CN117373066A - Pedestrian re-identification method and system based on Yun Bian searching federal deep learning method - Google Patents
Pedestrian re-identification method and system based on Yun Bian searching federal deep learning method Download PDFInfo
- Publication number
- CN117373066A CN117373066A CN202311667337.3A CN202311667337A CN117373066A CN 117373066 A CN117373066 A CN 117373066A CN 202311667337 A CN202311667337 A CN 202311667337A CN 117373066 A CN117373066 A CN 117373066A
- Authority
- CN
- China
- Prior art keywords
- edge
- network model
- depth network
- cloud
- model
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 39
- 238000013135 deep learning Methods 0.000 title claims abstract description 16
- 230000006870 function Effects 0.000 claims abstract description 44
- 238000012549 training Methods 0.000 claims abstract description 28
- 238000005457 optimization Methods 0.000 claims abstract description 23
- 230000002776 aggregation Effects 0.000 claims abstract description 16
- 238000004220 aggregation Methods 0.000 claims abstract description 16
- 239000013598 vector Substances 0.000 claims description 5
- 238000006243 chemical reaction Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 3
- 239000000284 extract Substances 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 238000010801 machine learning Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 238000011176 pooling Methods 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000006116 polymerization reaction Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06N3/098—Distributed learning, e.g. federated 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
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- 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
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Medical Informatics (AREA)
- Bioethics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Mathematical Physics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Computer Hardware Design (AREA)
- Computer Security & Cryptography (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a pedestrian re-identification method and system based on Yun Bian searching federal deep learning method, which relates to the technical field of machine learning, and comprises the following steps: s1, initializing a global depth network model by a cloud, and S2, issuing the global depth network model to edge equipment by the cloud; s3, the edge equipment builds an overall optimization objective function by utilizing the personalized initialization function, and performs edge depth network model training; s4, the cloud end carries out weighted average aggregation on the edge depth network model weight parameters to update the cloud end global depth network model; s5, repeating the steps S2 to S4 to the maximum times, and taking the cloud global depth network model generated in the last time as a pedestrian re-identification model; s6, utilizing the pedestrian re-identification model to realize pedestrian re-identification. According to the invention, on the premise of protecting data privacy, each edge device is enabled to initialize the self network in a personalized manner according to the local data characteristics, and the performance of the edge depth network and the cloud global depth network model in federal learning is improved.
Description
Technical Field
The invention relates to technologies in the fields of distributed computing, machine learning, data privacy protection and re-identification and the like, in particular to a pedestrian re-identification method and system based on a cloud edge search federal deep learning method.
Background
In the traditional pedestrian re-identification, because a centralized data processing method is adopted, the sensitive information of the pedestrian can be exposed in a central server, and the artificial intelligence technology is suddenly advanced in the aspect of virtual image generation, so that the privacy and safety protection in the pedestrian re-identification are increasingly important. Federal learning, a training framework for data isolation with model sharing, has shown great potential in protecting data privacy. However, in the conventional federal learning method, all edge devices generally use the same cloud initialization depth network model to perform local training, which may cause that the global model cannot sufficiently adapt to the local data characteristics of each device due to the data distribution difference between the devices, so that the performance difference of the model on different edge devices is large, and further performance accuracy is reduced. Compared with the common image classification task, in application of pedestrian re-identification in the public safety field, the difference of data sets of different edge devices is larger, because of a plurality of variables such as camera view angle difference, camera dynamic range difference, data mode difference (infrared light and visible light), non-overlapping of pedestrian identity types and the like, how to skillfully weigh privacy safety of data and accuracy of a model in frame design and implementation of federal learning, ensure that accurate identification can still be realized while protecting the privacy safety of data, and become challenges which must be considered in the field.
Disclosure of Invention
The invention aims to solve the problem that the existing pedestrian re-identification technology does not balance the data privacy security and the accuracy of a model.
The technical scheme adopted for solving the technical problems is as follows: the pedestrian re-identification method based on the cloud edge search federal deep learning method comprises the following steps:
s1, initializing a global depth network model by a cloud;
s2, the cloud transmits the global depth network model to the edge equipment;
s3, the edge equipment builds an overall optimization objective function by utilizing the personalized initialization function, and performs edge depth network model training;
s4, the cloud end carries out weighted average aggregation on the edge depth network model weight parameters to update the cloud end global depth network model;
s5, repeating the steps S2 to S4 to the maximum times, and taking the cloud global depth network model generated in the last time as a pedestrian re-identification model;
s6, utilizing a pedestrian re-identification model to realize pedestrian re-identification;
the edge device utilizes the personalized initialization function to construct an overall optimization objective function as follows:
wherein,representing a set of edge depth network model weight parameters and cloud global depth network model weight parameters, wherein +.>Weight parameters representing the respective edge depth network model, < +.>The method comprises the steps of representing weight parameters of a cloud global depth network model; />Representing model weight parameter combination factor set, +.>Model weight parameter combination factors representing respective edge devices; />Representing the number of edge devices; />As a loss function; />Indicate->Data set corresponding to the respective edge device, +.>Representing data samples in the corresponding dataset, including local pedestrian images and corresponding tags; />Representing the total amount of all edge device dataset samples; />Indicate->Weight parameters of historical versions of the edge depth network model; />Representing the edge depth network model personalization initialization function.
Preferably, the edge device performs edge depth network model training based on an overall optimization objective function, and the method comprises the following steps:
the edge equipment acquires local pedestrian images as a local data set, wherein each image is provided with an identity mark as a label;
carrying out personalized initialization on the edge depth network model based on the model weight parameter combination factors to obtain an initial edge depth network model;
adjusting model weight parameter combination factors by using an iterative optimization strategy based on gradient descent;
training an initial edge depth network model by utilizing a local data set;
and uploading the trained edge depth network model weight parameters to the cloud.
Preferably, the personalized initialization is specifically: for the firstThe weight parameters of the historical version of the edge depth network model and the weight parameters of the cloud global depth network model version stored by the edge devices are subjected to probability combination,representing cloud global depth network model ++>Probability of weight parameters of the individual network layers, +.>Representing edge depth network model->Probability of historical version weight parameter of individual network layer,/->Expressed as:
wherein,is a learnable parameter, < >>And->Edge depth network model +.>Historical version of personal network layer and cloud global depth network +.>Model weight parameter combination factors for the individual network layers.
Preferably, the weight parameter is expressed as:
wherein,indicate->Personal edge depth network model>Weight parameters of the individual network layers,/->Then indicate->Remain at->Personal edge depth network model>Weight parameters of historical versions of the individual network layers; />Then represent cloud global depth network model +.>Weight parameters of the individual network layers; />Indicate->Personal edge depth network model>Model weight parameter combination factors of the individual network layers; />Representing the +.f obtained after the conversion of the model weight parameter combination factor>Personal edge depth network model>Probabilities of historical version weight parameters for individual network layers,cloud global depth network model obtained after conversion of model weight parameter combination factors>Probability of individual network layer weight parameters.
Preferably, the model weight parameter combination factor is adjusted by using an iterative optimization strategy based on gradient descent, and the model weight parameter is fixed when the model weight parameter combination factor is optimized, and the optimization process is expressed as:
wherein,is->The method comprises the steps that a set of model weight parameter combination factors of all network layers of the edge depth network model is formed; />Is->Data set corresponding to the respective edge device, +.>Indicating that the loss function is +.>The calculated loss; />Is a deflection guide; />Is the learning step length; />Indicate->Weight parameters for the individual edge depth network model.
Preferably, the formula of the weighted average aggregation is expressed as:
wherein,indicate->Weight parameters of the individual edge depth network model, +.>Weight parameter representing cloud global depth network model, < ->Representing the total amount of all edge device dataset samples, +.>Indicate->And the data sets corresponding to the edge devices.
Preferably, the pedestrian re-identification is realized by using a pedestrian re-identification model, which specifically comprises: the pedestrian re-identification model receives the target image and the image to be checked, extracts the feature vectors of the target image and the image to be checked, calculates the Euclidean distance between the target image and the image to be checked, and carries out ascending order on the image to be checked according to the Euclidean distance, and the front is arrangedAnd outputting the image to be checked as an identification result of the pedestrian re-identification model.
The invention also provides a pedestrian re-identification system based on Yun Bian searching federal deep learning method, comprising:
the cloud end initializes the global depth network model;
the cloud end transmits the global depth network model to the edge equipment;
the edge equipment builds a general optimization objective function by utilizing the personalized initialization function and carries out edge depth network model training;
the aggregation module is used for carrying out weighted average aggregation on the edge depth network model weight parameters by the cloud to update the cloud global depth network model;
the circulation module is used for repeating the operation steps of the distribution module, the training module and the aggregation module to the maximum times, and the cloud global depth network model generated in the last time is used as a pedestrian re-identification model;
the identification module is used for realizing pedestrian re-identification by utilizing the pedestrian re-identification model;
in the training module, the edge device constructs an overall optimization objective function by utilizing the personalized initialization function as follows:
wherein,representing a set of edge depth network model weight parameters and cloud global depth network model weight parameters, wherein +.>Weight parameters representing the respective edge depth network model, < +.>The method comprises the steps of representing weight parameters of a cloud global depth network model; />Representing model weight parameter combination factor set, +.>Model weight parameter combination factors representing respective edge devices; />Representing the number of edge devices; />As a loss function; />Indicate->Data set corresponding to the respective edge device, +.>Representing data samples in the corresponding dataset, including local pedestrian images and corresponding tags; />Representing the total amount of all edge device dataset samples; />Indicate->Weight parameters of historical versions of the edge depth network model; />Representing the edge depth network model personalization initialization function.
The invention has the following beneficial effects: the edge depth network model and the cloud global depth network model in federal deep learning training are searched for personalized initialization of the edge depth network model, a model weight parameter combination factor is updated by using an iterative optimization strategy based on gradient descent, and the contribution degree of the weight parameter of the cloud global depth network model and the weight parameter of the edge depth network model in personalized initialization of the edge depth network model is calculated by the combination factor, so that each edge device can initialize own network in a personalized manner according to the local data characteristic on the premise of protecting data privacy, and the performance of the edge depth network and the cloud global depth network model in federal learning is improved.
The present invention will be described in further detail with reference to the drawings and examples, but the present invention is not limited to the examples.
Drawings
FIG. 1 is a diagram of steps in a method according to an embodiment of the present invention;
FIG. 2 is a flow chart of training an edge depth network model according to an embodiment of the present invention;
FIG. 3 is a flowchart of updating a cloud global depth network model according to an embodiment of the present invention;
fig. 4 is a system configuration diagram of an embodiment of the present invention.
Detailed Description
Referring to fig. 1, a method step diagram of an embodiment of the present invention includes the following steps:
s101, initializing a global depth network model by a cloud;
s102, the cloud transmits the global depth network model to edge equipment;
s103, the edge equipment builds an overall optimization objective function by utilizing the personalized initialization function, and performs edge depth network model training;
s104, the cloud end carries out weighted average aggregation on the edge depth network model weight parameters to update a cloud end global depth network model;
s105, repeating the steps S102 to S104 for the maximum times, and taking the cloud global depth network model generated in the last time as a pedestrian re-identification model;
s106, utilizing the pedestrian re-identification model to realize pedestrian re-identification.
Specifically, referring to fig. 2, a flowchart for training an edge depth network model according to an embodiment of the present invention includes:
the edge equipment acquires local pedestrian images as a local data set, wherein each image is provided with an identity mark as a label;
carrying out personalized initialization on the edge depth network model based on the model weight parameter combination factors to obtain an initial edge depth network model; the personalized initialization: for the followingThe>The edge equipment stores the weight parameters and cloud of the historical version of the edge depth network modelProbability combination of weight parameters of end global depth network model,/->Probability of representing weight parameters of cloud global depth network model,/->A probability representing a historical version weight parameter of the edge depth network model;
and the edge equipment builds an overall optimization objective function by utilizing the personalized initialization function, and performs edge depth network model training. Iterative training of model weight parameter combination factors using gradient descent based iterative optimization strategyAnd training the initial edge depth network model by utilizing the local data set, updating model weight parameters and recording model weight parameter combination factors and historical versions of the edge depth network model.
Specifically, in S106, a test stage is further included, where a part of images are obtained from the local dataset as a test set, the test set is divided into a target image and an image to be checked, the pedestrian re-recognition model obtained in S105 is used for extracting features (taking a general depth network model res net50 as an example, the step is to take the output of a global pooling layer (Global Average Pooling, GAP) in the pedestrian re-recognition model as feature vectors of the target image and the image to be checked, calculate the euclidean distance between the target image and the image to be checked based on the extracted feature vectors, and perform ascending order according to the euclidean distance, where the image to be checked that is ranked at the front is the image most likely to have the same identity attribute as the target image, that is, as a result of recognition of the pedestrian re-recognition model, and perform further artificial verification on the suspected image according to actual needs.
Specifically, referring to fig. 3, for a flowchart of updating a cloud global depth network model according to an embodiment of the present invention, an edge device includes edge device 1 to edge deviceThe flow can be divided into the following four parts:
(1) issuing a global model: the cloud uses ResNet50 as a global depth network model, initializes the global depth network model and issues the global depth network model to the edge equipment;
(2) local training: the edge equipment performs personalized initialization on the edge depth network model and trains by utilizing a local data set, loss is a Loss function related to pedestrian re-identification, a cross entropy classification Loss function and a ternary Loss function which are most commonly used for pedestrian re-identification can be adopted, and a full connection layer (FC) used by the cross entropy classification Loss function is completely localized in consideration of different pedestrian categories of the local data set of each edge equipment and does not participate in model convergence to the cloud;
(3) uploading model weight parameters: the edge equipment uploads the trained edge depth network model weight parameters to the cloud;
(4) polymerization: cloud end connects edge device 1 to edge deviceUploading->And carrying out weighted average aggregation on the weight parameters of the edge depth network model to obtain a new global depth network model. And the global depth network model performs feature extraction on the data set to obtain feature vectors for pedestrian re-identification.
Referring to fig. 4, a system structure diagram of an embodiment of the present invention includes:
the initialization module 401 is used for cloud initialization of the global depth network model;
the distribution module 402 is used for transmitting the global depth network model to the edge equipment by the cloud;
the training module 403 is used for constructing an overall optimization objective function by the edge equipment through the personalized initialization function and training an edge depth network model;
the aggregation module 404, which performs weighted average aggregation on the edge depth network model weight parameters by the cloud to update the cloud global depth network model;
the circulation module 405, repeating the operation steps of the distribution module, the training module and the aggregation module to the maximum number of times, and taking the cloud global depth network model generated last time as a pedestrian re-identification model;
the identifying module 406 realizes the pedestrian re-identification by using the pedestrian re-identification model;
in the training module 403, the edge device constructs an overall optimization objective function by using the personalized initialization function as follows:
wherein,representing a set of edge depth network model weight parameters and cloud global depth network model weight parameters, wherein +.>Weight parameters representing the respective edge depth network model, < +.>The method comprises the steps of representing weight parameters of a cloud global depth network model; />Representing model weight parameter combination factor set, +.>Model weight parameter combination factors representing respective edge devices; />Representing the number of edge devices; />As a loss function; />Indicate->The data sets corresponding to the respective edge devices,representing data samples in the corresponding dataset, including local pedestrian images and corresponding tags; />Representing the total amount of all edge device dataset samples; />Indicate->Weight parameters of historical versions of the edge depth network model; />Representing the edge depth network model personalization initialization function.
Therefore, the edge depth network model and the cloud global depth network model in the federal deep learning training are searched, and the weight parameters of the edge depth network model and the cloud global depth network model are combined with probability by using the model weight parameter combination factors, so that each edge device can initialize the self network in a personalized manner according to the local data characteristics on the premise of protecting the data privacy, and the performance of the edge depth network and the cloud global depth network model in the federal learning is improved.
The foregoing is only illustrative of the present invention and is not to be construed as limiting thereof, but rather as various modifications, equivalent arrangements, improvements, etc., within the spirit and principles of the present invention.
Claims (8)
1. A pedestrian re-identification method based on cloud edge search federal deep learning method is characterized by comprising the following steps:
s1, initializing a global depth network model by a cloud;
s2, the cloud transmits the global depth network model to the edge equipment;
s3, the edge equipment builds an overall optimization objective function by utilizing the personalized initialization function, and performs edge depth network model training;
s4, the cloud end carries out weighted average aggregation on the edge depth network model weight parameters to update the cloud end global depth network model;
s5, repeating the steps S2 to S4 to the maximum times, and taking the cloud global depth network model generated in the last time as a pedestrian re-identification model;
s6, utilizing a pedestrian re-identification model to realize pedestrian re-identification;
the edge device utilizes the personalized initialization function to construct an overall optimization objective function as follows:
；
wherein,representing a set of edge depth network model weight parameters and cloud global depth network model weight parameters, wherein +.>Weight parameters representing the respective edge depth network model, < +.>The method comprises the steps of representing weight parameters of a cloud global depth network model; />Representing model weight parameter combination factor set, +.>Model weight parameter combination factors representing respective edge devices; />Representing the number of edge devices;as a loss function; />Indicate->Data set corresponding to the respective edge device, +.>Representing data samples in the corresponding dataset, including local pedestrian images and corresponding tags; />Representing the total amount of all edge device dataset samples; />Represent the firstWeight parameters of historical versions of the edge depth network model; />Representing the edge depth network model personalization initialization function.
2. The pedestrian re-recognition method based on the cloud edge search federal deep learning method according to claim 1, wherein the edge device performs edge deep network model training based on an overall optimization objective function, and the method comprises the following steps:
the edge equipment acquires local pedestrian images as a local data set, wherein each image is provided with an identity mark as a label;
carrying out personalized initialization on the edge depth network model based on the model weight parameter combination factors to obtain an initial edge depth network model;
adjusting model weight parameter combination factors by using an iterative optimization strategy based on gradient descent;
training an initial edge depth network model by utilizing a local data set;
and uploading the trained edge depth network model weight parameters to the cloud.
3. The pedestrian re-identification method based on the cloud edge search federal deep learning method according to claim 2, wherein the personalized initialization is specifically: for the firstCarrying out probability combination on the weight parameters of the historical version of the edge depth network model and the weight parameters of the cloud global depth network model version stored by the edge devices, wherein the weight parameters are +.>Representing cloud global depth network model ++>Probability of weight parameters of the individual network layers, +.>Representing edge depth network model->Probability of historical version weight parameter of individual network layer,/->Expressed as:
；
wherein,is a learnable parameter, < >>And->Edge depth network model +.>Historical version of personal network layer and cloud global depth network model +.>Model weight parameter combination factors for the individual network layers.
4. The pedestrian re-recognition method based on the cloud edge search federal deep learning method according to claim 3, wherein the weight parameter is expressed as:
；
wherein,represents the kth edge depth network model +.>Weight parameters of the individual network layers,/->Then indicate->Remain at->Personal edge depth network model>Weight parameters of historical versions of the individual network layers; />Then represent cloud global depth network model +.>Weight parameters of the individual network layers; />Indicate->Personal edge depth network model>Model weight parameter combination factors of the individual network layers; />Representing the +.f obtained after the conversion of the model weight parameter combination factor>Personal edge depth network model>Probability of historical version weight parameter of individual network layer,/->Cloud global depth network model obtained after conversion of model weight parameter combination factors>Probability of individual network layer weight parameters.
5. The pedestrian re-recognition method based on the cloud edge search federal deep learning method according to claim 2, wherein the model weight parameter combination factor is adjusted by using an iterative optimization strategy based on gradient descent, and the optimization process is expressed as:
；
wherein,a set of model weight parameter combination factors for all network layers of the kth edge depth network model; />Is->Data set corresponding to the respective edge device, +.>Indicating that the loss function is +.>The calculated loss; />Is a deflection guide; />Is the learning step length; />And the weight parameter of the kth edge depth network model is represented.
6. The pedestrian re-recognition method based on the cloud edge search federal deep learning method according to claim 1, wherein the formula of the weighted average aggregation is expressed as:
；
wherein,weight parameter representing kth edge depth network model,/->Weight parameter representing cloud global depth network model, < ->Representing the total amount of all edge device dataset samples, +.>Indicate->And the data sets corresponding to the edge devices.
7. The pedestrian re-identification method based on the cloud edge search federal deep learning method according to claim 1, wherein the pedestrian re-identification is realized by using a pedestrian re-identification model, specifically: the pedestrian re-identification model receives the target image and the image to be checked, extracts feature vectors of the target image and the image to be checked, calculates Euclidean distance between the target image and the image to be checked, carries out ascending order on the image to be checked according to the Euclidean distance, and outputs the first M images to be checked as identification results of the pedestrian re-identification model.
8. A pedestrian re-identification system based on cloud edge search federal deep learning method is characterized by comprising the following steps:
the cloud end initializes the global depth network model;
the cloud end transmits the global depth network model to the edge equipment;
the edge equipment builds a general optimization objective function by utilizing the personalized initialization function and carries out edge depth network model training;
the aggregation module is used for carrying out weighted average aggregation on the edge depth network model weight parameters by the cloud to update the cloud global depth network model;
the circulation module is used for repeating the operation steps of the distribution module, the training module and the aggregation module to the maximum times, and the cloud global depth network model generated in the last time is used as a pedestrian re-identification model;
the identification module is used for realizing pedestrian re-identification by utilizing the pedestrian re-identification model;
in the training module, the edge device constructs an overall optimization objective function by utilizing the personalized initialization function as follows:
；
wherein,representing a set of edge depth network model weight parameters and cloud global depth network model weight parameters, wherein +.>Weight parameters representing the respective edge depth network model, < +.>The method comprises the steps of representing weight parameters of a cloud global depth network model; />Representing model weight parameter combination factor set, +.>Model weight parameter combination factors representing respective edge devices; />Representing edge devicesIs the number of (3);as a loss function; />Indicate->Data set corresponding to the respective edge device, +.>Representing data samples in the corresponding dataset, including local pedestrian images and corresponding tags; />Representing the total amount of all edge device dataset samples; />Represent the firstWeight parameters of historical versions of the edge depth network model; />Representing the edge depth network model personalization initialization function.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311667337.3A CN117373066B (en) | 2023-12-07 | 2023-12-07 | Pedestrian re-identification method and system based on Yun Bian searching federal deep learning method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311667337.3A CN117373066B (en) | 2023-12-07 | 2023-12-07 | Pedestrian re-identification method and system based on Yun Bian searching federal deep learning method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117373066A true CN117373066A (en) | 2024-01-09 |
| CN117373066B CN117373066B (en) | 2024-03-12 |
Family
ID=89404492
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311667337.3A Active CN117373066B (en) | 2023-12-07 | 2023-12-07 | Pedestrian re-identification method and system based on Yun Bian searching federal deep learning method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117373066B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111190487A (en) * | 2019-12-30 | 2020-05-22 | 中国科学院计算技术研究所 | Method for establishing data analysis model |
| CN113902021A (en) * | 2021-10-13 | 2022-01-07 | 北京邮电大学 | High-energy-efficiency clustering federal edge learning strategy generation method and device |
| CN114219097A (en) * | 2021-11-30 | 2022-03-22 | 华南理工大学 | Federal learning training and prediction method and system based on heterogeneous resources |
| CN115062710A (en) * | 2022-06-22 | 2022-09-16 | 西安电子科技大学 | Federal learning classification model training method based on depth certainty strategy gradient |
| CN115170565A (en) * | 2022-09-06 | 2022-10-11 | 浙商银行股份有限公司 | Image fraud detection method and device based on automatic neural network architecture search |
| CN115277689A (en) * | 2022-04-29 | 2022-11-01 | 国网天津市电力公司 | Yun Bianwang network communication optimization method and system based on distributed federal learning |
| US20220374776A1 (en) * | 2021-12-27 | 2022-11-24 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and system for federated learning, electronic device, and computer readable medium |
| CN116386853A (en) * | 2023-04-10 | 2023-07-04 | 南京邮电大学 | Intelligent medical-oriented deep separable convolution dual-aggregation federal learning method |
-
2023
- 2023-12-07 CN CN202311667337.3A patent/CN117373066B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111190487A (en) * | 2019-12-30 | 2020-05-22 | 中国科学院计算技术研究所 | Method for establishing data analysis model |
| CN113902021A (en) * | 2021-10-13 | 2022-01-07 | 北京邮电大学 | High-energy-efficiency clustering federal edge learning strategy generation method and device |
| CN114219097A (en) * | 2021-11-30 | 2022-03-22 | 华南理工大学 | Federal learning training and prediction method and system based on heterogeneous resources |
| US20220374776A1 (en) * | 2021-12-27 | 2022-11-24 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and system for federated learning, electronic device, and computer readable medium |
| CN115277689A (en) * | 2022-04-29 | 2022-11-01 | 国网天津市电力公司 | Yun Bianwang network communication optimization method and system based on distributed federal learning |
| CN115062710A (en) * | 2022-06-22 | 2022-09-16 | 西安电子科技大学 | Federal learning classification model training method based on depth certainty strategy gradient |
| CN115170565A (en) * | 2022-09-06 | 2022-10-11 | 浙商银行股份有限公司 | Image fraud detection method and device based on automatic neural network architecture search |
| CN116386853A (en) * | 2023-04-10 | 2023-07-04 | 南京邮电大学 | Intelligent medical-oriented deep separable convolution dual-aggregation federal learning method |
Non-Patent Citations (1)
| Title |
|---|
| 陈翔: "联邦学习框架下的自动行人轨迹预测模型设计研究", pages 1 - 67 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117373066B (en) | 2024-03-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110110642B (en) | Pedestrian re-identification method based on multi-channel attention features | |
| CN111325115B (en) | Cross-modal countervailing pedestrian re-identification method and system with triple constraint loss | |
| CN105740894B (en) | Semantic annotation method for hyperspectral remote sensing image | |
| CN107132516B (en) | A kind of Radar range profile's target identification method based on depth confidence network | |
| Xiang et al. | Cross-modality person re-identification based on dual-path multi-branch network | |
| CN108960184B (en) | Pedestrian re-identification method based on heterogeneous component deep neural network | |
| CN106203483B (en) | A kind of zero sample image classification method based on semantic related multi-modal mapping method | |
| CN110555390A (en) | pedestrian re-identification method, device and medium based on semi-supervised training mode | |
| CN109299707A (en) | A kind of unsupervised pedestrian recognition methods again based on fuzzy depth cluster | |
| CN109190521B (en) | Construction method and application of face recognition model based on knowledge purification | |
| WO2019230666A1 (en) | Feature amount extraction device, method, and program | |
| CN110569779A (en) | Pedestrian attribute identification method based on pedestrian local and overall attribute joint learning | |
| CN106250925B (en) | A kind of zero Sample video classification method based on improved canonical correlation analysis | |
| CN112633071A (en) | Pedestrian re-identification data domain adaptation method based on data style decoupling content migration | |
| CN110751027A (en) | Pedestrian re-identification method based on deep multi-instance learning | |
| CN113065409A (en) | Unsupervised pedestrian re-identification method based on camera distribution difference alignment constraint | |
| CN115862055A (en) | Pedestrian re-identification method and device based on comparison learning and confrontation training | |
| CN114579794A (en) | Multi-scale fusion landmark image retrieval method and system based on feature consistency suggestion | |
| CN117373066B (en) | Pedestrian re-identification method and system based on Yun Bian searching federal deep learning method | |
| Priyanka et al. | Artificial intelligence impact on pattern classification in association with IoT for advanced applications | |
| CN116070696A (en) | Cross-domain data deep migration method, device, equipment and storage medium | |
| CN116071544A (en) | Image description prediction method oriented to weak supervision directional visual understanding | |
| CN116186513A (en) | Vibration signal identification method based on one-dimensional convolutional neural network | |
| US20240185555A1 (en) | Method, device, and storage medium for targeted adversarial discriminative domain adaptation | |
| CN115497105A (en) | Multi-modal hate cause detection method based on multi-task learning network |
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 |