CN110097103A - Based on the semi-supervision image classification method for generating confrontation network - Google Patents
Based on the semi-supervision image classification method for generating confrontation network Download PDFInfo
- Publication number
- CN110097103A CN110097103A CN201910323702.6A CN201910323702A CN110097103A CN 110097103 A CN110097103 A CN 110097103A CN 201910323702 A CN201910323702 A CN 201910323702A CN 110097103 A CN110097103 A CN 110097103A
- Authority
- CN
- China
- Prior art keywords
- network
- training
- sample
- classification
- arbiter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2155—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the incorporation of unlabelled data, e.g. multiple instance learning [MIL], semi-supervised techniques using expectation-maximisation [EM] or naïve labelling
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Abstract
The invention discloses a kind of based on the semi-supervision image classification method for generating confrontation network, mainly solve the problems, such as that existing unsupervised learning nicety of grading is low and semi-supervised learning needs a large amount of accurate labels, implementation step are as follows: 1) choose and download standard picture training sample and test sample;2) setting network supervised learning relevant parameter, and build one and network is fought by the generation that generator network, arbiter network and subsidiary classification device form side by side;3) it is trained using stochastic gradient descent method to confrontation network is generated;4) test sample to be sorted is input in trained generation confrontation network model, exports the classification of image to be detected.The present invention improves the image classification accuracy of unsupervised learning, and can obtain good image classification effect on the sample set for containing only a small amount of accurate mark sample, can be used for target classification in actual scene.
Description
Technical field
The invention belongs to technical field of image processing, further relate to a kind of semi-supervision image classification method, can be used for
Target classification in actual scene.
Background technique
The main task of image classification can be achieved on the detection that the target in the image to input carries out classification, Jin Erzhun
Really set the goal belonging classification really.As people are to the understanding that deepens continuously of computer vision field, image classification is at this
Field is used widely and is developed, and currently exists a large amount of sorting algorithm to realize image classification.For tape label
Image classification task based on supervised learning has developed more mature, and preferable classification can be obtained on standard data set
Precision, but in actual application scenarios, the training dataset of network is not easy to obtain, such as in terms of medical diagnosis, number
According to the acquisition of sample with regard to relatively difficult, the expert that the image that secondary acquisition precisely marks requires experience is labeled, this
Sample can expend huge time cost and human cost.Supervised learning itself needs largely to have the training number accurately marked
According to the quality and quantity of training data will have a direct impact on the nicety of grading of image classification task, this results in being difficult to obtain
Supervised learning can not obtain preferable completion classification task in the application scenarios of training sample, and the training sample marked on a small quantity can be led
Cause over-fitting.And semi-supervised learning is suitable for few in labeled data, and uses under the scene more than training data, has in use
It can get preferable training effect on the basis of limitation labeled data, reduce the degree of dependence to labeled data, greatly enrich
The application scenarios of classification task.
IanJ.Goodfellow, JeanPouget-Abadie, MehdiMirza, BingXu, David Ware-
The paper that Farley, SherjilOzair, AaronCourville, YoshuaBengio are delivered at it
Disclosed in " GenerativeAdversarialNets " it is a kind of utilize convolutional neural networks extract characteristics of image it is two stage
Image generates and classification method.This method respectively generates network and differentiates network first by establishing two network structures,
Middle generation network is responsible for receiving a noise vector, maps it onto a false sample image, by this false sample image
It is input in arbiter network together with authentic specimen image, by differentiating that network carries out the true and false to authentic specimen and false sample
Determine, characteristic pattern is classified using class probability.Although this method realizes the classification to natural image, still, the party
The shortcoming that method still has is to mention the characteristics of image of true picture and false image in the last layer neural network
It takes, there is no sufficiently accurate trained label due to only carrying out unsupervised learning sample set, lead to not be exactly found classification boundary,
So that nicety of grading is not ideal enough, it is not able to satisfy the requirement of realistic objective classification.
Patent document " Classification of Polarimetric SAR Image based on generation confrontation network of the Xian Electronics Science and Technology University in its application
Method " discloses in (applying date: on 08 25th, 2017, application number: 201710742716.2, publication number: 107563428A)
A kind of Classification of Polarimetric SAR Image method based on generation confrontation network.This method uses polarimetric synthetic aperture radar to be sorted
The coherence matrix of SAR image uses filter window size for 7 × 7 Lee's Lee filter, carries out to coherence matrix as input
Filtering, the coherence matrix after being denoised extract 9 dimensional features as a sample from the coherence matrix after denoising, will denoise
All samples in coherence matrix afterwards generate a sample set, and 10% sample is randomly selected from the sample set of each ground species
This conduct has exemplar collection, gives birth to 90% sample remaining in the sample set of each ground species as unlabeled exemplars collection training
At network and confrontation network.But there is still a need for 10% marker samples to divide as training sample set image for this method
Class training, although being suitable for the classification of SAR image, to medical image, natural image is sufficient amount of in the presence of being difficult to obtain
Accurate mark training sample problem, greatly limits its application range.
Summary of the invention
It is a kind of based on half prison for generating confrontation network it is an object of the invention in view of the above shortcomings of the prior art, propose
Image classification method is superintended and directed, increases a subsidiary classification device network on arbiter network to contain only a small amount of precise marking data
Training sample classification in obtain better nicety of grading.
Realizing the technical solution of the object of the invention is: firstly, using standard picture categorized data set mnist hand-written data
Collection, but only to wherein 200 accurate training labels of data sample setting, then build a generation confrontation network of network and be arranged every
Layer parameter is trained with training dataset to confrontation network is generated, and is obtained trained based on half prison for generating confrontation network
Sorter network model is superintended and directed, finally image to be detected containing target is input in trained generation confrontation network model, is extracted
The identification to target category is completed while target signature, implementation step includes the following:
(1) it chooses and downloads image classification standard exercise sample set, and carry out normalization operation;
(2) supervised learning parameter is set: i.e. first for the training number of tags of normalization training sample set setting supervised learning
Amount;Reset the sample size of network training each time;Then it is set by the sample size of number of labels and each network training
Determine the flag bit of supervised learning;
(3) it builds one and network is fought by the generation that generator network, arbiter network and subsidiary classification device form side by side,
Wherein, generator network is set as 4 layers, and arbiter network is set as 5 layers, and subsidiary classification device is set as 5 layers;
(4) generation confrontation network is trained:
Training sample set after normalization is input to the arbiter network generated in confrontation network by (4a), extracts each sample
The feature of initial target image in this, and all features are formed into characteristics of image figure, the last layer by arbiter network is defeated
1 unit vector out, for determining the classification of characteristic image;
(4b) receives a noise vector by generator network, and maps it onto one 28 × 28 characteristic pattern, then
By this characteristic pattern and true training sample together as the input of arbiter network, the arbiter net after training for the first time is obtained
Network parameter;
(4c) updates generator network parameter using the arbiter network parameter after training for the first time, then fixes updated
Generator network parameter trains arbiter network again, successively repetitive exercise arbiter network and generator network;
(4d) utilizes generator gradient updating formula, and characteristic matching item is added, and generates after calculating repetitive exercise each time
Device network weight;
(4e) utilizes supervised learning and unsupervised loss function, and each classification results value of computational discrimination device network and classification are tied
Error amount between the true value of fruit;
(4f) utilizes stochastic gradient descent method, and increases label smooth operation to training sample set, updates generator network
With the weight of each node of arbiter network, until arbiter network is for the nicety of grading of authentic specimen and false sample
Reach nash banlance, obtains trained generation confrontation network model;
(5) classify to image to be detected:
Image containing target to be sorted is input in trained generation confrontation network model, the general of class option is exported
Rate value, the highest class option of select probability value are exported as classification results.
The invention has the following advantages over the prior art:
First, improve the accuracy to image classification
Unsupervised training is only carried out in the prior art, does not have the training data of accurate label in training sample, it can not be accurate
The boundary between class object is divided, and the present invention uses semi-supervised learning classification method, in original base for generating confrontation network
Increase a subsidiary classification device on plinth and be used for image classification task, and inputs a small amount of precise marking data in training, these essences
True flag data can more accurately divide the boundary between class object, to more accurately classify to image.
Second, need accurate exemplar quantity less
In existing semi-supervised training technique, it is still desirable to more accurate exemplar, but some practical application areas
There is a problem of that accurate exemplar obtains difficulty, be unable to get a large amount of accurate labeled data, thus is lacking accurate enough mark
It was easy to appear conjunction phenomenon in the case where note training sample, can not accurately complete image classification problem, and the present invention is using more
The gradient of optimization declines update method, can compared with the prior art when processing contains the data sample of a small amount of flag data
Reach more accurately image classification accuracy, possesses wide applicability.
Detailed description of the invention
Fig. 1 is implementation flow chart of the invention;
Fig. 2 is the simulation result diagram with the present invention to mnist data set,
Fig. 3 is the simulation result diagram with the present invention to cifar10 data set.
Specific embodiment
The embodiment of the present invention and effect are described in further detail with reference to the accompanying drawing.
Referring to Fig.1, to the specific steps of the present invention are as follows.
Step 1, it chooses and downloads image classification standard exercise sample set.
Standard picture categorized data set mnist hand-written data collection is downloaded, and operation is normalized to data sample, is used for
Network model training;
Standard picture categorized data set cifar10 data set is downloaded, and operation is normalized to data sample, is used for net
Network model training.
Step 2, supervised learning parameter is set.
The training samples number that training sample is concentrated is counted, the control total training samples number of supervised learning label data Zhan
Percentage, the total training samples number percent value of label data Zhan is higher, and model training precision is better, but some practical applications are led
It is difficult to obtain sufficient amount of accurate label data in domain, the specific total training samples number percent value of label data N Zhan is by reality
Border applicable cases determine, choose 200 training samples to 50000 training samples in this example and are accurately marked, Zhan Zongxun
Practicing sample percentage value is 0.4%;
According to Experimental Hardware performance and model training time comprehensive consideration, the sample size of each network training is set
Size chooses each training samples number size=50 based on this Case Simulation platform;
By the sample size of training number of labels and each network training, the flag bit effective degree of supervised learning is calculated
n;
N=N/size,
Wherein, n indicates the flag bit effective degree of supervised learning, and N indicates the accurate label data of supervised learning.
Step 3, a generation confrontation network is built.
(3a) builds the arbiter network being made of 4 layers of convolutional layer and 1 full articulamentum, and each layer parameter is as follows:
Level 1 volume lamination, the sum of Feature Mapping figure are set as 64, and the scale of convolution kernel is set as 5 × 5 sections
Point, step-length 2, activation primitive are lrelu function;
Level 2 volume lamination, the sum of Feature Mapping figure are set as 128, and the scale of convolution kernel is set as 3 × 3 sections
Point, step-length 2, activation primitive are lrelu function;
3rd layer of convolutional layer, the sum of Feature Mapping figure are set as 256, and the scale of convolution kernel is set as 3 × 3 sections
Point, step-length 2, activation primitive are lrelu function;
4th layer of convolutional layer, the sum of Feature Mapping figure are set as 256, and the scale of convolution kernel is set as 3 × 3 sections
Point, step-length 2, activation primitive are lrelu function;
5th layer is full articulamentum, and the sum of Feature Mapping figure is set as 11.
(3b) builds the generator network being made of 1 full articulamentum and 4 layers of convolutional layer, and each layer parameter is as follows:
First layer is full articulamentum, and the sum of Feature Mapping figure is set as 256;
Second layer convolutional layer, the sum of Feature Mapping figure are set as 256, and the scale of convolution kernel is set as 5 × 5 sections
Point, step-length 2, activation primitive are relu function;
Third layer convolutional layer, the sum of Feature Mapping figure are set as 128, and the scale of convolution kernel is set as 5 × 5 sections
Point, step-length 2, activation primitive are relu function;
4th layer of convolutional layer, the sum of Feature Mapping figure are set as 64, and the scale of convolution kernel is set as 5 × 5 sections
Point, step-length 2, activation primitive are relu function;
(3c) increases a subsidiary classification device network on arbiter network, for inciting somebody to action in the classification task of data sample
The classification task of K class data is extended to K+1 class data sorting task, wherein K+1 class sample is the void that generator network generates
Fault image;
By above-mentioned arbiter network, generator network, subsidiary classification device network these three network parallel arrangeds, composition is generated
Fight network.
Step 4, training generates confrontation network.
To generating the training of confrontation network at present mainly using the method for arbiter network and generator network alternately training,
This step is also to be accomplished by using the method for arbiter network and the alternating training of generator network
Training sample set is input in the arbiter network generated in confrontation network by (4a), passes through arbiter network establishment
4 layers of convolutional layer, 4 convolution operations are carried out to the training sample set of input, extract each initial target image in each sample
Feature, by all features constitute characteristics of image figure;
Noise vector is input in the generator network generated in confrontation network by (4b), passes through generator network establishment
1 full articulamentum and 4 layers of convolutional layer carry out 4 convolution operations to the training sample set of input, and generator network will export one
The false sample of size identical as authentic specimen fixes the parameter value of each node of generator network for generator network at this time
Input of the false picture and authentic specimen of generation together as arbiter network;
(4c) differentiates authentic specimen image and false sample image by arbiter network:
Using generator network losses function, generator network losses functional value is calculated:
tmin=EZ~P (z)Log (1-D (g (z))+f,
Wherein, tminIndicate generator network losses functional value, EZ~P (z)Indicate that sample obeys the mathematic expectaion of a certain distribution
Value, z indicate that the noise vector of a certain distribution of obedience, D (G (z)) indicate that arbiter network generates image for generator network
Differentiate that probability value, G (z) indicate that the sample image that generator network generates, f indicate characteristic matching value;
F={ [ni(r)-ni(f)]*[ni(r)-ni(f)] },
Wherein, ni(r) the true picture mapping value of the i-th layer network of arbiter, n are indicatedi(f) the i-th layer network of arbiter is indicated
Vitua limage mapping value, optimize the gradient of generator network model by characteristic matching item f, so that arbiter network is each
The result of layer is as similar as possible;
The loss function value of (4d) computational discrimination device;
(4d1) calculates the penalty values of supervised learning when supervised learning flag bit is effective by supervised learning loss function,
Network inputs are the training sample with accurate label at this time, and the subsidiary classification device network in arbiter network, which exercises supervision, learns instruction
Practice, calculates the network monitoring of subsidiary classification device and learn loss function value;
The subsidiary classification device supervised learning loss function is as follows:
Wherein, LsIndicate the loss function value of supervised learning,Indicate that sample obeys the mathematics phase of a certain distribution
Prestige value, x indicate to obey the authentic specimen vector of a certain distribution, Pdata(x, y) indicates to obey the authentic specimen probability of a certain distribution
Density function, the categorization values of y presentation class task, Pmodel(y | x, y < K+1) indicate K+1 class class probability model;
(4d2) calculates the damage of unsupervised learning by unsupervised learning loss function when supervised learning flag bit is invalid
Mistake value, calculation formula are as follows:
Wherein, LunsIndicate the loss function value of unsupervised learning,Indicate that authentic specimen obeys a certain distribution
Mathematical expectation, x indicate to obey the authentic specimen vector of a certain distribution, Pmodel(y=K+1 | x) it indicates to obey the general of a certain distribution
Rate density model, EZ~noiseIndicate that false sample obeys the mathematical expectation of a certain distribution, z indicates to obey a certain distribution
Noise vector.
(4d3) is according to supervised learning loss function value and unsupervised learning loss function value, the damage of computational discrimination device network
Functional value is lost, calculation formula is as follows:
Lloss=Luns+Ls
Wherein, LlossIndicate the loss function value of arbiter network, LunsIndicate the damage of the unsupervised learning of arbiter network
Lose functional value, LsIndicate the loss function value of the supervised learning of arbiter network.
(4e) utilizes stochastic gradient descent method, and the weight of each node of more newly-generated confrontation network convolutional layer obtains
Trained generation fights network:
(4e1) selects a number at random in (0,0.1) range, uses the number as each node in generation confrontation network of network
Initial weight;
The initial weight of each node is fought each node in network by (4e2)
Current weight;
(4e3) randomly selects 50 sample images from training sample concentration and is input in arbiter network, and sets training
Sample label, Lables indicate the data label value after unilateral label is smooth;
Lables=Lable*a
Wherein, Lable indicates the data label value of certain training in arbiter network, and a indicates unilateral label smoothing factor,
Unilateral smoothing factor is set in this example as 0.9, and unilateral label smoothing method is added and makes arbiter classification boundaries smooth, prevents
Arbiter easily finds the boundary distinguished between authentic specimen and false sample completely, reaches Continuous optimization arbiter gradient network
Purpose;
The Vitua limage Mixed design that (4e4) generates authentic specimen and generator network into arbiter network, according to
Arbiter network losses function seeks the current weight of each node in arbiter network by stochastic gradient descent method
Local derviation obtains the gradient value of each node current weight in arbiter network;
(4e5) according to the following formula, the weight in computational discrimination device network after each node updates;
Wherein,Indicate the weight in arbiter network after t-th of node updates, StIt indicates in arbiter network t-th
The current weight of node, ξ indicate learning rate, are set as 0.0001 in this example, Δ StIndicate t-th of node in arbiter network
Current weight gradient value.
(4e6) executes (4e3) using the weight after each node updates as after current weight, and successively alternating iteration training is sentenced
Other device network and generator network obtain trained generation confrontation network model until network reaches nash banlance.
Step 5, classify to detection image.
Image containing target to be sorted is input in trained generation confrontation network model, possible class label is exported
Parameter is chosen the highest classification parameter of classification value parameter and is exported as classification results.
Effect of the invention is described further below with reference to emulation experiment.
1. emulation experiment condition:
The hardware test platform of emulation experiment of the present invention is: CPU intelCorei5-6500, dominant frequency 3.2GHz, interior
Deposit 8GB, GPU NVIDIATITANXp;Software platform is: Ubuntu16.04LTS, 64 bit manipulation systems, python3.5.
2. emulation content analysis of simulation result:
Emulation 1, using method of the invention to 200 in 50000 training samples of mnist hand-written data collection accurate marks
Note training sample and residue 49800 are emulated without label training sample, as a result such as Fig. 2, wherein result figure abscissa
Iteration represents network training number, result figure ordinate testacc representative image nicety of grading.As it is clear from fig. 2 that this hair
It is bright when 200 accurate mark samples are used only, the just image classification accuracy of acquirement 96%.
Emulation 2, using method of the invention to 200 in 50000 training samples of cifar10 data set accurate marks
Training sample and residue 49800 are emulated without label training sample, as a result such as Fig. 3, wherein result figure abscissa
Iteration represents network training number, result figure ordinate testacc representative image nicety of grading.It can be seen from figure 3 that this hair
It is bright when 200 accurate mark samples are used only, the just image classification accuracy of acquirement 86%.
To sum up, 200 tape label data are only used only in 50000 training samples and can obtain preferably by the present invention
Image classification accuracy significantly reduces the training set threshold of pattern classification task, possesses better applicability multi-field.
Claims (8)
1. a kind of based on the semi-supervision image classification method for generating confrontation network, which is characterized in that fight network using generating
Minimax game reaches the nash banlance of network training;Using the subsidiary classification device network in arbiter network to detection mesh
Mark completes classification task, realizes that step includes the following:
(1) it chooses and downloads image classification standard exercise sample set, and carry out normalization operation;
(2) supervised learning parameter is set: i.e. first for the training number of labels of normalization training sample set setting supervised learning;Again
Set the sample size of network training each time;Then supervision is set by the sample size of number of labels and each network training
The flag bit of study;
(3) it builds one and network is fought by the generation that generator network, arbiter network and subsidiary classification device form side by side,
In, generator network is set as 5 layers, and arbiter network is set as 4 layers, and subsidiary classification device is set as 5 layers;
(4) generation confrontation network is trained:
Training sample set after normalization is input to the arbiter network generated in confrontation network by (4a), is extracted in each sample
The feature of initial target image, and all features are formed into characteristics of image figure, 1 is exported by the last layer of arbiter network
Unit vector, for determining the classification of characteristic image;
(4b) receives a noise vector by generator network, and maps it onto a characteristic pattern, then by this characteristic pattern
Arbiter network parameter with true training sample together as the input of arbiter network, after obtaining training for the first time;
(4c) updates generator network parameter using the arbiter network parameter after training for the first time, then fixes updated generation
Device network parameter trains arbiter network again, successively repetitive exercise arbiter network and generator network;
(4d) utilizes generator gradient updating formula, and characteristic matching item is added, and calculates generator net after repetitive exercise each time
Network weight;
(4e) utilizes supervised learning and unsupervised loss function, each classification results value of computational discrimination device network and classification results
Error amount between true value;
(4f) utilizes stochastic gradient descent method, and increases label smooth operation to training sample set, update generator network and sentence
The weight of each node of other device network, until nicety of grading of the arbiter network for authentic specimen and false sample reaches
Nash banlance obtains trained generation confrontation network model;
(5) classify to image to be detected:
Image containing target to be sorted is input in trained generation confrontation network model, the probability of class option is exported
Value, the highest class option of select probability value are exported as classification results.
2. according to the method described in claim 1, wherein for the training of normalization training sample set setting supervised learning in (2)
Number of labels is to control the percentage of the total training data of supervised learning label data Zhan, root by counting total training samples number
The training number of labels of supervised learning is calculated according to percentages.
3. according to the method described in claim 1, wherein being set in (2) by the sample size of number of labels and each network training
The flag bit for determining supervised learning, is accomplished by
One supervised learning symbol position flag is set to distinguish the sample training of supervised learning and unsupervised learning, works as progress
When supervised learning training, setting supervised learning symbol position is effective status 1, the loss of arbiter network query function supervised learning
Function, when carrying out unsupervised learning training, setting supervised learning symbol position is invalid state 0;
The loss function of arbiter network query function unsupervised learning, and pass through the sample of tape label sample size and each network training
The flag bit effective degree of this quantity calculating supervised learning.
4. according to the method described in claim 1, the wherein arbiter network in (3), by 4 layers of convolutional layer and 1 layer of full articulamentum
Composition, each layer parameter are as follows:
Level 1 volume lamination, the sum of Feature Mapping figure are set as 64, and the scale of convolution kernel is set as 5 × 5 nodes, step
A length of 2, activation primitive is lrelu function;
Level 2 volume lamination, the sum of Feature Mapping figure are set as 128, and the scale of convolution kernel is set as 3 × 3 nodes, step
A length of 2, activation primitive is lrelu function;
3rd layer of convolutional layer, the sum of Feature Mapping figure are set as 256, and the scale of convolution kernel is set as 3 × 3 nodes, step
A length of 2, activation primitive is lrelu function;
4th layer of convolutional layer, the sum of Feature Mapping figure are set as 256, and the scale of convolution kernel is set as 3 × 3 nodes, step
A length of 2, activation primitive is lrelu function;
5th layer of full articulamentum, the sum of Feature Mapping figure are set as 11.
5. according to the method described in claim 1, wherein generator network in (3), by 1 layer of full articulamentum and 4 layers of convolutional layer group
At each layer parameter is as follows:
The full articulamentum of first layer, the sum of Feature Mapping figure are set as 256;
Second layer convolutional layer, the sum of Feature Mapping figure are set as 256, and the scale of convolution kernel is set as 5 × 5 nodes,
Step-length is 2, and activation primitive is relu function;
Third layer convolutional layer, the sum of Feature Mapping figure are set as 128, and the scale of convolution kernel is set as 5 × 5 nodes,
Step-length is 2, and activation primitive is relu function;
4th layer of convolutional layer, the sum of Feature Mapping figure are set as 64, and the scale of convolution kernel is set as 5 × 5 nodes, step
A length of 2, activation primitive is relu function;
Layer 5 convolutional layer, the sum of Feature Mapping figure are set as 1, and the scale of convolution kernel is set as 1 × 1 node, step
A length of 1, activation primitive is relu function.
6. structure is identical as arbiter network, is used for according to the method described in claim 1, wherein subsidiary classification device in (3)
The classification task of K class data is extended to K+1 class data sorting task in the classification task of data sample, wherein K+1 class
Data are the Vitua limages that generator network generates.
7. leading to according to the method described in claim 1, wherein calculating generator network weight after repetitive exercise each time in (4d)
Following formula is crossed to carry out:
EZ~P (z)Log (1-D (G (z))),
Wherein, EZ~P (z)Indicate that sample obeys the mathematical expectation of a certain distribution, z indicates to obey the noise vector of a certain distribution, P
(z) indicate that the noise probability density function of a certain distribution of obedience, D (G (z)) indicate that arbiter network generates generator network
The differentiation probability value of image, G (z) indicate the sample image that generator network generates.
8. according to the method described in claim 1, wherein calculating each classification results value of arbiter network and classification knot in (4e)
Error amount between the true value of fruit realizes that steps are as follows:
The loss function value of (4e1) calculating supervised learning:
Wherein, LsIndicate the loss function value of supervised learning,Indicate that sample obeys the mathematical expectation of a certain distribution,
X indicates to obey the authentic specimen vector of a certain distribution, Pdata(x, y) indicates to obey the authentic specimen probability density letter of a certain distribution
Number, the categorization values of y presentation class task, Pmodel(y | x, y < K+1) indicate K+1 class class probability model.
The loss function value of (4e2) calculating unsupervised learning:
Wherein, LunsIndicate the loss function value of unsupervised learning,Indicate that authentic specimen obeys the mathematics of a certain distribution
Desired value, x indicate to obey the authentic specimen vector of a certain distribution, Pmodel(y=K+1 | x) indicate that the probability for obeying a certain distribution is close
Spend model, EZ~noiseIndicate that false sample obeys the mathematical expectation of a certain distribution, z indicates to obey the noise of a certain distribution
Vector;
(4e3) is according to (4e1) and (4e2) as a result, the error amount L of computational discrimination device networkloss:
Lloss=Luns+Ls。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910323702.6A CN110097103A (en) | 2019-04-22 | 2019-04-22 | Based on the semi-supervision image classification method for generating confrontation network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910323702.6A CN110097103A (en) | 2019-04-22 | 2019-04-22 | Based on the semi-supervision image classification method for generating confrontation network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110097103A true CN110097103A (en) | 2019-08-06 |
Family
ID=67445433
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910323702.6A Pending CN110097103A (en) | 2019-04-22 | 2019-04-22 | Based on the semi-supervision image classification method for generating confrontation network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110097103A (en) |
Cited By (70)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110413924A (en) * | 2019-07-18 | 2019-11-05 | 广东石油化工学院 | A kind of Web page classification method of semi-supervised multiple view study |
CN110555474A (en) * | 2019-08-28 | 2019-12-10 | 上海电力大学 | photovoltaic panel fault detection method based on semi-supervised learning |
CN110569842A (en) * | 2019-09-05 | 2019-12-13 | 江苏艾佳家居用品有限公司 | Semi-supervised learning method for GAN model training |
CN110602078A (en) * | 2019-09-04 | 2019-12-20 | 南京邮电大学 | Application encryption traffic generation method and system based on generation countermeasure network |
CN110617966A (en) * | 2019-09-23 | 2019-12-27 | 江南大学 | Bearing fault diagnosis method based on semi-supervised generation countermeasure network |
CN110689086A (en) * | 2019-10-08 | 2020-01-14 | 郑州轻工业学院 | Semi-supervised high-resolution remote sensing image scene classification method based on generating countermeasure network |
CN110823574A (en) * | 2019-09-30 | 2020-02-21 | 安徽富煌科技股份有限公司 | Fault diagnosis method based on semi-supervised learning deep countermeasure network |
CN110827258A (en) * | 2019-10-31 | 2020-02-21 | 西安交通大学 | Screening device and method for diabetic retinopathy based on semi-supervised antagonistic learning |
CN110853703A (en) * | 2019-10-16 | 2020-02-28 | 天津大学 | Semi-supervised learning prediction method for protein secondary structure |
CN110941829A (en) * | 2019-11-27 | 2020-03-31 | 北京电子科技学院 | Large-scale hardware Trojan horse library generation system and method based on generation countermeasure network |
CN110991636A (en) * | 2019-11-14 | 2020-04-10 | 东软医疗系统股份有限公司 | Training method and device of generative confrontation network, image enhancement method and equipment |
CN110993094A (en) * | 2019-11-19 | 2020-04-10 | 中国科学院深圳先进技术研究院 | Intelligent auxiliary diagnosis method and terminal based on medical images |
CN111028941A (en) * | 2019-12-10 | 2020-04-17 | 南京医科大学 | TKA preoperative clinical staging intelligent evaluation method based on generation of antagonistic network |
CN111046900A (en) * | 2019-10-25 | 2020-04-21 | 重庆邮电大学 | Semi-supervised generation confrontation network image classification method based on local manifold regularization |
CN111046673A (en) * | 2019-12-17 | 2020-04-21 | 湖南大学 | Countermeasure generation network for defending text malicious samples and training method thereof |
CN111079893A (en) * | 2019-11-05 | 2020-04-28 | 深圳大学 | Method and device for obtaining generator network for interference fringe pattern filtering |
CN111160415A (en) * | 2019-12-12 | 2020-05-15 | 江苏艾佳家居用品有限公司 | Generative model training method for layer-by-layer inverse recognition model |
CN111178260A (en) * | 2019-12-30 | 2020-05-19 | 山东大学 | Modulation signal time-frequency diagram classification system based on generation countermeasure network and operation method thereof |
CN111191654A (en) * | 2019-12-30 | 2020-05-22 | 重庆紫光华山智安科技有限公司 | Road data generation method and device, electronic equipment and storage medium |
CN111260755A (en) * | 2020-01-19 | 2020-06-09 | 金陵科技学院 | Digital tie-dyeing pattern generation method based on deep learning and digital image processing |
CN111292220A (en) * | 2020-01-19 | 2020-06-16 | 西北工业大学 | Target camouflage image generation method for target image recognition of synthetic aperture radar |
CN111368912A (en) * | 2020-03-03 | 2020-07-03 | 中国福利会国际和平妇幼保健院 | Method and system for forming classifier model |
CN111382686A (en) * | 2020-03-04 | 2020-07-07 | 上海海事大学 | Lane line detection method based on semi-supervised generation confrontation network |
CN111398955A (en) * | 2020-03-13 | 2020-07-10 | 中国科学院电子学研究所苏州研究院 | SAR image sidelobe removing method based on generation of antagonistic neural network |
CN111461264A (en) * | 2020-05-25 | 2020-07-28 | 南京大学 | Scalable modular image recognition method based on generation countermeasure network |
CN111489300A (en) * | 2020-03-11 | 2020-08-04 | 天津大学 | Screen image Moire removing method based on unsupervised learning |
CN111488911A (en) * | 2020-03-15 | 2020-08-04 | 北京理工大学 | Image entity extraction method based on Mask R-CNN and GAN |
CN111522958A (en) * | 2020-05-28 | 2020-08-11 | 泰康保险集团股份有限公司 | Text classification method and device |
CN111582474A (en) * | 2020-04-24 | 2020-08-25 | 中科物栖(北京)科技有限责任公司 | Neural network structure detection method, and training method and device of structure detection model |
CN111612081A (en) * | 2020-05-25 | 2020-09-01 | 深圳前海微众银行股份有限公司 | Recognition model training method, device, equipment and storage medium |
CN111625697A (en) * | 2020-05-12 | 2020-09-04 | 深圳赋乐科技有限公司 | Network data identification filtering method for generating countermeasure network based on condition |
CN111626317A (en) * | 2019-08-14 | 2020-09-04 | 广东省智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
CN111695276A (en) * | 2020-05-11 | 2020-09-22 | 东南大学 | Finite element model proxy modeling method based on generation countermeasure learning |
CN111709318A (en) * | 2020-05-28 | 2020-09-25 | 西安理工大学 | High-resolution remote sensing image classification method based on generation countermeasure network |
CN111709470A (en) * | 2020-06-08 | 2020-09-25 | 北京百度网讯科技有限公司 | Image generation method, apparatus, device and medium |
CN111832650A (en) * | 2020-07-14 | 2020-10-27 | 西安电子科技大学 | Image classification method based on generation of confrontation network local aggregation coding semi-supervision |
CN111832516A (en) * | 2020-07-22 | 2020-10-27 | 西安电子科技大学 | Video behavior identification method based on unsupervised video representation learning |
CN111860593A (en) * | 2020-06-15 | 2020-10-30 | 北京华电天仁电力控制技术有限公司 | Fan blade fault detection method based on generation countermeasure network |
CN112052893A (en) * | 2020-09-01 | 2020-12-08 | 西安电子科技大学 | Semi-supervised image classification method based on generation countermeasure network |
CN112100908A (en) * | 2020-08-31 | 2020-12-18 | 西安工程大学 | Garment design method for generating confrontation network based on multi-condition deep convolution |
CN112101404A (en) * | 2020-07-24 | 2020-12-18 | 西安电子科技大学 | Image classification method and system based on generation countermeasure network and electronic equipment |
CN112199887A (en) * | 2020-09-30 | 2021-01-08 | 山东大学 | Semi-supervised resistivity inversion method and system based on countermeasure generation network and pseudo-labeling |
CN112381148A (en) * | 2020-11-17 | 2021-02-19 | 华南理工大学 | Semi-supervised image classification method based on random regional interpolation |
CN112381116A (en) * | 2020-10-21 | 2021-02-19 | 福州大学 | Self-supervision image classification method based on contrast learning |
CN112434716A (en) * | 2019-08-21 | 2021-03-02 | 中国科学院声学研究所 | Underwater target data amplification method and system based on conditional adversarial neural network |
CN112488209A (en) * | 2020-11-25 | 2021-03-12 | 南京大学 | Incremental image classification method based on semi-supervised learning |
CN112613494A (en) * | 2020-11-19 | 2021-04-06 | 北京国网富达科技发展有限责任公司 | Power line monitoring abnormity identification method and system based on deep countermeasure network |
CN112668655A (en) * | 2020-12-30 | 2021-04-16 | 中山大学 | Method for detecting out-of-distribution image based on generation of confrontation network uncertainty attention enhancement |
CN112766381A (en) * | 2021-01-22 | 2021-05-07 | 西安电子科技大学 | Attribute-guided SAR image generation method under limited sample |
CN112784930A (en) * | 2021-03-17 | 2021-05-11 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
CN112865866A (en) * | 2021-01-20 | 2021-05-28 | 重庆邮电大学 | Visible light PAM system nonlinear compensation method based on GSN |
CN112906811A (en) * | 2021-03-09 | 2021-06-04 | 西安电子科技大学 | Automatic classification method for images of engineering vehicle-mounted equipment based on Internet of things architecture |
CN112966740A (en) * | 2021-03-05 | 2021-06-15 | 西安邮电大学 | Small sample hyperspectral image classification method based on core sample adaptive expansion |
CN112990342A (en) * | 2021-04-08 | 2021-06-18 | 重庆大学 | Semi-supervised SAR target recognition method |
CN113128620A (en) * | 2021-05-11 | 2021-07-16 | 北京理工大学 | Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship |
CN113221948A (en) * | 2021-04-13 | 2021-08-06 | 复旦大学 | Digital slice image classification method based on countermeasure generation network and weak supervised learning |
CN113255789A (en) * | 2021-05-31 | 2021-08-13 | 西安电子科技大学 | Video quality evaluation method based on confrontation network and multi-tested electroencephalogram signals |
CN113366494A (en) * | 2019-01-29 | 2021-09-07 | 辉达公司 | Method for few-sample unsupervised image-to-image conversion |
CN113361801A (en) * | 2021-06-28 | 2021-09-07 | 嘉兴国电通新能源科技有限公司 | Hybrid rolling short-term load prediction method based on GAN network and semi-supervised regression |
CN113706380A (en) * | 2021-08-20 | 2021-11-26 | 西安交通大学 | Method and system for generating countermeasure network based on weak supervision to improve seismic data resolution |
CN113884290A (en) * | 2021-09-28 | 2022-01-04 | 江南大学 | Voltage regulator fault diagnosis method based on self-training semi-supervised generation countermeasure network |
CN114024721A (en) * | 2021-10-13 | 2022-02-08 | 国网浙江省电力有限公司宁波供电公司 | Traffic classification identification method based on transmission channel quality sequencing |
CN114201632A (en) * | 2022-02-18 | 2022-03-18 | 南京航空航天大学 | Label noisy data set amplification method for multi-label target detection task |
CN114299348A (en) * | 2022-02-21 | 2022-04-08 | 山东力聚机器人科技股份有限公司 | Image classification method and device based on restoration self-supervision task |
CN114462558A (en) * | 2022-04-13 | 2022-05-10 | 南昌工程学院 | Data-augmented supervised learning image defect classification method and system |
CN114863225A (en) * | 2022-07-06 | 2022-08-05 | 腾讯科技(深圳)有限公司 | Image processing model training method, image processing model generation device, image processing equipment and image processing medium |
CN114898159A (en) * | 2022-06-01 | 2022-08-12 | 西北工业大学 | SAR image interpretability feature extraction method for generating confrontation network based on decoupling representation |
CN115127192A (en) * | 2022-05-20 | 2022-09-30 | 中南大学 | Semi-supervised water chilling unit fault diagnosis method and system based on graph neural network |
CN115292728A (en) * | 2022-07-15 | 2022-11-04 | 浙江大学 | Image data privacy protection method based on generation countermeasure network |
TWI800767B (en) * | 2020-11-16 | 2023-05-01 | 財團法人工業技術研究院 | Anomaly detection apparatus and anomaly detection method based on generative adversarial networks |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102096825A (en) * | 2011-03-23 | 2011-06-15 | 西安电子科技大学 | Graph-based semi-supervised high-spectral remote sensing image classification method |
US8837819B1 (en) * | 2012-04-05 | 2014-09-16 | Google Inc. | Systems and methods for facilitating identification of and interaction with objects in a video or image frame |
CN107563428A (en) * | 2017-08-25 | 2018-01-09 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on generation confrontation network |
CN108171700A (en) * | 2018-01-12 | 2018-06-15 | 西安电子科技大学 | Medical image pulmonary nodule detection method based on confrontation network |
CN108470187A (en) * | 2018-02-26 | 2018-08-31 | 华南理工大学 | A kind of class imbalance question classification method based on expansion training dataset |
CN108492298A (en) * | 2018-04-13 | 2018-09-04 | 西安电子科技大学 | Based on the multispectral image change detecting method for generating confrontation network |
CN108520282A (en) * | 2018-04-13 | 2018-09-11 | 湘潭大学 | A kind of sorting technique based on Triple-GAN |
CN108564115A (en) * | 2018-03-30 | 2018-09-21 | 西安电子科技大学 | Semi-supervised polarization SAR terrain classification method based on full convolution GAN |
CN108805188A (en) * | 2018-05-29 | 2018-11-13 | 徐州工程学院 | A kind of feature based recalibration generates the image classification method of confrontation network |
CN108875818A (en) * | 2018-06-06 | 2018-11-23 | 西安交通大学 | Based on variation from code machine and confrontation network integration zero sample image classification method |
WO2018225061A1 (en) * | 2017-06-04 | 2018-12-13 | De-Identification Ltd. | System and method for image de-identification |
CN109190665A (en) * | 2018-07-30 | 2019-01-11 | 国网上海市电力公司 | A kind of general image classification method and device based on semi-supervised generation confrontation network |
-
2019
- 2019-04-22 CN CN201910323702.6A patent/CN110097103A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102096825A (en) * | 2011-03-23 | 2011-06-15 | 西安电子科技大学 | Graph-based semi-supervised high-spectral remote sensing image classification method |
US8837819B1 (en) * | 2012-04-05 | 2014-09-16 | Google Inc. | Systems and methods for facilitating identification of and interaction with objects in a video or image frame |
WO2018225061A1 (en) * | 2017-06-04 | 2018-12-13 | De-Identification Ltd. | System and method for image de-identification |
CN107563428A (en) * | 2017-08-25 | 2018-01-09 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on generation confrontation network |
CN108171700A (en) * | 2018-01-12 | 2018-06-15 | 西安电子科技大学 | Medical image pulmonary nodule detection method based on confrontation network |
CN108470187A (en) * | 2018-02-26 | 2018-08-31 | 华南理工大学 | A kind of class imbalance question classification method based on expansion training dataset |
CN108564115A (en) * | 2018-03-30 | 2018-09-21 | 西安电子科技大学 | Semi-supervised polarization SAR terrain classification method based on full convolution GAN |
CN108492298A (en) * | 2018-04-13 | 2018-09-04 | 西安电子科技大学 | Based on the multispectral image change detecting method for generating confrontation network |
CN108520282A (en) * | 2018-04-13 | 2018-09-11 | 湘潭大学 | A kind of sorting technique based on Triple-GAN |
CN108805188A (en) * | 2018-05-29 | 2018-11-13 | 徐州工程学院 | A kind of feature based recalibration generates the image classification method of confrontation network |
CN108875818A (en) * | 2018-06-06 | 2018-11-23 | 西安交通大学 | Based on variation from code machine and confrontation network integration zero sample image classification method |
CN109190665A (en) * | 2018-07-30 | 2019-01-11 | 国网上海市电力公司 | A kind of general image classification method and device based on semi-supervised generation confrontation network |
Cited By (116)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113366494A (en) * | 2019-01-29 | 2021-09-07 | 辉达公司 | Method for few-sample unsupervised image-to-image conversion |
CN110413924A (en) * | 2019-07-18 | 2019-11-05 | 广东石油化工学院 | A kind of Web page classification method of semi-supervised multiple view study |
CN111626317B (en) * | 2019-08-14 | 2022-01-07 | 广东省科学院智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
CN111626317A (en) * | 2019-08-14 | 2020-09-04 | 广东省智能制造研究所 | Semi-supervised hyperspectral data analysis method based on double-flow conditional countermeasure generation network |
CN112434716B (en) * | 2019-08-21 | 2024-03-29 | 中国科学院声学研究所 | Underwater target data amplification method and system based on condition countermeasure neural network |
CN112434716A (en) * | 2019-08-21 | 2021-03-02 | 中国科学院声学研究所 | Underwater target data amplification method and system based on conditional adversarial neural network |
CN110555474B (en) * | 2019-08-28 | 2023-06-02 | 上海电力大学 | Photovoltaic panel fault detection method based on semi-supervised learning |
CN110555474A (en) * | 2019-08-28 | 2019-12-10 | 上海电力大学 | photovoltaic panel fault detection method based on semi-supervised learning |
CN110602078A (en) * | 2019-09-04 | 2019-12-20 | 南京邮电大学 | Application encryption traffic generation method and system based on generation countermeasure network |
CN110602078B (en) * | 2019-09-04 | 2022-12-20 | 南京邮电大学 | Application encryption traffic generation method and system based on generation countermeasure network |
CN110569842A (en) * | 2019-09-05 | 2019-12-13 | 江苏艾佳家居用品有限公司 | Semi-supervised learning method for GAN model training |
CN110569842B (en) * | 2019-09-05 | 2022-08-12 | 江苏艾佳家居用品有限公司 | Semi-supervised learning method for GAN model training |
CN110617966A (en) * | 2019-09-23 | 2019-12-27 | 江南大学 | Bearing fault diagnosis method based on semi-supervised generation countermeasure network |
CN110823574A (en) * | 2019-09-30 | 2020-02-21 | 安徽富煌科技股份有限公司 | Fault diagnosis method based on semi-supervised learning deep countermeasure network |
CN110689086B (en) * | 2019-10-08 | 2020-09-25 | 郑州轻工业学院 | Semi-supervised high-resolution remote sensing image scene classification method based on generating countermeasure network |
CN110689086A (en) * | 2019-10-08 | 2020-01-14 | 郑州轻工业学院 | Semi-supervised high-resolution remote sensing image scene classification method based on generating countermeasure network |
CN110853703A (en) * | 2019-10-16 | 2020-02-28 | 天津大学 | Semi-supervised learning prediction method for protein secondary structure |
CN111046900A (en) * | 2019-10-25 | 2020-04-21 | 重庆邮电大学 | Semi-supervised generation confrontation network image classification method based on local manifold regularization |
CN111046900B (en) * | 2019-10-25 | 2022-10-18 | 重庆邮电大学 | Semi-supervised generation confrontation network image classification method based on local manifold regularization |
CN110827258B (en) * | 2019-10-31 | 2022-02-11 | 西安交通大学 | Device and method for screening diabetic retinopathy based on counterstudy |
CN110827258A (en) * | 2019-10-31 | 2020-02-21 | 西安交通大学 | Screening device and method for diabetic retinopathy based on semi-supervised antagonistic learning |
CN111079893A (en) * | 2019-11-05 | 2020-04-28 | 深圳大学 | Method and device for obtaining generator network for interference fringe pattern filtering |
CN111079893B (en) * | 2019-11-05 | 2023-05-09 | 深圳大学 | Acquisition method and device for generator network for interference fringe pattern filtering |
CN110991636A (en) * | 2019-11-14 | 2020-04-10 | 东软医疗系统股份有限公司 | Training method and device of generative confrontation network, image enhancement method and equipment |
CN110991636B (en) * | 2019-11-14 | 2023-07-04 | 东软医疗系统股份有限公司 | Training method and device of generated countermeasure network, image enhancement method and device |
CN110993094A (en) * | 2019-11-19 | 2020-04-10 | 中国科学院深圳先进技术研究院 | Intelligent auxiliary diagnosis method and terminal based on medical images |
CN110941829B (en) * | 2019-11-27 | 2023-03-10 | 北京电子科技学院 | Large-scale hardware Trojan horse library generation system and method based on generation countermeasure network |
CN110941829A (en) * | 2019-11-27 | 2020-03-31 | 北京电子科技学院 | Large-scale hardware Trojan horse library generation system and method based on generation countermeasure network |
CN111028941A (en) * | 2019-12-10 | 2020-04-17 | 南京医科大学 | TKA preoperative clinical staging intelligent evaluation method based on generation of antagonistic network |
CN111160415B (en) * | 2019-12-12 | 2022-08-12 | 江苏艾佳家居用品有限公司 | Generative model training method for layer-by-layer inverse recognition model |
CN111160415A (en) * | 2019-12-12 | 2020-05-15 | 江苏艾佳家居用品有限公司 | Generative model training method for layer-by-layer inverse recognition model |
CN111046673A (en) * | 2019-12-17 | 2020-04-21 | 湖南大学 | Countermeasure generation network for defending text malicious samples and training method thereof |
CN111191654B (en) * | 2019-12-30 | 2023-03-24 | 重庆紫光华山智安科技有限公司 | Road data generation method and device, electronic equipment and storage medium |
CN111191654A (en) * | 2019-12-30 | 2020-05-22 | 重庆紫光华山智安科技有限公司 | Road data generation method and device, electronic equipment and storage medium |
CN111178260A (en) * | 2019-12-30 | 2020-05-19 | 山东大学 | Modulation signal time-frequency diagram classification system based on generation countermeasure network and operation method thereof |
CN111178260B (en) * | 2019-12-30 | 2023-04-07 | 山东大学 | Modulation signal time-frequency diagram classification system based on generation countermeasure network and operation method thereof |
CN111260755A (en) * | 2020-01-19 | 2020-06-09 | 金陵科技学院 | Digital tie-dyeing pattern generation method based on deep learning and digital image processing |
CN111292220B (en) * | 2020-01-19 | 2021-09-28 | 西北工业大学 | Target camouflage image generation method for target image recognition of synthetic aperture radar |
CN111292220A (en) * | 2020-01-19 | 2020-06-16 | 西北工业大学 | Target camouflage image generation method for target image recognition of synthetic aperture radar |
CN111368912A (en) * | 2020-03-03 | 2020-07-03 | 中国福利会国际和平妇幼保健院 | Method and system for forming classifier model |
CN111368912B (en) * | 2020-03-03 | 2023-12-19 | 中国福利会国际和平妇幼保健院 | Forming method and system of classifier model |
CN111382686A (en) * | 2020-03-04 | 2020-07-07 | 上海海事大学 | Lane line detection method based on semi-supervised generation confrontation network |
CN111382686B (en) * | 2020-03-04 | 2023-03-24 | 上海海事大学 | Lane line detection method based on semi-supervised generation confrontation network |
CN111489300B (en) * | 2020-03-11 | 2022-07-08 | 天津大学 | Screen image Moire removing method based on unsupervised learning |
CN111489300A (en) * | 2020-03-11 | 2020-08-04 | 天津大学 | Screen image Moire removing method based on unsupervised learning |
CN111398955A (en) * | 2020-03-13 | 2020-07-10 | 中国科学院电子学研究所苏州研究院 | SAR image sidelobe removing method based on generation of antagonistic neural network |
CN111488911A (en) * | 2020-03-15 | 2020-08-04 | 北京理工大学 | Image entity extraction method based on Mask R-CNN and GAN |
CN111582474A (en) * | 2020-04-24 | 2020-08-25 | 中科物栖(北京)科技有限责任公司 | Neural network structure detection method, and training method and device of structure detection model |
CN111582474B (en) * | 2020-04-24 | 2023-08-25 | 中科物栖(南京)科技有限公司 | Neural network structure detection method, training method and training device of structure detection model |
CN111695276A (en) * | 2020-05-11 | 2020-09-22 | 东南大学 | Finite element model proxy modeling method based on generation countermeasure learning |
CN111625697A (en) * | 2020-05-12 | 2020-09-04 | 深圳赋乐科技有限公司 | Network data identification filtering method for generating countermeasure network based on condition |
CN111612081A (en) * | 2020-05-25 | 2020-09-01 | 深圳前海微众银行股份有限公司 | Recognition model training method, device, equipment and storage medium |
CN111461264A (en) * | 2020-05-25 | 2020-07-28 | 南京大学 | Scalable modular image recognition method based on generation countermeasure network |
CN111612081B (en) * | 2020-05-25 | 2024-04-02 | 深圳前海微众银行股份有限公司 | Training method, device, equipment and storage medium for recognition model |
CN111522958A (en) * | 2020-05-28 | 2020-08-11 | 泰康保险集团股份有限公司 | Text classification method and device |
CN111709318B (en) * | 2020-05-28 | 2023-03-24 | 西安理工大学 | High-resolution remote sensing image classification method based on generation countermeasure network |
CN111709318A (en) * | 2020-05-28 | 2020-09-25 | 西安理工大学 | High-resolution remote sensing image classification method based on generation countermeasure network |
CN111709470B (en) * | 2020-06-08 | 2023-10-03 | 北京百度网讯科技有限公司 | Image generation method, device, equipment and medium |
CN111709470A (en) * | 2020-06-08 | 2020-09-25 | 北京百度网讯科技有限公司 | Image generation method, apparatus, device and medium |
CN111860593A (en) * | 2020-06-15 | 2020-10-30 | 北京华电天仁电力控制技术有限公司 | Fan blade fault detection method based on generation countermeasure network |
CN111832650B (en) * | 2020-07-14 | 2023-08-01 | 西安电子科技大学 | Image classification method based on generation of antagonism network local aggregation coding semi-supervision |
CN111832650A (en) * | 2020-07-14 | 2020-10-27 | 西安电子科技大学 | Image classification method based on generation of confrontation network local aggregation coding semi-supervision |
CN111832516A (en) * | 2020-07-22 | 2020-10-27 | 西安电子科技大学 | Video behavior identification method based on unsupervised video representation learning |
CN111832516B (en) * | 2020-07-22 | 2023-08-18 | 西安电子科技大学 | Video behavior recognition method based on unsupervised video representation learning |
CN112101404A (en) * | 2020-07-24 | 2020-12-18 | 西安电子科技大学 | Image classification method and system based on generation countermeasure network and electronic equipment |
CN112101404B (en) * | 2020-07-24 | 2024-02-09 | 西安电子科技大学 | Image classification method and system based on generation countermeasure network and electronic equipment |
CN112100908A (en) * | 2020-08-31 | 2020-12-18 | 西安工程大学 | Garment design method for generating confrontation network based on multi-condition deep convolution |
CN112100908B (en) * | 2020-08-31 | 2024-03-22 | 西安工程大学 | Clothing design method for generating countermeasure network based on multi-condition deep convolution |
CN112052893A (en) * | 2020-09-01 | 2020-12-08 | 西安电子科技大学 | Semi-supervised image classification method based on generation countermeasure network |
CN112199887A (en) * | 2020-09-30 | 2021-01-08 | 山东大学 | Semi-supervised resistivity inversion method and system based on countermeasure generation network and pseudo-labeling |
CN112381116B (en) * | 2020-10-21 | 2022-10-28 | 福州大学 | Self-supervision image classification method based on contrast learning |
CN112381116A (en) * | 2020-10-21 | 2021-02-19 | 福州大学 | Self-supervision image classification method based on contrast learning |
TWI800767B (en) * | 2020-11-16 | 2023-05-01 | 財團法人工業技術研究院 | Anomaly detection apparatus and anomaly detection method based on generative adversarial networks |
CN112381148A (en) * | 2020-11-17 | 2021-02-19 | 华南理工大学 | Semi-supervised image classification method based on random regional interpolation |
CN112381148B (en) * | 2020-11-17 | 2022-06-14 | 华南理工大学 | Semi-supervised image classification method based on random regional interpolation |
CN112613494A (en) * | 2020-11-19 | 2021-04-06 | 北京国网富达科技发展有限责任公司 | Power line monitoring abnormity identification method and system based on deep countermeasure network |
CN112488209B (en) * | 2020-11-25 | 2024-02-20 | 南京大学 | Incremental picture classification method based on semi-supervised learning |
CN112488209A (en) * | 2020-11-25 | 2021-03-12 | 南京大学 | Incremental image classification method based on semi-supervised learning |
CN112668655B (en) * | 2020-12-30 | 2023-08-29 | 中山大学 | Out-of-distribution image detection method based on generating attention enhancement against network uncertainty |
CN112668655A (en) * | 2020-12-30 | 2021-04-16 | 中山大学 | Method for detecting out-of-distribution image based on generation of confrontation network uncertainty attention enhancement |
CN112865866B (en) * | 2021-01-20 | 2022-04-05 | 重庆邮电大学 | Visible light PAM system nonlinear compensation method based on GSN |
CN112865866A (en) * | 2021-01-20 | 2021-05-28 | 重庆邮电大学 | Visible light PAM system nonlinear compensation method based on GSN |
CN112766381B (en) * | 2021-01-22 | 2023-01-24 | 西安电子科技大学 | Attribute-guided SAR image generation method under limited sample |
CN112766381A (en) * | 2021-01-22 | 2021-05-07 | 西安电子科技大学 | Attribute-guided SAR image generation method under limited sample |
CN112966740A (en) * | 2021-03-05 | 2021-06-15 | 西安邮电大学 | Small sample hyperspectral image classification method based on core sample adaptive expansion |
CN112966740B (en) * | 2021-03-05 | 2024-03-29 | 西安邮电大学 | Small sample hyperspectral image classification method based on core sample self-adaptive expansion |
CN112906811B (en) * | 2021-03-09 | 2023-04-18 | 西安电子科技大学 | Automatic classification method for images of engineering vehicle-mounted equipment based on Internet of things architecture |
CN112906811A (en) * | 2021-03-09 | 2021-06-04 | 西安电子科技大学 | Automatic classification method for images of engineering vehicle-mounted equipment based on Internet of things architecture |
CN112784930A (en) * | 2021-03-17 | 2021-05-11 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
CN112784930B (en) * | 2021-03-17 | 2022-03-04 | 西安电子科技大学 | CACGAN-based HRRP identification database sample expansion method |
CN112990342B (en) * | 2021-04-08 | 2023-09-19 | 重庆大学 | Semi-supervised SAR target recognition method |
CN112990342A (en) * | 2021-04-08 | 2021-06-18 | 重庆大学 | Semi-supervised SAR target recognition method |
CN113221948A (en) * | 2021-04-13 | 2021-08-06 | 复旦大学 | Digital slice image classification method based on countermeasure generation network and weak supervised learning |
CN113221948B (en) * | 2021-04-13 | 2022-08-05 | 复旦大学 | Digital slice image classification method based on countermeasure generation network and weak supervised learning |
CN113128620A (en) * | 2021-05-11 | 2021-07-16 | 北京理工大学 | Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship |
CN113128620B (en) * | 2021-05-11 | 2022-10-21 | 北京理工大学 | Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship |
CN113255789A (en) * | 2021-05-31 | 2021-08-13 | 西安电子科技大学 | Video quality evaluation method based on confrontation network and multi-tested electroencephalogram signals |
CN113255789B (en) * | 2021-05-31 | 2023-01-24 | 西安电子科技大学 | Video quality evaluation method based on confrontation network and multi-tested electroencephalogram signals |
CN113361801A (en) * | 2021-06-28 | 2021-09-07 | 嘉兴国电通新能源科技有限公司 | Hybrid rolling short-term load prediction method based on GAN network and semi-supervised regression |
CN113706380B (en) * | 2021-08-20 | 2022-12-09 | 西安交通大学 | Method and system for improving seismic data resolution based on weak supervision generation countermeasure network |
CN113706380A (en) * | 2021-08-20 | 2021-11-26 | 西安交通大学 | Method and system for generating countermeasure network based on weak supervision to improve seismic data resolution |
CN113884290A (en) * | 2021-09-28 | 2022-01-04 | 江南大学 | Voltage regulator fault diagnosis method based on self-training semi-supervised generation countermeasure network |
CN113884290B (en) * | 2021-09-28 | 2022-08-02 | 江南大学 | Voltage regulator fault diagnosis method based on self-training semi-supervised generation countermeasure network |
CN114024721A (en) * | 2021-10-13 | 2022-02-08 | 国网浙江省电力有限公司宁波供电公司 | Traffic classification identification method based on transmission channel quality sequencing |
CN114201632B (en) * | 2022-02-18 | 2022-05-06 | 南京航空航天大学 | Label noisy data set amplification method for multi-label target detection task |
CN114201632A (en) * | 2022-02-18 | 2022-03-18 | 南京航空航天大学 | Label noisy data set amplification method for multi-label target detection task |
CN114299348A (en) * | 2022-02-21 | 2022-04-08 | 山东力聚机器人科技股份有限公司 | Image classification method and device based on restoration self-supervision task |
CN114462558A (en) * | 2022-04-13 | 2022-05-10 | 南昌工程学院 | Data-augmented supervised learning image defect classification method and system |
CN115127192B (en) * | 2022-05-20 | 2024-01-23 | 中南大学 | Semi-supervised water chilling unit fault diagnosis method and system based on graph neural network |
CN115127192A (en) * | 2022-05-20 | 2022-09-30 | 中南大学 | Semi-supervised water chilling unit fault diagnosis method and system based on graph neural network |
CN114898159A (en) * | 2022-06-01 | 2022-08-12 | 西北工业大学 | SAR image interpretability feature extraction method for generating confrontation network based on decoupling representation |
CN114898159B (en) * | 2022-06-01 | 2024-03-08 | 西北工业大学 | SAR image interpretability feature extraction method for generating countermeasure network based on decoupling characterization |
CN114863225A (en) * | 2022-07-06 | 2022-08-05 | 腾讯科技(深圳)有限公司 | Image processing model training method, image processing model generation device, image processing equipment and image processing medium |
CN114863225B (en) * | 2022-07-06 | 2022-10-04 | 腾讯科技(深圳)有限公司 | Image processing model training method, image processing model generation device, image processing model equipment and image processing model medium |
CN115292728B (en) * | 2022-07-15 | 2023-08-04 | 浙江大学 | Image data privacy protection method based on generation countermeasure network |
CN115292728A (en) * | 2022-07-15 | 2022-11-04 | 浙江大学 | Image data privacy protection method based on generation countermeasure network |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110097103A (en) | Based on the semi-supervision image classification method for generating confrontation network | |
CN107563428B (en) | Based on the Classification of Polarimetric SAR Image method for generating confrontation network | |
CN110210486A (en) | A kind of generation confrontation transfer learning method based on sketch markup information | |
CN109815801A (en) | Face identification method and device based on deep learning | |
CN110532859A (en) | Remote Sensing Target detection method based on depth evolution beta pruning convolution net | |
CN110334765A (en) | Remote Image Classification based on the multiple dimensioned deep learning of attention mechanism | |
CN109118564A (en) | A kind of three-dimensional point cloud labeling method and device based on fusion voxel | |
CN110163127A (en) | A kind of video object Activity recognition method from thick to thin | |
CN106909924A (en) | A kind of remote sensing image method for quickly retrieving based on depth conspicuousness | |
CN104376326B (en) | A kind of feature extracting method for image scene identification | |
CN108710863A (en) | Unmanned plane Scene Semantics dividing method based on deep learning and system | |
CN108109160A (en) | It is a kind of that interactive GrabCut tongue bodies dividing method is exempted from based on deep learning | |
CN107239514A (en) | A kind of plants identification method and system based on convolutional neural networks | |
CN106156793A (en) | Extract in conjunction with further feature and the classification method of medical image of shallow-layer feature extraction | |
CN108564115A (en) | Semi-supervised polarization SAR terrain classification method based on full convolution GAN | |
CN107492095A (en) | Medical image pulmonary nodule detection method based on deep learning | |
CN106651915B (en) | The method for tracking target of multi-scale expression based on convolutional neural networks | |
CN106600595A (en) | Human body characteristic dimension automatic measuring method based on artificial intelligence algorithm | |
CN105760900B (en) | Hyperspectral image classification method based on neighbour's propagation clustering and sparse Multiple Kernel Learning | |
CN104239902B (en) | Hyperspectral image classification method based on non local similitude and sparse coding | |
CN106611423B (en) | SAR image segmentation method based on ridge ripple filter and deconvolution structural model | |
CN107644415A (en) | A kind of text image method for evaluating quality and equipment | |
CN106682696A (en) | Multi-example detection network based on refining of online example classifier and training method thereof | |
CN108416353A (en) | Crop field spike of rice fast partition method based on the full convolutional neural networks of depth | |
CN110490227A (en) | A kind of few sample image classification method based on Feature Conversion |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190806 |