CN108416370B - Image classification method and device based on semi-supervised deep learning and storage medium - Google Patents
Image classification method and device based on semi-supervised deep learning and storage medium Download PDFInfo
- Publication number
- CN108416370B CN108416370B CN201810132654.8A CN201810132654A CN108416370B CN 108416370 B CN108416370 B CN 108416370B CN 201810132654 A CN201810132654 A CN 201810132654A CN 108416370 B CN108416370 B CN 108416370B
- Authority
- CN
- China
- Prior art keywords
- label
- semi
- deep learning
- sample
- training
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Abstract
The invention discloses an image classification method, device and storage medium based on semi-supervised deep learning, wherein the method comprises the following steps: obtaining a label training image sample and a non-label training image sample to obtain a label training set; carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation; and carrying out image identification classification based on the models of semi-supervised deep learning and unlabelled sample class estimation. The invention can utilize the identification information hidden in the non-label training data and the high separability of the current depth characteristic, and can more effectively and accurately utilize the unmarked sample, thereby obtaining better image identification performance.
Description
Technical Field
The invention relates to the technical field of image recognition, in particular to an image classification method and device based on semi-supervised deep learning and a storage medium.
Background
In recent years, Deep Learning (Deep Learning) based techniques have achieved great success in the field of computer vision, such as face recognition and object classification, wherein representative Deep Learning methods are CNN (convolutional neural network), RNN (recurrent neural network), Autoencoder, GAN (generative countermeasure network), and the like.
However, in practical applications, since it takes a lot of time and labor to label samples, in real life, there are usually a lot of unlabeled samples, and the method of using the unlabeled samples to improve the final recognition effect is called semi-supervised learning.
In order to better utilize the discrimination information of the non-label training data and the high discrimination of the depth features, the work based on the deep semi-supervised learning has been intensively studied by researchers. For example, there are methods of semi-supervised deep kernel learning by integrating convolutional neural networks and gaussian probability models, training semi-supervised deep learning models by minimizing the loss of both supervised and unsupervised functions, and so on. While the semi-supervised deep learning approach is able to learn high-level representation features, it ignores how to more effectively exploit the high discriminative power of non-labeled exemplars.
Disclosure of Invention
The invention provides an image classification method, device and storage medium based on semi-supervised deep learning, aiming at establishing a unified model for semi-supervised deep learning and unlabelled sample class estimation, so that unlabelled samples can be more effectively and accurately utilized, and the image identification effect is improved.
In order to achieve the above object, the present invention provides an image classification method based on semi-supervised deep learning, comprising the following steps:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
and performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Optionally, the step of performing convolutional neural network training in combination with deep learning and semi-supervised learning to establish a unified model of semi-supervised deep learning and unlabeled sample class estimation includes:
inputting and training a convolutional neural network model by the label training set, and extracting the depth characteristics of the label training image samples and the non-label training image samples;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
and adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
Optionally, the step of performing image recognition classification based on the model of semi-supervised deep learning and unlabelled sample class estimation comprises:
and performing image character recognition and image target classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Optionally, the step of obtaining a label training image sample and a non-label training image sample to obtain a label training set further includes:
initializing the label training image sample and non-label training image sample data.
Optionally, the step of initializing the label training image samples and non-label training image samples comprises:
initializing a probability matrix P0The method comprises the following steps: and constructing a matrix of n x l of matrix elements (m, k), wherein n is the total number of the label training images and the non-label training images, l is the total class number, and the matrix elements (m, k) represent the probability that the mth training image belongs to the kth class, wherein m is less than or equal to n, and k is less than or equal to l.
Optionally, the step of performing category estimation on the non-label image sample after the depth feature is extracted to obtain a category estimation result includes:
constructing a similar matrix T according to the depth characteristics of the label training image sample and the non-label training image sample;
according to the initialized probability matrix P0And the similar matrix T is used for carrying out class estimation on the non-label image sample after the depth features are extracted to obtain a final probability matrix P.
Optionally, the step of adding the non-labeled image sample with the reliability meeting the preset condition to the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation includes:
obtaining a credible sample in the final probability matrix P;
and adding all the credible samples as new label data into the training of the convolutional neural network to retrain the convolutional neural network so as to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
The embodiment of the invention also provides an image classification device based on semi-supervised deep learning, which comprises a memory, a processor and an image classification program based on semi-supervised deep learning, wherein the image classification program based on semi-supervised deep learning is stored on the memory, and when the image classification program based on semi-supervised deep learning is operated by the processor, the following operations are realized:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
and performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Optionally, the semi-supervised deep learning based image classification program when executed by the processor further implements the following operations:
inputting and training a convolutional neural network model by the label training set, and extracting the depth characteristics of the label training image samples and the non-label training image samples;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
and adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
An embodiment of the present invention further provides a computer-readable storage medium, where a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the computer program implements the steps of the method described above.
According to the image classification method, device and storage medium based on semi-supervised deep learning, the label training set is obtained by obtaining the label training image sample and the non-label training image sample; carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation; and performing image recognition and classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation, so that a unified model of the semi-supervised deep learning and the unlabelled sample class estimation is established by combining the deep learning and the semi-supervised learning, a large number of unlabelled samples can be more effectively and accurately utilized, and the final image recognition effect is improved.
Compared with the prior art, the invention has the following advantages:
1. compared with the fully supervised deep learning, the method effectively utilizes the non-label sample data to strengthen the training of the network model, and avoids the condition of low final identification precision caused by the limitation of the number of label training samples in practice. Secondly, for semi-supervised learning, namely estimation of non-label samples, the invention adopts the depth characteristic with higher semantics, so that the estimation of the non-label samples is more accurate and reliable;
2. the invention learns a united and unified model, so that the probability estimation of the non-label data and the learning of the deep network are alternately and iteratively carried out, the deep characteristics can be effectively utilized, meanwhile, the discriminative information in the non-label training data can be discovered, and the reliable non-label samples can be added into the learning of the network through an effective semi-supervision method.
Therefore, compared with the traditional technical scheme, the method can utilize the identification information hidden in the non-label training data and the high separability of the current depth feature, so that better identification performance is obtained.
Drawings
FIG. 1 is a schematic flowchart of a first embodiment of an image classification method based on semi-supervised deep learning according to the present invention;
FIG. 2 is a schematic diagram of the operation of a convolutional neural network in an embodiment of the present invention;
FIG. 3 is a flowchart illustrating a second embodiment of the image classification method based on semi-supervised deep learning according to the present invention;
FIG. 4 is a schematic diagram of the main process flow of an embodiment of the present invention;
FIG. 5 is a flow chart illustrating the general training phase in an embodiment of the present invention;
fig. 6 is a schematic diagram of a hardware operating environment according to an embodiment of the present invention.
The implementation, functional features and advantages of the objects of the present invention will be further explained with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
The main solution of the embodiment of the invention is as follows: obtaining a label training set by obtaining a label training image sample and a non-label training image sample; carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation; and performing image recognition and classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation, so that a unified model of the semi-supervised deep learning and the unlabelled sample class estimation is established by combining the deep learning and the semi-supervised learning, a large number of unlabelled samples can be more effectively and accurately utilized, and the final image recognition effect is improved.
Specifically, referring to fig. 1, fig. 1 is a flowchart illustrating a first embodiment of an image classification method based on semi-supervised deep learning according to the present invention.
As shown in fig. 1, an image classification method based on semi-supervised deep learning according to a first embodiment of the present invention includes:
step S101, obtaining a label training image sample and a non-label training image sample to obtain a label training set;
the embodiment of the invention aims to better combine deep learning and semi-supervised learning and establish a unified model for semi-supervised deep learning and unlabelled sample class estimation, so that a large number of unlabelled samples can be more effectively and accurately utilized, and the final image identification effect is improved, and the method mainly comprises the following two aspects of application: character recognition and object classification.
In the scheme of the embodiment of the invention, the high identifiability of the features under the training of the convolutional neural network, namely the small intra-class distance and the large inter-class distance, is mainly utilized, meanwhile, the label propagation is used as a semi-supervised frame, the class estimation is carried out on the non-label samples after the deep features are extracted, and the non-label samples with high credibility are added into the training of the convolutional network again.
The embodiment of the invention provides the following models:
wherein x isiRepresents the training data of the ith label,
indicates the category, x, to which the ith label training data belongsjRepresents the jth non-labeled training sample,
representing the category in which the jth non-labeled sample was evaluated,
representing the corresponding probability. n islIs the total number of label samples, nuIs the total number of non-labeled samples. The function f represents the learning of convolution features in a convolutional neural network,
representing the loss of the classification phase in a convolutional neural network. f and
the role in a typical convolutional neural network is shown in fig. 2 by the dashed box:
firstly, obtaining a label training image sample and a non-label training image sample to obtain a label training set, wherein the label training image sample and the non-label training image sample in the label training set can be embodied in a matrix mode.
Step S102, carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
specifically, inputting and training a convolutional neural network model by the label training set, and extracting the depth features of the label training image sample and the non-label training image sample;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
and adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
And step S103, carrying out image identification and classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Then, image recognition classification can be carried out based on the models of the semi-supervised deep learning and the unlabelled sample class estimation. And performing image character recognition and image target classification mainly based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Compared with the traditional technical scheme, the method can better utilize the identification information hidden in the non-label training data and can also utilize the high separability of the current depth feature, thereby obtaining better identification performance.
As shown in fig. 3, an image classification method based on semi-supervised deep learning according to a second embodiment of the present invention is based on the embodiment shown in fig. 1, and in step S101: obtaining a label training image sample and a non-label training image sample, and after obtaining a label training set, further comprising:
and S100, initializing the label training image sample and the non-label training image sample.
The implementation of the method of the embodiment of the invention is divided into the following two major stages: stage 1 is an initialization data stage, stage 2 is a training stage, and the main processing flow is shown in fig. 4.
Initializing the label training image sample and the non-label training image sample data comprises the following steps:
initializing a probability matrix P0The method comprises the following steps: and constructing a matrix of n x l of matrix elements (m, k), wherein n is the total number of the label training images and the non-label training images, l is the total class number, and the matrix elements (m, k) represent the probability that the mth training image belongs to the kth class, wherein m is less than or equal to n, and k is less than or equal to l.
And then, inputting the label training set and training a convolutional neural network model, and extracting the depth features of the label training image samples and the non-label training image samples. And performing class estimation on the non-label image sample after the depth features are extracted.
The method comprises the following steps of carrying out class estimation on a non-label image sample after the depth features are extracted to obtain a class estimation result, wherein the step of carrying out class estimation on the non-label image sample after the depth features are extracted comprises the following steps:
constructing a similar matrix T according to the depth characteristics of the label training image sample and the non-label training image sample;
according to the initialized probability matrix P0And the similar matrix T is used for carrying out class estimation on the non-label image sample after the depth features are extracted to obtain a final probability matrix P.
Then, obtaining a credible sample in the final probability matrix P; and adding all the credible samples as new label data into the training of the convolutional neural network to retrain the convolutional neural network so as to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
The specific process is as follows:
phase 1-initializing data
Initializing a probability matrix P0: and constructing an n-l matrix, wherein n is the total number of the label training images and the non-label training images, l is the total number of the classes, and matrix elements (m, k) represent the probability (m is less than or equal to n, k is less than or equal to l) that the mth training image (possibly the label image or the non-label image) belongs to the kth class.
Stage 2-training stage
In the training phase, three parts are mainly included:
1: obtaining depth features, updating function f and
2: estimating the class of non-labeled exemplars: updating the probability matrix P;
3: add authentic unlabeled sample: the convolutional network is retrained.
The flow of the overall training phase is shown in fig. 5:
first, the depth feature is obtained, namely, the updating function f and
the class of the non-labeled exemplars is then estimated: updating the probability matrix P; finally, add authentic unlabeled samples: the convolutional network is retrained.
The concrete implementation is as follows:
(1) obtaining depth features
After the training set of labels is obtained, the convolutional neural network may be trained. The embodiment of the invention adopts some convolution neural networks which exist currently, for example, LeNet network is adopted for character recognition.
After the convolutional neural network is trained, the corresponding depth features can be obtained by inputting label and non-label training data.
(2) Estimating classes of non-labeled exemplars
After obtaining the depth features of the labeled training data and the non-labeled training data, then a similarity matrix may be constructed.
Specifically, the embodiment adopts a formula representing the similarity between training samples i and j, calculates the distance between any two samples by the euclidean distance, and if the distance between any two samples is closer, the weight between them is larger, and the mathematical formula is:
wherein x isiOr xjRepresenting training sample i and training sample j, with sample dimension D, and σ being a parameter. After the weight matrix W is obtained, through further normalization, the sum of any row of the weight matrix W is 1, and the specific mathematical formula is as follows:
after normalization, T is a symmetric matrix.
By initializing the data phase, a probability matrix P has been obtained0And combining the similarity matrix T to obtain a final probability matrix P through the following two steps:
firstly, propagation calculation of the label is carried out by the following specific formula:
Pt+1＝T*Pt；
here, P ist+1Represents PtThe symbol' indicates a dot product operation.
Since the calculation of r changes the fact that the probability of the label data in the corresponding category is 1, it is necessary to reset the probability of the obtained probability matrix in the label data portion, i.e. the calculation is performed
Wherein the content of the first and second substances,
represents the initialization matrix P0The portion of the tag data.
(3) Adding authentic unlabeled exemplars
By estimating the class of the non-labeled sample, a final probability matrix is obtained
For convenience of presentation, P is substituted. If a certain non-labeled sample gets a higher result in P, the non-labeled sample is considered to be credible, i.e. more likely to belong to the corresponding class, for example, if P (i, k) ═ 1, the sample i is considered to belong to the kth class with the probability of 1, and therefore, the sample i is a credible sample.
And then, all the credible samples are taken as new label data and added into the training of the convolutional neural network, and the convolutional neural network is retrained.
According to the scheme, the label training set is obtained by obtaining the label training image sample and the non-label training image sample; carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation; and performing image recognition and classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation, so that a unified model of the semi-supervised deep learning and the unlabelled sample class estimation is established by combining the deep learning and the semi-supervised learning, a large number of unlabelled samples can be more effectively and accurately utilized, and the final image recognition effect is improved.
Compared with the prior art, the invention has the following advantages:
1. compared with the fully supervised deep learning, the method effectively utilizes the non-label sample data to strengthen the training of the network model, and avoids the condition of low final identification precision caused by the limitation of the number of label training samples in practice. Secondly, for semi-supervised learning, namely estimation of non-label samples, the invention adopts the depth characteristic with higher semantics, so that the estimation of the non-label samples is more accurate and reliable;
2. the invention learns a united and unified model, so that the probability estimation of the non-label data and the learning of the deep network are alternately and iteratively carried out, the deep characteristics can be effectively utilized, meanwhile, the discriminative information in the non-label training data can be discovered, and the reliable non-label samples can be added into the learning of the network through an effective semi-supervision method.
Therefore, compared with the traditional technical scheme, the method can utilize the identification information hidden in the non-label training data and the high separability of the current depth feature, so that better identification performance is obtained.
The embodiment of the present invention will be further described with reference to specific experimental results as follows:
the present invention is compared to existing fully supervised dictionary learning techniques (FDDL, CRC), semi-supervised dictionary learning techniques (DSSDL), and fully supervised convolutional neural networks (LeNet as described in the experiments below) which compare the recognition accuracy of character recognition and target classification on two major visual tasks.
In order to compare objectivity and justice, a Mnist character library and a CIFAR-10 target classification database which are standard databases are still adopted in the experiment.
Character recognition:
the Mnist character library contains 10 categories, 6 ten thousand training images in total, and 1 ten thousand test images. To meet the practical situation that the number of non-label samples is far more than the number of label samples, the detailed experiment is set as follows:
a. for each category in the data set, randomly selecting 100 character images as a training set;
b. randomly selecting two ten thousand images from the rest training samples as a non-label training set;
c. each set of experiments was repeated ten times and the final average recognition rate and standard deviation were calculated.
And (3) target classification:
the Cifar-10 database contains 10 classes, 6 million 32 x 32-sized color images, 5 million training images, and 1 million test images, with 6 thousand images for each class. The same settings as in the character recognition experiment were used.
The comparison of the recognition accuracy of the two tasks is summarized in the following table 1, and as can be seen from table 1, the recognition accuracy of the invention is higher than that of other experimental methods under two different tasks, especially a target classification task, and the recognition accuracy is much higher than that of other methods for learning the fully supervised dictionary, including only a deep learning method LeNet, because the method for learning the fully supervised dictionary is very dependent on the number of label training samples, and when the number of labels is less, the discriminative performance of the learned classifier is weak.
TABLE 1 comparison of recognition accuracy (%) between the two tasks
Compared with the prior art, the method adopts a uniform semi-supervised deep learning model, effectively integrates the estimation of the non-label samples and the high-discriminative deep learning method, estimates the types of the non-label samples through the high-discriminative characteristic of the deep characteristics, takes the reliable non-label samples as the label training data, trains the deep learning network again, equivalently increases the number of samples of the label training data, and better utilizes the non-label samples existing in practice; in addition, the invention effectively utilizes the label propagation method in the semi-supervised learning, so that the learning of the depth features and the semi-supervised learning are alternately carried out, namely, the label propagation is not based on the traditional image features but on the depth features with high-level semantics, and therefore, the estimation of the non-label samples is more accurate.
In addition, an embodiment of the present invention further provides an image classification apparatus based on semi-supervised deep learning, including a memory, a processor, and an image classification program based on semi-supervised deep learning stored on the memory, where when the image classification program based on semi-supervised deep learning is executed by the processor, the following operations are implemented:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
and performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Specifically, as shown in fig. 6, the image classification apparatus based on semi-supervised deep learning of the present embodiment may include: a processor 1001, such as a CPU, a network interface 1004, a user interface 1003, a memory 1005, a communication bus 1002. Wherein a communication bus 1002 is used to enable connective communication between these components. The user interface 1003 may include a Display screen (Display), an input unit such as a Keyboard (Keyboard), and the optional user interface 1003 may also include a standard wired interface, a wireless interface. The network interface 1004 may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface). The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a storage device separate from the processor 1001.
Those skilled in the art will appreciate that the device configuration shown in fig. 6 does not constitute a limitation of the device and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components.
As shown in fig. 6, a memory 1005, which is a kind of computer storage medium, may include therein an operating device, a network communication module, a user interface module, and an image classification program based on semi-supervised deep learning.
In the apparatus shown in fig. 6, the network interface 1004 is mainly used for connecting to a network server and performing data communication with the network server; the user interface 1003 is mainly used for interacting with a user terminal and receiving an instruction input by a user; and the processor 1001 may be configured to invoke an image classification program based on semi-supervised deep learning stored in the memory 1005 and perform the following operations:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
and performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Further, the processor 1001 may be further configured to invoke an image classification program based on semi-supervised deep learning stored in the memory 1005, and perform the following operations:
inputting and training a convolutional neural network model by the label training set, and extracting the depth characteristics of the label training image samples and the non-label training image samples;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
and adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
Further, the processor 1001 may be further configured to invoke an image classification program based on semi-supervised deep learning stored in the memory 1005, and perform the following operations:
and performing image character recognition and image target classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Further, the processor 1001 may be further configured to invoke an image classification program based on semi-supervised deep learning stored in the memory 1005, and perform the following operations:
initializing the label training image samples and non-label training image sample data, including:
initializing a probability matrix P0The method comprises the following steps: constructing a matrix of n x l of matrix elements (m, k), n being the total number of labeled training images and unlabeled training images, l being the total number of classes, represented by matrix elements (m, k)Is the probability that the mth training image belongs to the kth class, wherein m is less than or equal to n, and k is less than or equal to l.
Further, the processor 1001 may be further configured to invoke an image classification program based on semi-supervised deep learning stored in the memory 1005, and perform the following operations:
constructing a similar matrix T according to the depth characteristics of the label training image sample and the non-label training image sample;
according to the initialized probability matrix P0And a similar matrix T, carrying out category estimation on the non-label image sample after the depth features are extracted to obtain a final probability matrix P;
obtaining a credible sample in the final probability matrix P;
and adding all the credible samples as new label data into the training of the convolutional neural network to retrain the convolutional neural network so as to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
Furthermore, the present invention also provides a computer-readable storage medium, on which a computer program is stored, which when executed by the processor implements the following operations:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
and performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Further, the computer program when executed by the processor further performs the following:
inputting and training a convolutional neural network model by the label training set, and extracting the depth characteristics of the label training image samples and the non-label training image samples;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
and adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
Further, the computer program when executed by the processor further performs the following:
and performing image character recognition and image target classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
Further, the computer program when executed by the processor further performs the following:
initializing the label training image samples and non-label training image sample data, including:
initializing a probability matrix P0The method comprises the following steps: and constructing a matrix of n x l of matrix elements (m, k), wherein n is the total number of the label training images and the non-label training images, l is the total class number, and the matrix elements (m, k) represent the probability that the mth training image belongs to the kth class, wherein m is less than or equal to n, and k is less than or equal to l.
Further, the computer program when executed by the processor further performs the following:
constructing a similar matrix T according to the depth characteristics of the label training image sample and the non-label training image sample;
according to the initialized probability matrix P0And the similar matrix T is used for carrying out class estimation on the non-label image sample after the depth features are extracted to obtain a final probability matrix P.
Further, the computer program when executed by the processor further performs the following:
obtaining a credible sample in the final probability matrix P;
and adding all the credible samples as new label data into the training of the convolutional neural network to retrain the convolutional neural network so as to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
Compared with the prior art, the image classification method, the image classification device and the storage medium based on semi-supervised deep learning provided by the invention have the advantages that a label training set is obtained by obtaining a label training image sample and a non-label training image sample; carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation; and performing image recognition and classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation, so that a unified model of the semi-supervised deep learning and the unlabelled sample class estimation is established by combining the deep learning and the semi-supervised learning, a large number of unlabelled samples can be more effectively and accurately utilized, and the final image recognition effect is improved.
Compared with the prior art, the invention has the following advantages:
1. compared with the fully supervised deep learning, the method effectively utilizes the non-label sample data to strengthen the training of the network model, and avoids the condition of low final identification precision caused by the limitation of the number of label training samples in practice. Secondly, for semi-supervised learning, namely estimation of non-label samples, the invention adopts the depth characteristic with higher semantics, so that the estimation of the non-label samples is more accurate and reliable;
2. the invention learns a united and unified model, so that the probability estimation of the non-label data and the learning of the deep network are alternately and iteratively carried out, the deep characteristics can be effectively utilized, meanwhile, the discriminative information in the non-label training data can be discovered, and the reliable non-label samples can be added into the learning of the network through an effective semi-supervision method.
Therefore, compared with the traditional technical scheme, the method can utilize the identification information hidden in the non-label training data and the high separability of the current depth feature, so that better identification performance is obtained.
The above description is only for the preferred embodiment of the present invention and is not intended to limit the scope of the present invention, and all equivalent structures or flow transformations made by the present specification and drawings, or applied directly or indirectly to other related arts, are included in the scope of the present invention.
Claims (5)
1. An image classification method based on semi-supervised deep learning is characterized by comprising the following steps:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation;
the step of performing convolutional neural network training by combining deep learning and semi-supervised learning and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation comprises the following steps of:
inputting and training a convolutional neural network model by the label training set, and extracting the depth characteristics of the label training image samples and the non-label training image samples;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation;
the step of obtaining the label training image sample and the non-label training image sample to obtain the label training set further comprises:
initializing the label training image sample and non-label training image sample data;
the step of initializing the label training image samples and non-label training image sample data comprises:
initializing a probability matrix P0The method comprises the following steps: constructing a matrix element(m, k) n x l matrix, n being the total number of tagged and non-tagged training images, l being the total number of classes, matrix elements (m, k) representing the probability that the mth training image belongs to the kth class, wherein m is less than or equal to n, k is less than or equal to l;
the step of performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result comprises the following steps:
constructing a similar matrix T according to the depth characteristics of the label training image sample and the non-label training image sample;
according to the initialized probability matrix P0And the similar matrix T is used for carrying out class estimation on the non-label image sample after the depth features are extracted to obtain a final probability matrix P.
2. The method of claim 1, wherein the step of image recognition classification based on the model of semi-supervised deep learning and unlabeled sample class estimation comprises:
and performing image character recognition and image target classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation.
3. The method according to claim 1, wherein the step of adding the unlabeled image sample with the reliability meeting the preset condition to the convolutional neural network again for training to obtain the model for semi-supervised deep learning and unlabeled sample class estimation according to the class estimation result comprises:
obtaining a credible sample in the final probability matrix P;
and adding all the credible samples as new label data into the training of the convolutional neural network to retrain the convolutional neural network so as to obtain a model for semi-supervised deep learning and unlabelled sample class estimation.
4. An image classification device based on semi-supervised deep learning, which is characterized by comprising a memory, a processor and an image classification program based on semi-supervised deep learning, which is stored in the memory, wherein when the image classification program based on semi-supervised deep learning is executed by the processor, the following operations are realized:
obtaining a label training image sample and a non-label training image sample to obtain a label training set;
carrying out convolutional neural network training on the label training set by combining deep learning and semi-supervised learning, and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation;
performing image identification classification based on the models of the semi-supervised deep learning and the unlabelled sample class estimation;
the step of performing convolutional neural network training by combining deep learning and semi-supervised learning and establishing a unified model of semi-supervised deep learning and unlabelled sample class estimation comprises the following steps of:
inputting and training a convolutional neural network model by the label training set, and extracting the depth characteristics of the label training image samples and the non-label training image samples;
performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result;
adding the non-label image samples with the credibility meeting the preset conditions into the convolutional neural network again for training according to the class estimation result to obtain a model for semi-supervised deep learning and unlabelled sample class estimation;
the step of obtaining the label training image sample and the non-label training image sample to obtain the label training set further comprises:
initializing the label training image sample and non-label training image sample data;
the step of initializing the label training image samples and non-label training image sample data comprises:
initializing a probability matrix P0The method comprises the following steps: constructing a matrix of n x l of matrix elements (m, k), n being the total number of labeled training images and unlabeled training images, l being the total number of classes, the matrix elements (m, k) representing the probability that the mth training image belongs to the kth class, which isIn the formula, m is less than or equal to n, and k is less than or equal to l;
the step of performing category estimation on the non-label image sample after the depth features are extracted to obtain a category estimation result comprises the following steps:
constructing a similar matrix T according to the depth characteristics of the label training image sample and the non-label training image sample;
according to the initialized probability matrix P0And the similar matrix T is used for carrying out class estimation on the non-label image sample after the depth features are extracted to obtain a final probability matrix P.
5. A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, which computer program, when being executed by a processor, carries out the steps of the method according to any one of claims 1-3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810132654.8A CN108416370B (en) | 2018-02-07 | 2018-02-07 | Image classification method and device based on semi-supervised deep learning and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810132654.8A CN108416370B (en) | 2018-02-07 | 2018-02-07 | Image classification method and device based on semi-supervised deep learning and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108416370A CN108416370A (en) | 2018-08-17 |
| CN108416370B true CN108416370B (en) | 2022-03-22 |
Family
ID=63127085
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810132654.8A Active CN108416370B (en) | 2018-02-07 | 2018-02-07 | Image classification method and device based on semi-supervised deep learning and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108416370B (en) |
Families Citing this family (59)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109389080B (en) * | 2018-09-30 | 2022-04-19 | 西安电子科技大学 | Hyperspectral image classification method based on semi-supervised WGAN-GP |
| CN111008294B (en) * | 2018-10-08 | 2023-06-20 | 阿里巴巴集团控股有限公司 | Traffic image processing and image retrieval method and device |
| CN109359684B (en) * | 2018-10-17 | 2021-10-29 | 苏州大学 | Fine-grained vehicle type identification method based on weak supervision positioning and subcategory similarity measurement |
| CN109146944B (en) * | 2018-10-30 | 2020-06-26 | 浙江科技学院 | Visual depth estimation method based on depth separable convolutional neural network |
| CN109389180A (en) * | 2018-10-30 | 2019-02-26 | 国网四川省电力公司广元供电公司 | A power equipment image-recognizing method and inspection robot based on deep learning |
| CN109657697B (en) * | 2018-11-16 | 2023-01-06 | 中山大学 | Classification optimization method based on semi-supervised learning and fine-grained feature learning |
| CN109872162B (en) * | 2018-11-21 | 2023-06-16 | 创新先进技术有限公司 | Wind control classification and identification method and system for processing user complaint information |
| CN109522961B (en) * | 2018-11-23 | 2022-09-13 | 中山大学 | Semi-supervised image classification method based on dictionary deep learning |
| CN109800781A (en) * | 2018-12-07 | 2019-05-24 | 北京奇艺世纪科技有限公司 | A kind of image processing method, device and computer readable storage medium |
| CN109711319B (en) * | 2018-12-24 | 2023-04-07 | 安徽高哲信息技术有限公司 | Method and system for establishing imperfect grain image recognition sample library |
| CN109741346B (en) * | 2018-12-30 | 2020-12-08 | 上海联影智能医疗科技有限公司 | Region-of-interest extraction method, device, equipment and storage medium |
| CN111414930B (en) * | 2019-01-07 | 2023-10-27 | 中国移动通信有限公司研究院 | Deep learning model training method and device, electronic equipment and storage medium |
| CN109522973A (en) * | 2019-01-17 | 2019-03-26 | 云南大学 | Medical big data classification method and system based on production confrontation network and semi-supervised learning |
| CN111507362B (en) * | 2019-01-30 | 2023-12-19 | 中强光电股份有限公司 | Neural network training method, neural network-based classification method and device |
| KR20200107555A (en) * | 2019-03-08 | 2020-09-16 | 에스케이텔레콤 주식회사 | Apparatus and method for analysing image, and method for generating image analysis model used therefor |
| CN111753863A (en) * | 2019-04-12 | 2020-10-09 | 北京京东尚科信息技术有限公司 | Image classification method and device, electronic equipment and storage medium |
| CN110147456B (en) * | 2019-04-12 | 2023-01-24 | 中国科学院深圳先进技术研究院 | Image classification method and device, readable storage medium and terminal equipment |
| CN110175653A (en) * | 2019-05-29 | 2019-08-27 | 北京达佳互联信息技术有限公司 | Method, apparatus, equipment and the storage medium of image recognition |
| CN110189323B (en) * | 2019-06-05 | 2022-12-13 | 深圳大学 | Breast ultrasound image lesion segmentation method based on semi-supervised learning |
| CN110245721B (en) * | 2019-06-25 | 2023-09-05 | 深圳市腾讯计算机系统有限公司 | Training method and device for neural network model and electronic equipment |
| CN110363231B (en) * | 2019-06-27 | 2023-01-06 | 平安科技（深圳）有限公司 | Abnormity identification method and device based on semi-supervised deep learning and storage medium |
| CN112215248A (en) * | 2019-07-11 | 2021-01-12 | 深圳先进技术研究院 | Deep learning model training method and device, electronic equipment and storage medium |
| CN110598737B (en) * | 2019-08-06 | 2023-02-24 | 深圳大学 | Online learning method, device, equipment and medium of deep learning model |
| CN110728295B (en) * | 2019-09-02 | 2022-05-24 | 深圳中科保泰空天技术有限公司 | Semi-supervised landform classification model training and landform graph construction method |
| CN110796192B (en) * | 2019-10-29 | 2023-03-28 | 深圳大学 | Image classification method and device based on Internet social contact system |
| CN111161848B (en) * | 2019-10-31 | 2023-08-29 | 杭州深睿博联科技有限公司 | Focus marking method and device for CT image and storage medium |
| CN110880117A (en) * | 2019-10-31 | 2020-03-13 | 北京三快在线科技有限公司 | False service identification method, device, equipment and storage medium |
| CN110866564B (en) * | 2019-11-22 | 2023-04-25 | 上海携程国际旅行社有限公司 | Season classification method, system, electronic device and medium for multiple semi-supervised images |
| CN110909803B (en) * | 2019-11-26 | 2023-04-18 | 腾讯科技（深圳）有限公司 | Image recognition model training method and device and computer readable storage medium |
| US11449711B2 (en) * | 2020-01-02 | 2022-09-20 | Applied Materials Isreal Ltd. | Machine learning-based defect detection of a specimen |
| CN111222648B (en) * | 2020-01-15 | 2023-09-26 | 深圳前海微众银行股份有限公司 | Semi-supervised machine learning optimization method, device, equipment and storage medium |
| CN111539443B (en) * | 2020-01-22 | 2024-02-09 | 北京小米松果电子有限公司 | Image recognition model training method and device and storage medium |
| CN111353549B (en) * | 2020-03-10 | 2023-01-31 | 创新奇智(重庆)科技有限公司 | Image label verification method and device, electronic equipment and storage medium |
| CN111488840A (en) * | 2020-04-15 | 2020-08-04 | 桂林电子科技大学 | Human behavior classification method based on multi-task learning model |
| CN111523597B (en) * | 2020-04-23 | 2023-08-25 | 北京百度网讯科技有限公司 | Target recognition model training method, device, equipment and storage medium |
| CN111612062B (en) * | 2020-05-20 | 2023-11-24 | 清华大学 | Method and system for constructing graph semi-supervised learning model |
| CN111582404B (en) * | 2020-05-25 | 2024-03-26 | 腾讯科技（深圳）有限公司 | Content classification method, device and readable storage medium |
| CN111723209B (en) * | 2020-06-28 | 2023-04-25 | 上海携旅信息技术有限公司 | Semi-supervised text classification model training method, text classification method, system, equipment and medium |
| CN111783142B (en) * | 2020-07-06 | 2021-10-08 | 北京字节跳动网络技术有限公司 | Data protection method, device, server and medium |
| CN111967576B (en) * | 2020-07-22 | 2022-09-02 | 长春工程学院 | Geochemical data processing method and system based on deep learning |
| CN111899254A (en) * | 2020-08-12 | 2020-11-06 | 华中科技大学 | Method for automatically labeling industrial product appearance defect image based on semi-supervised learning |
| CN112200245A (en) * | 2020-10-10 | 2021-01-08 | 深圳市华付信息技术有限公司 | Image classification method based on semi-supervision |
| CN112417986B (en) * | 2020-10-30 | 2023-03-10 | 四川天翼网络股份有限公司 | Semi-supervised online face recognition method and system based on deep neural network model |
| CN112381810A (en) * | 2020-11-20 | 2021-02-19 | 平安科技（深圳）有限公司 | Pneumonia image identification method and device and computing equipment |
| CN112766323A (en) * | 2020-12-30 | 2021-05-07 | 清华大学 | Image identification method and device |
| CN112651467A (en) * | 2021-01-18 | 2021-04-13 | 第四范式（北京）技术有限公司 | Training method and system and prediction method and system of convolutional neural network |
| CN112861926B (en) * | 2021-01-18 | 2023-10-31 | 平安科技（深圳）有限公司 | Coupled multi-task feature extraction method and device, electronic equipment and storage medium |
| CN112801153B (en) * | 2021-01-19 | 2022-08-26 | 南京邮电大学 | Semi-supervised image classification method and system of image embedded with LBP (local binary pattern) features |
| CN112836755B (en) * | 2021-02-05 | 2024-04-16 | 中国科学院深圳先进技术研究院 | Sample image generation method and system based on deep learning |
| CN112862809B (en) * | 2021-03-09 | 2023-07-18 | 中央财经大学 | Spatial resolution enhancement method based on weak supervision deep learning, terminal equipment and computer readable storage medium |
| CN113095162B (en) * | 2021-03-24 | 2023-05-23 | 杭州电子科技大学 | Spectrum sensing method based on semi-supervised deep learning |
| CN112990374B (en) * | 2021-04-28 | 2023-09-15 | 平安科技（深圳）有限公司 | Image classification method, device, electronic equipment and medium |
| CN113221995B (en) * | 2021-04-30 | 2023-06-30 | 中南民族大学 | Data classification method, device and apparatus based on semi-supervised deep classification algorithm |
| CN113191332B (en) * | 2021-05-27 | 2023-06-27 | 安阳工学院 | Target detection method and device based on deep learning confidence estimation |
| CN113435488B (en) * | 2021-06-17 | 2023-11-07 | 深圳大学 | Image sampling probability improving method and application thereof |
| CN113420705B (en) * | 2021-07-02 | 2023-04-07 | 中国船舶重工集团公司第七二四研究所 | Semi-supervised signal analysis method based on bidirectional embedding characteristics |
| CN113408652B (en) * | 2021-07-12 | 2023-12-12 | 匀熵科技(无锡)有限公司 | Semi-supervised learning image classification method based on group representation features |
| CN114037868B (en) * | 2021-11-04 | 2022-07-01 | 杭州医策科技有限公司 | Image recognition model generation method and device |
| CN114612717B (en) * | 2022-03-09 | 2023-05-26 | 四川大学华西医院 | AI model training label generation method, training method, using method and equipment |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
| CN104794489A (en) * | 2015-04-23 | 2015-07-22 | 苏州大学 | Deep label prediction based inducing type image classification method and system |
| WO2016103651A1 (en) * | 2014-12-22 | 2016-06-30 | 日本電気株式会社 | Information processing system, information processing method and recording medium |
| CN106096557A (en) * | 2016-06-15 | 2016-11-09 | 浙江大学 | A kind of semi-supervised learning facial expression recognizing method based on fuzzy training sample |
| CN106874862A (en) * | 2017-01-24 | 2017-06-20 | 复旦大学 | People counting method based on submodule technology and semi-supervised learning |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9075824B2 (en) * | 2012-04-27 | 2015-07-07 | Xerox Corporation | Retrieval system and method leveraging category-level labels |
| US9589210B1 (en) * | 2015-08-26 | 2017-03-07 | Digitalglobe, Inc. | Broad area geospatial object detection using autogenerated deep learning models |
| CN107578060B (en) * | 2017-08-14 | 2020-12-29 | 电子科技大学 | Method for classifying dish images based on depth neural network capable of distinguishing areas |
-
2018
- 2018-02-07 CN CN201810132654.8A patent/CN108416370B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101853400A (en) * | 2010-05-20 | 2010-10-06 | 武汉大学 | Multiclass image classification method based on active learning and semi-supervised learning |
| CN104318242A (en) * | 2014-10-08 | 2015-01-28 | 中国人民解放军空军工程大学 | High-efficiency SVM active half-supervision learning algorithm |
| WO2016103651A1 (en) * | 2014-12-22 | 2016-06-30 | 日本電気株式会社 | Information processing system, information processing method and recording medium |
| CN104794489A (en) * | 2015-04-23 | 2015-07-22 | 苏州大学 | Deep label prediction based inducing type image classification method and system |
| CN106096557A (en) * | 2016-06-15 | 2016-11-09 | 浙江大学 | A kind of semi-supervised learning facial expression recognizing method based on fuzzy training sample |
| CN106874862A (en) * | 2017-01-24 | 2017-06-20 | 复旦大学 | People counting method based on submodule technology and semi-supervised learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108416370A (en) | 2018-08-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108416370B (en) | Image classification method and device based on semi-supervised deep learning and storage medium | |
| CN107526799B (en) | Knowledge graph construction method based on deep learning | |
| EP2806374B1 (en) | Method and system for automatic selection of one or more image processing algorithm | |
| CN108288051B (en) | Pedestrian re-recognition model training method and device, electronic equipment and storage medium | |
| US10963685B2 (en) | Generating variations of a known shred | |
| EP3048561A1 (en) | Method and system to perform text-to-image queries with wildcards | |
| EP2808827A1 (en) | System and method for OCR output verification | |
| US20140219563A1 (en) | Label-embedding for text recognition | |
| US11915500B2 (en) | Neural network based scene text recognition | |
| CN110413773B (en) | Intelligent text classification method, device and computer readable storage medium | |
| WO2020238353A1 (en) | Data processing method and apparatus, storage medium, and electronic apparatus | |
| CN111898550B (en) | Expression recognition model building method and device, computer equipment and storage medium | |
| JP2022141931A (en) | Method and device for training living body detection model, method and apparatus for living body detection, electronic apparatus, storage medium, and computer program | |
| CN113051914A (en) | Enterprise hidden label extraction method and device based on multi-feature dynamic portrait | |
| Tang et al. | Zero-shot learning by mutual information estimation and maximization | |
| CN110705490A (en) | Visual emotion recognition method | |
| Lv et al. | Chinese character CAPTCHA recognition based on convolution neural network | |
| CN110414622B (en) | Classifier training method and device based on semi-supervised learning | |
| Wang et al. | Improved knowledge distillation for training fast low resolution face recognition model | |
| CN113255557B (en) | Deep learning-based video crowd emotion analysis method and system | |
| CN114694150B (en) | Method and system for improving generalization capability of digital image classification model | |
| Kumar et al. | Extraction of texture and geometrical features from informative facial regions for sign language recognition | |
| CN114373097A (en) | Unsupervised image classification method, terminal equipment and storage medium | |
| Xue et al. | Incremental zero-shot learning based on attributes for image classification | |
| CN113505229B (en) | Entity relationship extraction model training method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |