CN109784399A - Based on the multi-source image target association method for improving dictionary learning - Google Patents
Based on the multi-source image target association method for improving dictionary learning Download PDFInfo
- Publication number
- CN109784399A CN109784399A CN201910028307.5A CN201910028307A CN109784399A CN 109784399 A CN109784399 A CN 109784399A CN 201910028307 A CN201910028307 A CN 201910028307A CN 109784399 A CN109784399 A CN 109784399A
- Authority
- CN
- China
- Prior art keywords
- source image
- data set
- target association
- dictionary learning
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 26
- 230000006870 function Effects 0.000 claims abstract description 13
- 239000010410 layer Substances 0.000 claims description 14
- 230000006872 improvement Effects 0.000 claims description 8
- 239000011159 matrix material Substances 0.000 claims description 5
- 241001269238 Data Species 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 4
- 238000003062 neural network model Methods 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 3
- 239000011229 interlayer Substances 0.000 claims description 2
- 230000004927 fusion Effects 0.000 abstract description 4
- 238000012545 processing Methods 0.000 abstract description 4
- 238000013528 artificial neural network Methods 0.000 abstract description 3
- 230000000694 effects Effects 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 abstract description 2
- 238000005259 measurement Methods 0.000 abstract description 2
- 238000000691 measurement method Methods 0.000 abstract 1
- 238000005457 optimization Methods 0.000 description 7
- 230000004069 differentiation Effects 0.000 description 3
- 238000005352 clarification Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000012549 training Methods 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 230000015572 biosynthetic process Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000010191 image analysis Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
Landscapes
- Image Analysis (AREA)
Abstract
本发明公开了基于改进字典学习的多源图像目标关联方法，属于情报数据处理领域，主要解决现有多源情报数据融合中，图像目标关联先识别再关联存在的信息损失、步骤冗余的问题。首先收集多种信源对同一场景或同一目标的图像数据集，形成原始数据集；利用改进字典学习方法对多源图像进行统一稀疏表示，通过在目标函数中引入标签信息，增加字典集特征的表征判别能力；构建神经网络对各图像的稀疏表示和标签信息进行学习，得到关联和非关联目标间距离度量标准，代替传统距离度量方式，完成关联判别模型的建立。该方法充分利用图像的特征信息，避免了现有方法的步骤冗余，具有模型生成速度快、信息损失少、实用效果好等优点。
The invention discloses a multi-source image target association method based on improved dictionary learning, belongs to the field of intelligence data processing, and mainly solves the problems of information loss and redundant steps existing in the existing multi-source intelligence data fusion in which image target association is first identified and then associated . First, collect image data sets of the same scene or the same target from multiple sources to form the original data set; use the improved dictionary learning method to perform a unified sparse representation of multi-source images, and introduce label information into the objective function to increase the characteristics of the dictionary set. Characterize the discriminative ability; build a neural network to learn the sparse representation and label information of each image, and obtain the distance measurement standard between associated and non-associated targets, instead of the traditional distance measurement method, to complete the establishment of the association discrimination model. The method makes full use of the feature information of the image, avoids the redundant steps of the existing methods, and has the advantages of fast model generation, less information loss, and good practical effect.
Description
Technical field
The present invention is under the jurisdiction of information data process field, is related to the life of the target association discrimination model between multi-source heterogeneous image
At suitable for multi-source information fusion treatment link.
Background technique
The acquisition of image category information generally has the characteristics that investigative range is wide, revisiting period is long, positioning accuracy is poor, can be used for
A wide range of early warning early period during early warning detection, can increase the phase of target by the target association to a variety of image category informations
It closes and measures dot density, improve measurement information renewal time, improve the positioning accuracy of target.Target association master between image category information
If the correlation according to clarification of objective information is associated judgement under certain space-time restriction.It is main in practical applications
It is to carry out indirect association, first with target signature information, identifies target category model, it is whether consistent further according to target type model
Be associated differentiation, but this has the following disadvantages: interrelating effect is accurate too dependent on the completeness and algorithm for identifying library
Property；Target signature information is identified as classification information and arrives association results again, increases information process loss.Currently, multi-source image
The research of fusion is that image co-registration is realized from Pixel-level mostly to obtain more detailed information, convenient for further to image
Analysis, processing and understand, but it has higher requirements to the time of input picture and spatial registration, and computationally intensive, and when processing disappears
Time-consuming is long, it is difficult to reach requirement of real-time；And due to characteristic information dimension height, the feature of foreign peoples's information source between multi-source image information
Information difference is big, rarely has at present from the angle of feature-based fusion and carries out correlative study.
Summary of the invention
The purpose of the present invention is to propose to the multi-source image target association methods based on improvement dictionary learning, from image feature level
It sets out, compares multi-source image clarification of objective information by extracting, to judge whether image is derived from same target, construct multi-source figure
As the mapping relations between target；It aims to solve the problem that and first identifies again associated information damage in current existing image object correlating method
The problem of mistake, step redundancy.
Multi-source image target association method of the present invention based on improvement dictionary learning, can be mainly divided into four ranks
Section, data set is collected and pretreatment stage, unified wordbook generation phase, representative learning stage and classification learning stage, specifically
Including following technical measures, it is data set collection and pretreatment stage first, collects plurality of information resources to Same Scene or same mesh
Target image data set, and handmarking's analysis is carried out to multi-source image data set, store image data set and handmarking's knot
Fruit is formed with label multi-source image target association raw data set.Unified wordbook generation phase, on the basis of dictionary learning
On, it is further proposed that the method for improving dictionary learning, to increase the characterization discriminating power of the generated feature of dictionary learning, and utilizes
It improves dictionary learning method and concentrates all image datas to carry out dictionary learning initial data, generate the common dictionary of multi-source image
Collection.Representative learning stage, each multi-source that multi-source image target association initial data is concentrated using multi-source image common wordbook
Image data carries out unified rarefaction representation, generates the sparse coefficient collection of each multi-source image data, the label concentrated with initial data
Information is collectively formed multi-source image target association and differentiates data set.The classification learning stage establishes the differentiation of multi-source image target association
Model, learnt using data and label of the multilayer neural network model to multi-source image target association differentiation data concentration,
Training, finally obtains the distance metric between associated objects and dereferenced target, thus complete instead of traditional distance metric mode
At the foundation of association discrimination model.
Multi-source image target association method proposed by the present invention based on improvement dictionary learning can be based on measured data, directly
The association method of discrimination between training generation multi-source image target is connect, avoids and first identifies again in associated conventional method to identification library
Information requirements are high, the serious equal traditional problems of information loss, have that model formation speed is fast, information loss is few, good practical effect etc.
Advantage, multi-source image target association method generated can be directly used for solving more in practical information data processing without debugging
Source information data merges problem.
Detailed description of the invention
Fig. 1 is based on the multi-source image target association method flow diagram for improving dictionary learning.
Specific embodiment
Multi-source image target association method and technology scheme proposed by the present invention based on improvement dictionary learning includes following step
It is rapid:
Step 1: data set is collected and pretreatment stage, collects plurality of information resources to the picture number of Same Scene or same target
Handmarking's analysis is carried out according to collection, and to multi-source image data set, stores image data set and handmarking as a result, being formed with mark
Sign multi-source image target association raw data set S；
Step 1.1: mainly including visible images, multispectral image, infrared image, SAR image to plurality of information resources image
In two kinds of images collected extensively, and guarantee between the multi-source image for target association to be for Same Scene, comprising same
A kind of target or corresponding target；
Step 1.2: having label multi-source image target association raw data set S={ (yi,yj',aij),i∈(0,N),j∈
(0, N') }, wherein yiAnd yj' indicate that the data of any two kinds of information sources indicate, aij=0 or 1, it is handmarking as a result, aij=0 table
Show yiAnd yj' it is dereferenced target, aij=1 indicates yiAnd yj' it is associated objects, N and N' are respectively that two kinds of multi-source image collection include
Number of elements.
Step 2: unified wordbook generation phase, using the method for improvement dictionary learning to all figures in raw data set S
As data carry out dictionary learning, the generation common wordbook D=[d of multi-source image1,d2,…dk]；
Step 2.1: dictionary learning is the sample Y=[y for common dense expression1,…,yn]∈Rn×N, find one it is excessively complete
Standby wordbook D=[d1,…,dK]∈Rn×KFormer dense sample is expressed as linear group of each element in wordbook by (K > n)
It closes, guarantees the coefficient X=[x of linear combination1,…,xN]∈RK×NIt is sparse as far as possible, and the degree of information loss is controlled, to make to learn
Habit task is simplified, and reduces the complexity of model, is seen shown in formula (1)；
Step 2.2: to make the sparse coefficient obtained that there is certain classification capacity, dictionary learning is further improved,
On the basis of its objective function, increase a label to limitation, the objective function for improving dictionary learning is obtained, as shown in formula (2)；
Wherein, α and β is hyper parameter, for controlling the weight of error representated by items；Section 3 is to have sparse coefficient
The limit entry of classification capacity, when target is to (yi,yj') label be associated objects, i.e. same category when, Mij=1；When target pair
(yi,yj') label be dereferenced target, i.e., when different classes of, Mij=-1；Other situations, such as (yi,yj') it is same target,
That is when i=j, Mij=0, see shown in formula (3)；
N=diag { N1,…,NN, it is diagonal matrix, each diagonal element is the sum of each column element in matrix M,L=N-M；；
Step 2.3: objective function (2) has two parameters of D and X, while being optimized for non-convex optimization problem to parameter D and X,
Need to fix parametric solution another parameter, thus the optimization problem of objective function can be converted to following two formula
(4), the iteration optimization of (5):
Wherein, formula (4) is that fixed wordbook D optimizes sparse coefficient X one by one, and formula (5) is fixed sparse system
Several pairs of wordbooks optimize, and the optimization of formula (5) can use K-SVD progress, update the element and right in wordbook one by one
Answer sparse coefficient；To formula (4), due to there is a presence of L1 norm, objective function be not it is continuously differentiable, searched using characteristic symbol
Rope algorithm is corresponded to the sign vector θ of sparse coefficient by grey iterative generation, converted objective function to standard, unconfined
Double optimization problem, therefore, the gradient of calculation formula (4) are shown in that formula (6) is shown,
Formula (6) gradient is enabled to be equal to zero, can obtain sparse coefficient as a result, sees shown in formula (7),
Step 2.3.1；When generating wordbook to formula (5) using K-SVD algorithm, every wheel iteration only updates a dictionary member
Plain dk, keep other elements constant, find a best dkAnd update corresponding sparse coefficient(For the K of coefficient matrix X
Row, the kth element d with wordbookkIt is corresponding), so that formula (5) is reached minimum；
As shown in formula (8), directly to EkCarry out singular value decomposition (SVD) Lai Gengxin dkWithIt is likely to occur sparse coefficientNot sparse phenomenon, i.e., it is updatedBefore updateThe position of nonzero element is inconsistent, therefore, only retains sparse system
NumberIn nonzero value, EkOnly retain dkWithItem after the product of middle non-zero position is formedIt is rightSingular value decomposition is carried out to obtain
It arrives,And it enablesIt is updated to the first row of U, the product conduct of the first row and Δ (1,1) of VUpdated value,
Complete the update of a dictionary element；
Step 2.3.2；For formula (4), characteristic symbol searching algorithm is dilute to generate by iteration one group of active set of update
Sparse coefficient symbol, the element in active set are the number of sparse coefficient, specifically includes the following steps:
Step 2.3.2.1；Initialization, characteristic symbol vectorWherein θiIt is sparse that ∈ { -1,0,1 } respectively indicates correspondence
Coefficient xiSymbol, active set activeset:={ } be empty set；
Step 2.3.2.2；The sign of each non-zero sparse coefficient and the element of active set are set, is selected conducive to target letter
The sparse coefficient number that number (4) reduces is active set element, it may be assumed that
WhenWhen, if θi=-1, activeset:={ i } ∪
activeset；
WhenWhen, if θi=1, activeset:={ i } ∪
activeset；
Step 2.3.2.3；After generating active set, respectively to being selected in wordbook D, sparse coefficient X and characteristic symbol vector θ
