CN113378718A - Action identification method based on generation of countermeasure network in WiFi environment - Google Patents
Action identification method based on generation of countermeasure network in WiFi environment Download PDFInfo
- Publication number
- CN113378718A CN113378718A CN202110654354.8A CN202110654354A CN113378718A CN 113378718 A CN113378718 A CN 113378718A CN 202110654354 A CN202110654354 A CN 202110654354A CN 113378718 A CN113378718 A CN 113378718A
- Authority
- CN
- China
- Prior art keywords
- layer
- data
- model
- generation
- csi
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention provides a method for identifying actions based on generation of a countermeasure network in a WiFi environment, which comprises the following steps: preprocessing received WiFi data to be identified to obtain input data corresponding to the CSI data to be identified; inputting the input data to a generation countermeasure network model for action recognition. The problem of among the prior art marking work is complicated, and the discernment performance descends when less and the model is applied to new user of mark data is solved. The invention overcomes the limitation of the traditional identification technology on conditions such as illumination and the like, avoids the risk of privacy disclosure, and can still accurately identify different human actions under the conditions of small samples and target change.
Description
Technical Field
The invention belongs to the technical field of wireless sensing and human-computer interaction, and particularly relates to a method for recognizing actions based on a generated countermeasure network in a WiFi (wireless fidelity) environment.
Background
In recent years, the motion recognition technology is mature day by day, plays an important role in the field of human-computer interaction, and promotes the progress of the fields of smart home, virtual/augmented reality, health monitoring and the like. Traditional motion recognition mainly uses cameras, special dedicated hardware and wearable sensors. Among them, vision-based methods are limited and lighting conditions and risk privacy disclosure. The dedicated special hardware based approach uses a general purpose software radio peripheral to obtain activity information, which is too costly to deploy on a large scale. The sensor-based approach requires the user to carry the device, which is inconvenient to use.
With the development and popularization of wireless communication technology, a WiFi-based motion recognition technology becomes a research hotspot, and motion recognition is performed by acquiring channel state information CSI capable of depicting environmental changes. The CSI describes the attenuation degree of the WiFi signal on each propagation path, and the CSI is influenced by the action of people in the environment when being propagated, so that the action of people can be analyzed by collecting the CSI of the receiving end, and the action is identified by combining signal processing and deep learning technology. The method is not limited by illumination conditions, does not need to carry any equipment, and is easy for large-scale deployment.
However, most of the existing WiFi-based motion recognition technologies require a large amount of tag data, and since each user has a different shape and different motion habits, the recognition performance is reduced when the model is applied to a new user.
Disclosure of Invention
The invention aims to: aiming at the problems in the prior art, the action identification method based on the generation countermeasure network in the WiFi environment is provided, and the problems that the marking work is complicated, the identification performance is reduced when marking data are less and the model is applied to a new user in the prior art are solved.
A method for recognizing actions based on generation of countermeasure networks in a WiFi environment comprises the following steps:
step 1: extracting an amplitude value of the channel state information from the WiFi signal;
step 2: and preprocessing the data, and linearly interpolating the CSI to unify the shape of the CSI. Making a data set, selecting data of one user to be divided into a test set and unmarked data, and taking the data of the other users as a training set;
and step 3: inputting the data set into a generation countermeasure network;
and 4, step 4: returning to the step 3 for iteration according to the accuracy of the generation of the confrontation network output and the loss function value, and continuously optimizing the network so as to enable the loss function to reach the minimum value;
preferably, step 3 further comprises:
the generation of the confrontation network model is: a generator and a discriminator;
the generator model includes one input layer and three hidden layers:
an input layer: random Gaussian noise of 100 dimensions is taken as input; adopting a full connection layer, wherein the output dimension is 25 multiplied by 5 multiplied by 512, a batch standardization layer and an activation function layer, and the activation function is SoftPlus;
first hidden layer: using one deconvolution layer, the convolution kernel size is 5 × 5(kernel _ size is 5 × 5), the number of feature maps is 256(filters is 256), the step size is 2 × 1(stride is 2 × 1), and the padding mode is same; adopting a batch standardization layer and an activation function layer, wherein the activation function is SoftPlus;
second hidden layer: using one deconvolution layer, the convolution kernel size is 5 × 5(kernel _ size ═ 5 × 5), the number of feature maps is 128(filters ═ 128), the step size is 2 × 3(stride ═ 2 × 3), and the padding mode is same (padding ═ same); adopting a batch standardization layer and an activation function layer, wherein the activation function is SoftPlus;
third hidden layer: with one deconvolution layer, the convolution kernel size is 5 × 5(kernel _ size ═ 5 × 5), the number of profiles is 3(filters ═ 3), the step size is 2 × 2(stride ═ 2 × 2), and the activation function is Tanh.
The discriminator model comprises an input layer, three hidden layers, a softmax layer:
an input layer: the CSI amplitude value with the dimension of 200 multiplied by 30 multiplied by 3 is taken as input;
first hidden layer: a convolution layer is adopted, the size of a convolution kernel is 3 × 3(kernel _ size is 3 × 3), the number of feature maps is 96(filters is 96), the step size is 5 × 2(stride is 5 × 2), and the activation function is LEAKYRELU; a Gaussian noise layer;
second hidden layer: using one convolution layer, the convolution kernel size is 3 × 3(kernel _ size ═ 3 × 3), the number of feature maps is 192(filters ═ 192), the step size is 5 × 2(stride ═ 5 × 2), and the activation function is leakyreu; a Gaussian noise layer;
third hidden layer: a convolution layer is adopted, the size of a convolution kernel is 3 × 3(kernel _ size is 3 × 3), the number of feature maps is 192(filters are 192), and an activation function is LEAKYRELU; a nin layer, the number of feature maps is 192(filters 192), and the activation function is LEAKYRELU; one maximum pooling layer, 6 × 6 sampling area, with step size of 1(stride ═ 1).
The learning rate was set to 0.0006;
the optimization method is Adam algorithm;
the verification method adopts leave-one-subject-out validation and 5-Fold cross-validation;
training a neural network, feeding back parameters after each training, and continuously optimizing;
and outputting the trained network model after iteration, wherein the network model at the moment is applied to action recognition.
Advantageous effects
The invention discloses a method for identifying actions based on a generation countermeasure network in a WiFi (wireless fidelity) environment, and provides an improved generation countermeasure network for semi-supervised learning, which is applied to action identification. The invention reduces the requirement of a large amount of marked data in the WiFi-based activity recognition by adopting semi-supervised learning, solves the problem of reduced recognition performance under the condition of small samples by utilizing the characteristic that the generated countermeasure network can generate data to expand the number of samples, and shows good performance for the action recognition of a new user by learning the action characteristics of the new user through unmarked data.
Drawings
FIG. 1 is a general flow diagram of the present invention.
FIG. 2 is a diagram of the training process of the present invention for generating a confrontation network model.
Fig. 3 is a diagram of a generator architecture for generating a countermeasure network model of the present invention.
Fig. 4 is a diagram of a discriminator structure for generating a countermeasure network model of the present invention.
Detailed Description
The present disclosure and embodiments are further explained in detail in the following description, taken in conjunction with the accompanying drawings, for the purpose of facilitating the understanding of the disclosure by those skilled in the art. The following examples are presented only to illustrate and explain the present invention, but not to limit the present invention. The invention is capable of other and different embodiments and its several details are capable of modification in various, obvious aspects all without departing from the spirit and scope of the present invention.
The invention receives and sends WiFi signals through a group of transmitting terminal and receiving terminal equipment, and channel state information of the WiFi signals is obtained by utilizing Linux802.11n CSI Tool. When a user acts in the environment, interference is generated on the propagation path of the WiFi signal, so that the WiFi signal is changed correspondingly. The invention realizes action recognition by analyzing the change of the WiFi CSI.
Embodiments of the present invention will be described below with reference to the drawings.
Fig. 1 is a general flowchart of an action recognition method based on generation of a countermeasure network in a WiFi environment, which is specifically implemented as:
step 1: and preprocessing the received WiFi CSI data to obtain input data corresponding to the WiFi CSI data to be identified.
Preferably, the preprocessing the received WiFi CSI data to be identified to obtain input data corresponding to the WiFi CSI data to be identified includes: extracting an amplitude value in the WiFi CSI data to be identified; performing linear interpolation from the extracted CSI values, unifying the shapes of the CSI, and filling the lengths of all actions to be equal; the data are divided into marked data and unmarked data, wherein the data of one user are selected to be divided into unmarked data and a test set, and the data of the other users are used as a training set and input as input data to generate the confrontation network model.
Step 2: inputting the input data to a generation countermeasure network model for action recognition.
The generating of the antagonistic network model comprises: a generator model for generating dummy samples to augment the data; and the discriminator model is used for extracting characteristics of the input data, carrying out countermeasure training with the generator model, and classifying to obtain the action recognition class probability.
FIG. 2 is a diagram of the training process of the present invention for generating a confrontation network model. Inputting the input data into a training process for generating an confrontation network model for action recognition, wherein the training process comprises the following steps:
using 100-dimensional Gaussian random noise as the input of a generator, learning a real data sample to fit noise distribution, and generating a false data sample which is as real as possible; and outputting the training set, the unlabeled data and the false samples generated by the generator through the confrontation training with the generator and the sample distribution of the learning real data to obtain a k +1 class sample type, wherein the k class real sample type and the k +1 th class are used for generating the false samples.
Fig. 3 is a diagram of a generator architecture for generating a countermeasure network model of the present invention. The generator network of the invention consists of 3 hidden layers, each using SoftPlus as activation function, as shown in the following equation:
softplus(x)＝log(1+exp(x))
the last layer of the generator uses Tanh as the activation function, as shown in the following equation:
fig. 4 is a diagram of a discriminator structure for generating a countermeasure network model of the present invention. The discriminator network of the invention consists of 3 hidden layers, each using LeakyReLU as the activation function, as shown in the following equation:
the last layer of the discriminator uses softmax as output to classify k +1 classes as shown in the following formula:
the discriminator loss function is shown below:
L＝Lsupervised+Lunsupervised
setting the hyper-parameters: the learning rate is set to 0.0006, and the optimization method is Adam algorithm.
The verification method comprises the following steps: using leave-one-subject-out-validation and 5-Fold cross-validation;
the above description is provided for the preferred embodiments of the present invention, so that those skilled in the art can understand the technical solutions of the present invention, however, these embodiments are only examples, and the specific embodiments of the present invention are not considered to be limited to the description of these embodiments, and simple deductions and changes can be made without departing from the inventive concept of the present invention, and all of them should fall within the protection scope of the present invention.
Claims (4)
1. A method for recognizing actions based on generation of countermeasure networks in a WiFi environment is characterized by comprising the following steps:
step 1: obtaining an amplitude value of received Channel State Information (CSI) of the WiFi, and transposing to obtain an N multiplied by 30 multiplied by Nr multiplied by Nt matrix, wherein N represents the number of data packets included in one motion sample, Nr represents the number of antennas of receiving equipment, and Nt represents the number of antennas of sending equipment;
step 2: performing linear interpolation on the CSI, and normalizing the CSI into a matrix of 200 multiplied by 30 multiplied by Nr multiplied by Nt as data;
and step 3: making a labeled data set from one part of data and corresponding labels, inputting the data set into a generated countermeasure network for training without making the labels of the other part of data as unlabeled data sets;
and 4, step 4: and applying the trained generated confrontation network model to action recognition.
2. The method for recognizing actions based on generation of countermeasure network in WiFi environment of claim 1, wherein said generation of countermeasure network model in step 3 comprises:
a generator model for generating false CSI sample data for a Gaussian random noise vector of input 100 dimensions;
the discriminator model extracts features from the input labeled and unlabeled data sets, and performs a countertraining with the generator and classifies the data.
3. The method for recognizing actions based on generation of countermeasure networks in a WiFi environment as claimed in claim 2, wherein the model of the generator comprises: an input layer, a batch normalization layer, a deconvolution layer, a SoftPlus layer and a Tanh layer; the model of the discriminator includes: the device comprises an input layer, a convolution layer, a LEAKYRELU layer, a Gaussian noise layer, a NiN layer, a maximum pooling layer, a full connection layer and a SoftMax layer.
4. The method for recognizing actions based on generation of countermeasure network in WiFi environment of claim 2, wherein the training process for generating countermeasure network model comprises:
generating, with a generator, CSI dummy samples;
inputting CSI false samples and labeled data sets and unlabeled data sets by a discriminator to carry out feature extraction, and carrying out countermeasure training with a generator; updating discriminator model parameters LDAnd the generator model parameters LG；
Wherein, the Adam algorithm is adopted to carry out comparison on the identifier model parameter LDAnd the generator model parameters LGAnd (6) optimizing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110654354.8A CN113378718A (en) | 2021-06-10 | 2021-06-10 | Action identification method based on generation of countermeasure network in WiFi environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110654354.8A CN113378718A (en) | 2021-06-10 | 2021-06-10 | Action identification method based on generation of countermeasure network in WiFi environment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113378718A true CN113378718A (en) | 2021-09-10 |
Family
ID=77574041
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110654354.8A Pending CN113378718A (en) | 2021-06-10 | 2021-06-10 | Action identification method based on generation of countermeasure network in WiFi environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113378718A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115913415A (en) * | 2022-11-09 | 2023-04-04 | 华工未来科技(江苏)有限公司 | RIS (remote location system) assisted WIFI (Wireless Fidelity) signal action identification method and device and storage medium |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN108769993A (en) * | 2018-05-15 | 2018-11-06 | 南京邮电大学 | Based on the communication network abnormal user detection method for generating confrontation network |
| CN109190524A (en) * | 2018-08-17 | 2019-01-11 | 南通大学 | A kind of human motion recognition method based on generation confrontation network |
| WO2020038548A1 (en) * | 2018-08-20 | 2020-02-27 | Telefonaktiebolaget Lm Ericsson (Publ) | Improving immune system of sites using generative adversarial networks and reinforcement learning |
| CA3076646A1 (en) * | 2019-03-22 | 2020-09-22 | Royal Bank Of Canada | System and method for generation of unseen composite data objects |
| EP3748544A1 (en) * | 2019-06-03 | 2020-12-09 | IMRA Europe S.A.S. | Mixture distribution estimation for future prediction |
| CN112633377A (en) * | 2020-12-24 | 2021-04-09 | 电子科技大学 | Human behavior prediction method and system based on generation of confrontation network |
| US20210125075A1 (en) * | 2019-10-24 | 2021-04-29 | Lg Electronics Inc. | Training artificial neural network model based on generative adversarial network |
-
2021
- 2021-06-10 CN CN202110654354.8A patent/CN113378718A/en active Pending
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108520199A (en) * | 2018-03-04 | 2018-09-11 | 天津大学 | Based on radar image and the human action opener recognition methods for generating confrontation model |
| CN108769993A (en) * | 2018-05-15 | 2018-11-06 | 南京邮电大学 | Based on the communication network abnormal user detection method for generating confrontation network |
| CN109190524A (en) * | 2018-08-17 | 2019-01-11 | 南通大学 | A kind of human motion recognition method based on generation confrontation network |
| WO2020038548A1 (en) * | 2018-08-20 | 2020-02-27 | Telefonaktiebolaget Lm Ericsson (Publ) | Improving immune system of sites using generative adversarial networks and reinforcement learning |
| CA3076646A1 (en) * | 2019-03-22 | 2020-09-22 | Royal Bank Of Canada | System and method for generation of unseen composite data objects |
| EP3748544A1 (en) * | 2019-06-03 | 2020-12-09 | IMRA Europe S.A.S. | Mixture distribution estimation for future prediction |
| US20210125075A1 (en) * | 2019-10-24 | 2021-04-29 | Lg Electronics Inc. | Training artificial neural network model based on generative adversarial network |
| CN112633377A (en) * | 2020-12-24 | 2021-04-09 | 电子科技大学 | Human behavior prediction method and system based on generation of confrontation network |
Non-Patent Citations (1)
| Title |
|---|
| 刘希文等: "一种基于CSI的人体动作计数与识别方法", 北京邮电大学学报, vol. 43, no. 5, pages 105 - 111 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115913415A (en) * | 2022-11-09 | 2023-04-04 | 华工未来科技(江苏)有限公司 | RIS (remote location system) assisted WIFI (Wireless Fidelity) signal action identification method and device and storage medium |
| CN115913415B (en) * | 2022-11-09 | 2024-02-02 | 华工未来科技(江苏)有限公司 | WIFI signal action recognition method and device based on RIS assistance and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112036433B (en) | CNN-based Wi-Move behavior sensing method | |
| US11907847B2 (en) | Operating machine-learning models on different platforms | |
| CN111027487B (en) | Behavior recognition system, method, medium and equipment based on multi-convolution kernel residual error network | |
| Zhang et al. | WiFi-based cross-domain gesture recognition via modified prototypical networks | |
| CN110674875A (en) | Pedestrian motion mode identification method based on deep hybrid model | |
| CN111954250B (en) | Lightweight Wi-Fi behavior sensing method and system | |
| CN114423034B (en) | Indoor personnel action recognition method, system, medium, equipment and terminal | |
| CN104700100A (en) | Feature extraction method for high spatial resolution remote sensing big data | |
| Yu et al. | Fingerprint extraction and classification of wireless channels based on deep convolutional neural networks | |
| CN110353693A (en) | A kind of hand-written Letter Identification Method and system based on WiFi | |
| Moshiri et al. | CSI-based human activity recognition using convolutional neural networks | |
| CN113378718A (en) | Action identification method based on generation of countermeasure network in WiFi environment | |
| CN114980122A (en) | Small sample radio frequency fingerprint intelligent identification system and method | |
| CN112380903B (en) | Human body activity recognition method based on WiFi-CSI signal enhancement | |
| Zhong et al. | A climate adaptation device-free sensing approach for target recognition in foliage environments | |
| CN115276857B (en) | Full-blind spectrum detection method based on combination of Cholesky decomposition and convolutional neural network | |
| CN114764575B (en) | Multi-modal data classification method based on deep learning and time sequence attention mechanism | |
| Tian et al. | Small CSI Samples‐Based Activity Recognition: A Deep Learning Approach Using Multidimensional Features | |
| CN115438691A (en) | Small sample gesture recognition method based on wireless signals | |
| CN113869238A (en) | Cognitive Internet of vehicles intelligent frequency spectrum sensing method and system | |
| CN114998731A (en) | Intelligent terminal navigation scene perception identification method | |
| CN114724245A (en) | CSI-based incremental learning human body action identification method | |
| Rawat et al. | A Classifier Approach using Deep Learning for Human Activity Recognition | |
| CN113642457A (en) | Cross-scene human body action recognition method based on antagonistic meta-learning | |
| Yang et al. | Deep learning and unsupervised domain adaptation for WiFi-based sensing |
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 |