset of performance amount |
| CN107451614A (en) * | 2017-08-01 | 2017-12-08 | 西安电子科技大学 | The hyperspectral classification method merged based on space coordinates with empty spectrum signature |
| CN107463965A (en) * | 2017-08-16 | 2017-12-12 | 湖州易有科技有限公司 | Fabric attribute picture collection and recognition methods and identifying system based on deep learning |
-
2018
- 2018-04-25 CN CN201810378739.4A patent/CN108564132A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102855488A (en) * | 2011-06-30 | 2013-01-02 | 北京三星通信技术研究有限公司 | Three-dimensional gesture recognition method and system |
| CN103345645A (en) * | 2013-06-27 | 2013-10-09 | 复旦大学 | Commodity image category forecasting method based on online shopping platform |
| CN103544963A (en) * | 2013-11-07 | 2014-01-29 | 东南大学 | Voice emotion recognition method based on core semi-supervised discrimination and analysis |
| CN106503746A (en) * | 2016-11-03 | 2017-03-15 | 哈尔滨工业大学 | A kind of Fault Diagnosis of Aeroengines method based on offset of performance amount |
| CN107451614A (en) * | 2017-08-01 | 2017-12-08 | 西安电子科技大学 | The hyperspectral classification method merged based on space coordinates with empty spectrum signature |
| CN107463965A (en) * | 2017-08-16 | 2017-12-12 | 湖州易有科技有限公司 | Fabric attribute picture collection and recognition methods and identifying system based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 罗凯旋 等: "评估几种降维分类器应用于生物质谱数据的性能", 《中国科学:生命科学》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20220076003A1 (en) * | 2020-09-04 | 2022-03-10 | Hitachi, Ltd. | Action recognition apparatus, learning apparatus, and action recognition method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN102831447B (en) | Method for identifying multi-class facial expressions at high precision | |
| CN109165538B (en) | Bar code detection method and device based on deep neural network | |
| CN101930549B (en) | Second generation curvelet transform-based static human detection method | |
| CN1333366C (en) | On-line hand-written Chinese characters recognition method based on statistic structural features | |
| Larios et al. | Haar random forest features and SVM spatial matching kernel for stonefly species identification | |
| CN102722713B (en) | Handwritten numeral recognition method based on lie group structure data and system thereof | |
| CN106056082B (en) | A kind of video actions recognition methods based on sparse low-rank coding | |
| CN110503054B (en) | Text image processing method and device | |
| CN103618918A (en) | Method and device for controlling display of smart television | |
| CN104680144A (en) | Lip language recognition method and device based on projection extreme learning machine | |
| CN101383008A (en) | Image classification method based on visual attention model | |
| CN103336942A (en) | Traditional Chinese painting identification method based on Radon BEMD (bidimensional empirical mode decomposition) transformation | |
| CN102208020A (en) | Human face recognition method based on optimal dimension scale cutting criterion | |
| CN110245697B (en) | Surface contamination detection method, terminal device and storage medium | |
| CN110717426A (en) | Garbage classification method based on domain adaptive learning, electronic equipment and storage medium | |
| CN103839078A (en) | Hyperspectral image classifying method based on active learning | |
| CN108898269A (en) | Electric power image-context impact evaluation method based on measurement | |
| Deore et al. | A survey on offline signature recognition and verification schemes | |
| Kanda et al. | A deep learning-based recognition technique for plant leaf classification | |
| CN103020614A (en) | Human movement identification method based on spatio-temporal interest point detection | |
| Rigaud et al. | What do we expect from comic panel extraction? | |
| CN112036520A (en) | Panda age identification method and device based on deep learning and storage medium | |
| CN111680577A (en) | Face detection method and device | |
| CN108564132A (en) | A method of classified to depth characteristic based on integrated supporting vector machine | |
| CN106295478A (en) | A kind of image characteristic extracting method and device |
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 | ||
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: Room 612, Building 5, No. 998 Wenyi West Road, Yuhang District, Hangzhou City, Zhejiang Province, 311100 Applicant after: Hangzhou flash Mdt InfoTech Ltd Address before: Room 612, Building 5, No. 998 Wenyi West Road, Yuhang District, Hangzhou City, Zhejiang Province, 311100 Applicant before: Hangzhou flash press information Polytron Technologies Inc |
|
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 310000 Room 608, Building No. 998 Wenyi West Road, Wuchang Street, Yuhang District, Hangzhou City, Zhejiang Province Applicant after: Hangzhou flash Mdt InfoTech Ltd Address before: Room 612, Building 5, No. 998 Wenyi West Road, Yuhang District, Hangzhou City, Zhejiang Province, 311100 Applicant before: Hangzhou flash Mdt InfoTech Ltd |
|
| CB02 | Change of applicant information | ||
| CB02 | Change of applicant information |
Address after: 310000 Room 608, building 5, No. 998, Wenyi West Road, Wuchang Street, Yuhang District, Hangzhou City, Zhejiang Province Applicant after: Flash it Co., Ltd Address before: 310000 Room 608, building 5, No. 998, Wenyi West Road, Wuchang Street, Yuhang District, Hangzhou City, Zhejiang Province Applicant before: HANGZHOU SECSMART INFORMATION TECHNOLOGY Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180921 |