CN116340849B - Non-contact type cross-domain human activity recognition method based on metric learning - Google Patents
Non-contact type cross-domain human activity recognition method based on metric learning Download PDFInfo
- Publication number
- CN116340849B CN116340849B CN202310556403.3A CN202310556403A CN116340849B CN 116340849 B CN116340849 B CN 116340849B CN 202310556403 A CN202310556403 A CN 202310556403A CN 116340849 B CN116340849 B CN 116340849B
- Authority
- CN
- China
- Prior art keywords
- data
- activity
- encoder
- csi
- support set
- 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
- 230000000694 effects Effects 0.000 title claims abstract description 110
- 238000000034 method Methods 0.000 title claims abstract description 60
- 238000012549 training Methods 0.000 claims abstract description 30
- 238000000605 extraction Methods 0.000 claims abstract description 13
- 238000007781 pre-processing Methods 0.000 claims abstract description 4
- 239000000523 sample Substances 0.000 claims description 19
- 230000000875 corresponding effect Effects 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 11
- 239000013598 vector Substances 0.000 claims description 8
- 230000005540 biological transmission Effects 0.000 claims description 4
- 238000005259 measurement Methods 0.000 claims description 4
- 230000002159 abnormal effect Effects 0.000 claims description 2
- 239000013074 reference sample Substances 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims 1
- 230000001502 supplementing effect Effects 0.000 claims 1
- 230000009471 action Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 6
- 230000008569 process Effects 0.000 description 6
- 239000011159 matrix material Substances 0.000 description 4
- 230000006399 behavior Effects 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000036544 posture Effects 0.000 description 2
- 238000002679 ablation Methods 0.000 description 1
- 230000001133 acceleration Effects 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000005452 bending Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000010924 continuous production Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000003550 marker Substances 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
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
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
-
- 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
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Image Analysis (AREA)
Abstract
A non-contact cross-domain human activity recognition method based on metric learning is characterized in that data enhancement is carried out on collected activity data by adopting a self-encoder, and then recognition of new activity categories which are not in a training set is completed by adopting the metric learning. The method comprises the following specific steps: acquiring corresponding wireless signal data when personnel are active in an indoor environment, and extracting CSI original data from the wireless signal data; performing data preprocessing, data interpolation, unifying data length and data denoising on original CSI data; performing data enhancement on the data with known activity types by using a self-encoder, and expanding a data set; and (3) using the expanded data set training feature extraction network to input the to-be-identified activity data and the support set into the feature extraction network to obtain corresponding features, and using a metric learning method to compare the to-be-identified data features with the support set data features one by one so as to judge the activity type. The method can realize higher recognition precision for untrained activity types, and enhances generalization and robustness.
Description
Technical Field
The invention relates to the field of activity recognition, in particular to a non-contact type cross-domain human activity recognition method based on metric learning.
Background
Wi-Fi based human activity recognition (Human Activity Recognition, HAR) aims to recognize human activity using Wi-Fi signals. Wi-Fi-based human body recognition is receiving more and more attention in the aspects of building human-computer interaction, intelligent monitoring and intrusion detection intelligent application because special equipment and additional deployment cost are not required to be worn by users. However, the existing Wi-Fi aware public data set is very deficient due to the large amount of manpower and material resources required to collect channel state information (ChannelState Information, CSI) data. Thus, wi-Fi perception does not have enough marker data to train a machine learning model with good performance. Meanwhile, the wireless signal is affected by physical environment and human behavior, and similar signal fluctuation may be caused by different action types of human bodies (positions, postures or actions) in the transmission process, which certainly complicates the activity recognition process. Particularly in practical applications, the system may need to identify some activity types that are never seen during the training phase, which is a significant challenge.
