CN104392412B - Compressed sensing signal recovery method based on evolution orthogonal matching pursuit - Google Patents
Compressed sensing signal recovery method based on evolution orthogonal matching pursuit Download PDFInfo
- Publication number
- CN104392412B CN104392412B CN201410571392.7A CN201410571392A CN104392412B CN 104392412 B CN104392412 B CN 104392412B CN 201410571392 A CN201410571392 A CN 201410571392A CN 104392412 B CN104392412 B CN 104392412B
- Authority
- CN
- China
- Prior art keywords
- individual
- lambda
- population
- gene activity
- compressed sensing
- 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
Landscapes
- Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
Abstract
The invention discloses a kind of signal recovery method based on evolution orthogonal matching pursuit, mainly in solution compressed sensing, traditional tracing algorithm is excessively greedy, and backdating capability is poor and recovers the low problem of accuracy rate.Its technical scheme is:The framework of evolutionary computation is incorporated into compressed sensing signal and recovers central;The problem that atom is selected is converted into the process of the population optimizing based on heuristic search;In conjunction with the dependency of observation error in conventional greedy tracing algorithm and atom, define a kind of activity functions to measure the selected probability of each atom;By activity functions, have devised intersection and the mutation operator of weak greediness, so that more atom is possible to be chosen, increased the reachable tree of atom searching during signal recovers.Experiment shows, the compressed sensing for signal is recovered, and the present invention has higher recovery probability and less restoration errors than traditional greedy tracing algorithm, can be used for the one-dimensional signal and two dimensional image signal recovery problem under low sampling rate random observation.
Description
Technical field
The invention belongs to signal processing technology field, particularly to a kind of compressed sensing signal recovery method.Can be used for one
The dimensional signal and two dimensional image signal recovery problem under low sampling rate random observation.
Background technology
Compressed sensing is a kind of efficient signal acquisition means, and it only leads to too small amount of observation and can recover original letter
Number.The basic assumption of compressed sensing is:Signal can be organized the sparsely linear expression of base by certain, and that is, most of atomic are
Zero, only only a few atomic is non-zero.Under this assumption, by meeting the observation of certain condition, you can extensive exactly
Appear again signal.Because compressed sensing is non-convex underdetermined problem, it is difficult to provide a closed solution form.Such issues that solution, is the most frequently used
Technology be greedy follow the trail of, classical algorithm has match tracing, orthogonal matching pursuit, subspace is followed the trail of, the orthogonal coupling of segmentation chases after
Track and regularization orthogonal matching pursuit.
The basic thought of orthogonal matching pursuit algorithm is:Initialization observation error first is observation vector, in each iteration
In, the inner product of each atom in calculating observation error and compressed sensing matrix;Then, select the former of corresponding inner product maximum absolute value
Son is in the middle of supported collection, and calculates the coefficient of atom in supported collection with method of least square;Finally, update observation error, enter
Next iteration.By this iterative process, so that observation error is gradually reduced, realize primary signal is approached.
Subspace tracing algorithm is improved by orthogonal matching pursuit algorithm, and it introduces back trace technique can not to eliminate
The atom of letter, its basic thought is:Assume atom nonzero coefficient number be K, first, initialization observation error be observation to
Amount, and calculating observation error and the inner product of each atom in compressed sensing matrix, K atom of inner product maximum absolute value is added
To in supported collection, then seek the coefficient of this K atom using method of least square, and update observation error;In each iteration,
Again in calculating observation error and compressed sensing matrix each atom inner product, and the K atom by inner product maximum absolute value
It is added in supported collection, this results in the supported collection of 2K atom;Then, solve this 2K atom using method of least square
Coefficient;Afterwards, by K reservation of 2K atomic maximum absolute value in supported collection, remaining atom is deleted, such
Supported collection to a K atom;Method of least square is recycled to solve the coefficient of this K atom;Finally, update observation by mistake
Difference, enters next iteration.
