CN115017351B - Lightweight industrial picture classification method and system based on federal small sample learning - Google Patents
Lightweight industrial picture classification method and system based on federal small sample learning Download PDFInfo
- Publication number
- CN115017351B CN115017351B CN202210611106.XA CN202210611106A CN115017351B CN 115017351 B CN115017351 B CN 115017351B CN 202210611106 A CN202210611106 A CN 202210611106A CN 115017351 B CN115017351 B CN 115017351B
- Authority
- CN
- China
- Prior art keywords
- small sample
- classifier
- training
- industrial
- federal
- 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
- 238000000034 method Methods 0.000 title claims abstract description 68
- 238000012549 training Methods 0.000 claims abstract description 82
- 230000008520 organization Effects 0.000 claims description 17
- 238000005516 engineering process Methods 0.000 claims description 8
- 230000006870 function Effects 0.000 claims description 8
- 238000005094 computer simulation Methods 0.000 claims description 6
- 238000011176 pooling Methods 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 239000011159 matrix material Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000004891 communication Methods 0.000 description 6
- 238000013527 convolutional neural network Methods 0.000 description 4
- 238000002955 isolation Methods 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/50—Information retrieval; Database structures therefor; File system structures therefor of still image data
- G06F16/55—Clustering; Classification
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Image Analysis (AREA)
Abstract
Lightweight industrial picture classification method and system based on federal small sample learning, comprising the following steps: modeling an industrial picture classification system based on a federal small sample learning framework; task of classifying small sample picturesModeling is carried out; constructing a small sample classifier f based on a lightweight relational network, which consists of an embedding module l θ and a relational module g φ; task for classifying pictures based on small samplesConstructing a training method of a small sample classifier f; based on a training method of the small sample classifier, constructing a side cloud cooperative deployment and training method of the small sample classifier; the method for collaborative deployment and training of the edge cloud based on the small sample classifier carries out model reasoning locally. The invention fully utilizes heterogeneous and mutually isolated limited sample sets of different industrial institutions to realize effective training of the overall small sample picture classifier, is also suitable for external clients with samples which are not seen during training, and supports the industrial institutions with limited samples to establish a data-oriented picture classifier.
Description
Technical Field
The invention belongs to the field of distributed intelligence, and particularly relates to a lightweight industrial picture classification method and system based on federal small sample learning.
Background
Currently, image classification based on deep convolutional neural networks (Deep Convolutional Neural Network, DCNN) is widely used in different industrial scenarios. In the existing method, a DCNN classifier with intensive computation is deployed on a cloud end, and each industrial organization uploads an original picture for classification. However, this approach may reveal the privacy of the industrial institution. In addition, cloud-on-large-scale industrial pictures bring about a high network communication burden. In order to solve the problem, the federal learning technology is paid general attention to, and for the problem of classifying industrial pictures, different industrial institutions can be used as clients to uniformly coordinate and cooperatively train the global DCNN classifier by the cloud without uploading the original pictures. However, the sample scarcity problem faced by the actual industrial institution (including insufficient sample numbers and sample class isolation) limits the application of existing data-intensive federal learning methods in industrial picture classification scenarios. Aiming at the problem of sample scarcity, each industrial organization can adopt the existing single-machine small sample learning technology to train to obtain a small sample picture classifier on the local basis of limited data, but due to strict data isolation among the industrial organizations, a model obtained by training cannot be generalized on other types of samples, and in this case, the industrial organization with extremely limited samples cannot establish the available picture classifier. In addition, practical industrial institutions often have heterogeneous and limited computing and network resources that cannot effectively support existing resource-intensive DCNN picture classification methods.
Disclosure of Invention
The invention aims to provide a lightweight industrial picture classification method and system based on federal small sample learning, which are used for solving the problems that an industrial organization with extremely limited samples cannot establish a usable picture classifier, and an actual industrial organization usually has heterogeneous and limited computing and network resources and cannot effectively support the existing resource-intensive DCNN picture classification method.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
the lightweight industrial picture classification method based on federal small sample learning comprises the following steps:
Modeling an industrial picture classification system based on a federal small sample learning framework, wherein the system comprises a cloud server c and a group of industrial institutions epsilon with the same interests, and each e i epsilon independently holds a local sample set
For each industrial organization in the constructed system model, classifying tasks for small sample picturesModeling, each task contains a support set/>And query set/>
Constructing a small sample classifier f based on a lightweight relational network, which consists of an embedding module l θ and a relational module g φ, wherein θ and phi are parameters of functions l and g;
Task for classifying pictures based on small samples Constructing a training method of a small sample classifier f;
Constructing an edge cloud collaborative deployment and training method of a small sample classifier based on a small sample classifier training method, wherein a complete small sample picture classifier based on a lightweight relational network is deployed on c, an embedded module is deployed on all e i epsilon, and training is carried out on the classifier along discrete time points t=0, 1, … and T based on the training method in the step 4, wherein T is the preset model iteration times;
small sample classifier based edge cloud collaborative deployment and training method, deployment and training of constructed small sample classifier After training is finished, each external client/>, is initializedModel/>Model reasoning is performed locally.
Further, the industrial picture classification system is modeled by utilizing the federal small sample learning technology, privacy of each industrial organization is guaranteed, meanwhile, global small sample picture classifier training of industrial picture islands which are cross-heterogeneous and limited in samples is supported, and the obtained global picture classifier is not only suitable for federal model training participation clients, but also suitable for external clients without class samples during the holding training period.
Further, the specific operation is as follows: industrial picture classification system modeling based on federal small sample learning, the system comprising a cloud server c and a set of industrial institutions epsilon=e i of equal interest, wherein Each e i e epsilon holds its own local sample set/>, separatelyWherein e i holds/>A class sample is used to determine the class of samples, For sample j on e i, i+.j,/>, for all e i,ej εC can receive the message sent by epsilon but not access the sample set, epsilon can receive the message sent by c but the epsilon cannot be communicated, and a federal small sample picture classifier is arranged to train the participating client set/>External client set/>
Further, the specific operation of modeling the small sample picture classification task T is: c-way K-shot picture classification task on classifier fContaining support sets/>And query set/> Containing C x K tagged samples,/>Containing C x K' samples to be classified (, requiring f to be based on/>)Pair/>Classification is performed.
Further, a small sample classifier f based on a lightweight relational network is constructed by using a relational network technology.
Further, the specific operation is as follows: constructing a multi-layer CNN model f, which consists of an embedding module l θ and a relation module g φ, wherein theta and phi are parameters of functions l and g, and l θ is used for extracting an input image sample feature map, and comprises two convolution blocks, namely a convolution layer containing 32 3×3 filters, a batch normalization layer, a ReLU nonlinear layer and a2×2 maximum pooling layer; g φ determines whether two input samples belong to the same class based on the similarity of the feature map, and the output size H of the last maximum pooling layer is 32 and 32 multiplied by 3 for single and three-channel images, wherein the two input samples comprise the two convolution blocks, an 8-dimensional full-connection layer based on ReLU and a 1-dimensional full-connection layer based on Sigmoid.
Further, the training method of the small sample classifier f comprises the following specific operations: small sample picture classification task based onPair of arbitrary samples/>Obtaining a feature map l θ(xa)、lθ(xb based on the embedding module), calculating a relationship score/>, based on the relationship moduleWherein/>For feature graph connection operation, the relation matrix is adopted to organize the relation scores of all sample pairs, and the classifier loss is achievedWherein:
the model parameters θ *,φ* are updated by random gradient descent.
Furthermore, a side cloud collaborative deployment and training method of the small sample picture classifier based on the lightweight relational network is constructed by utilizing a side cloud collaborative intelligent technology.
Further, the method comprises the following steps:
step 5.1, setting the current iteration times t≡0, and initializing a cloud model Wherein θ init、φinit is a random initialization parameter, initializing each participating client/>Model parameters of (2)Each participating client e i∈εi organizes the local task set/>Updating the current iteration times t≡1;
Step 5.2, each participating client e i∈εi slave Random extraction task/>Obtaining all pairs of samplesFeature map/>Uploading all feature graphs and corresponding sample labels to c;
Step 5.3, for each participating client e i∈εi, calculate based on step 4, c UpdatingWherein α is a random gradient descent learning rate, returning θ g (t+1) to each participating client e i∈εi;
step 5.4, update per participating client e i∈εi
Step 5.5 the current iteration number T _ T +1 is updated, if T is less than or equal to T, repeating the steps 5.2 to 5.4;
step 5.6, outputting training results
Further, a lightweight industrial picture classification system based on federal small sample learning, comprising:
the system modeling module is used for modeling an industrial picture classification system based on a federal small sample learning framework, the system comprises a cloud server c and a group of industrial institutions epsilon with the same interests, and each e i epsilon independently holds a local sample set
A small sample picture classification task modeling module for classifying tasks of small sample pictures for each industrial organization in the constructed system modelModeling, each task contains a support set/>And query set/>
The classifier construction module is used for constructing a small sample classifier f based on a lightweight relational network, which consists of an embedding module l θ and a relational module g φ, wherein θ and phi are parameters of functions l and g;
training module for classifying tasks based on small sample pictures Constructing a training method of a small sample classifier f; constructing an edge cloud collaborative deployment and training method of a small sample classifier based on a small sample classifier training method, wherein a complete small sample picture classifier based on a lightweight relational network is deployed on c, an embedded module is deployed on all e i epsilon, and training is carried out on the classifier along discrete time points t=0, 1, … and T based on the training method in the step 4, wherein T is the preset model iteration times;
The output reasoning module is used for deploying and training the constructed small sample classifier based on the edge cloud collaborative deployment and training method of the small sample classifier After training is finished, each external client/>, is initializedModel/>Model reasoning is performed locally.
Compared with the prior art, the invention has the following technical effects:
According to the invention, by establishing the industrial picture classification system model based on federal small sample learning, heterogeneous and mutually isolated limited sample sets of different industrial institutions are fully utilized while the privacy of the industrial institutions is not violated, so that the effective training of the global small sample picture classifier is realized, the obtained global model is not only suitable for federal model training participation clients, but also suitable for external clients without class samples during the holding training period, and the industrial institutions supporting the limitation of the samples establish a data-oriented picture classifier.
Further, by constructing a small sample picture classifier based on a lightweight relational network, the calculation and storage cost of an industrial organization with limited resources in the model training process is obviously reduced by utilizing a more lightweight model structural design while the model classification precision is ensured.
Further, by establishing the edge cloud collaborative deployment and training method of the small sample picture classifier based on the lightweight relational network, communication lightweight embedded modules are deployed only at the client side of the industrial institution while the splitting precision of the model is ensured, and the communication overhead of the industrial institution with limited resources in the model training process is obviously reduced.
Drawings
FIG. 1 is a schematic representation of the practice of the process herein;
FIG. 2 is a logic flow diagram of the method herein;
Detailed Description
The invention is described in further detail below with reference to the accompanying drawings.
Referring to fig. 1, the invention provides a lightweight industrial picture classification method based on federal small sample learning, comprising the following steps:
Step 1, modeling a federal small sample industrial picture classification system, the system comprising a cloud server c and a set of industry institutions (federal learning clients) epsilon=e i of equal interest (desiring co-training of a shared global picture classifier f g), wherein Each e i epsilon independently holds a respective local sample setWherein e i holds/>Class sample,/> For the j-th sample on e i, i+.epsilon.j,/>, for all e i,ej ε, i+.epsilon.j due to class isolation between actual industrial institutionsC can receive the message sent by epsilon but not access the sample set, epsilon can receive the message sent by c but the epsilon cannot be communicated, and a federal small sample picture classifier is arranged to train the participating client set/>External client set/>
Step 2, modeling a small sample picture classification task, and performing C-way K-shot picture classification task on a classifier fContaining support sets/>And query set/> Containing C x K tagged samples (K samples for each of class C)/>Containing C x K 'samples to be classified (K' samples from each of the above C classes), requiring f to be based on/>Pair/>Classifying;
Step 3, a small sample picture classifier based on a lightweight relational network is a multi-layer CNN model f and consists of an embedding module l θ and a relational module g φ, wherein θ and φ are parameters of functions l and g, l θ extracts an input image sample feature map and comprises two convolution blocks, wherein each convolution block comprises a convolution layer containing 32 3×3 filters, a batch normalization layer, a ReLU nonlinear layer and a 2×2 maximum pooling layer; g φ determines whether two input samples belong to the same class based on the similarity (relation score) of the feature map, wherein the two input samples comprise the two convolution blocks, an 8-dimensional full-connection layer based on ReLU and a 1-dimensional full-connection layer based on Sigmoid, and for single-channel and three-channel images, the output size H of the last maximum pooling layer is 32 and 32 multiplied by 3;
Step 4, training a small sample picture classifier based on a lightweight relational network, and classifying tasks based on the small sample pictures in step 2 Pair of arbitrary samples/>Obtaining a feature map l θ(xa)、lθ(xb based on the embedding module), calculating a relationship score/>, based on the relationship moduleWherein/>For feature graph connection operation, the relation matrix is adopted to organize the relation scores of all sample pairs, and the classifier loss is achievedWherein:
updating the model parameter theta *,φ* through random gradient descent;
Step 5, edge cloud collaborative deployment and training of a small sample picture classifier based on a lightweight relational network, in order to reduce calculation, communication and storage costs of a client in the federal model training process, deploying the complete small sample picture classifier based on the lightweight relational network on c, deploying an embedded module on all e i epsilon, and training the classifier along discrete time points t=0, 1, …, wherein T is a preset model iteration number, and step 5 comprises the following steps:
step 5.1, setting the current iteration times t≡0, and initializing a cloud model Wherein θ init、φinit is a random initialization parameter, initializing each participating client/>Model parameters of (2)Each participating client e i∈εi organizes the local task set/>Updating the current iteration times t≡1;
Step 5.2, each participating client e i∈εi slave Random extraction task/>Based on step 4, all pairs of samples were obtained/>Feature map/>Uploading all feature graphs and corresponding sample labels to c;
Step 5.3, for each participating client e i∈εi, calculate based on step 4, c UpdatingWherein α is a random gradient descent learning rate, returning θ g (t+1) to each participating client e i∈εi;
step 5.4, update per participating client e i∈εi
Step 5.5 the current iteration number T _ T +1 is updated, if T is less than or equal to T, repeating the steps 5.2 to 5.4;
step 5.6, outputting training results
Step 6, initializing each external clientModel/>Model reasoning is performed locally.
Referring to fig. 2, the invention provides a lightweight industrial picture classification method based on federal small sample learning, wherein a logic architecture main body of the lightweight industrial picture classification method comprises industrial picture classification system modeling based on federal small sample learning, small sample picture classifier construction based on a lightweight relational network, and edge cloud collaborative deployment and training of the small sample picture classifier based on the lightweight relational network. In order to solve the problem that a sample-limited industrial organization cannot effectively establish a data-driven industrial picture classifier, an industrial picture classification system based on federal small sample learning is established, and global small sample picture classifier training of an industrial picture island crossing isomerism and sample limitation is supported; constructing a small sample picture classifier based on a lightweight relational network for reducing the calculation and storage cost of an industrial organization with limited resources in the model training process; in order to reduce the communication overhead of the industrial institutions with limited resources in the model training process, a side cloud collaborative deployment and training method of a small sample picture classifier based on a lightweight relational network is constructed.
In still another embodiment of the present invention, a lightweight industrial picture classification system based on federal small sample learning is provided, which can be used to implement the lightweight industrial picture classification method based on federal small sample learning, and specifically, the system includes:
the system modeling module is used for modeling an industrial picture classification system based on a federal small sample learning framework, the system comprises a cloud server c and a group of industrial institutions epsilon with the same interests, and each e i epsilon independently holds a local sample set
A small sample picture classification task modeling module for classifying tasks of small sample pictures for each industrial organization in the constructed system modelModeling, each task contains a support set/>And query set/>
The classifier construction module is used for constructing a small sample classifier f based on a lightweight relational network, which consists of an embedding module l θ and a relational module g φ, wherein θ and phi are parameters of functions l and g;
training module for classifying tasks based on small sample pictures Constructing a training method of a small sample classifier f; constructing an edge cloud collaborative deployment and training method of a small sample classifier based on a small sample classifier training method, wherein a complete small sample picture classifier based on a lightweight relational network is deployed on c, an embedded module is deployed on all e i epsilon, and training is carried out on the classifier along discrete time points t=0, 1, … and T based on the training method in the step 4, wherein T is the preset model iteration times;
The output reasoning module is used for deploying and training the constructed small sample classifier based on the edge cloud collaborative deployment and training method of the small sample classifier After training is finished, each external client/>, is initializedModel/>Model reasoning is performed locally.
The invention solves the problem that the available data-driven picture classifier can not be established by a scarce industrial organization with sample isolation under the strong privacy protection requirement. The method can effectively train the global small sample picture classifier across heterogeneous and sample-limited industrial picture islands on the premise of ensuring the privacy of an industrial organization, and the obtained global model is not only suitable for federal model training participation clients, but also suitable for external clients without class samples during the holding training period. According to the small sample picture classifier based on the lightweight relational network and the edge cloud collaborative deployment and training method thereof, the computing, communication and storage costs of the industrial institutions with limited resources can be obviously reduced while the classification precision of the models is ensured.
Finally, it should be noted that: the above embodiments are only for illustrating the technical aspects of the present invention and not for limiting the same, and although the present invention has been described in detail with reference to the above embodiments, it should be understood by those of ordinary skill in the art that: modifications and equivalents may be made to the specific embodiments of the invention without departing from the spirit and scope of the invention, which is intended to be covered by the claims.
Claims (10)
1. The lightweight industrial picture classification method based on federal small sample learning is characterized by comprising the following steps of:
Modeling an industrial picture classification system based on a federal small sample learning framework, wherein the system comprises a cloud server c and a group of industrial institutions epsilon with the same interests, and each e i epsilon independently holds a local sample set
For each industrial organization in the constructed system model, classifying tasks for small sample picturesModeling, each task contains a support set/>And query set/>
Constructing a small sample classifier f based on a lightweight relational network, which consists of an embedding module l θ and a relational module g φ, wherein θ and phi are parameters of functions l and g;
Task for classifying pictures based on small samples Constructing a training method of a small sample classifier f;
Constructing an edge cloud collaborative deployment and training method of a small sample classifier based on a small sample classifier training method, wherein a complete small sample picture classifier based on a lightweight relational network is deployed on c, an embedded module is deployed on all e i epsilon, and training is carried out on the classifier along discrete time points t=0, 1, … and T based on the training method in the step 4, wherein T is the preset model iteration times;
small sample classifier based edge cloud collaborative deployment and training method, deployment and training of constructed small sample classifier After training is finished, each external client/>, is initializedModel/>Model reasoning is performed locally.
2. The method for classifying lightweight industrial pictures based on federal small sample learning according to claim 1, wherein an industrial picture classification system is modeled by using federal small sample learning technology, privacy of each industrial organization is guaranteed while global small sample picture classifier training across heterogeneous and sample limited industrial picture islands is supported, and the obtained global picture classifier is suitable for not only federal model training participation clients but also external clients who have no class sample during training.
3. The lightweight industrial picture classification method based on federal small sample learning of claim 2, characterized by the specific operations of: industrial picture classification system modeling based on federal small sample learning, the system comprising a cloud server c and a set of industrial institutions epsilon=e i of equal interest, wherein Each e i e epsilon holds its own local sample set/>, separatelyWherein e i holds/>Class sample,/> For sample j on e i, i+.j,/>, for all e i,ej εC can receive the message sent by epsilon but not access the sample set, epsilon can receive the message sent by c but the epsilon cannot be communicated, and a federal small sample picture classifier is arranged to train the participating client set/>External client set/>
4. The federal small sample learning-based lightweight industrial picture classification method according to claim 1, wherein the specific operation of modeling the small sample picture classification task T is: c-way K-shot picture classification task on classifier fContaining support sets/>And query set/> Containing C x K tagged samples,/>Containing C x K' samples to be classified (, requiring f to be based on/>)Pair/>Classification is performed.
5. The method for classifying light-weight industrial pictures based on federal small sample learning according to claim 1, wherein the small sample classifier f based on the light-weight relational network is constructed by using a relational network technology.
6. The lightweight industrial picture classification method based on federal small sample learning of claim 5, characterized by the specific operations of: constructing a multi-layer CNN model f, which consists of an embedding module l θ and a relation module g φ, wherein theta and phi are parameters of functions l and g, and l θ is used for extracting an input image sample feature map, and comprises two convolution blocks, namely a convolution layer containing 32 3×3 filters, a batch normalization layer, a ReLU nonlinear layer and a 2×2 maximum pooling layer; g φ determines whether two input samples belong to the same class based on the similarity of the feature map, and the output size H of the last maximum pooling layer is 32 and 32 multiplied by 3 for single and three-channel images, wherein the two input samples comprise the two convolution blocks, an 8-dimensional full-connection layer based on ReLU and a 1-dimensional full-connection layer based on Sigmoid.
7. The lightweight industrial picture classification method based on federal small sample learning of claim 1, wherein the training method of the small sample classifier f specifically operates as follows: small sample picture classification task based onFor any sample pairObtaining a feature map l θ(xa)、lθ(xb based on the embedding module), calculating a relationship score/>, based on the relationship moduleWherein/>For feature graph connection operation, the relation matrix is adopted to organize the relation scores of all sample pairs, and the classifier loss/>Wherein:
the model parameters θ *,φ* are updated by random gradient descent.
8. The lightweight industrial picture classification method based on federal small sample learning of claim 1, wherein the edge cloud collaborative deployment and training method of the lightweight relational network-based small sample picture classifier is constructed by utilizing an edge cloud collaborative intelligent technology.
9. The federal small sample learning-based lightweight industrial picture classification method according to claim 8, comprising the steps of:
step 5.1, setting the current iteration times t≡0, and initializing a cloud model Wherein θ init、φinit is a random initialization parameter, initializing each participating client/>Model parameters of (2)Each participating client e i∈εi organizes the local task set/>Updating the current iteration times t≡1;
Step 5.2, each participating client e i∈εi slave Random extraction task/>Obtaining all pairs of samplesFeature map/>Uploading all feature graphs and corresponding sample labels to c;
Step 5.3, for each participating client e i∈εi, calculate based on step 4, c UpdatingWherein α is a random gradient descent learning rate, returning θ g (t+1) to each participating client e i∈εi;
step 5.4, update per participating client e i∈εi
Step 5.5 the current iteration number T _ T +1 is updated, if T is less than or equal to T, repeating the steps 5.2 to 5.4;
step 5.6, outputting training results
10. Lightweight industrial picture classification system based on federal small sample study, characterized by comprising:
the system modeling module is used for modeling an industrial picture classification system based on a federal small sample learning framework, the system comprises a cloud server c and a group of industrial institutions epsilon with the same interests, and each e i epsilon independently holds a local sample set
A small sample picture classification task modeling module for classifying tasks of small sample pictures for each industrial organization in the constructed system modelModeling, each task contains a support set/>And query set/>
The classifier construction module is used for constructing a small sample classifier f based on a lightweight relational network, which consists of an embedding module l θ and a relational module g φ, wherein θ and phi are parameters of functions l and g;
training module for classifying tasks based on small sample pictures Constructing a training method of a small sample classifier f; constructing an edge cloud collaborative deployment and training method of a small sample classifier based on a small sample classifier training method, wherein a complete small sample picture classifier based on a lightweight relational network is deployed on c, an embedded module is deployed on all e i epsilon, and training is carried out on the classifier along discrete time points t=0, 1, … and T based on the training method in the step 4, wherein T is the preset model iteration times;
The output reasoning module is used for deploying and training the constructed small sample classifier based on the edge cloud collaborative deployment and training method of the small sample classifier After training is finished, each external client/>, is initializedModel of (2)Model reasoning is performed locally.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210611106.XA CN115017351B (en) | 2022-05-31 | 2022-05-31 | Lightweight industrial picture classification method and system based on federal small sample learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210611106.XA CN115017351B (en) | 2022-05-31 | 2022-05-31 | Lightweight industrial picture classification method and system based on federal small sample learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115017351A CN115017351A (en) | 2022-09-06 |
CN115017351B true CN115017351B (en) | 2024-05-10 |
Family
ID=83071882
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210611106.XA Active CN115017351B (en) | 2022-05-31 | 2022-05-31 | Lightweight industrial picture classification method and system based on federal small sample learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115017351B (en) |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021189906A1 (en) * | 2020-10-20 | 2021-09-30 | 平安科技(深圳)有限公司 | Target detection method and apparatus based on federated learning, and device and storage medium |
WO2021190638A1 (en) * | 2020-11-24 | 2021-09-30 | 平安科技(深圳)有限公司 | Federated modelling method based on non-uniformly distributed data, and related device |
CN113469234A (en) * | 2021-06-24 | 2021-10-01 | 成都卓拙科技有限公司 | Network flow abnormity detection method based on model-free federal meta-learning |
CN113518007A (en) * | 2021-07-06 | 2021-10-19 | 华东师范大学 | Multi-internet-of-things equipment heterogeneous model efficient mutual learning method based on federal learning |
-
2022
- 2022-05-31 CN CN202210611106.XA patent/CN115017351B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2021189906A1 (en) * | 2020-10-20 | 2021-09-30 | 平安科技(深圳)有限公司 | Target detection method and apparatus based on federated learning, and device and storage medium |
WO2021190638A1 (en) * | 2020-11-24 | 2021-09-30 | 平安科技(深圳)有限公司 | Federated modelling method based on non-uniformly distributed data, and related device |
CN113469234A (en) * | 2021-06-24 | 2021-10-01 | 成都卓拙科技有限公司 | Network flow abnormity detection method based on model-free federal meta-learning |
CN113518007A (en) * | 2021-07-06 | 2021-10-19 | 华东师范大学 | Multi-internet-of-things equipment heterogeneous model efficient mutual learning method based on federal learning |
Non-Patent Citations (2)
Title |
---|
基于小样本学习的高光谱遥感图像分类算法;张婧;袁细国;;聊城大学学报(自然科学版);20200804(第06期);全文 * |
联邦学习安全与隐私保护研究综述;周俊;方国英;吴楠;;西华大学学报(自然科学版);20200710(第04期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN115017351A (en) | 2022-09-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021190451A1 (en) | Method and apparatus for training image processing model | |
US10248664B1 (en) | Zero-shot sketch-based image retrieval techniques using neural networks for sketch-image recognition and retrieval | |
CN109635917A (en) | A kind of multiple agent Cooperation Decision-making and training method | |
WO2022001805A1 (en) | Neural network distillation method and device | |
WO2021164750A1 (en) | Method and apparatus for convolutional layer quantization | |
JP7383803B2 (en) | Federated learning using heterogeneous model types and architectures | |
CN112396106B (en) | Content recognition method, content recognition model training method, and storage medium | |
CN112016601B (en) | Network model construction method based on knowledge graph enhanced small sample visual classification | |
WO2022111617A1 (en) | Model training method and apparatus | |
CN112200266B (en) | Network training method and device based on graph structure data and node classification method | |
CN113469373A (en) | Model training method, system, equipment and storage medium based on federal learning | |
WO2022001724A1 (en) | Data processing method and device | |
CN115587633A (en) | Personalized federal learning method based on parameter layering | |
CN114091667A (en) | Federal mutual learning model training method oriented to non-independent same distribution data | |
WO2022179586A1 (en) | Model training method, and device associated therewith | |
CN114710330B (en) | Anomaly detection method based on heterogeneous layered federated learning | |
CN110097177A (en) | A kind of network pruning method based on pseudo- twin network | |
CN117236421B (en) | Large model training method based on federal knowledge distillation | |
WO2022052782A1 (en) | Image processing method and related device | |
CN116664930A (en) | Personalized federal learning image classification method and system based on self-supervision contrast learning | |
WO2022013879A1 (en) | Federated learning using heterogeneous labels | |
CN115331069A (en) | Personalized image classification model training method based on federal learning | |
CN108959664A (en) | Distributed file system based on picture processor | |
CN115018039A (en) | Neural network distillation method, target detection method and device | |
CN113284227A (en) | Distributed motion inference structure method for large-scale aerial images |
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 |