Conventional Wi-Fi based identification generally adopts a supervised method, only predefined activities can be identified, and a large amount of data is required to learn enough prior knowledge, and the complex training process of the Wi-Fi based identification exacerbates the cost of model training, so that the overall overhead is greatly increased. These activity recognition methods first collect a large number of tagged samples and then train a conventional machine learning model to recognize the type of activity. However, when it is required to identify a new activity type which does not occur in the training process, since the activity type of the new data is different from the activity type of the data in the training set, it is required to identify the new activity data across the types, and at this time, the identification performance of the model of the above method may be seriously degraded. To address this problem, recent efforts have collected some new types of activity data that are not present in the training set to retrain the model. However, the process of retraining undoubtedly increases the overhead of model training.
In the field of computer vision, metric learning is widely used to identify unseen activities and scenes. Metric learning problems are generally described as an optimization problem that optimizes some objective function that measures data similarity. Mapping data from the original vector space to the hidden space results in activity-related features. For the above-mentioned problems, since the data set of Wi-Fi CSI is very small, there may be a case where it is necessary to identify an unknown type of activity in an actual scene.
Disclosure of Invention
The invention aims to provide a non-contact cross-domain human activity recognition method based on metric learning, which uses a self-encoder to expand an activity data set to enrich the data set and improve recognition accuracy, and then realizes the function of recognizing new activity categories which do not appear in a training set by comparing similarity between data based on the metric learning method.
A non-contact cross-domain human activity recognition method based on metric learning comprises the following steps:
step 1, acquiring corresponding wireless signal data when personnel are moving in an indoor environment, and extracting an original signal of CSI from the wireless signal data;
step 2, carrying out data preprocessing on the original CSI data acquired in the step 1, carrying out data interpolation to unify the length, and denoising the data to remove noise signals caused by hardware equipment;
step 3, using the self-encoder to carry out data enhancement on the data processed in the step 2, and generating data relevant to corresponding activities so as to enlarge a data set;
and 4, establishing an activity recognition model by using a method based on measurement learning, training the data generated in the step 3 and the original data, learning the similarity between the data, training an activity related characteristic extraction model, comparing the characteristics of the data to be recognized and the support set data, and recognizing the unknown type of activity.
The invention has the beneficial effects that:
(1) And comparing the data characteristics of the unknown type of activity to be identified with the data characteristics of the support set one by using a measurement learning method, thereby realizing the identification of the unknown type of activity.
(2) The method solves the problem of unknown type activity recognition in wireless perception, can realize higher recognition precision for untrained activity types, does not need extra training cost, and enhances the robustness of an activity recognition network.
(3) Inspired by the metric learning core idea, the network is trained by the criterion that the similar samples are closer and the dissimilar samples are farther, and the identification of the unknown type of activity can be realized without complex training strategies.
Drawings
Fig. 1 is a flow chart of a method for identifying an unknown type of activity based on channel state information in an embodiment of the present invention.
Fig. 2 is a network architecture diagram of an unknown activity recognition network in an embodiment of the present invention.
FIG. 3 is a schematic diagram of cross-domain activity recognition effects in a laboratory scenario in an embodiment of the present invention.
Fig. 4 is a schematic diagram of a cross-domain activity recognition effect in a conference room scenario in an embodiment of the present invention.
Fig. 5 is a schematic diagram of a cross-domain activity recognition effect in a bedroom scenario in an embodiment of the present invention.
Fig. 6 is a schematic diagram of a cross-domain activity recognition effect in a corridor scenario in an embodiment of the present invention.
Fig. 7 is a schematic diagram showing the effect of the data enhancement method on the activity recognition accuracy in the embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further described in detail below with reference to the attached drawings.
As shown in fig. 1, the invention provides a non-contact type cross-domain human activity recognition method based on metric learning, which comprises the following steps:
step 1: the devices were deployed in four different daily life scenarios (laboratory, conference room, office, bedroom) using a computer equipped with an Intel 5300 network card and three antennas as the receiver of the experiment, and a computer equipped with an Intel 5300 network card and one antenna as the transmitter of the experiment. Wherein the straight line distance between the transmitter and the receiver is 3 meters and the height is 1 meter from the ground. The transceiver receives and transmits data packets during the activities of the volunteers, and collects personnel activity related data. CSI raw data for human activity recognition is extracted from the acquired activity data using linux802.11 CSI tools.
