CN115393666A - Small sample expansion method and system based on prototype completion in image classification - Google Patents
Small sample expansion method and system based on prototype completion in image classification Download PDFInfo
- Publication number
- CN115393666A CN115393666A CN202210923952.5A CN202210923952A CN115393666A CN 115393666 A CN115393666 A CN 115393666A CN 202210923952 A CN202210923952 A CN 202210923952A CN 115393666 A CN115393666 A CN 115393666A
- Authority
- CN
- China
- Prior art keywords
- data
- features
- prototype
- support
- base class
- 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
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
Abstract
The invention provides a small sample expansion method and a small sample expansion system based on prototype completion in image classification, which relate to the technical field of sample processing in image classification.
Description
Technical Field
The invention relates to the technical field of sample processing in image classification, in particular to a small sample expansion method and system based on prototype completion in image classification.
Background
The image classification problem is a problem that after an image is input, a description of the image content is output in a classification manner. The method is characterized in that different types of images are distinguished according to semantic information of the images, is an important basic problem in computer vision, and is also the basis of other high-level vision tasks such as image detection, image segmentation, object tracking, behavior analysis and the like. The image classification is applied to various fields, including face recognition, intelligent video analysis and the like in the security field, traffic scene recognition in the traffic field, content-based image retrieval and automatic album classification in the internet field, image recognition in the medical field and the like.
In the past, image classification was based on artificial features or simple machine learning methods, which have the disadvantages of insufficient accuracy, a large number of samples being required, and the need for partial manual design. With the development of deep learning, means for image classification make a stepwise progress, but there is no improvement in the demand for the number of samples, and the deep learning method still requires a huge number of samples to meet the training requirements, because the performance is better and the classification accuracy is high in the case that the number of samples is sufficient, but in the case that the number of samples is rare, overfitting is caused and the classification accuracy is sharply reduced. However, in many scenarios, the acquisition and labeling of samples need to consume excessive resources, and sufficient training samples often include a certain amount of noise interference.
Under the setting of small samples, many methods are to train a deep network by using small samples, for example, as a construction method of an image classification model and an image classification method are disclosed in the prior art, a neural network is firstly established, then a limited image data set is used for training the neural network to form the image classification model, and finally the image classification model is used for completing an image classification task.
Disclosure of Invention
In order to solve the problems that excessive network parameters are easily introduced and universality is poor in the conventional method for classifying images by training a neural network by using small samples, the invention provides a small sample expansion method and system based on prototype completion in image classification.
In order to achieve the technical effects, the technical scheme of the invention is as follows:
a method for prototype-completion-based small sample augmentation in image classification, the method comprising the steps of:
s1, collecting an original sample image data set, and dividing the original sample image data set into a base class data set, a support set and a query set according to the image data type in the original sample image data set;
s2, selecting a feature extractor, training the feature extractor by using a base class data set, and respectively extracting data features of the base class data set, data features of a support set and data features of a query set by using the trained feature extractor;
s3, calculating the distance between the data features of the support set and the data features of all base class data sets based on a spatial metric distance function, and determining K samples which are most similar to the data features of the support set according to the distance;
s4, carrying out weighted calculation on the similarity of the K samples and the data features of the support set to obtain prototype features of the support set;
s5, generating a pseudo sample set based on the prototype features of the support set;
and S6, using the generated data characteristics of the pseudo sample set and the support set as training data, using the data characteristics of the query set as test data, and performing image classification.
In the technical scheme, an original sample image data set is collected firstly, the original sample image data set is divided into a base class data set, a support set and a query set, the features of three types of data sets are extracted respectively based on a feature extractor, samples are mapped to a feature space from the space, prototype feature completion is carried out in the feature space based on the similarity degree of the sample features to obtain prototype features, a small number of samples can be subjected to prototype completion and generation of a pseudo sample set only by a feature library and a feature extractor of the base class data set as a whole, the defect of small sample image classification is greatly overcome, extra parameters needing training are not introduced, the time and equipment requirements of subsequent image classification are saved, and compared with a 'black box' neural network model, the interpretability is extremely strong.
Preferably, in step S1, assuming that the image data categories in the original image data set are Z-class, the base class data set belongs to X-class, and the support set and the query set belong to Y-class, the following are satisfied:
X+Y＝Z，
preferably, in step S2, the feature extractor is selected as a Wide ResNet network, a loss function of the Wide ResNet network is set in the process of training the Wide ResNet network by using the base class data set, a weight parameter of the Wide ResNet network is updated in a back propagation manner until the loss function converges, the network parameter is fixed after the training is completed, and the trained feature extractor extracts the data feature of the set target task.
Preferably, taking the mean value of the ith class data feature of the base class data set as a general term of the class data feature, the following requirements are met:
wherein, mu 1i Mean, x, representing the i-th class data feature of the base class dataset j Representing the jth data characteristic in the ith data characteristic of the base class data set; n is i Representing the number of ith class data features of the base class data set;
let the mean of the data features of the support data set be μ 2 In μ 2 As a general term for supporting data characteristics of a data set, let a space metric distance function be collectively referred to as f d () Then based on space measurement distance function, calculating the data characteristics of support set and the number of all base class data setsAccording to the characteristic distance, the following requirements are met:
d i ＝f d (μ i ,x s )
wherein d is i Representing the distance between the data characteristics of the support set and the ith type data characteristics of the base class data set; finally, a set of distances D is obtained, denoted as:
D＝{d 1 ,d 2 ,...,d i ,...,d q }
and q represents the number of the types of the data features of the base class data set, the values of all the distance elements in the distance set D are arranged from small to large, and the data features of the base class data set corresponding to the first K distance elements are selected according to the arrangement sequence and serve as the K samples with the most similar data features of the support set.
Preferably, after the distances between the data features of the support set and the data features of all base class data sets are obtained, the distances are normalized, and the expression is as follows:
or
Wherein, d' i Represents the value after the distance normalization and the distance normalization,
the maximum value of the distances is represented,
representing the minimum in distance.
Preferably, in step S4, the similarity between the K samples and the support set data feature, i.e. the mean μ between the K samples and the support set data feature 2 By using a spatial metric distance function f d () And solving, wherein on the premise of distance normalization, an expression for performing weighting calculation is as follows:
wherein e is a natural constant, w s The weight for supporting the characteristics of the set data is preset according to the number of the samples in the supporting set, w' g Represents the weight, d ', of the g-th sample of the K samples' g Representing the distance of the g sample of the K samples from the support set data features;
the expression of the prototype features is:
wherein n is K Representing the sum of the number of similar samples and the number of support set data features, and mu' represents a prototype feature;
the covariance matrix C' corresponding to the prototype feature is:
wherein, C g And the covariance matrix represents the characteristic covariance matrix of the g-th similar sample data in the K samples, and alpha is a hyperparameter.
Preferably, in step S5, the generated set of pseudo samples is D y The expression is:
where y represents the newly generated category,
representing newly generated pseudo samples, obeying a gaussian distribution.
Preferably, in step S6, a classifier is selected, the generated data features of the pseudo sample set and the support set are used as training data together, the classifier is trained, then the data features of the query set are used as test data, and the trained classifier is tested, so as to complete image classification.
Preferably, the classifier is a linear regression classification or support vector machine.
The application provides a small sample expansion system based on prototype completion in image classification, which comprises:
the original sample image acquisition and division unit is used for acquiring an original sample image data set and dividing the original sample image data set into a base class data set, a support set and a query set according to the image data type in the original sample image data set;
the feature extraction unit is used for selecting a feature extractor, training the feature extractor by utilizing the base class data set, and respectively extracting data features of the base class data set, data features of the support set and data features of the query set by utilizing the trained feature extractor;
the similar sample determining unit is used for calculating the distances between the data characteristics of the support set and the data characteristics of all base class data sets based on a spatial metric distance function, and determining K samples most similar to the data characteristics of the support set according to the distances;
the prototype feature determination unit is used for carrying out weighted calculation on the similarity between the K samples and the data features of the support set to obtain prototype features of the support set;
the pseudo sample set generating unit generates a pseudo sample set based on the prototype features of the support set;
and the image classification unit is used for performing image classification by using the generated pseudo sample set and the support set data characteristics as training data and using the query set data characteristics as test data.
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
the invention provides a small sample expansion method and a small sample expansion system based on prototype completion in image classification, which are characterized by firstly collecting an original sample image data set, dividing the original sample image data set into a base class data set, a support set and a query set, respectively extracting the characteristics of three classes of data sets based on a characteristic extractor, mapping a sample to a characteristic space from the space, completing the prototype characteristics based on the similarity degree of the sample characteristics in the characteristic space to obtain the prototype characteristics, and performing prototype completion and generation of a pseudo sample set on a small number of samples by using a characteristic library and the characteristic extractor of the base class data set as a whole.
Drawings
Fig. 1 is a schematic flow chart of a small sample expansion method based on prototype completion in image classification according to embodiment 1 of the present invention;
fig. 2 is a block diagram showing an implementation process of a small sample expansion based on prototype completion in image classification proposed in embodiment 1 of the present invention.
Fig. 3 shows a TSNE visualization proposed in embodiment 2 of the present invention;
fig. 4 is a block diagram of a small sample expansion system based on prototype completion in image classification according to embodiment 3 of the present invention.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the patent;
for better illustration of the present embodiment, certain parts of the drawings may be omitted, enlarged or reduced, and do not represent actual dimensions;
it will be understood by those skilled in the art that certain well-known descriptions of the figures may be omitted.
The technical solution of the present invention is further described below with reference to the accompanying drawings and examples.
The positional relationships depicted in the drawings are for illustrative purposes only and are not to be construed as limiting the present patent;
example 1
As shown in fig. 1, this embodiment proposes a small sample expansion method based on prototype completion in image classification, which includes the following steps:
s1, collecting an original sample image data set, and dividing the original sample image data set into a base class data set, a support set and a query set according to the image data type in the original sample image data set;
in step S1, the original image dataset, that is, the small sample image dataset, is to be expanded, where the "image data category in the original image dataset" refers to image data with a certain number of categories in the acquired original image dataset, for example, the original image dataset is a mobile phone image, the collection contains mobile phone images of different brands, each brand is respectively used as a category, the image data categories in the original image dataset are set to be Z categories, the base class dataset belongs to X categories, the support set and the query set belong to Y categories, and the following conditions are satisfied:
X+Y＝Z，
here, of all the categories Z in the original image dataset, the category to which the base class dataset belongs and the category described in the latter two do not belong to the same category, and the support set and the query set belong to the same category.
S2, selecting a feature extractor, training the feature extractor by using a base class data set, and respectively extracting data features of the base class data set, data features of a support set and data features of a query set by using the trained feature extractor;
in step S2, the feature extractor is selected as a Wide ResNet network, a loss function of the Wide ResNet network is set in the process of training the Wide ResNet network by using the base class data set, a weight parameter of the Wide ResNet network is updated in a back propagation manner until the loss function converges, the network parameter is fixed after the training is completed, and the trained feature extractor extracts the data feature of the set target task.
In consideration of the universality problem in practical application, a pre-trained feature extractor is adopted for feature extraction.
S3, calculating the distance between the data features of the support set and the data features of all base class data sets based on a spatial metric distance function, and determining K samples which are most similar to the data features of the support set according to the distance;
taking the average value of the ith class data characteristics of the base class data set as a general name of the class data characteristics, the following requirements are met:
wherein, mu 1i Mean, x, representing the ith class data feature of the base class data set j Representing the jth data characteristic in the ith type data characteristic of the base class data set; n is i Representing the number of ith class data features of the base class data set;
further, the corresponding covariance matrix is expressed as:
let the mean of the data features of the support data set be μ 2 In μ 2 As a general term for supporting data features of a data set, let a spatial metric distance function be collectively referred to as f d () And when the distances between the data features of the support set and the data features of all the base class data sets are calculated based on the spatial metric distance function, the following requirements are met:
d i ＝f d (μ i ,x s )
wherein d is i Representing the distance between the data characteristics of the support set and the ith type data characteristics of the base class data set; finally, a set of distances D is obtained, denoted as:
D＝{d 1 ,d 2 ,...,d i ,...,d q }
and q represents the number of the types of the data features of the base class data set, the values of all the distance elements in the distance set D are arranged from small to large, and the data features of the base class data set corresponding to the first K distance elements are selected according to the arrangement sequence and serve as the K samples with the most similar data features of the support set.
Here, the spatial metric distance function may be different specific metric functions, and considering that the implementation process is to establish and apply to different spatial metric distance functions, after obtaining distances between the support set data features and all base class data set data features, the distance is normalized, the implementation method is applicable to different spatial metric functions, taking the L1 similarity as an example, and if the similarity is higher and the function value is larger, the formula is adopted, where the expression is:
in the function using JS divergence as an example, the higher the similarity, the lower the function value, the following may be used:
wherein, d' i Represents the value after the distance normalization and the distance normalization,
the maximum value of the distances is represented,
representing the minimum in distance.
S4, carrying out weighted calculation on the similarity of the K samples and the data characteristics of the support set to obtain prototype characteristics of the support set;
the sum of the distances is 1 by utilizing the softmax function, and in step S4, the similarity between the K samples and the data feature of the support set, namely the mean value mu of the K samples and the data feature of the support set 2 By using a spatial metric distance function f d () And solving, wherein on the premise of distance normalization, an expression for performing weighting calculation is as follows:
wherein e is a natural constant, w s For the weighting of the support set data characteristics, it is generally set that there is only one or five support set data samples, and w in the case of only one support set sample s Set to 1.5,5 support set samplesIs set to 2.5, and is preset according to the number of the support concentrated samples, w' g Represents the weight, d ', of the g-th sample of the K samples' g Representing the distance of the g sample of the K samples from the support set data features;
the expression of the prototype features is:
wherein n is K Representing the sum of the number of similar samples and the number of features of the support set data, and mu' represents the prototype feature;
the covariance matrix C' corresponding to the prototype feature is:
wherein, C g And the covariance matrix represents the characteristic covariance matrix of the g-th similar sample data in the K samples, and alpha is a hyperparameter.
S5, generating a pseudo sample set based on the prototype features of the support set;
in step S5, the generated set of pseudo samples is D y The expression is:
where y represents the newly generated category,
representing newly generated pseudo samples, obeying a gaussian distribution.
And S6, using the generated data characteristics of the pseudo sample set and the support set as training data, using the data characteristics of the query set as test data, and performing image classification.
In this step, a classifier is selected, which may be a linear regression classification or a support vector machine, and in actual implementation, one of the classifiers is selected, and the generated pseudo sample set and the support set data features are used as training data together, the classifier is trained, then the query set data features are used as test data, and the trained classifier is tested to complete image classification.
In this embodiment, referring to fig. 2, an original sample image dataset is first collected, the original sample image dataset is divided into a base class dataset, a support set and a query set, corresponding to the base class image, the support set image and the query set image in fig. 2, features of three classes of datasets are respectively extracted based on a feature extractor after detection, samples are mapped to a feature space from the space, prototype feature completion is performed in the feature space based on the similarity of the sample features, prototype features are obtained, a small number of samples can be subjected to prototype completion and generation of a pseudo sample set as a whole only by using a feature library and a feature extractor of the base class dataset, the defect of small sample image classification is made up to a great extent, no additional parameters needing to be trained are introduced, time and equipment requirements of subsequent image classification are saved, and interpretability is stronger compared with a "black box" neural network model.
Example 2
In this embodiment, the validity of the method provided by the present application is verified through a specific experiment, where the miniImagenet and CUB data sets used in the experiment are small sample data sets, and the miniImagenet data set includes 100 classes including 600 images with a size of 84 × 84 pixels. It is divided into 64 base classes, 16 verification classes and 20 new classes. The CUB data set is a bird image data set containing 200 birds, and has a total of 11,788 images and a size of 84 × 84 pixels. It is divided into 100 base classes, 50 verification classes and 50 new classes.
The types of comparison algorithms involved are: optimization-based methods, metric-based methods, generation-based methods. The comparative results are shown in Table 1.
TABLE 1
From the results of table 1, it can be found that the classification performance of the method proposed in the present application is superior to that of other comparative methods. The effectiveness of the invention can be verified through the above simulation experiments, fig. 3 is a dimension reduction visualization diagram of a pseudo sample set generated by using the method provided by the present application, and as can be seen from fig. 3, the generated pseudo sample clusters are all distributed around the gaussian sample of the support set, and there are obvious discrimination boundaries between classes.
Example 3
As shown in fig. 4, the present embodiment provides a small sample expansion system based on prototype completion in image classification, the system includes:
the original sample image acquisition and division unit is used for acquiring an original sample image data set and dividing the original sample image data set into a base class data set, a support set and a query set according to the image data type in the original sample image data set;
the feature extraction unit is used for selecting a feature extractor, training the feature extractor by using a base class data set, and respectively extracting data features of the base class data set, data features of a support set and data features of a query set by using the trained feature extractor;
the similar sample determining unit is used for calculating the distances between the data features of the support set and the data features of all base class data sets based on a spatial metric distance function, and determining K samples which are most similar to the data features of the support set according to the distances;
the prototype feature determining unit is used for performing weighted calculation on the similarity between the K samples and the data features of the support set to obtain prototype features of the support set;
the pseudo sample set generating unit generates a pseudo sample set based on the prototype features of the support set;
and the image classification unit is used for performing image classification by using the generated pseudo sample set and the support set data characteristics as training data and using the query set data characteristics as test data.
It should be understood that the above-described embodiments of the present invention are merely examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention. Other variations and modifications will be apparent to persons skilled in the art in light of the above description. And are neither required nor exhaustive of all embodiments. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention should be included in the protection scope of the claims of the present invention.
Claims (10)
1. A small sample expansion method based on prototype completion in image classification, the method comprising the steps of:
s1, collecting an original sample image data set, and dividing the original sample image data set into a base class data set, a support set and a query set according to the image data type in the original sample image data set;
s2, selecting a feature extractor, training the feature extractor by using a base class data set, and respectively extracting data features of the base class data set, data features of a support set and data features of a query set by using the trained feature extractor;
s3, calculating the distance between the data features of the support set and the data features of all base class data sets based on a spatial metric distance function, and determining K samples which are most similar to the data features of the support set according to the distance;
s4, carrying out weighted calculation on the similarity of the K samples and the data characteristics of the support set to obtain prototype characteristics of the support set;
s5, generating a pseudo sample set based on the prototype features of the support set;
and S6, using the generated pseudo sample set and the support set data characteristics as training data together, using the query set data characteristics as test data, and performing image classification.
2. The method for expanding a small sample based on prototype completion in image classification according to claim 1, wherein in step S1, assuming that the image data classes in the original sample image data set are Z-class, the base class data set belongs to X-class, the support set and the query set belong to Y-class, and the following are satisfied:
3. the small sample expansion method based on prototype completion in image classification as claimed in claim 1, wherein in step S2, the feature extractor is selected as Wide ResNet network, in the process of training Wide ResNet network by using base class data set, the loss function of Wide ResNet network is set, the weight parameter of Wide ResNet network is updated by using back propagation mode until the loss function converges, the network parameter is fixed after training is completed, and the trained feature extractor extracts the data feature of the set target task.
4. The method for small sample expansion based on prototype completion in image classification according to claim 3, wherein the average value of the ith class data feature of the base class data set is taken as a general term of the class data feature, so that the following conditions are satisfied:
wherein, mu 1i Mean, x, representing the i-th class data feature of the base class dataset j Representing the jth data characteristic in the ith data characteristic of the base class data set; n is i Representing the number of ith class data features of the base class data set;
let the mean of the data features of the support data set be μ 2 In μ 2 As a general term for supporting data features of a data set, let a spatial metric distance function be collectively referred to as f d () Then, when the distances between the data features of the support set and the data features of all base class data sets are calculated based on the spatial metric distance function, the following requirements are satisfied:
d i ＝f d (μ i ,x s )
wherein d is i Data characteristics and base classes for a presentation support setDistance of class i data features of the dataset; finally, a set of distances D is obtained, denoted as:
D＝{d 1 ,d 2 ,...,d i ,...,d q }
and q represents the number of the types of the data features of the base class data set, the values of all the distance elements in the distance set D are arranged from small to large, and the data features of the base class data set corresponding to the first K distance elements are selected according to the arrangement sequence and serve as the K samples with the most similar data features of the support set.
5. The small sample expansion method based on prototype completion in image classification according to claim 4, characterized in that after obtaining the distance between the data features of the support set and the data features of all base class data sets, the distance is normalized, and the expression is:
or
Wherein d is i ' denotes a value after the distance normalization,
the maximum value of the distances is represented,
representing the minimum in distance.
6. The prototype-completion-based small sample expansion method in image classification as claimed in claim 5, wherein in step S4, similarity between K samples and the data feature of the support set, that is, mean value μ of K samples and the data feature of the support set 2 By measuring the distance function in spaceNumber f d () And solving, wherein on the premise of distance normalization, an expression for performing weighting calculation is as follows:
wherein e is a natural constant, w s The weight for supporting the characteristics of the set data is preset according to the number of the samples in the supporting set, w' g Represents the weight, d ', of the g-th sample of the K samples' g Representing the distance between the g sample of the K samples and the data feature of the support set;
the expression of the prototype features is:
wherein n is K Representing the sum of the number of similar samples and the number of support set data features, and mu' represents a prototype feature;
the covariance matrix C' corresponding to the prototype feature is:
wherein, C g And the covariance matrix represents the characteristic covariance matrix of the g-th similar sample data in the K samples, and alpha is a hyperparameter.
7. The method for small sample expansion based on prototype completion in image classification according to claim 6, wherein in step S5, the generated pseudo sample set is D y The expression is:
where y represents the newly generated category,
representing newly generated pseudo samples, obeying a gaussian distribution.
8. The small sample expansion method based on prototype completion in image classification as claimed in claim 7, wherein in step S6, a classifier is selected, the generated pseudo sample set and the support set data features are used together as training data to train the classifier, then the query set data features are used as test data to test the trained classifier, and the image classification is completed.
9. The method of claim 8, wherein the classifier is a linear regression classification or a support vector machine.
10. A prototype-completion-based small sample augmentation system in image classification, the system comprising:
the original sample image acquisition and division unit is used for acquiring an original sample image data set and dividing the original sample image data set into a base class data set, a support set and a query set according to the image data type in the original sample image data set;
the feature extraction unit is used for selecting a feature extractor, training the feature extractor by using a base class data set, and respectively extracting data features of the base class data set, data features of a support set and data features of a query set by using the trained feature extractor;
the similar sample determining unit is used for calculating the distances between the data features of the support set and the data features of all base class data sets based on a spatial metric distance function, and determining K samples which are most similar to the data features of the support set according to the distances;
the prototype feature determining unit is used for performing weighted calculation on the similarity between the K samples and the data features of the support set to obtain prototype features of the support set;
the pseudo sample set generating unit generates a pseudo sample set based on the prototype features of the support set;
and the image classification unit is used for performing image classification by using the generated pseudo sample set and the support set data characteristics as training data and using the query set data characteristics as test data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210923952.5A CN115393666A (en) | 2022-08-02 | 2022-08-02 | Small sample expansion method and system based on prototype completion in image classification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210923952.5A CN115393666A (en) | 2022-08-02 | 2022-08-02 | Small sample expansion method and system based on prototype completion in image classification |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115393666A true CN115393666A (en) | 2022-11-25 |
Family
ID=84118165
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210923952.5A Pending CN115393666A (en) | 2022-08-02 | 2022-08-02 | Small sample expansion method and system based on prototype completion in image classification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115393666A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116168257A (en) * | 2023-04-23 | 2023-05-26 | 安徽大学 | Small sample image classification method, device and storage medium based on sample generation |
| CN117078853A (en) * | 2023-08-18 | 2023-11-17 | 广东工业大学 | Workpiece defect sample amplification method based on digital twin body and storage medium |
-
2022
- 2022-08-02 CN CN202210923952.5A patent/CN115393666A/en active Pending
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116168257A (en) * | 2023-04-23 | 2023-05-26 | 安徽大学 | Small sample image classification method, device and storage medium based on sample generation |
| CN117078853A (en) * | 2023-08-18 | 2023-11-17 | 广东工业大学 | Workpiece defect sample amplification method based on digital twin body and storage medium |
| CN117078853B (en) * | 2023-08-18 | 2024-03-19 | 广东工业大学 | Workpiece defect sample amplification method based on digital twin body and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Gosselin et al. | Revisiting the fisher vector for fine-grained classification | |
| CN112396027A (en) | Vehicle weight recognition method based on graph convolution neural network | |
| CN108288051B (en) | Pedestrian re-recognition model training method and device, electronic equipment and storage medium | |
| CN103425996B (en) | A kind of large-scale image recognition methods of parallel distributed | |
| CN111652317B (en) | Super-parameter image segmentation method based on Bayes deep learning | |
| CN115393666A (en) | Small sample expansion method and system based on prototype completion in image classification | |
| Kadam et al. | Detection and localization of multiple image splicing using MobileNet V1 | |
| CN109410184B (en) | Live broadcast pornographic image detection method based on dense confrontation network semi-supervised learning | |
| CN114998220B (en) | Tongue image detection and positioning method based on improved Tiny-YOLO v4 natural environment | |
| CN110751027B (en) | Pedestrian re-identification method based on deep multi-instance learning | |
| CN112949740B (en) | Small sample image classification method based on multilevel measurement | |
| CN109670423A (en) | A kind of image identification system based on deep learning, method and medium | |
| CN105844299B (en) | A kind of image classification method based on bag of words | |
| CN108830302B (en) | Image classification method, training method, classification prediction method and related device | |
| Dong et al. | Scene-oriented hierarchical classification of blurry and noisy images | |
| CN108960013B (en) | Pedestrian re-identification method and device | |
| CN113780335B (en) | Small sample commodity image classification method, device, equipment and storage medium | |
| CN115203408A (en) | Intelligent labeling method for multi-modal test data | |
| CN111598144B (en) | Training method and device for image recognition model | |
| CN113627522A (en) | Image classification method, device and equipment based on relational network and storage medium | |
| CN114782752A (en) | Small sample image grouping classification method and device based on self-training | |
| Pan et al. | Leukocyte image segmentation using novel saliency detection based on positive feedback of visual perception | |
| WO2010134539A1 (en) | Feature value generating device, feature value generating method, feature value generating program, class determining device, class determining method, and class determining program | |
| CN111652265A (en) | Robust semi-supervised sparse feature selection method based on self-adjusting graph | |
| Anggoro et al. | Classification of Solo Batik patterns using deep learning convolutional neural networks algorithm |
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 |