Number element in active set corresponds to generating subsetFormula (4) can be converted into be seen without limitation double optimization problem
Shown in formula (9),
It enables the derivative of formula (9) be equal to 0, can obtain optimal
Step 3: the representative learning stage, using the common wordbook D of multi-source image to multi-source image target association initial data
Multi-source image data in collection S carry out unified rarefaction representation, the sparse coefficient collection X of each multi-source image data are generated, with original number
Multi-source image target association is collectively formed according to the label information of concentration and differentiates data set S'；
Step 3.1: using the wordbook D that improvement dictionary learning generates in step 2 to multi-source image target association original number
Rarefaction representation is carried out according to the multi-source image data in collection S, using formula (4), fixed wordbook D carries out one by one sparse coefficient X
Optimization, generates the sparse coefficient of each multi-source image data；
Step 3.2: being collectively formed in conjunction with the sparse coefficient of label information and each multi-source image data in raw data set S
Multi-source image target association differentiates data set S'={ (xi,xj',aij), i ∈ (0, N), j ∈ (0, N') }, wherein xiAnd xj' be
The rarefaction representation of the two kinds of information sources generated in 3.1, aij=0 or 1, it marks for the Central Plains S somebody's work as a result, aij=0 indicates xiWith
xj' it is dereferenced target, aij=1 indicates xiAnd xj' it is associated objects.
Step 4: the classification learning stage establishes multi-source image target association discrimination model, utilizes multilayer neural network model
Data and label in data set S', which are learnt, train, to be differentiated to multi-source image target association, neural network association is obtained and sentences
Other model；
Step 4.1: the network structure of multilayer neural network model uses three layers of full connection structure, and input layer is two any more
The L2 distance of source image data sparse coefficient is shown in that wherein n is characterized dimension shown in formula (10), for 256 dimensions, the second node layer
Number is 512, and third layer node number is 256, and in addition to output layer, each interlayer is all made of dropout strategy to avoid mistake
Fitting；
Step 4.2: defeated since the problem of whether being associated with two targets is converted into binary classification discrimination
Node layer number is set as 1 out, and the activation primitive of the output layer of the multilayer neural network of use uses sigmoid function, sees public affairs
Shown in formula (11), export as the number of 0~1 range, 1 to represent two image datas associated, 0 represent it is uncorrelated between two image datas
Connection is exported closer to 1, and association probability is bigger, conversely, association probability is smaller；The activation primitive of each middle layer of network structure uses
Relu function is shown in shown in formula (12).
sigmoid(wTX+b)=1/1+exp [- (wTx+b)] (11)
Claims (5)
1. based on the multi-source image target association method for improving dictionary learning, which comprises the following steps:
Step 1, data set collection and pretreatment stage
Plurality of information resources is collected to the image data set of Same Scene or same target, and multi-source image data set is manually marked
It scores analysis, stores image data set and handmarking as a result, being formed with label multi-source image target association raw data set S；
Step 2, unified wordbook generation phase
Dictionary learning is carried out to image datas all in raw data set S using the method for improving dictionary learning, generates multi-source figure
As common wordbook D=[d1,d2,…dk]；
Step 3, the representative learning stage
The multi-source image data in multi-source image target association raw data set S are carried out using multi-source image common wordbook D
Unified rarefaction representation, generates the sparse coefficient collection X of each multi-source image data, and the label information concentrated with initial data is collectively formed
Multi-source image target association differentiates data set S'；
Step 4, the classification learning stage
Multi-source image target association discrimination model is established, is differentiated in data set S' using multi-source image target association in step 3
Data and label are learnt, are trained, and trained association discrimination model is obtained.
2. as described in claim 1 based on the multi-source image target association method for improving dictionary learning, which is characterized in that described
Improvement dictionary learning in step 2 is to increase a label on the basis of former dictionary learning objective function to limitation, obtain such as public affairs
Objective function shown in formula (1)；
Wherein, α and β is hyper parameter；N=diag { N1,…,NN, it is diagonal matrix, each diagonal element is each column element in matrix M
The sum of,L=N-M；The definition of M is shown in that formula (2) is shown,
3. as described in claim 1 based on the multi-source image target association method for improving dictionary learning, which is characterized in that described
Step 1 specifically includes following sub-step:
Step 1.1, to two kinds of images in plurality of information resources image visible images, multispectral image, infrared image, SAR image
It is collected extensively, and guarantee is for being for Same Scene between the multi-source image of target association, includes same class target or right
Answer target；
Step 1.2, there is label multi-source image target association raw data set S={ (yi,yj',aij),i∈(0,N),j∈(0,
N') }, wherein yiAnd yj' indicate that the data of any two kinds of information sources indicate, aij=0 or 1, it is handmarking as a result, aij=0 indicates
yiAnd yj' it is dereferenced target, aij=1 indicates yiAnd yj' it is associated objects, N and N' are respectively that two kinds of multi-source image collection include
Number of elements.
4. as described in claim 1 based on the multi-source image target association method for improving dictionary learning, which is characterized in that described
Step 3 specifically includes following sub-step:
Step 3.1, using the wordbook D that improvement dictionary learning generates in step 2 to multi-source image target association raw data set S
In multi-source image data carry out rarefaction representation, generate the sparse coefficient of each multi-source image data；
Step 3.2, multi-source is collectively formed in conjunction with the sparse coefficient of label information and each multi-source image data in raw data set S
Image object association differentiates data set S'={ (xi,xj',aij), i ∈ (0, N), j ∈ (0, N') }, wherein xiAnd xj' it is in 3.1
The sparse coefficient of the two kinds of information sources generated, aij=0 or 1, it marks for the Central Plains S somebody's work as a result, aij=0 indicates xiAnd xj' be
Dereferenced target, aij=1 indicates xiAnd xj' it is associated objects.
5. as described in claim 1 based on the multi-source image target association method for improving dictionary learning, which is characterized in that described
Multi-source image target association discrimination model in step 4 is the multilayer neural network model using three layers of full connection structure, input
Layer is the L2 distance of two any multi-source image Sparse coefficients, and second layer node number is 512, and third layer node number is
256, in addition to output layer, each interlayer is all made of dropout strategy, and activation primitive uses relu function；Output layer node number
It is set as 1, activation primitive uses sigmoid function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910028307.5A CN109784399A (en) | 2019-01-11 | 2019-01-11 | Based on the multi-source image target association method for improving dictionary learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910028307.5A CN109784399A (en) | 2019-01-11 | 2019-01-11 | Based on the multi-source image target association method for improving dictionary learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109784399A true CN109784399A (en) | 2019-05-21 |
Family
ID=66500374
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910028307.5A Pending CN109784399A (en) | 2019-01-11 | 2019-01-11 | Based on the multi-source image target association method for improving dictionary learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109784399A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582356A (en) * | 2020-04-30 | 2020-08-25 | 中国人民解放军海军航空大学 | A unified representation and association method of target information |
| CN111738310A (en) * | 2020-06-04 | 2020-10-02 | 科大讯飞股份有限公司 | Material classification method and device, electronic equipment and storage medium |
| CN111855208A (en) * | 2020-07-02 | 2020-10-30 | 南京航空航天大学 | A method for extracting weak fault signals of rotating machinery based on order analysis and sparse coding |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544507A (en) * | 2013-10-15 | 2014-01-29 | 中国矿业大学 | High spectral data dimension reduction method based on pairwise constraint discriminant analysis-nonnegative sparse divergence |
| US20140072209A1 (en) * | 2012-09-13 | 2014-03-13 | Los Alamos National Security, Llc | Image fusion using sparse overcomplete feature dictionaries |
| CN105808706A (en) * | 2016-03-06 | 2016-07-27 | 中国人民解放军国防科学技术大学 | Space object identification method based on application ontology |
| US9613439B1 (en) * | 2015-10-16 | 2017-04-04 | The United States Of America, As Represented By The Secretary Of The Navy | System and method for a hierarchical Bayesian-map approach for solving inverse problems |
| CN107358249A (en) * | 2017-06-07 | 2017-11-17 | 南京师范大学 | The hyperspectral image classification method of dictionary learning is differentiated based on tag compliance and Fisher |
| CN108133232A (en) * | 2017-12-15 | 2018-06-08 | 南京航空航天大学 | A kind of Radar High Range Resolution target identification method based on statistics dictionary learning |
| CN108198147A (en) * | 2018-01-02 | 2018-06-22 | 昆明理工大学 | A kind of method based on the multi-source image fusion denoising for differentiating dictionary learning |
| CN108229505A (en) * | 2018-02-05 | 2018-06-29 | 南京邮电大学 | Image classification method based on FISHER multistage dictionary learnings |
| US10055853B1 (en) * | 2017-08-07 | 2018-08-21 | Standard Cognition, Corp | Subject identification and tracking using image recognition |
| CN108734115A (en) * | 2018-05-03 | 2018-11-02 | 电子科技大学 | A kind of radar target identification method based on the consistent dictionary learning of label |
| CN108846384A (en) * | 2018-07-09 | 2018-11-20 | 北京邮电大学 | Merge the multitask coordinated recognition methods and system of video-aware |
| CN109063752A (en) * | 2018-07-17 | 2018-12-21 | 华北水利水电大学 | The method for sorting of the multiple dimensioned real-time stream of multi-source higher-dimension neural network based |
-
2019
- 2019-01-11 CN CN201910028307.5A patent/CN109784399A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140072209A1 (en) * | 2012-09-13 | 2014-03-13 | Los Alamos National Security, Llc | Image fusion using sparse overcomplete feature dictionaries |
| CN103544507A (en) * | 2013-10-15 | 2014-01-29 | 中国矿业大学 | High spectral data dimension reduction method based on pairwise constraint discriminant analysis-nonnegative sparse divergence |
| US9613439B1 (en) * | 2015-10-16 | 2017-04-04 | The United States Of America, As Represented By The Secretary Of The Navy | System and method for a hierarchical Bayesian-map approach for solving inverse problems |
| CN105808706A (en) * | 2016-03-06 | 2016-07-27 | 中国人民解放军国防科学技术大学 | Space object identification method based on application ontology |
| CN107358249A (en) * | 2017-06-07 | 2017-11-17 | 南京师范大学 | The hyperspectral image classification method of dictionary learning is differentiated based on tag compliance and Fisher |
| US10055853B1 (en) * | 2017-08-07 | 2018-08-21 | Standard Cognition, Corp | Subject identification and tracking using image recognition |
| CN108133232A (en) * | 2017-12-15 | 2018-06-08 | 南京航空航天大学 | A kind of Radar High Range Resolution target identification method based on statistics dictionary learning |
| CN108198147A (en) * | 2018-01-02 | 2018-06-22 | 昆明理工大学 | A kind of method based on the multi-source image fusion denoising for differentiating dictionary learning |
| CN108229505A (en) * | 2018-02-05 | 2018-06-29 | 南京邮电大学 | Image classification method based on FISHER multistage dictionary learnings |
| CN108734115A (en) * | 2018-05-03 | 2018-11-02 | 电子科技大学 | A kind of radar target identification method based on the consistent dictionary learning of label |
| CN108846384A (en) * | 2018-07-09 | 2018-11-20 | 北京邮电大学 | Merge the multitask coordinated recognition methods and system of video-aware |
| CN109063752A (en) * | 2018-07-17 | 2018-12-21 | 华北水利水电大学 | The method for sorting of the multiple dimensioned real-time stream of multi-source higher-dimension neural network based |
Non-Patent Citations (7)
| Title |
|---|
| MAOGUO GONG 等: "Coupled Dictionary Learning for Change Detection From Multisource Data", 《IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING》 * |
| XIONG ZHENYU 等: "Object-Level Matching for Multi-source Image Using Improved Dictionary Learning Algorithm", 《INTERNATIONAL SYMPOSIUM ON ARTIFICIAL INTELLIGENCE AND ROBOTICS》 * |
| YONGLI XU 等: "A New Method Based on Two-Stage Detection Mechanism for Detecting Ships in High-Resolution SAR Images", 《2017 INTERNATIONAL CONFERENCE ON ELECTRONIC INFORMATION TECHNOLOGY AND COMPUTER ENGINEERING (EITCE 2017)》 * |
| 吴烨 等: "面向高效检索的多源地理空间数据关联模型", 《计算机学报》 * |
| 周杏芳: "基于多源图像融合的铁路异物侵限检测方法研究", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
| 尹海涛: "面向图像融合和图像复原的稀疏表示研究", 《中国博士学位论文全文数据库 信息科技辑》 * |
| 张宝华: "基于多尺度变换和稀疏表示的多源图像融合算法研究", 《中国博士论文全文数据库 信息科技辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111582356A (en) * | 2020-04-30 | 2020-08-25 | 中国人民解放军海军航空大学 | A unified representation and association method of target information |
| CN111582356B (en) * | 2020-04-30 | 2022-03-29 | 中国人民解放军海军航空大学 | Target information unified representation and association method |
| CN111738310A (en) * | 2020-06-04 | 2020-10-02 | 科大讯飞股份有限公司 | Material classification method and device, electronic equipment and storage medium |
| CN111738310B (en) * | 2020-06-04 | 2023-12-01 | 科大讯飞股份有限公司 | Material classification method, device, electronic equipment and storage medium |
| CN111855208A (en) * | 2020-07-02 | 2020-10-30 | 南京航空航天大学 | A method for extracting weak fault signals of rotating machinery based on order analysis and sparse coding |
| CN111855208B (en) * | 2020-07-02 | 2021-09-21 | 南京航空航天大学 | Rotary machine weak fault signal extraction method based on order analysis and sparse coding |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110222653B (en) | A Behavior Recognition Method of Skeleton Data Based on Graph Convolutional Neural Network | |
| CN110414368B (en) | Unsupervised pedestrian re-identification method based on knowledge distillation | |
| CN108960140B (en) | Pedestrian re-identification method based on multi-region feature extraction and fusion | |
| CN105138998B (en) | Pedestrian based on the adaptive sub-space learning algorithm in visual angle recognition methods and system again | |
| CN109559320A (en) | Realize that vision SLAM semanteme builds the method and system of figure function based on empty convolution deep neural network | |
| CN110728694B (en) | Long-time visual target tracking method based on continuous learning | |
| Balaska et al. | Unsupervised semantic clustering and localization for mobile robotics tasks | |
| CN111325115A (en) | Countermeasures cross-modal pedestrian re-identification method and system with triple constraint loss | |
| CN111814845B (en) | A pedestrian re-identification method based on multi-tributary fusion model | |
| CN110084211B (en) | A method of action recognition | |
| CN113743544A (en) | Cross-modal neural network construction method, pedestrian retrieval method and system | |
| CN109784399A (en) | Based on the multi-source image target association method for improving dictionary learning | |
| CN116863177B (en) | An object view distillation method for general multi-view object clustering | |
| CN110399828A (en) | A vehicle re-identification method based on multi-angle deep convolutional neural network | |
| CN113947101A (en) | An unsupervised person re-identification method and system based on softened similarity learning | |
| CN114241226A (en) | Three-dimensional point cloud semantic segmentation method based on multi-neighborhood characteristics of hybrid model | |
| CN109299676A (en) | A Visual Tracking Method Combining Classification and Domain Adaptation | |
| CN116805051A (en) | Dual convolution dynamic domain adaptive equipment fault diagnosis method based on attention mechanism | |
| CN112329842A (en) | A multi-view and multi-label classification method based on "view-marker" contribution and consistency | |
| CN114881155B (en) | Fruit image classification method based on deep migration learning | |
| CN114550210A (en) | Person Re-identification Method Based on Modality Adaptive Mixing and Invariant Convolution Decomposition | |
| CN115100246A (en) | Cross-modal retrieval method and system for language-vision target tracking | |
| CN112699954B (en) | Closed loop detection method based on deep learning and bag-of-word model | |
| Li et al. | Study of face orientation recognition based on neural network | |
| CN112001345B (en) | Few-sample human behavior identification method and system based on feature transformation measurement 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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190521 |
|
| WD01 | Invention patent application deemed withdrawn after publication |