CN115049876A - Image classification method and system based on discriminant atom embedded semi-supervised network - Google Patents
Image classification method and system based on discriminant atom embedded semi-supervised network Download PDFInfo
- Publication number
- CN115049876A CN115049876A CN202210667551.8A CN202210667551A CN115049876A CN 115049876 A CN115049876 A CN 115049876A CN 202210667551 A CN202210667551 A CN 202210667551A CN 115049876 A CN115049876 A CN 115049876A
- Authority
- CN
- China
- Prior art keywords
- discriminant
- atom
- image data
- classification
- semi
- 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
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- 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
-
- 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
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
- G06V10/7625—Hierarchical techniques, i.e. dividing or merging patterns to obtain a tree-like representation; Dendograms
-
- 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
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Databases & Information Systems (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the field of image classification, and provides an image classification method and system based on a discriminant atom embedded semi-supervised network, which comprises the steps of obtaining input image data; preprocessing is carried out based on input image data to obtain initialized discriminant atoms of each subclass; comparing and predicting the input image and the initialized discriminant atoms by using a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculating a subclass and a class corresponding to the maximum score as a prediction classification result; according to the method, a classical semi-supervised learning network mean-teacher is used as a basic framework, distinguishing characteristics of distinguishing atomic learning are introduced, a relational dual network structure is constructed, each sub-network comprises a characteristic extractor and a relational learning device, basic distinguishing atomic embedding and an algorithm of a relational dual semi-supervised learning idea can meet deep learning requirements of an image classification task.
Description
Technical Field
The invention belongs to the technical field of image classification systems, and particularly relates to an image classification method and system based on a discriminant atom embedded semi-supervised network.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
Image classification is an important field of computer vision, and has a wide range of application scenarios, such as identity verification, automatic disease diagnosis, target identification, and the like. The existing image classification method can obtain better performance to a certain extent, and can be divided into the following two methods according to the complexity of a model:
the method based on traditional shallow machine learning comprises the following steps:
conventional machine learning classification methods typically include a feature extraction module and a classifier module. The Feature extraction module may extract useful Feature information of texture features such as edges, colors, sizes, etc. of the target region using algorithms such as Principal Component Analysis (PCA), organized texture map (HOG), Scale-innovative Feature Transform (SIFT), etc., and the Classifier module may classify images using Bayesian Classifier, Random Forest, Support Vector Machine (SVM), etc. However, these methods often need to rely on the prior information inherent in the task, resulting in poor model robustness, and the algorithm performance depends on the characteristic information provided by the doctor.
The method based on deep learning comprises the following steps:
in recent years, deep learning has made a significant breakthrough in the task of image recognition and classification. In view of this, researchers have used deep learning in the field of image classification. The more classical depth classification network models are VGG, inclusion V3, Resnet, Densenet and the like. The method has a complex structure, and can learn more accurate feature representation based on big data, so that the method becomes a mainstream image classification method.
Compared with the traditional shallow method, the deep learning model can achieve larger improvement, but still faces larger challenges in certain tasks, and limits the improvement of classification performance. On one hand, certain tasks have the characteristics of small inter-class difference and large intra-class difference, and the existing deep learning method is difficult to learn distinctive information and reduces the classification precision. On the other hand, the labeled data in some tasks are very limited (such as medical image processing), so that the model is difficult to learn enough knowledge, and the generalization capability of the model is reduced.
Disclosure of Invention
In order to solve the problems, the invention provides an image classification method and system based on a discriminant atom embedded semi-supervised network, and in consideration of the fact that semi-supervised learning is an effective idea for solving the problem of limited labeled data, the invention takes a classic semi-supervised learning network mean-teacher as a basic framework, introduces discriminant atom learning discriminant features and constructs a relationship dual structure, and each sub-network comprises a feature extractor and a relationship learner. The relation learner can learn the relation characteristics of the input image characteristics and the discriminant atomic characteristics, and the representation capability of the characteristics is improved. In order to improve the distinctiveness of the features, a distinctiveness loss function is introduced, so that the distinguishing atoms have strong distinctiveness, and the distinctiveness of the learned features is improved.
According to some embodiments, a first aspect of the present invention provides an image classification method based on a discriminative atom embedding semi-supervised network, which adopts the following technical solutions:
the image classification method based on the discriminant atom embedded semi-supervised network comprises the following steps:
acquiring input image data;
preprocessing is carried out based on input image data to obtain initialized discriminant atoms of each subclass;
comparing and predicting the input image and the initialized discriminant atoms by using a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculating a subclass and a class corresponding to the maximum score as a prediction classification result;
the training process of the semi-supervised network image classification model comprises the following steps:
acquiring a sample image data set and a sample initialization discriminant atom;
randomly adding noise to the sample image data set to obtain a noise-added sample image data set;
inputting the sample image data set and the sample initialization discriminant atom into a student model for classification, and inputting the noise-added sample image data set and the sample initialization discriminant atom into a teacher model for classification;
and obtaining total loss based on the classification prediction probability, continuously training to enable the total loss to show a descending trend until the training turns reach a set value or the total loss shows a stable trend, and storing the network model when the minimum loss value is obtained so as to obtain the trained semi-supervised network image classification model.
Further, the acquiring process of the initialized discriminant atom includes:
acquiring an image data set for hierarchical clustering;
dividing the clustered image data set into K subclasses;
and acquiring the cluster center of each subclass as an initialized discriminant atom of each subclass.
Further, the sample image dataset comprises a labeled sample image dataset and an unlabeled sample image dataset;
the noisy sample image dataset comprises a labeled noisy sample image dataset and a non-labeled noisy sample image dataset.
Further, the classification process of the student model comprises the following steps:
performing feature extraction on the sample image dataset and the sample initialization discriminant atoms to obtain a sample image dataset vector and a discriminant atom vector;
performing discriminative learning on the sample image data set vector and the discriminative atom vector;
performing feature splicing on the sample image data set vector and the discriminant atom vector to obtain spliced total vector features;
and inputting the spliced total vector features into a relation learning device to obtain a relation score of the sample image data set vector and the discriminant atom vector, namely the prediction probability of each subclass of the sample image.
Further, the classification process of the teacher model comprises the following steps:
performing feature extraction on the image dataset of the noise-added sample and the initialized sample discriminative atom to obtain a vector of the image dataset of the noise-added sample and a vector of the discriminative atom;
performing feature splicing on the denoised sample image data set vector and the discriminant atom vector to obtain spliced total vector features;
and inputting the spliced total vector characteristics into a relation learning device to obtain a relation score of the noisy sample image data set and the discriminant atom vector, namely the prediction probability of each subclass of the noisy sample image.
Further, calculating the cross entropy loss between the student model and the real label type by using the prediction probability of the labeled sample image data in the student model;
calculating the fine-grained consistency loss of the student model and the teacher model by utilizing the subclass prediction probability of the unlabeled sample image data and the subclass prediction probability of the unlabeled noisy sample image data in the teacher model;
calculating the marking consistency loss of the student model and the teacher model by using the prediction probability of the unlabelled sample image data and the prediction probability of the unlabelled noisy sample image data in the teacher model;
calculating the discriminative loss of the image data set vector with the label and the discriminative atom vector in the student model;
and summing the cross entropy loss, the fine-grained consistency loss weighting, the marked consistency loss weighting and the discriminative loss weighting to obtain the total loss.
Furthermore, the feature extraction module for extracting features is composed of four convolutional layers and two shallow convolutional neural networks with maximum pooling layers;
the relationship learner is comprised of two volume blocks and two maximum pooling layers and two fully connected layers.
According to some embodiments, a second aspect of the present invention provides an image classification system based on a discriminative atom embedding semi-supervised network, which adopts the following technical solutions:
the image classification system based on the discriminant atom embedded semi-supervised network comprises:
a data acquisition module configured to acquire input image data;
a data processing module configured to perform preprocessing based on input image data to obtain initialized discriminant atoms for each subclass;
the image classification module is configured to compare and predict the input image and the initialized discriminant atoms by utilizing a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculate a subclass and a class corresponding to the maximum score as a prediction classification result;
the training process of the semi-supervised network image classification model comprises the following steps:
acquiring a sample image data set and a sample initialization discriminant atom;
randomly adding noise to the sample image data set to obtain a noise-added sample image data set;
inputting the sample image data set and the sample initialization discriminant atom into a student model for classification, and inputting the noise-added sample image data set and the sample initialization discriminant atom into a teacher model for classification;
and obtaining total loss based on the classification prediction probability, continuously training to enable the total loss to show a descending trend until the training turns reach a set value or the total loss shows a stable trend, and storing the network model when the minimum loss value is obtained so as to obtain the trained semi-supervised network image classification model.
According to some embodiments, a third aspect of the invention provides a computer-readable storage medium.
A computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the steps in the discrimination atom-based embedding semi-supervised network image classification method as described in the first aspect above.
According to some embodiments, a fourth aspect of the invention provides a computer apparatus.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps of the discrimination atom embedding semi-supervised network based image classification method as described in the first aspect above when executing the program.
Compared with the prior art, the invention has the beneficial effects that:
the invention provides an image classification method and system based on a semi-supervised network with embedded discriminant atoms. On one hand, the method introduces discriminant atoms as class template ideas of each subclass, introduces the relational features of the dual-network learning input images and the atoms, improves the representation of learned features, introduces discriminant loss, further improves the discriminant of the learned features, and solves the problem that the classification precision is low due to the fact that the similarity in the classes with large similarity among the classes is small due to the lack of the feature discriminant.
On the other hand, the invention utilizes the semi-supervised learning idea aiming at the characteristic of difficult acquisition of labeled data in the image classification field, can complete the deep learning training process by using a small amount of labeled data, and has good learning effect. The algorithm of the semi-supervised learning idea of basic discriminant atom embedding and relationship duality can meet the deep learning requirement of an image classification task. The idea can also be applied to other fields with the problem to help other fields to better complete deep learning tasks.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate exemplary embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
FIG. 1 is a flowchart of an image classification method based on discriminative atom embedding semi-supervised network according to an embodiment of the present invention;
fig. 2 is a schematic diagram of a semi-supervised network image classification model framework according to an embodiment of the present invention.
Detailed Description
The invention is further described with reference to the following figures and examples.
It is to be understood that the following detailed description is exemplary and is intended to provide further explanation of the invention as claimed. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the invention. As used herein, the singular forms "a", "an" and "the" are intended to include the plural forms as well, and it should be understood that when the terms "comprises" and/or "comprising" are used in this specification, they specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof, unless the context clearly indicates otherwise.
The embodiments and features of the embodiments of the present invention may be combined with each other without conflict.
Example one
As shown in fig. 1-2, the present embodiment provides an image classification method based on a discriminative atom embedding semi-supervised network, including:
acquiring input image data;
preprocessing is carried out based on input image data to obtain initialized discriminant atoms of each subclass;
comparing and predicting the input image and the initialized discriminant atoms by using a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculating a subclass and a class corresponding to the maximum score as a prediction classification result;
the training process of the semi-supervised network image classification model comprises the following steps:
acquiring a sample image data set and a sample initialization discriminant atom;
randomly adding noise to the sample image data set to obtain a noise-added sample image data set;
inputting the sample image data set and the sample initialization discriminant atom into a student model for classification, and inputting the noise-added sample image data set and the sample initialization discriminant atom into a teacher model for classification;
and obtaining total loss based on the classification prediction probability, continuously training to enable the total loss to show a descending trend until the training turns reach a set value or the total loss shows a stable trend, and storing the network model when the minimum loss value is obtained so as to obtain the trained semi-supervised network image classification model.
As shown in fig. 1, the method of this embodiment includes the following steps:
step S1: acquiring image data and initializing discriminative atoms, comprising:
step S1.1: the image data set is first hierarchically clustered. Hierarchical clustering is a common unsupervised clustering method, similarity is measured by calculating the distance between data points, two data points with the closest distance are merged to generate a nested clustering tree, and a tree diagram can obviously show the hierarchical structure of the nested clustering tree. Common distance measurement methods include: manhattan distance, Euclidean distance, Chebyshev distance. The invention utilizes a hierarchical clustering algorithm based on Euclidean distance to cluster the labeled data sets.
Where x, y are each points of an n-dimensional space.
Because image categories in some tasks have the characteristics of small inter-class difference and large intra-class difference, in order to enable a model to distinguish CNV types more easily, the method divides a clustered data set into K sub-classes, and selects related images as initialized discriminant atoms.
K＝mC，m>1&& m∈N + (2)
Wherein K is the number of the types of the scratch molecules, C is the number of the types of the original data, and m is any natural number more than one.
Each class is divided into m subclasses, so that class balance is guaranteed, and the distinguishing performance of characteristics is improved, so that the subclasses can be distinguished by the model more easily.
Step S1.2: discriminative atomic initialization
And selecting a proper image as an initialized discriminant atom according to the clustering result and the dividing result in the step S1.1. The discriminant atom is the class center of the subclass and represents the characteristics of the subclass. The initial discriminative atom is obtained from the cluster center of the hierarchical cluster. The cluster center of each subclass represents the discriminative atom of each subclass, respectively.
In addition, randomly adding noise to the image data set to obtain a noise-added image data set;
the image dataset comprises a tagged image dataset and an untagged image dataset;
the noisy image dataset comprises a tagged noisy image dataset and a non-tagged noisy image dataset.
Step S2: the method comprises the following steps of predicting by using a pre-trained semi-supervised network image classification model to obtain a comparison score of each class, and calculating the class corresponding to the maximum score as a prediction classification result, wherein the method comprises the following steps:
step S2.1: feature extraction
For the image data obtained in step S1, the key feature information vector of the image is extracted by the feature extraction module f (x, phi) (where x is the input vector and phi is the model parameter) to obtain the image data set vector, and the discriminative atomic vector is also obtained
(where K represents the number of subcategories).
The tagged image dataset vector is
(where n represents the number of labeled data) and unlabeled image dataset vector of
(where N-N represents the amount of unlabeled data).
The feature extraction module is composed of four convolutional layers and two shallow convolutional neural networks with maximum pooling layers. The feature vectors are extracted from the same network model and thus belong to the same feature space. In the process, a Student (Student) model and a Teacher (Teacher) model are carried out simultaneously, the feature extraction modules are the same, and input data are different. The image input by the Student (Student) model is an original image, and the image input by the Teacher (Teacher) model is a randomly-noisy image.
And carrying out random noise addition operation on the labeled data and the unlabeled data, wherein the noise comprises the random combination of four modes of displacement, image brightness change, contrast and saturation. The change values of the displacement value, the image brightness, the contrast and the saturation all adopt random numbers in a certain range.
Step S2.2: feature stitching
Vector X of data set obtained in S2.1 l 、X u And respectively splicing the discriminant atom vector V, namely splicing each input data set vector with the discriminant atom vector. The purpose of splicing is to more conveniently learnThe relationship between the two is studied. The Student (Student) model and Teacher (Teacher) model for this step are the same.
Step S2.3: discriminative learning
And performing differentiated learning on the input image and the initialized discriminant atoms after passing through a feature extraction module of the network. And dynamically adjusting the discriminant atoms to be the centers of the feature vectors corresponding to the labeled subclasses, and utilizing discriminability loss to shorten the distance between the labeled samples and the corresponding discriminant atoms so as to further enhance the discriminability of the features. The distinguishing atom is an important basis for learning and classifying the relation learner, and directly influences the performance of the model. This step was only performed on Student (Student) models.
Step S2.4: relationship learning
The relationship learner is comprised of two volume blocks and two max pooling layers and two full connections. And inputting the total vector after feature splicing into a relation learning device g (y, phi), so that a relation score of the vector of the input image and each judgment atom vector can be obtained, namely the prediction probability of each sub-category of the input image is obtained. Knowing the subcategories and the affiliations of the subcategories, the prediction probability of the subcategories can be used for obtaining the prediction probability of the subcategories, and the Student (Student) model and the Teacher (Teacher) model are the same. The output prediction probability has not only the result of the presence of the tag data but also the result of the absence of the tag data.
Step S3: acquiring the total loss of the semi-supervised network image classification model, comprising the following steps:
step S3.1: calculating cross entropy loss
Tagged data X with all Student modules l Predicting probability, calculating it and true label class y l A cross entropy loss is performed.
Wherein, y i Is x i Theta is a parameter of the Student model, f θ Is the Student model. The classification loss can effectively ensure correct parameters of network learning, and the labeled parameters are utilizedThe data constrains the network.
Step S3.2: calculating fine-grained consistency loss
Unlabeled data X using all Student modules u Subclass of prediction probabilities and unlabeled data X of all Teacher modules u The probability of the subclass prediction of (2) and the loss of consistency of the two are calculated.
Wherein the content of the first and second substances,
″
is the network model that outputs the subclass results, θ is the parameter of the Teacher model, and η are random noise. The fine-grained consistency loss is a main innovation point provided by the invention, and beneficial information can be strongly restricted from being mined from the non-label data by the network if the subclass results of the non-label data predictions output by the two branches of the network are kept consistent as much as possible.
Step S3.3: computing tag consistency loss
Unlabeled data X using all Student modules u Prediction probability of (3) and unlabeled data X of all Teacher modules u And calculating the loss of consistency between the two.
The consistency loss of the mark is to restrict the classification result output by two branches of the network, and the information can be obtained from the label-free data by the weak restriction network.
Step S3.4: calculating discriminability loss
Tagged data X with Student module l And a discriminative atom vector V, calculating the loss of both.
Wherein the content of the first and second substances,
is x i The corresponding sub-class label category of the sub-class label,
is that
Discriminant atoms of the corresponding class. The discriminative loss is mainly used for restricting a discriminative learning module of the Student branch in the network and enhancing the discriminative performance.
Step S3.5: joint training
Loss of cross entropy Loss c Loss of fine grain consistency Loss FJ Weighted, mark Loss of consistency Loss J Weighted and discriminative Loss v Weighting together as total Loss c +λLoss FJ +ωLoss J +μLoss v And (wherein lambda, omega and mu are hyper-parameters), continuously training to enable Loss to show a descending trend until the training turns reach a set value or the Loss shows a steady trend, and storing the network model with the minimum Loss value as a trained semi-supervised network image classification model.
As shown in fig. 1, the system in the dashed box corresponding to the figure is a system module that mainly performs a classification function, wherein the feature vector module uses the network model f (x, Φ) in step S2.1, and the score-for-contrast module uses the network model g (y, Φ) in step S2.4. They have been trained to determine the appropriate network parameters φ, φ.
The user inputs image data to be tested into the classification system, three processes of feature vector extraction, comparison score acquisition and prediction category calculation are automatically carried out in the classification system, and then the prediction category is output to interact with the user.
Example two
The embodiment provides an image classification system based on the embedding of discriminant atoms into a semi-supervised network,
a data acquisition module configured to acquire input image data;
a data processing module configured to perform preprocessing based on input image data to obtain initialized discriminant atoms for each subclass;
the image classification module is configured to compare and predict the input image and the initialized discriminant atoms by utilizing a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculate a subclass and a class corresponding to the maximum score as a prediction classification result;
the training process of the semi-supervised network image classification model comprises the following steps:
acquiring a sample image data set and a sample initialization discriminant atom;
randomly adding noise to the sample image data set to obtain a noise-added sample image data set;
inputting the sample image data set and the sample initialization discriminant atom into a student model for classification, and inputting the noise-added sample image data set and the sample initialization discriminant atom into a teacher model for classification;
and obtaining total loss based on the classification prediction probability, continuously training to enable the total loss to show a descending trend until the training turns reach a set value or the total loss shows a stable trend, and storing the network model when the minimum loss value is obtained so as to obtain the trained semi-supervised network image classification model.
The modules are the same as the corresponding steps in the implementation example and application scenarios, but are not limited to the disclosure of the first embodiment. It should be noted that the modules described above as part of a system may be implemented in a computer system such as a set of computer-executable instructions.
In the foregoing embodiments, the descriptions of the embodiments have different emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
The proposed system can be implemented in other ways. For example, the above-described system embodiments are merely illustrative, and for example, the division of the above-described modules is merely a logical division, and in actual implementation, there may be other divisions, for example, multiple modules may be combined or integrated into another system, or some features may be omitted, or not executed.
EXAMPLE III
The present embodiment provides a computer-readable storage medium, on which a computer program is stored, which when executed by a processor implements the steps in the discrimination atom embedding-based semi-supervised network image classification method as described in the first embodiment above.
Example four
The embodiment provides a computer device, which includes a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the image classification method based on the discriminant atom embedded semi-supervised network as described in the first embodiment.
As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above may be implemented by a computer program, which may be stored in a computer readable storage medium and executed by a computer to implement the processes of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), or the like.
Although the embodiments of the present invention have been described with reference to the accompanying drawings, it is not intended to limit the scope of the present invention, and it should be understood by those skilled in the art that various modifications and variations can be made without inventive efforts by those skilled in the art based on the technical solution of the present invention.
Claims (10)
1. The image classification method based on the discriminant atom embedded semi-supervised network is characterized by comprising the following steps of:
acquiring input image data;
preprocessing is carried out based on input image data to obtain initialized discriminant atoms of each subclass;
comparing and predicting the input image and the initialized discriminant atoms by using a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculating a subclass and a class corresponding to the maximum score as a prediction classification result;
the training process of the semi-supervised network image classification model comprises the following steps:
acquiring a sample image data set and a sample initialization discriminant atom;
randomly adding noise to the sample image data set to obtain a noise-added sample image data set;
inputting the sample image data set and the sample initialization discriminant atom into a student model for classification, and inputting the noise-added sample image data set and the sample initialization discriminant atom into a teacher model for classification;
and obtaining total loss based on the classification prediction probability, continuously training to enable the total loss to show a descending trend until the training turns reach a set value or the total loss shows a stable trend, and storing the network model when the minimum loss value is obtained so as to obtain the trained semi-supervised network image classification model.
2. The image classification method based on the discriminant atom embedded semi-supervised network as claimed in claim 1, wherein the process of acquiring the initialized discriminant atom is as follows:
acquiring an image data set for hierarchical clustering;
dividing the clustered image data set into K subclasses;
and acquiring the cluster center of each subclass as an initialized discriminant atom of each subclass.
3. The discriminative atom-embedding semi-supervised network-based image classification method of claim 1, wherein the sample image dataset comprises a labeled sample image dataset and an unlabeled sample image dataset;
the noisy sample image dataset comprises a labeled noisy sample image dataset and a non-labeled noisy sample image dataset.
4. The image classification method based on the discriminative atom embedded semi-supervised network as claimed in claim 3, wherein the classification process of the student model comprises:
performing feature extraction on the sample image dataset and the sample initialization discriminant atoms to obtain a sample image dataset vector and a discriminant atom vector;
performing discriminative learning on the sample image data set vector and the discriminative atom vector;
performing feature splicing on the sample image data set vector and the discriminant atom vector to obtain spliced total vector features;
and inputting the spliced total vector features into a relation learning device to obtain a relation score of the sample image data set vector and the discriminant atom vector, namely the prediction probability of each subclass of the sample image.
5. The image classification method based on the discriminative atom embedded semi-supervised network as claimed in claim 3, wherein the classification process of the teacher model comprises:
performing feature extraction on the image dataset of the noise-added sample and the initialized sample discriminative atom to obtain a vector of the image dataset of the noise-added sample and a vector of the discriminative atom;
performing feature splicing on the noisy sample image data set vector and the discriminant atom vector to obtain spliced total vector features;
and inputting the spliced total vector characteristics into a relation learning device to obtain a relation score of the noisy sample image data set and the discriminant atom vector, namely the prediction probability of each subclass of the noisy sample image.
6. The image classification method based on the discriminative atom embedded semi-supervised network as claimed in claim 4 or 5, wherein the cross entropy loss with the real label class is calculated by using the prediction probability of the labeled sample image data in the student model;
calculating the fine-grained consistency loss of the student model and the teacher model by utilizing the subclass prediction probability of the unlabeled sample image data and the subclass prediction probability of the unlabeled noisy sample image data in the teacher model;
calculating the marking consistency loss of the student model and the teacher model by using the prediction probability of the unlabelled sample image data and the prediction probability of the unlabelled noisy sample image data in the teacher model;
calculating the discriminative loss of the image data set vector with the label and the discriminative atom vector in the student model;
and summing the cross entropy loss, the fine-grained consistency loss weighting, the marked consistency loss weighting and the discriminative loss weighting to obtain the total loss.
7. The image classification method based on the discriminative atom-embedded semi-supervised network as claimed in claim 4 or 5, wherein the feature extraction module for feature extraction is composed of four convolutional layers and two shallow convolutional neural networks of maximum pooling layer;
the relationship learner is comprised of two volume blocks and two maximum pooling layers and two fully connected layers.
8. Image classification system based on discriminative atom embedding semi-supervised network is characterized by comprising:
a data acquisition module configured to acquire input image data;
a data processing module configured to perform preprocessing based on input image data to obtain initialized discriminant atoms for each subclass;
the image classification module is configured to compare and predict the input image and the initialized discriminant atoms by utilizing a pre-trained semi-supervised network image classification model to obtain a comparison score of each subclass, and calculate a subclass and a class corresponding to the maximum score as a prediction classification result;
the training process of the semi-supervised network image classification model comprises the following steps:
acquiring a sample image data set and a sample initialization discriminant atom;
randomly adding noise to the sample image data set to obtain a noise-added sample image data set;
inputting the sample image data set and the sample initialization discriminant atom into a student model for classification, and inputting the noise-added sample image data set and the sample initialization discriminant atom into a teacher model for classification;
and obtaining total loss based on the classification prediction probability, continuously training to enable the total loss to show a descending trend until the training turns reach a set value or the total loss shows a stable trend, and storing the network model when the minimum loss value is obtained so as to obtain the trained semi-supervised network image classification model.
9. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for classifying an image based on the embedding of discriminative atoms into a semi-supervised network according to any one of claims 1 to 7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor when executing the program implements the steps in the method for image classification based on the embedding of discriminative atoms into semi-supervised networks according to any of claims 1 to 7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210667551.8A CN115049876A (en) | 2022-06-14 | 2022-06-14 | Image classification method and system based on discriminant atom embedded semi-supervised network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210667551.8A CN115049876A (en) | 2022-06-14 | 2022-06-14 | Image classification method and system based on discriminant atom embedded semi-supervised network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115049876A true CN115049876A (en) | 2022-09-13 |
Family
ID=83161794
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210667551.8A Pending CN115049876A (en) | 2022-06-14 | 2022-06-14 | Image classification method and system based on discriminant atom embedded semi-supervised network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115049876A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115272777A (en) * | 2022-09-26 | 2022-11-01 | 山东大学 | Semi-supervised image analysis method for power transmission scene |
| CN115311605A (en) * | 2022-09-29 | 2022-11-08 | 山东大学 | Semi-supervised video classification method and system based on neighbor consistency and contrast learning |
| CN116385809A (en) * | 2023-06-05 | 2023-07-04 | 山东第一医科大学附属省立医院(山东省立医院) | MRI brain tumor classification method and system based on semi-supervised learning |
-
2022
- 2022-06-14 CN CN202210667551.8A patent/CN115049876A/en active Pending
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115272777A (en) * | 2022-09-26 | 2022-11-01 | 山东大学 | Semi-supervised image analysis method for power transmission scene |
| CN115272777B (en) * | 2022-09-26 | 2022-12-23 | 山东大学 | Semi-supervised image analysis method for power transmission scene |
| CN115311605A (en) * | 2022-09-29 | 2022-11-08 | 山东大学 | Semi-supervised video classification method and system based on neighbor consistency and contrast learning |
| CN116385809A (en) * | 2023-06-05 | 2023-07-04 | 山东第一医科大学附属省立医院(山东省立医院) | MRI brain tumor classification method and system based on semi-supervised learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Galleguillos et al. | Context based object categorization: A critical survey | |
| Joseph et al. | Facial emotion detection using modified eyemap–mouthmap algorithm on an enhanced image and classification with tensorflow | |
| CN115049876A (en) | Image classification method and system based on discriminant atom embedded semi-supervised network | |
| Bougourzi et al. | Deep learning based face beauty prediction via dynamic robust losses and ensemble regression | |
| Guo et al. | Facial expression recognition using ELBP based on covariance matrix transform in KLT | |
| Zou et al. | Chronological classification of ancient paintings using appearance and shape features | |
| Liu et al. | Facial age estimation using a multi-task network combining classification and regression | |
| Singh et al. | Image Understanding-a Brief Review of Scene Classification and Recognition. | |
| Tan et al. | Rapid fine-grained classification of butterflies based on FCM-KM and mask R-CNN fusion | |
| Kamarajugadda et al. | Extract features from periocular region to identify the age using machine learning algorithms | |
| Padala et al. | Effect of input noise dimension in GANs | |
| Yu et al. | Fine-grained comparisons with attributes | |
| Lu et al. | A survey of object co-segmentation | |
| Puthenputhussery et al. | Sparse representation based complete kernel marginal fisher analysis framework for computational art painting categorization | |
| Zhang et al. | Facial beauty analysis based on features prediction and beautification models | |
| Zhao et al. | Multi-view dimensionality reduction via subspace structure agreement | |
| Padala et al. | Effect of input noise dimension in GANs | |
| Shen et al. | Face identification using reference-based features with message passing model | |
| Manisha et al. | Effect of input noise dimension in gans | |
| Liang et al. | Random forest with suppressed leaves for Hough voting | |
| Zhou et al. | Comparative analysis of the performance of complex texture clustering driven by computational intelligence methods using multiple clustering models | |
| CN112241680A (en) | Multi-mode identity authentication method based on vein similar image knowledge migration network | |
| Iqbal et al. | Effect of Laplacian smoothing stochastic gradient descent with angular margin softmax loss on face recognition | |
| Lou et al. | Learning perceptual texture similarity and relative attributes from computational features | |
| Dharavath et al. | A Parallel Deep Learning Approach for Age Invariant Face Recognition system |
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 |