Step 2: CSI (channel state information) characterizes the propagation space of a signal, and different activities can lead to different multipath propagation of the signal.
The wireless signal transmitted by the transmitting end is influenced by physical environment or human factors (position, posture and action) in the transmission process, so that multiple paths such as direct radiation, reflection and scattering are formed to propagate, and multipath effect is generated. When personnel are active, the signal at the receiving end changes. The signal received by the receiver at this point reflects multipath variations due to personnel activity. The CSI contains amplitude and phase information on each subcarrier, characterizes the spatial propagation of the signal, and can reflect the signal variation. The signal received by the receiver can be described as:
wherein Representing the CSI information for a channel matrix; the received and transmitted signal vectors are respectivelyAnd；is additive white gaussian noise.
Since radio frequency signals propagate through multiple paths from a transmitter to a receiver in an indoor environment, CSI is a superposition of all path signals. The channel matrix can thus be expressed as:
here, theIs the number of paths that are to be followed,is the complex attenuation of the light and the light,is the firstThe propagation length of the path of the strip,is the wavelength. The model can obtain that when indoor personnel are active, signals are reflected by human bodies, and the signal path length is changed, so that the channel state matrix is influenced.
Step 2-1: and (3) carrying out data interpolation, wherein due to the influence of equipment hardware and a channel, a packet loss phenomenon can occur in the data transmission process, and the CSI data is complemented by using cubic spline interpolation according to a time stamp, so that the length of the CSI data is uniform. The interpolation function is directly provided in the interpolation complement Matlab, so that the method is a common data processing method.
Step 2-2: data denoising, and for human daily activity behaviors, the amplitude of the CSI fluctuates between 30 and 60 and Hz. The invention utilizes a Butterworth low-pass filter to remove the CSI data in the step 2-1, removes high-frequency noise in the environment, and utilizes a Hample filter to remove abnormal values in the CSI data to obtain preprocessed data.
Step 3: to generate more corresponding activity data from the collected data, a multi-self encoder module is designed, wherein each sub-module is dedicated to one activity type. The overall framework of the data enhancement module is shown in fig. 2 (a), forFirst, theType of activity, use the firstAn automatic encoder generates relevant activity data. Wherein each automatic encoder is an encoder-decoder structure. First, the data is compressed by an encoder, the characteristic information related to the activity is extracted, and the input is encoded into normal distribution. Suppose that data is generatedThe probability of (2) isThe method comprises the steps of carrying out a first treatment on the surface of the The encoder is at the input ofThe probability of outputting the potential vector z under the condition of (2) is. The extracted features and gaussian random noise are then input to a decoder to reconstruct the data, wherein the decoder is described as。
Finally, a loss function is introduced during the training phase. The goal is to train the probability of the encoder outputting the potential vector zProbability of approximating generated potential vector z, wherein Compliance withIs a normal distribution of (c). The encoder loss is defined as the Kullback-Leibler divergence,andrespectively represent categoriesNormal distribution of the distribution and variance of the distribution.
Output mixing noise of encoderReconstructing an input in an input decoder, wherein To be a true value, minimizing the reconstruction loss to be
Then, the overall loss function of the data enhancement module is defined as:
a large amount of synthetic data similar to but different from the training data may be generated by the data enhancement module to augment the training set.
Step 4: for unknown activity types, they cannot be mapped directly to known tags. Inspired by the measurement and learning thought, the simple classification problem can be converted into the comparison problem, namely the similarity between the comparison samples, by learning the similarity between the two samples. The invention thus proposes an activity recognition module.
Step 4-1: the data set is partitioned.
Based on step 3, an enhanced extended data set is obtained, and the data set is used as a training set. Thereafter, byLabeling training sets in the form of (1), wherein positive samplesAnd a reference sampleWith the same class labels, and negative samplesThen there is a different activity category label. Thus, the training data set may be represented as, wherein Corresponds to the firstSamples.
For active data of an unknown type, a set of data (one class for each sample) is randomly selected from the active data set of the unknown type to generate a support set. Describing a support set as, wherein To randomly extract samplesThe corresponding label is used to identify the label,corresponding to the number of unknown types of activity.
Step 4-2: the feature extractor is trained to extract activity-related features.
The overall framework of the feature extraction module is shown in fig. 2 (b). In particular, the activity recognition method consists of three modules that use the same feed-forward network and share the same parameters. The activity recognition method has three inputs and two corresponding intermediate output values, which represent the L2 distance of the two inputs; to output a comparison operator from the model,Softmaxa function is applied to the output to create a ratiometric; the feedforward network mainly comprises six network layers, namely CNN- & gtLSTM- & gtCNN- & gtLinear; the CNN layer learns deep characteristic signals of CSI, which are formed by BatchNorm- & gtCNN- & gtMaxpool; the LSTM layer learns the correlation of the CSI signal in the time domain, specifically, the correlation is from BatchNorm to LSTM to Dropout; the Linear layer Linear is composed of Linear- & gt Relu- & gt Linear, and the extracted data is mapped into the feature space.
The aim of the training is to keep the data distance between the same category as small as possible and the data distance between different categories as large as possible. Loss is therefore defined as:
wherein ,
in the formula, the function E () refers to a feature extractor; the goal is to get the loss of the feature extraction network towards 0, furthermore define，Representation ofAndthe euclidean distance between the two,representation ofAndeuclidean distance between them.
Three feed forward networks are updated simultaneously using a back propagation algorithm. Finally, the training set is input into a feature extractor, and a feature extractor is obtained through training to extract the features related to the activities in the CSI data.
Step 4-3: and (5) activity identification.
As shown in fig. 2, the input support setAnd the target sample data is sent to a feature extraction network to obtain corresponding activity features. In the activity recognition process, as shown in fig. 2 (c), the similarity between the characteristic values of the target sample and the characteristic values of all samples in the support set is measured, and the label corresponding to the support set sample with the largest similarity is selected as the label of the target sample。
wherein Refers to the label corresponding to the sample in the support set, < >>A finger support set, i refers to the ith data in the support set; support set sample and target sample->Cosine similarity between->Is that
The identification of the unknown type of activity is accomplished through the above steps.
In order to evaluate the robustness of the method in different environments, the proposed method is implemented in 4 indoor scenarios (laboratory, conference room, bedroom and living room) with various complex wireless environments. Each scene uses two notebook computers (Think-pad X200) as transceivers, both devices are equipped with Intel 5300 cards, and Linux802.11 n CSI Tool is installed for collecting CSI data. The number of transmitting end antennas is nt=1, and the number of receiving end antennas is nr=3. CSI data of 1×3×30=90 subcarriers can be obtained by collecting personnel activity signals, wherein each transceiver antenna pair has 30 subcarriers. In the method, the signal frequency is 5.8 GHz, the bandwidth is 20MHz, and the sampling frequency is 200Hz.
To assess the adaptability of the method to different people, in each experimental scenario 7 volunteers of different heights and weights were recruited to complete 8 activities (standing, sitting down, bending down, travelling, hand-engaging and turning around). The data for each scene is divided into a training set (containing four activity types), a testing set (containing four unseen activity types), and a support set (containing only one sample of each activity).
The accuracy is used to evaluate the performance of the system.
The identification effect of the invention is as follows:
1. and the overall recognition accuracy rate.
The proposed method is verified in four scenarios. As can be seen from the following table, compared with some existing networks, the identification accuracy of the method is obviously improved. Specifically, compared with CNN and MatNet activity recognition accuracy, the activity recognition method has the advantage that the activity recognition accuracy is remarkably improved; in addition, the accuracy rate of the method provided by the invention under 4 life scenes is improved by 9.31%, 5.25%, 10.22% and 21.33% respectively compared with the transfer learning method. This is mainly due to the fact that the method provided by the invention mines the relation between the data, and the similarity between the data and the support set data is compared to identify the sample, so that the sample can be distinguished more effectively.
TABLE 1
| Model | Laboratory | Meeting room | Bedroom | Parlor |
| CNN | 26.15 | 23.48 | 23.79 | 21.74 |
| MatNet | 55.35 | 52.87 | 53.10 | 52.61 |
| Transfer | 63.19 | 68.85 | 70.28 | 56.67 |
| WIUS | 72.50 | 74.10 | 80.50 | 78.00 |
2. New category activity recognition accuracy that does not appear in the training set.
The effectiveness of the activity recognition method of the present invention in different categories of activity is detailed by drawing a confusion matrix of the activity classification. As shown in fig. 3-6, the accuracy of recognition of the new category activity reaches 72.5%, 74.1%, 80.5% and 78%, respectively. Experimental results show that the probability of correct recognition of the unseen "standing" action is higher than the other actions. The reason is that when people stand up they first experience a rapid acceleration and then the speed drops to zero. In addition, the standing up action is a continuous process of approaching the transceiver pair, which is well distinguished from other activities. While both "waving" and "turning around" have a process of getting close to the transceiver pair after going far from the transceiver link, so they are easily misclassified into the category of the other party.
3. The impact of the data enhancement method on the accuracy of the activity recognition.
To verify the effectiveness of the data enhancement method proposed in the present method, ablation analysis was performed with and without cross-domain activity recognition of the data enhancement method. As shown in FIG. 7, the unknown activity recognition accuracy of the method is greatly improved after the data set is enhanced, and the average recognition accuracy of the unknown activity of the method is improved by 10.95% in four life scenes. Particularly in bedrooms and corridors, the expansion of the data set enables the method to have enough data to train the model, and the feature extraction capacity of the model is improved.
The above description is merely of preferred embodiments of the present invention, and the scope of the present invention is not limited to the above embodiments, but all equivalent modifications or variations according to the present disclosure will be within the scope of the claims.
Claims (4)
1. A non-contact cross-domain human activity recognition method based on metric learning is characterized in that: the method comprises the following steps:
step 1, acquiring corresponding wireless signal data when personnel are moving in an indoor environment, and extracting an original signal of CSI from the wireless signal data;
step 2, carrying out data preprocessing on the original CSI data acquired in the step 1, carrying out data interpolation to unify the length, and denoising the data to remove noise signals caused by hardware equipment;
step 3, using the self-encoder to carry out data enhancement on the data processed in the step 2, and generating data relevant to corresponding activities so as to enlarge a data set;
step 4, an activity recognition model is established by using a method based on measurement learning, the data generated in the step 3 and the original data are trained, the similarity between the data is learned, an activity related feature extraction model is trained, and features of the data to be recognized and the support set data are compared to recognize unknown types of activities;
in step 4, the activity recognition module comprises a feature extraction network and an activity classification module;
step 4-1: dividing the data set;
based on the step 3, obtaining an enhanced extended data set, and taking the data set as a training set; the training dataset is represented as train = { (x) 1 ，x 1+ ，x 1- )，(x 2 ，x 2+ ，x 2- )，...(x i ，x i+ ，x i- ) Positive sample x + Having the same class label as the reference sample x, while the negative sample x - Then there is a different activity category label;
randomly selecting a set of data from an active data set of unknown type to generate a support set; describing the support set as wherein />Is a data sample->The corresponding label, k, is the number of unknown types of activity;
step 4-2: training a feature extractor to extract activity-related features;
the feature extraction module consists of three modules which use the same feed-forward network and share the same parameters; the feature extraction module has three inputs and two corresponding intermediate output values, which represent the L2 distance of the two inputs; to output a comparison operator from the model, a Softmax function is applied to the output to create a ratiometric; the feedforward network mainly comprises six network layers, namely CNN- & gtLSTM- & gtCNN- & gtLinear; the CNN layer learns deep characteristic signals of CSI, which are formed by BatchNorm- & gtCNN- & gtMaxpool; the LSTM layer learns the correlation of the CSI signal in the time domain, specifically, the correlation is from BatchNorm to LSTM to Dropout; the Linear layer Linear consists of Linear, relu, and Linear, and the extracted data is mapped into the feature space;
the purpose of training is to make the data distance between the same class as small as possible and the data distance between different classes as large as possible, so the penalty is defined as:
wherein ,
in the formula, the function E () refers to a feature extractor; the goal is to make the loss of the feature extraction network approach 0, defining margin=1, d - Represents x - And Euclidean distance between x, d + Represents x + And x, and x;
step 4-3: activity recognition;
inputting the support set S and target sample data to a feature extraction network to obtain corresponding activity features; measuring the similarity between the characteristic values of the target sample and the characteristic values of all samples in the support set, and selecting the label corresponding to the support set sample with the maximum similarity as the label y of the target sample t ；
wherein ,the method refers to a label corresponding to a sample in a support set, S refers to the support set, and i refers to the ith data in the support set; support set sample and target sample x t Cosine similarity d between k The method comprises the following steps:
2. the non-contact cross-domain human activity recognition method based on metric learning of claim 1, wherein the method comprises the following steps: in step 1, linux802.11 CSI tools are used to extract the original signal of CSI from it.
3. The non-contact cross-domain human activity recognition method based on metric learning of claim 1, wherein the method comprises the following steps: in step 2, the data preprocessing method is as follows:
step 2-1: data interpolation, namely supplementing the CSI data by using cubic spline interpolation according to a time stamp for the packet loss phenomenon in the data transmission process;
step 2-2: denoising the data, namely, denoising the CSI data in the step 2-1 by using a Butterworth low-pass filter, removing high-frequency noise in the environment, and removing abnormal values in the CSI data by using a Hampel filter to obtain preprocessed data.
4. The non-contact cross-domain human activity recognition method based on metric learning of claim 1, wherein the method comprises the following steps: in step 3, the specific method for enhancing data is as follows:
generating associated activity data using automatic encoders, wherein each automatic encoder is an encoder-decoder structure; firstly, compressing data by an encoder, extracting characteristic information related to activities of the data, and encoding input into normal distribution N (0, sigma); assuming that the probability of generating data x is P (x); the probability of the potential vector z output by the encoder under the condition that the input is x is P (z|x); then inputting the extracted features and Gaussian random noise into a decoder to reconstruct the data;
finally, introducing a loss function in a training stage; the goal is to train the encoder to output a probability P (z|x) of the potential vector z that approximates the probability P (z) of generating the potential vector z, where z corresponds to a normal distribution of N (0, σ); the encoder loss is defined as Kullback-Leibler divergence, μ i and σi Respectively representing the expected and variance of the normal distribution of the category i;
L encoder ＝D KL (N(μ i ，σ i )，N(0，σ))；
the output of the encoder mixes noise n into the decoder to reconstruct the input x, whereFor a true value, minimizing the reconstruction loss is:
then, the overall loss function of the data enhancement module is defined as:
L aug ＝min(L encoder ，L decoder )。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310556403.3A CN116340849B (en) | 2023-05-17 | 2023-05-17 | Non-contact type cross-domain human activity recognition method based on metric learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310556403.3A CN116340849B (en) | 2023-05-17 | 2023-05-17 | Non-contact type cross-domain human activity recognition method based on metric learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116340849A CN116340849A (en) | 2023-06-27 |
| CN116340849B true CN116340849B (en) | 2023-08-15 |
Family
ID=86895199
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310556403.3A Active CN116340849B (en) | 2023-05-17 | 2023-05-17 | Non-contact type cross-domain human activity recognition method based on metric learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116340849B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110879989A (en) * | 2019-11-22 | 2020-03-13 | 四川九洲电器集团有限责任公司 | Ads-b signal target identification method based on small sample local machine learning model |
| CN113158905A (en) * | 2021-04-23 | 2021-07-23 | 佛山市南海区广工大数控装备协同创新研究院 | Pedestrian re-identification method based on attention mechanism |
| CN114998995A (en) * | 2022-06-13 | 2022-09-02 | 西安电子科技大学 | Cross-view-angle gait recognition method based on metric learning and space-time double-flow network |
| CN115146670A (en) * | 2022-05-30 | 2022-10-04 | 西安交通大学 | Radio frequency fingerprint identification method and system based on data enhancement and comparison learning |
| CN115937910A (en) * | 2022-12-01 | 2023-04-07 | 湖南工商大学 | Palm print image identification method based on small sample measurement network |
-
2023
- 2023-05-17 CN CN202310556403.3A patent/CN116340849B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110879989A (en) * | 2019-11-22 | 2020-03-13 | 四川九洲电器集团有限责任公司 | Ads-b signal target identification method based on small sample local machine learning model |
| CN113158905A (en) * | 2021-04-23 | 2021-07-23 | 佛山市南海区广工大数控装备协同创新研究院 | Pedestrian re-identification method based on attention mechanism |
| CN115146670A (en) * | 2022-05-30 | 2022-10-04 | 西安交通大学 | Radio frequency fingerprint identification method and system based on data enhancement and comparison learning |
| CN114998995A (en) * | 2022-06-13 | 2022-09-02 | 西安电子科技大学 | Cross-view-angle gait recognition method based on metric learning and space-time double-flow network |
| CN115937910A (en) * | 2022-12-01 | 2023-04-07 | 湖南工商大学 | Palm print image identification method based on small sample measurement network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116340849A (en) | 2023-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Yan et al. | WiAct: A passive WiFi-based human activity recognition system | |
| Ali et al. | Recognizing keystrokes using WiFi devices | |
| CN108629380B (en) | Cross-scene wireless signal sensing method based on transfer learning | |
| Wang et al. | WiCAR: WiFi-based in-car activity recognition with multi-adversarial domain adaptation | |
| Huang et al. | WiDet: Wi-Fi based device-free passive person detection with deep convolutional neural networks | |
| Sobron et al. | Device-free people counting in IoT environments: New insights, results, and open challenges | |
| WO2022037130A1 (en) | Network traffic anomaly detection method and apparatus, and electronic apparatus and storage medium | |
| Shi et al. | Human activity recognition using deep learning networks with enhanced channel state information | |
| CN104135327A (en) | Spectrum sensing method based on support vector machine | |
| CN111597991A (en) | Rehabilitation detection method based on channel state information and BilSTM-Attention | |
| CN110929842B (en) | Accurate intelligent detection method for non-cooperative radio signal burst time region | |
| Zhu et al. | NotiFi: A ubiquitous WiFi-based abnormal activity detection system | |
| CN114423034B (en) | Indoor personnel action recognition method, system, medium, equipment and terminal | |
| Hao et al. | CSI-HC: a WiFi-based indoor complex human motion recognition method | |
| CN110059612A (en) | A kind of gesture identification method and system that the position based on channel state information is unrelated | |
| Shi et al. | Towards environment-independent human activity recognition using deep learning and enhanced CSI | |
| Tsai et al. | Refined autoencoder-based CSI hidden feature extraction for indoor spot localization | |
| Cheng et al. | Deep learning network based spectrum sensing methods for OFDM systems | |
| Zhu et al. | Wi-ATCN: Attentional temporal convolutional network for human action prediction using WiFi channel state information | |
| Mo et al. | A deep learning-based human identification system with wi-fi csi data augmentation | |
| Zha et al. | LT-SEI: Long-tailed specific emitter identification based on decoupled representation learning in low-resource scenarios | |
| Shi et al. | Deep learning networks for human activity recognition with CSI correlation feature extraction | |
| Kabir et al. | CSI-IANet: An inception attention network for human-human interaction recognition based on CSI signal | |
| Meneghello et al. | Smartphone identification via passive traffic fingerprinting: A sequence-to-sequence learning approach | |
| CN114757224A (en) | Specific radiation source identification method based on continuous learning and combined feature extraction |
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 |