Although the innovatory algorithm based on orthogonal matching pursuit for the many obtains very big for signaling protein14-3-3 probability and error
Lifting, but due to the limitation such as matching pursuit algorithm itself is excessively greedy and hunting zone is limited, so that search is fallen into
Enter local optimum.
Content of the invention
It is an object of the invention to improving, traditional tracing algorithm is excessively greedy, the problem of backdating capability difference, to improve signal
Recovery probability and recover precision.It is a kind of compressed sensing signal recovery method based on evolution orthogonal matching pursuit, for right
The one-dimensional signal and two dimensional image signal recovery problem under low sampling rate random observation.The present invention is under the framework of evolutionary computation
To solve compressed sensing signal recovery problems, gene is introduced according to the dependency size of observation error in greedy algorithm and atom
The concept of activity, by the use of gene activity as the heuristic knowledge of whole population, to operate the intersection between Different Individual and change
Different, so that whole population being capable of constantly iteration optimizing.
The technical scheme is that:Initialize parent population first, calculate wherein all individual fitness and gene
Activity, and record optimum individual.Then the individuality in parent population is matched at random two-by-two, according to gene activity to every a pair
Individual enforcement crossover operation, the new individual obtaining is stored in progeny population, and calculates their fitness and gene activity.It
Afterwards, row variation is entered to each of progeny population individuality according to overall gene activity, often a variation individual will in time more
Its fitness new, gene activity and overall gene activity.Finally, according to fitness quality, select new in parent and filial generation
Parent population in next iteration, until iteration stopping.Its concrete steps includes as follows:
(1) input compressed sensing matrix DcsWith observation vector y, initialize one containing S individual parent population Pf=
{pi}1≤i≤S, put enumerator t=0;
(2) calculate parent populationFitness and gene activity, and to record the maximum individuality of fitness be optimum
Body pbest;
(3) to parent populationCarry out crossover operation, obtain progeny population
(4) calculate progeny populationFitness and gene activity;
(5) to progeny populationCarry out mutation operation, update in mutation process each individual fitness, gene activity and
Overall gene activity;
(6) from parent populationAnd progeny populationIn select new parent populationAnd update optimum individual
pbest, put enumerator t=t+1;
(7) set maximum iteration time as TmaxIf, t<Tmax, return to step (3);Otherwise, export optimal result.
Present invention incorporates evolutionary computation and greedy tracer technique, the solution of sparse coefficient in compressed sensing is converted into weak
Greedy heuristic population search procedure, and it is introduced into the dependency of observation error and atom in greedy algorithm as kind of a group hunting
Heuristic knowledge, it has the advantage that:
(A), the problem that conventional greedy tracing algorithm is absorbed in local optimum is improved based on the heuristic search of colony;
(B), the recovery probability for one-dimensional signal and recovery precision are significantly improved;
(C), for the recovery of two dimensional image, there is less restoration errors and less blocking effect.
It is demonstrated experimentally that for One-dimensional simulation signaling protein14-3-3, the present invention has higher extensive than traditional greedy tracing algorithm
Multiple probability and less restoration errors.For the recovery of two dimensional image, the present invention is higher than traditional greedy tracing algorithm has
Y-PSNR PSNR and less restoration errors.
Brief description
Fig. 1 is the overall flowchart of the present invention;
Fig. 2 is the one-dimensional signal recovery effects contrast that the present invention and other methods obey standard just too distribution for coefficient;
Fig. 3 is that the present invention obeys -1 to 1 equally distributed one-dimensional signal recovery effects contrast with other methods for coefficient;
Fig. 4 is that the present invention is contrasted for the recovery effects of two dimensional image Lena with other methods;
Fig. 5 is that the present invention is contrasted for the restoration errors of two dimensional image Lena with other methods;
Specific embodiment
With reference to Fig. 1, the present invention to implement step as follows:
Step 1, initialization one is containing 16 individual parent populations
1.1) input compressed sensing matrix DcsWith observation vector y, putting parent population is empty set
1.2) calculate compressed sensing matrix DcsMiddle atom and the dependency c of observation vector y:
Wherein, | | represent the absolute value seeking vector element;
1.3) by descending for the element in vectorial c sequence:Find a group index sequence { λ1,λ2,…,λNSo that c
[λ1]≥c[λ2]≥…≥c[λN], wherein c [λn] represent the λ of vectorial cnIndividual element, N is signal dimension;
1.4) generate a full 0 individuality pi=[0,0 ..., 0], puts pi[λi]=1, wherein pi[λi] represent individual pi?
λiIndividual gene;
1.5) by individual piIt is added in parent populationPut i=i+1;
1.6) if i≤16, return to step 1.4;Otherwise, export parent population
Step 2, calculates parent populationFitness and gene activity, and to record the maximum individuality of fitness be optimum
Individual pbest.
2.1) for parent populationIn each individual p, solve its sparse coefficient vector α, be calculated as follows:
Wherein αpRepresent the subvector that the position being 1 with gene in individual p is extracted from vectorial α for index,Represent with
In individual p, gene is the subvector that 0 position is index extracts from vectorial α,Represent the position being 1 with gene in individual p
For index from matrix DcsThe submatrix that the column vector of middle extraction is constituted, subscriptRepresent the pseudoinverse seeking matrix, 0 one element of expression
It is all 0 vector;
2.2) seek the fitness value of individual p, be calculated as follows:
Wherein, | | | |2For calculating 2 norms of vector;
2.3) calculate the gene activity of individual p, be calculated as follows:
Wherein U is normaliztion constant, is equal toThe maximum of middle element;
2.4) individual record maximum for fitness value is optimized individual pbest.
Step 3, to parent populationCarry out crossover operation, obtain progeny population
3.1) set progeny population as empty set:By parent populationIn individuality match two-by-two at random, obtain 8 right
Individual (pi,pj);
3.2) to every a pair individual (pi,pj), find the different index set Λ of gene under their same positions:
Wherein, pi[λ] represents individual piλ position gene;
3.3) find individual piAnd pjThe position of gene activity maximum in index ΛWith
Wherein, li[λ] represents individual piThe activity value of λ position gene;
3.4) exchange individual piAnd pjGene position:IfPutObtain new individual
pi', putObtain new individual p 'j;Otherwise, putObtain new individual pi', put?
To new individual p 'j;
3.5) by two new individuals pi' and p 'jIt is added to progeny populationRepeat step 3.2 to 3.5
Operation, until all individualities complete to intersect.
Step 4, calculates progeny populationFitness and gene activity.
To progeny populationIn all individualities, calculate their fitness and gene activity according to step 2.1 to 2.3.
Step 5, to progeny populationCarry out mutation operation, update each individual fitness in mutation process, gene is lived
Property and overall gene activity.
5.1) put i=1;
5.2) calculate progeny populationAverage degree of rarefication Kmean, it is calculated as follows:
Wherein p [j] represents j-th gene of individual p, and N is signal length;
5.3) calculate progeny populationOverall gene activity, be calculated as follows:
Wherein ljRepresent the gene activity of j-th individuality;
5.4) calculate individual piDegree of rarefication Ki:
5.5) judge individual piDegree of rarefication KiWith populationAverage degree of rarefication KmeanSize, and according to magnitude relationship
Carry out mutation operation;
5.5.1) if Ki≤Kmean, recording individual piMiddle genic value is 0 index set:
Λ=λ | pi[λ]=0,1≤λ≤N },
Find the position λ of overall gene activity L maximum in index set Λmax:
Put individual piλmaxIndividual gene is 1, i.e. pi[λmax]=1;
5.5.2) if Ki>Kmean, recording individual piMiddle genic value is 1 index set:
Λ=λ | pi[λ]=1,1≤λ≤N },
Find the position λ of overall gene activity L minima in index set Λmin:
Put individual piλminIndividual gene is 0, i.e. pi[λmin]=0;
5.6) according to step 2.1) to 2.3) more new individual piFitness value and gene activity, put i=i+1, if i≤
16, return to step 5.2);Otherwise, the progeny population after output variation
Step 6, from parent populationAnd progeny populationIn select new parent populationAnd update optimum
Body pbest, put enumerator t=t+1.
In parent populationAnd progeny populationSet in, the new father of the individuality composition that selects 16 fitness values maximum
For populationThe maximum individuality of fitness value in new parent population is updated to current optimum individual pbest, and put meter
Number device t=t+1.
Step 7, if maximum iteration time is Tmax=100, if t<Tmax, return to step 3;Otherwise, export optimal result.
The effect of the present invention can be illustrated by emulation experiment:
1. experiment condition
Microcomputer CPU used by experiment is Intel Core22.66GHz internal memory 4GB, and programming platform is Matlab2013.Experiment
The view data of middle employing is the Lena image in standard picture test library, and size is 512 × 512.
2. experiment content
The One-dimensional simulation signal that experiment 1 obeys standard normal distribution to nonzero coefficient recovers:
If signal length N=256, observe dimension M=100, the degree of rarefication of sparse coefficient vector α increases to 50 from 10, often
Individual degree of rarefication does 1000 independent experiments, and in testing every time, the nonzero coefficient of coefficient vector α is just being distributed very much random producing with standard
Raw, compressed sensing matrix DcsElement average be 0, standard deviation be 1/N Gauss distribution produce.Then observation distinct methods exist
The correct probability recovering signal and average restoration errors in 1000 experiments.With the present invention and 4 kinds of classical greediness tracing algorithms pair
Above-mentioned emulation signal is recovered, and 4 kinds of classical greediness tracing algorithms are respectively:Match tracing MP, orthogonal matching pursuit OMP, point
SP is followed the trail of in section orthogonal matching pursuit StOMP and subspace.Experimental result is as shown in Fig. 2 wherein:
Fig. 2 (a), obeys to nonzero coefficient for the present invention and described 4 kinds of classical greediness tracing algorithms under different degree of rarefications
The probability correlation curve that the signal of standard normal distribution accurately recovers,
Fig. 2 (b), obeys to nonzero coefficient for the present invention and described 4 kinds of classical greediness tracing algorithms under different degree of rarefications
The correlation curve of the signal averaging restoration errors of standard normal distribution.
Experiment 2 is obeyed -1 to 1 equally distributed One-dimensional simulation signal to nonzero coefficient and is recovered:
If signal length N=256, observe dimension M=100, the degree of rarefication of sparse coefficient vector α increases to 50 from 10, often
Individual degree of rarefication does 1000 independent experiments, in testing every time, the nonzero coefficient of coefficient vector α with -1 to 1 be uniformly distributed with
Machine produces, compressed sensing matrix DcsElement average be 0, standard deviation be 1/N Gauss distribution produce.Then observe not Tongfang
The method correct probability recovering signal and average restoration errors in 1000 experiments.Calculated with classical greedy tracking of the present invention and 4 kinds
Method is recovered to above-mentioned emulation signal, and 4 kinds of classical greediness tracing algorithms are respectively:Match tracing MP, orthogonal matching pursuit
SP is followed the trail of in OMP, segmentation orthogonal matching pursuit StOMP and subspace.Experimental result is as shown in figure 3, wherein:
Fig. 3 (a) be the present invention and described 4 kinds of classical greediness tracing algorithms under different degree of rarefications, nonzero coefficient is obeyed-
The probability correlation curve that 1 to 1 equally distributed signal accurately recovers,
Fig. 3 (b) be the present invention and described 4 kinds of classical greediness tracing algorithms under different degree of rarefications, nonzero coefficient is obeyed-
The correlation curve of 1 to 1 equally distributed signal averaging restoration errors.
As can be seen that for one-dimensional emulation signal, greedy follow the trail of classical compared to other 4 kinds is calculated from Fig. 2 and Fig. 3
Method, the present invention has higher recovery probability and less restoration errors.
Experiment 3 carries out splits' positions perception and recovers to two dimensional image Lena, compares the Y-PSNR PSNR of algorithms of different:
Lena image is divided into 8 × 8 fritter, setting observation dimension M=32, maximum iteration time is 16.Observing matrix
Element average in Φ is 0, and standard deviation is that the Gauss distribution of 1/N produces, and base dictionary Ψ adopts DCT dictionary, compressed sensing square
Battle array Dcs=Φ Ψ.Observe the restoration result of distinct methods, and compare their Y-PSNR PSNR.With the present invention and existing
Match tracing MP is recovered to image Lena with orthogonal matching pursuit OMP algorithm.Experimental result is as shown in figure 4, wherein:
Fig. 4 (a) is original Lena topography,
Fig. 4 (b) is the restoration result to Lena topography for the present invention, Y-PSNR PSNR=31.3dB,
Fig. 4 (c) is the restoration result to Lena topography for the MP algorithm, Y-PSNR PSNR=29.2dB,
Fig. 4 (d) is the restoration result to Lena topography for the OMP algorithm, Y-PSNR PSNR=29.6dB.
From fig. 4, it can be seen that for two dimensional image signaling protein14-3-3, the border of restoration result of the present invention closest to artwork,
Y-PSNR PSNR highest, and artificial blocking effect is minimum.
Experiment 4 carries out splits' positions perception and recovers to two dimensional image Lena, compares the restoration errors of algorithms of different:
Lena image is divided into 8 × 8 fritter, setting observation dimension M=32, maximum iteration time is 16.Observing matrix
Element average in Φ is 0, and standard deviation is that the Gauss distribution of 1/N produces, and base dictionary Ψ adopts DCT dictionary, compressed sensing square
Battle array Dcs=Φ Ψ.Observe the restoration errors figure of distinct methods.Relatively the present invention and subspace follow the trail of SP, match tracing MP with orthogonal
The restoration errors of match tracing OMP algorithm.Experimental result is as shown in figure 5, wherein:
Fig. 5 (a) is the restoration errors figure to Lena topography for the SP algorithm,
Fig. 5 (b) is the restoration errors figure to Lena topography for the present invention,
Fig. 5 (c) is the restoration errors figure to Lena topography for the MP algorithm,
Fig. 5 (d) is the restoration errors figure to Lena topography for the OMP algorithm.
From the point of view of Error Graph from Fig. 5, the error span of restoration result of the present invention is minimum.
To sum up, no matter in One-dimensional simulation signal or two dimensional image signaling protein14-3-3, the recovery effects of the present invention are all excellent
Classical greedy tracing algorithm in other 4.
Claims (5)
1. a kind of compressed sensing signal recovery method based on evolution orthogonal matching pursuit, its flow process is as follows:
(1) input compressed sensing matrix DcsWith observation vector y, initialize one containing S individual parent populationPut enumerator t=0;
(2) calculate parent populationFitness and gene activity, and to record the maximum individuality of fitness be optimum individual
pbest, calculate gene activity formula as follows:
Wherein U is normaliztion constant, is equal toThe maximum of middle element, y represents observation vector, DcsRepresent compression
Perception matrix, α represents sparse coefficient vector;
(3) to parent populationCarry out crossover operation, obtain progeny population
(4) calculate progeny populationFitness and gene activity;
(5) to progeny populationCarry out mutation operation, in mutation process, update each individual fitness, gene activity and entirety
Gene activity, the formula calculating overall gene activity is as follows:
Wherein ljRepresent the gene activity of j-th individuality, S represents the quantity containing individuality in population;
(6) from parent populationAnd progeny populationIn select new parent populationAnd update optimum individual pbest, put
Enumerator t=t+1;
(7) set maximum iteration time as TmaxIf, t < Tmax, return to step (3);Otherwise, export optimal result.
2. the compressed sensing signal recovery method based on evolution orthogonal matching pursuit according to claim 1, its step (1)
Described input compressed sensing matrix DcsWith observation vector y, initialize one containing S individual parent populationPut enumerator t=0 to comprise the following steps that:
(1.1) input compressed sensing matrix DcsWith observation vector y;
(1.2) putting parent population is empty setPut i=1;
(1.3) calculate compressed sensing matrix DcsMiddle atom and the dependency c of observation vector y:
Wherein, | | represent the absolute value seeking vector element;
(1.4) by descending for the element in vectorial c sequence:Find a group index { λ1,λ2,...λn...,λNSo that c
[λ1]≥c[λ2]≥...c[λn]...≥c[λN], wherein c [λn] represent the λ of vectorial cnIndividual element, N is signal dimension;
(1.5) generate a full 0 individuality pi=[0,0 ..., 0], puts pi[λi]=1, wherein pi[λi] represent individual piλiIndividual
Element;
(1.6) by individual piIt is added in parent populationPut i=i+1;
(1.7) set S as parent populationIn individual amount, if i≤S, return (1.5);Otherwise, export parent populationPut meter
Number device t=0.
3. the compressed sensing signal recovery method based on evolution orthogonal matching pursuit according to claim 1, its step (2)
Described calculating parent populationFitness and gene activity, and to record the maximum individuality of fitness be optimum individual pbest
Concretely comprise the following steps:
(2.1) for parent populationIn each individual p, solve its sparse coefficient vector α:
Wherein αpRepresent the subvector that the position being 1 with element in individual p is extracted from vectorial α for index,Represent with individual p
Middle element be 0 position be index from vectorial α extract subvector,Represent with element in individual p be 1 position as rope
Draw from matrix DcsThe submatrix that the column vector of middle extraction is constituted, subscriptRepresent the pseudoinverse seeking matrix, 0 one element of expression is all
0 vector;
(2.2) seek the fitness value of individual p:
Wherein, | | | |2For calculating 2 norms of vector;
(2.3) calculate the gene activity of individual p:
Wherein U is normaliztion constant, is equal toThe maximum of middle element.
4. the compressed sensing signal recovery method based on evolution orthogonal matching pursuit according to claim 1, its step (3)
Described to parent populationThe method carrying out crossover operation is as follows:
(3.1) putting progeny population is empty setBy parent populationIn individuality match two-by-two, obtain S/2 to individuality;
(3.2) to every a pair individual (pi,pj), find the different index set Λ of gene under their same positions, that is,:
(3.3) find individual piAnd pjThe position of gene activity maximum in index ΛWithI.e.:
Wherein li[λ] represents individual piThe gene activity value of λ position;
(3.4) exchange individual piAnd pjGene position:IfPutObtain new individual p 'i,
PutObtain new individual p 'j;Otherwise, putObtain new individual p 'i, putObtain new
Individual p 'j;
(3.5) by two new individual p 'iWith p 'jIt is added to progeny populationRepeat step (3.2) arrives (3.5)
Operation, until all individualities complete to intersect.
5. the compressed sensing signal recovery method based on evolution orthogonal matching pursuit according to claim 1, its step (5)
Described to progeny populationCarry out mutation operation, process is as follows:
(5.1) put i=1;
(5.2) calculate progeny populationAverage degree of rarefication Kmean:
Wherein p [j] represents j-th gene of individual p, and N is signal length;
(5.3) calculate progeny populationOverall gene activity, formula is as follows:
Wherein ljRepresent the gene activity of j-th individuality;
(5.4) calculate individual piDegree of rarefication Ki, formula is as follows:
(5.5) judge individual piDegree of rarefication KiWith populationAverage degree of rarefication KmeanSize, and carried out according to magnitude relationship
Mutation operation;
If (5.5.1) Ki≤Kmean, recording individual piMiddle genic value is 0 index set:
Λ=λ | pi[λ]=0,1≤λ≤N },
Find the position λ of overall gene activity L maximum in index set Λmax:
Put individual piλmaxIndividual gene is 1, i.e. pi[λmax]=1;
If (5.5.2) Ki> Kmean, recording individual piMiddle genic value is 1 index set:
Λ=λ | pi[λ]=1,1≤λ≤N },
Find the position λ of overall gene activity L minima in index set Λmin:
Put individual piλminIndividual gene is 0, i.e. pi[λmin]=0;
(5.6) more new individual piFitness value, gene activity, put i=i+1, if i≤S, return (5.2);Otherwise, output variation
Progeny population afterwards
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410571392.7A CN104392412B (en) | 2014-10-23 | 2014-10-23 | Compressed sensing signal recovery method based on evolution orthogonal matching pursuit |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410571392.7A CN104392412B (en) | 2014-10-23 | 2014-10-23 | Compressed sensing signal recovery method based on evolution orthogonal matching pursuit |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104392412A CN104392412A (en) | 2015-03-04 |
CN104392412B true CN104392412B (en) | 2017-03-01 |
Family
ID=52610310
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410571392.7A Active CN104392412B (en) | 2014-10-23 | 2014-10-23 | Compressed sensing signal recovery method based on evolution orthogonal matching pursuit |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104392412B (en) |
Families Citing this family (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108828525A (en) * | 2018-06-25 | 2018-11-16 | 大连大学 | The more sonic location systems of distributed microphone array in reverberation room |
CN109540279B (en) * | 2018-11-30 | 2020-03-24 | 厦门大学 | Reverse compression perception recovery method for undersampled dynamic signals in high-speed milling process of machine tool |
CN114740433B (en) * | 2022-04-27 | 2023-04-25 | 电子科技大学 | Time synchronization method based on compressed sensing under influence of multipath effect |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102148987B (en) * | 2011-04-11 | 2012-12-12 | 西安电子科技大学 | Compressed sensing image reconstructing method based on prior model and 10 norms |
CN103295198B (en) * | 2013-05-13 | 2015-11-25 | 西安电子科技大学 | Based on redundant dictionary and the sparse non-convex compressed sensing image reconstructing method of structure |
-
2014
- 2014-10-23 CN CN201410571392.7A patent/CN104392412B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN104392412A (en) | 2015-03-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
De Luna et al. | Identification of philippine herbal medicine plant leaf using artificial neural network | |
CN108170736A (en) | A kind of document based on cycle attention mechanism quickly scans qualitative method | |
CN105468677A (en) | Log clustering method based on graph structure | |
CN104992184A (en) | Multiclass image classification method based on semi-supervised extreme learning machine | |
CN106202999B (en) | Microorganism high-pass sequencing data based on different scale tuple word frequency analyzes agreement | |
CN109086826A (en) | Wheat Drought recognition methods based on picture depth study | |
CN104392412B (en) | Compressed sensing signal recovery method based on evolution orthogonal matching pursuit | |
CN108959841A (en) | A kind of drug targeting albumen effect prediction technique based on DBN algorithm | |
Shao et al. | Deep active learning for nucleus classification in pathology images | |
CN109325510B (en) | Image feature point matching method based on grid statistics | |
CN109697692A (en) | One kind being based on the similar feature matching method of partial structurtes | |
CN102142068A (en) | Method for detecting unknown malicious code | |
CN109948646A (en) | A kind of time series data method for measuring similarity and gauging system | |
CN110443372A (en) | A kind of transfer learning method and system based on entropy minimization | |
CN107346556A (en) | A kind of PET image reconstruction method based on block dictionary learning and sparse expression | |
CN102831614A (en) | Sequential medical image quick segmentation method based on interactive dictionary migration | |
CN105574265B (en) | Entire assembly model quantitative description towards model index | |
CN110019805A (en) | Article Topics Crawling method and apparatus and computer readable storage medium | |
CN107491508A (en) | A kind of data base querying time forecasting methods based on Recognition with Recurrent Neural Network | |
CN104484432A (en) | Color image searching method based on quaternion exponential moment | |
CN102799627A (en) | Data association method based on first-order logic and nerve network | |
CN117131449A (en) | Data management-oriented anomaly identification method and system with propagation learning capability | |
CN105701501B (en) | A kind of trademark image recognition methods | |
CN105069474B (en) | Semi-supervised learning high confidence level sample method for digging for audio event classification | |
CN104537392B (en) | A kind of method for checking object based on the semantic part study of identification |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |