CN107273908B - A kind of compressed sensing based target identification method - Google Patents
A kind of compressed sensing based target identification method Download PDFInfo
- Publication number
- CN107273908B CN107273908B CN201710548175.XA CN201710548175A CN107273908B CN 107273908 B CN107273908 B CN 107273908B CN 201710548175 A CN201710548175 A CN 201710548175A CN 107273908 B CN107273908 B CN 107273908B
- Authority
- CN
- China
- Prior art keywords
- target
- matrix
- dictionary
- kinds
- coefficient
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/457—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by analysing connectivity, e.g. edge linking, connected component analysis or slices
-
- 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
- G06V30/24—Character recognition characterised by the processing or recognition method
- G06V30/242—Division of the character sequences into groups prior to recognition; Selection of dictionaries
Abstract
The invention discloses a kind of compressed sensing based target identification methods, comprising steps of obtaining at least two classification target master sample figures；All kinds of clarification of objective atoms are obtained using characteristic atomic extracting method；Each characteristic atomic of target described in every class is diagonally rearranged into every classification target dictionary Ψ respectivelyp, and the dictionary Ψ that the dictionary arranged in parallel of all kinds of targets composition is comprehensive；Compression sampling, the sampled signal y compressed are carried out to original image x to be identified using calculation matrix Φ；In conjunction with comprehensive dictionary Ψ, calculation matrix Φ and sampled signal y, the sparse coefficient θ of original image to be identified is obtained by reconstruction calculations；Sparse coefficient θ is handled to obtain coefficient figure, Classification and Identification and counting are carried out according to line number locating for connected domain in coefficient figure and size, to realize the identification to all kinds of targets in original image.Target identification method proposed by the present invention, can be improved the speed of target identification in the picture, and multi-targets recognition may be implemented.
Description
Technical field
The present invention relates to image identification technical field more particularly to a kind of compressed sensing based target identification methods.
Background technique
Carrying out target identification in the picture is the mistake for being distinguished target specific in image using various programmed algorithms
Journey, and basis is provided using the target distinguished as being further processed, it, can be extensive in today of informatization and network
It is applied to many fields.Human eye speed when carrying out identifying some specific objective is often relatively slow, and similar target is carried out
Identification divides for a long time, will also result in aestheticly tired and gradually generates a large amount of wrong identifications, and machine recognition is used to know instead of human eye
Not, speed and reduction energy consumption can be improved with brain volume instead of human eye using computer calculation amount, for field of image recognition
Speech is very favorable, such as: the video frame picture of 1,000 width crossroads is identified, it is desirable that find out by wagon flow
Amount, hence it is evident that eye recognition is much conducive to using machine recognition；Likewise, if adding images steganalysis system to robot,
It is then equivalent to and is added to " eyes " to robot, be also very favorable for developing artificial intelligence technology.This field skill at present
Art personnel are made that many contributions in terms of target identification, and image recognition technology is not only applied to recognition of face, article by people
Identification etc., also applied in terms of, greatly facilitate people's lives.
And currently used images steganalysis technology it is general the disadvantage is that take a long time, speed it is slower, the reason is that tradition figure
As target identification technology needs following below scheme: image preprocessing, image segmentation, feature extraction and feature identification or matching；Need
The amount of image information to obtain to sampling, which carries out repeatedly processing, to come out required target identification, and existing image is known
Image acquisition process and target identification process are not separated, are unfavorable for the raising of speed.
The disclosure of background above technology contents is only used for auxiliary and understands design and technical solution of the invention, not necessarily
The prior art for belonging to present patent application, no tangible proof show above content present patent application the applying date
In disclosed situation, above-mentioned background technique should not be taken to the novelty and creativeness of evaluation the application.
Summary of the invention
In order to solve the above technical problems, the present invention proposes a kind of compressed sensing based target identification method, can be improved
The speed of target identification in the picture, and multi-targets recognition may be implemented.
In order to achieve the above object, the invention adopts the following technical scheme:
The invention discloses a kind of compressed sensing based target identification methods, comprising the following steps:
S1: at least two classification target master sample figures are obtained；
S2: according to the master sample figure of all kinds of targets, all kinds of targets are obtained using characteristic atomic extracting method
Characteristic atomic；
S3: each characteristic atomic of target described in every class is diagonally rearranged to the dictionary of target described in every class respectively
Ψp, and the dictionary Ψ that the dictionary arranged in parallel of all kinds of targets composition is comprehensive；
S4: compression sampling, the sampled signal y compressed are carried out to original image x to be identified using calculation matrix Φ；
S5: in conjunction with comprehensive dictionary Ψ, calculation matrix Φ and sampled signal y, institute to be identified is obtained by reconstruction calculations
State the sparse coefficient θ of original image；
S6: sparse coefficient θ being handled to obtain coefficient figure, according to line number locating for connected domain in the coefficient figure and
Size carries out Classification and Identification and counting, to realize the identification to all kinds of targets in the original image.
Preferably, step S1 is specifically included: target described at least two classes according to existing for known in image or video flowing mentions
Multiple target images of target described in every class are weighted the target image for taking out all kinds of targets respectively
To the master sample figure of target described in every class.
Preferably, wherein extracting the target image of all kinds of targets specifically: use morphological image method or people
Work identifies dividing method to extract the target image of all kinds of targets.
Preferably, step S4 further includes that sampled signal y is stored as data in memory.
Preferably, step S5 is specifically included: obtaining perception matrix A according to the product of comprehensive dictionary Ψ and calculation matrix Φ
The dilute of the original image is calculated by the reconstruction calculations formula y=A* θ of OMP algorithm in conjunction with sampled signal y in=Φ * Ψ
Sparse coefficient θ.
Preferably, the target identification method further includes step S7: sparse coefficient θ being multiplied with comprehensive dictionary Ψ, is obtained
To the reconstructed image x of acquisition1=Ψ * θ.
Preferably, it is described to extract specifically to use MOD algorithm or K-SVD algorithm for the characteristic atomic extracting method in step S2
Clarification of objective atom.
Preferably, the characteristic atomic extracting method in step S2 specifically includes:
S21: the objective matrix m of the master sample figure of all kinds of targets of input, and initialized: matrix n is initial
Turn to n=m, circulation mark i=0, the dictionary Ψ of the initial targetpFor empty set；
S22: assignment circulation mark i=i+1；
S23: the maximum kth column of two norms in all column of matrix n, the column element λ as extraction are foundi, whereinS is the columns of matrix n；
S24: the column element λ of all column and kth column in calculating matrix niOptimal ratio t, whereinJ refers to that the jth in matrix n arranges；
S25: the residual error r of matrix n, the residual error r respectively arranged in residual error r are updatedjCalculation formula be rj=nj-λi*tj；
S26: by the residual error r of column all in matrix njWith first threshold ξ1It is compared, if it exists rj< ξ1, execute step
S27, otherwise return step S23；
S27: matrix n is updated to n=r, and deletes nj, dictionary ΨpIt is updated to Ψp=[Ψp, λi]；
S28: by two norms of updated matrix n and second threshold ξ2It is compared, if | | n | | < ξ2, execute step
S29, otherwise return step S22；
S29: by dictionary ΨpIt is orthogonalized processing, exports characteristic atomic.
Preferably, further include S210 in step S2: the characteristic atomic combining target matrix m of output is subjected to OMP reconstruct meter
Calculation obtains corresponding coefficient, and the every row of corresponding coefficient is calculated two norms and is sized, according to coefficient magnitude and error
N row is effective before sets requirement judges, then the corresponding preceding N column of the characteristic atomic that corresponding judgement is extracted are defeated as final characteristic atomic
Out.
Preferably, orthogonalization process specifically uses Smith's orthogonalization process in step S29；Preferably, ξ1≤ξ2。
Compared with prior art, the beneficial effects of the present invention are: in target identification method of the invention, pass through spy first
Sign atom extracting method obtains all kinds of clarification of objective atoms, and every each characteristic atomic of classification target is diagonally rearranged respectively
Every classification target dictionary, then the dictionary that the dictionary arranged in parallel composition of all kinds of targets is comprehensive, by the way that characteristic atomic is carried out this
The particular arrangement of sample obtains coefficient figure in conjunction with sparse coefficient, identifies and counts directly in coefficient data, and door can be divided to sort out area
How many each target of every class target is separated, so as to carry out the identifying processing of multiple targets simultaneously；In addition compared to traditional
Target identification method carries out Classification and Identification without extracting area-of-interest progress matching primitives, directly specific by setting
Dictionary can obtain inhomogeneity target and be respectively at different coefficient regions, i.e., identification counting is carried out in coefficient data, process
Journey is simple, and calculation amount is small, improves the speed of target identification in the picture.
In further embodiment, reconstructed image can be also obtained parallel while carrying out target identification, parallel computation is mutual
It does not influence, Image Acquisition and target identification process is combined into one, target identification is carried out in image pick-up signal, further
The speed of target identification in the picture is improved, and improves the storage efficiency of data and reduces the consumption of hardware.More into
In the scheme of one step, the principles such as the limited isometry in compressive sensing theory are combined, for required identification target and background ring
The characteristics of border ga s safety degree, devises specific characteristic atomic extracting method, and by characteristic atomic combining shape by reasonable
At dictionary, the coefficient matrix being calculated in conjunction with calculation matrix with this dictionary can obtain better target identification effect.
Detailed description of the invention
Fig. 1 is the flow diagram of the compressed sensing based target identification method of the preferred embodiment of the present invention；
Fig. 2 is the flow diagram of the characteristic atomic extracting method of the preferred embodiment of the present invention.
Specific embodiment
Below against attached drawing and in conjunction with preferred embodiment, the invention will be further described.
As shown in Figure 1, the preferred embodiment of the present invention discloses a kind of compressed sensing based target identification method, including
Following steps:
S1: at least two classification target master sample figures are obtained；
According to existing at least two class targets known in some images or video flowing, extracted by morphological image method
Target image or manual identified are partitioned into the target image in image；Then the multiple target images of every classification target are distinguished
It is weighted to obtain every classification target master sample figure.
S2: according to the master sample figure of all kinds of targets, it is former that all kinds of clarifications of objective are obtained using characteristic atomic extracting method
Son；
Wherein it is former can to extract all kinds of clarifications of objective using MOD algorithm or K-SVD algorithm for characteristic atomic extracting method
The method told about hereinafter can also be used in son, further, characteristic atomic extracting method.
S3: every each characteristic atomic of classification target is diagonally rearranged into every classification target dictionary Ψ respectivelyp, and will be all kinds of
The comprehensive dictionary Ψ of the dictionary arranged in parallel composition of target；
S4: using calculation matrix Φ to original image x to be identified progress compression sampling, the sampled signal y compressed,
And sampled signal y is stored as data in memory；
S5: in conjunction with comprehensive dictionary Ψ, calculation matrix Φ and sampled signal y, original to be identified is obtained by reconstruction calculations
The sparse coefficient θ of beginning image；
Specifically, perception matrix A=Φ * Ψ is obtained according to the product of comprehensive dictionary Ψ and calculation matrix Φ, in conjunction with adopting
The sparse coefficient θ of original image is calculated by the reconstruction calculations formula y=A* θ of OMP algorithm by sample signal y.
S6: sparse coefficient θ being handled to obtain coefficient figure, is carried out according to line number locating for connected domain in coefficient figure and size
Classification and Identification and counting, to realize the identification to all kinds of targets in original image；
Specifically, sparse coefficient θ is filtered the processing such as binaryzation and obtains coefficient figure.
It in a further embodiment, further include that the sparse coefficient theta replication of original image is a in step S5；Step S6
In to be handled the sparse coefficient θ of a copy of it to obtain coefficient figure；The target identification method further includes step S7: will in addition
A sparse coefficient θ is multiplied with comprehensive dictionary Ψ, the reconstructed image x acquired1=Ψ * θ, wherein step S7 and step S6
It can carry out simultaneously.The reconstructed image obtained by step S7 is not lost compared with the image that traditional Nyquist method collects
Very, original image can be substituted and carry out relevant treatment.
In order to improve target identification speed and reduce the data volume of Image Acquisition, the present invention uses random Gaussian or Bernoulli Jacob
Calculation matrix carries out data compression acquisition, then collects compared to the storage of original image less data into memory, in this way
Mostly several times of image information can be acquired under same amount of storage；Then, by collected signal when needed from memory
It is middle to be reconstructed with OMP algorithm come since selected dictionary is the algorithm atom generated of designed, designed by particular arrangement
Composition (in every classification target dictionary each characteristic atomic be in diagonal line arrangement mode, then every classification target dictionary is arranged side by side
The comprehensive dictionary of column composition), therefore the sparse coefficient being calculated slightly is handled that just to divide door to distinguish every class target with sorting out each
How many target, while the sparse coefficient being calculated directly is multiplied with selected dictionary and can obtain and original graph image quality
Close reconstructed image is measured, and obtains reconstruct image and identifies that can carry out parallel processing is independent of each other with classification.
In a still further embodiment, the characteristic atomic extracting method in step S2 specifically includes:
S21: the objective matrix m of the master sample figure of all kinds of targets is inputted, and is initialized: matrix n is initialized as n
=m, circulation mark i=0, the dictionary Ψ of initial targetpFor empty set；
S22: assignment circulation mark i=i+1, which is the dictionary Ψ of carrying targetpHow many column contained；
S23: the maximum kth column of two norms in all column of matrix n, the column element λ as extraction are foundi, whereinS is the columns of matrix n, which is to obtain the maximum column of influence factor in matrix n；
S24: the column element λ of all column and kth column in calculating matrix niOptimal ratio tj, whereinJ refers to that the jth in matrix n arranges, which is every in the maximum column of calculating influence and matrix n
The related coefficient of column；
S25: the residual error r, the residual error r wherein respectively arranged in residual error r of matrix n are updatedjCalculation formula be rj=nj-λi*tj, should
Step is to leave out column all in matrix n to influence maximum column, so that the result that each column obtains is minimum (two norms)；
S26: by the residual error r of column all in matrix njWith first threshold ξ1It is compared, if it exists rj< ξ1, execute step
S27, otherwise return step S23；
S27: matrix n is updated to n=r, and deletes nj, dictionary ΨpIt is updated to Ψp=[Ψp, λi]；
Be in step S26 and step S27 by calculate each column leave out after residual error make comparisons with first threshold, judge whether
It is the column being affected, if it is can be used as a characteristic atomic, there are dictionary ΨpIn, otherwise select again.
S28: by two norms of updated matrix n and second threshold ξ2It is compared, if | | n | | < ξ2, execute step
S29, otherwise return step S22, wherein ξ1≤ξ2；
S29: by dictionary ΨpIt is orthogonalized processing (Smith's orthogonalization process can be used), exports characteristic atomic；
Step S28 and step S29 is to judge whether to have selected all characteristic atomics, if selected complete, by what is obtained
Dictionary ΨpIt is orthogonal to be exported as characteristic atomic, otherwise continue to select.
S210: the characteristic atomic combining target matrix m of output is subjected to OMP reconstruction calculations and obtains corresponding coefficient, by phase
The every row of the coefficient answered calculates two norms and is sized, and N row is effective before judging according to coefficient magnitude and error sets requirement,
The corresponding preceding N column of the characteristic atomic that then corresponding judgement is extracted are exported as final characteristic atomic.
In target identification method of the invention, it is former that all kinds of clarifications of objective are obtained by characteristic atomic extracting method first
Every each characteristic atomic of classification target is diagonally rearranged every classification target dictionary by son respectively, then by the dictionary of all kinds of targets
The comprehensive dictionary of arranged in parallel composition obtains coefficient in conjunction with sparse coefficient by the way that characteristic atomic is carried out such particular arrangement
Figure is identified directly in coefficient data and is counted, and can divide door distinguishes how many each target of every class target with sorting out, so as to
To carry out the identifying processing of multiple targets simultaneously；In addition compare traditional target identification method, without extract area-of-interest into
Row matching primitives carry out Classification and Identification, and directly can obtain inhomogeneity target by setting specific dictionary is respectively at difference
Coefficient region, treatment process is simple, and calculation amount is small, improves the speed of target identification in the picture.Wherein the present invention herein in connection with
The principles such as the limited isometry in compressive sensing theory, set for required identification target and the characteristics of background environment ga s safety degree
Specific characteristic atomic extracting method is counted, and by the way that characteristic atomic is formed dictionary by reasonable combination, with this dictionary
The coefficient matrix being calculated in conjunction with calculation matrix can obtain better target identification effect, and can obtain weight simultaneously
The preferable reconstruct image of structure quality, identification count and obtain reconstructed image and with parallel computation and can be independent of each other, the saving time.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
Specific implementation of the invention is only limited to these instructions.For those skilled in the art to which the present invention belongs, it is not taking off
Under the premise of from present inventive concept, several equivalent substitute or obvious modifications can also be made, and performance or use is identical, all answered
When being considered as belonging to protection scope of the present invention.
Claims (11)
1. a kind of compressed sensing based target identification method, which comprises the following steps:
S1: at least two classification target master sample figures are obtained；
S2: according to the master sample figure of all kinds of targets, the spy of all kinds of targets is obtained using characteristic atomic extracting method
Levy atom；
S3: each characteristic atomic of target described in every class is diagonally rearranged to the dictionary Ψ of target described in every class respectivelyp,
And the dictionary Ψ that the dictionary arranged in parallel of all kinds of targets composition is comprehensive；
S4: compression sampling, the sampled signal y compressed are carried out to original image x to be identified using calculation matrix Φ；
S5: in conjunction with comprehensive dictionary Ψ, calculation matrix Φ and sampled signal y, the original to be identified is obtained by reconstruction calculations
The sparse coefficient θ of beginning image；
S6: sparse coefficient θ being handled to obtain coefficient figure, according to line number and size locating for connected domain in the coefficient figure
Classification and Identification and counting are carried out, to realize the identification to all kinds of targets in the original image.
2. target identification method according to claim 1, which is characterized in that step S1 is specifically included: according to image or view
Target described in known existing at least two classes, extracts the target image of all kinds of targets, by target described in every class in frequency stream
Multiple target images be weighted to obtain the master sample figure of target described in every class respectively.
3. target identification method according to claim 2, which is characterized in that wherein extract the target of all kinds of targets
Image specifically: the target figure of all kinds of targets is extracted using morphological image method or manual identified dividing method
Picture.
4. target identification method according to claim 1, which is characterized in that step S4 further include using sampled signal y as
Data store in memory.
5. target identification method according to claim 1, which is characterized in that step S5 is specifically included: according to comprehensive word
The product of allusion quotation Ψ and calculation matrix Φ obtain perception matrix A=Φ * Ψ and pass through the reconstruct meter of OMP algorithm in conjunction with sampled signal y
Calculate the sparse coefficient θ that the original image is calculated in formula y=A* θ.
6. target identification method according to claim 1, which is characterized in that further include step S7: by sparse coefficient θ with it is comprehensive
The dictionary Ψ of conjunction is multiplied, the reconstructed image x acquired1=Ψ * θ.
7. target identification method according to any one of claims 1 to 6, which is characterized in that the characteristic atomic in step S2
Extracting method uses MOD algorithm or K-SVD algorithm specifically to extract the clarification of objective atom.
8. target identification method according to any one of claims 1 to 6, which is characterized in that the characteristic atomic in step S2
Extracting method specifically includes:
S21: the objective matrix m of the master sample figure of all kinds of targets of input, and initialized: matrix n is initialized as n
=m, circulation mark i=0, the dictionary Ψ of the initial targetpFor empty set；
S22: assignment circulation mark i=i+1；
S23: the maximum kth column of two norms in all column of matrix n, the column element λ as extraction are foundi, whereinS is the columns of matrix n；
S24: the column element λ of all column and kth column in calculating matrix niOptimal ratio t, whereinJ refers to that the jth in matrix n arranges；
S25: the residual error r of matrix n, the residual error r respectively arranged in residual error r are updatedjCalculation formula be rj=nj-λi*tj；
S26: by the residual error r of column all in matrix njWith first threshold ξ1It is compared, if it exists rj< ξ1, step S27 is executed, it is no
Then return step S23；
S27: matrix n is updated to n=r, and deletes nj, dictionary ΨpIt is updated to Ψp=[Ψp, λi]；
S28: by two norms of updated matrix n and second threshold ξ2It is compared, if | | n | | < ξ2, step S29 is executed, it is no
Then return step S22；
S29: by dictionary ΨpIt is orthogonalized processing, exports characteristic atomic.
9. target identification method according to claim 8, which is characterized in that further include S210 in step S2: by output
Characteristic atomic combining target matrix m carries out OMP reconstruction calculations and obtains corresponding coefficient, and the every row of corresponding coefficient is calculated two norms
It is sized, N row is effective before judging according to coefficient magnitude and error sets requirement, then the feature that corresponding judgement is extracted is former
The corresponding preceding N column of son are exported as final characteristic atomic.
10. target identification method according to claim 8, which is characterized in that orthogonalization process specifically uses in step S29
Smith's orthogonalization process.
11. target identification method according to claim 8, which is characterized in that ξ1≤ξ2。
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710548175.XA CN107273908B (en) | 2017-07-06 | 2017-07-06 | A kind of compressed sensing based target identification method |
| PCT/CN2017/098652 WO2019006835A1 (en) | 2017-07-06 | 2017-08-23 | Target recognition method based on compressed sensing |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710548175.XA CN107273908B (en) | 2017-07-06 | 2017-07-06 | A kind of compressed sensing based target identification method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107273908A CN107273908A (en) | 2017-10-20 |
| CN107273908B true CN107273908B (en) | 2019-09-06 |
Family
ID=60073082
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710548175.XA Active CN107273908B (en) | 2017-07-06 | 2017-07-06 | A kind of compressed sensing based target identification method |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN107273908B (en) |
| WO (1) | WO2019006835A1 (en) |
Families Citing this family (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108280818B (en) * | 2018-01-19 | 2020-04-03 | 清华大学深圳研究生院 | Rapid target imaging method and system based on compressed sensing |
| CN109033963B (en) * | 2018-06-22 | 2021-07-06 | 王连圭 | Multi-camera video cross-region human motion posture target recognition method |
| CN110472576A (en) * | 2019-08-15 | 2019-11-19 | 西安邮电大学 | A kind of method and device for realizing mobile human body Activity recognition |
| CN111796253B (en) * | 2020-09-01 | 2022-12-02 | 西安电子科技大学 | Radar target constant false alarm detection method based on sparse signal processing |
| CN112508089B (en) * | 2020-12-03 | 2023-10-31 | 国网山西省电力公司晋城供电公司 | Self-adaptive compressed sensing method for partial discharge signal compressed transmission |
| CN113093164B (en) * | 2021-03-31 | 2023-02-10 | 西安电子科技大学 | Translation-invariant and noise-robust radar image target identification method |
| CN113670435B (en) * | 2021-08-20 | 2023-06-23 | 西安石油大学 | Underground vibration measuring device and method based on compressed sensing technology |
| CN113922823B (en) * | 2021-10-29 | 2023-04-21 | 电子科技大学 | Social media information propagation graph data compression method based on constraint sparse representation |
| CN116541470B (en) * | 2023-07-07 | 2024-02-13 | 深圳创维智慧科技有限公司 | Synchronization method, device, equipment and medium of database read-only standby library |
| CN117041359B (en) * | 2023-10-10 | 2023-12-22 | 北京安视华业科技有限责任公司 | Efficient compression transmission method for information data |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104574450A (en) * | 2014-12-31 | 2015-04-29 | 南京邮电大学 | Image reconstruction method based on compressed sensing |
| CN105631478A (en) * | 2015-12-25 | 2016-06-01 | 天津科技大学 | Plant classification method based on sparse expression dictionary learning |
| CN106203374A (en) * | 2016-07-18 | 2016-12-07 | 清华大学深圳研究生院 | A kind of characteristic recognition method based on compressed sensing and system thereof |
| CN106203453A (en) * | 2016-07-18 | 2016-12-07 | 清华大学深圳研究生院 | A kind of based on compressed sensing biological with abiotic target identification method and system thereof |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10524686B2 (en) * | 2014-12-01 | 2020-01-07 | The Regents Of The University Of California | Diffusion reproducibility evaluation and measurement (DREAM)-MRI imaging methods |
| CN105844635B (en) * | 2016-03-21 | 2018-10-12 | 北京工业大学 | A kind of rarefaction representation depth image method for reconstructing based on structure dictionary |
| CN106157232B (en) * | 2016-06-30 | 2019-04-26 | 广东技术师范学院 | A kind of general steganalysis method of digital picture characteristic perception |
| CN106557784B (en) * | 2016-11-23 | 2020-05-01 | 上海航天控制技术研究所 | Rapid target identification method and system based on compressed sensing |
| CN106815806B (en) * | 2016-12-20 | 2020-01-10 | 浙江工业大学 | Single image SR reconstruction method based on compressed sensing and SVR |
-
2017
- 2017-07-06 CN CN201710548175.XA patent/CN107273908B/en active Active
- 2017-08-23 WO PCT/CN2017/098652 patent/WO2019006835A1/en active Application Filing
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104574450A (en) * | 2014-12-31 | 2015-04-29 | 南京邮电大学 | Image reconstruction method based on compressed sensing |
| CN105631478A (en) * | 2015-12-25 | 2016-06-01 | 天津科技大学 | Plant classification method based on sparse expression dictionary learning |
| CN106203374A (en) * | 2016-07-18 | 2016-12-07 | 清华大学深圳研究生院 | A kind of characteristic recognition method based on compressed sensing and system thereof |
| CN106203453A (en) * | 2016-07-18 | 2016-12-07 | 清华大学深圳研究生院 | A kind of based on compressed sensing biological with abiotic target identification method and system thereof |
Non-Patent Citations (1)
| Title |
|---|
| 基于稀疏性估计的图像压缩感知研究;杨森林;《西安文理学院学报( 自然科学版)》;20170331;第20卷(第2期);第25-28、40页 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107273908A (en) | 2017-10-20 |
| WO2019006835A1 (en) | 2019-01-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107273908B (en) | A kind of compressed sensing based target identification method | |
| CN108717568B (en) | A kind of image characteristics extraction and training method based on Three dimensional convolution neural network | |
| CN110119686B (en) | Safety helmet real-time detection method based on convolutional neural network | |
| CN105373777B (en) | A kind of method and device for recognition of face | |
| CN108491849A (en) | Hyperspectral image classification method based on three-dimensional dense connection convolutional neural networks | |
| Wang et al. | A deep symmetry convnet for stroke lesion segmentation | |
| CN110084318A (en) | A kind of image-recognizing method of combination convolutional neural networks and gradient boosted tree | |
| CN109389585A (en) | A kind of brain tissue extraction method based on full convolutional neural networks | |
| CN107424145A (en) | The dividing method of nuclear magnetic resonance image based on three-dimensional full convolutional neural networks | |
| CN106203374B (en) | A kind of characteristic recognition method and its system based on compressed sensing | |
| CN107622225B (en) | Face recognition method based on independent component analysis network | |
| CN107967674B (en) | Nuclear magnetic resonance image denoising method based on image block self-similarity prior | |
| CN110991554B (en) | Improved PCA (principal component analysis) -based deep network image classification method | |
| Song et al. | Feature extraction and target recognition of moving image sequences | |
| Asyhar et al. | Implementation LSTM Algorithm for Cervical Cancer using Colposcopy Data | |
| CN114842507A (en) | Reinforced pedestrian attribute identification method based on group optimization reward | |
| CN108090460A (en) | Expression recognition feature extraction algorithm based on multi-direction description of weber | |
| CN109165551B (en) | Expression recognition method for adaptively weighting and fusing significance structure tensor and LBP characteristics | |
| CN110135362A (en) | A kind of fast face recognition method based under infrared camera | |
| Jayalakshmi et al. | A study of Iris segmentation methods using fuzzy C-means and K-means clustering algorithm | |
| CN109766471A (en) | Skin disease image search method and system, storage medium, electronic equipment | |
| Shah et al. | Performance analysis of canny edge detection for illumination invariant facial expression recognition | |
| Hussain et al. | Face recognition using multiscale and spatially enhanced weber law descriptor | |
| Raouf et al. | Use of Discrete Cosine-based Stockwell Transform in the Binary Classification of Magnetic Resonance Images of Brain Tumor | |
| Abas | Diagnosing the leukemia using faster region based convolutional neural network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |