CN115100733A - RFID gesture recognition method, computer device, product and storage medium - Google Patents

RFID gesture recognition method, computer device, product and storage medium Download PDF

Info

Publication number
CN115100733A
CN115100733A CN202210480019.5A CN202210480019A CN115100733A CN 115100733 A CN115100733 A CN 115100733A CN 202210480019 A CN202210480019 A CN 202210480019A CN 115100733 A CN115100733 A CN 115100733A
Authority
CN
China
Prior art keywords
gesture
rfid
phase sequence
samples
gesture recognition
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
Application number
CN202210480019.5A
Other languages
Chinese (zh)
Inventor
张士庚
马子靖
王伟平
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Central South University
Original Assignee
Central South University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Central South University filed Critical Central South University
Priority to CN202210480019.5A priority Critical patent/CN115100733A/en
Publication of CN115100733A publication Critical patent/CN115100733A/en
Pending legal-status Critical Current

Links

Images

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
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06K—GRAPHICAL DATA READING; PRESENTATION OF DATA; RECORD CARRIERS; HANDLING RECORD CARRIERS
    • G06K17/00—Methods or arrangements for effecting co-operative working between equipments covered by two or more of main groups G06K1/00 - G06K15/00, e.g. automatic card files incorporating conveying and reading operations
    • G06K17/0022—Methods or arrangements for effecting co-operative working between equipments covered by two or more of main groups G06K1/00 - G06K15/00, e.g. automatic card files incorporating conveying and reading operations arrangements or provisious for transferring data to distant stations, e.g. from a sensing device
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Databases & Information Systems (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Human Computer Interaction (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses an RFID gesture recognition method, a computer device, a product and a storage medium, which are used for acquiring a phase sequence generated by a user gesture; preprocessing the phase sequence and identifying a gesture type; acquiring a spectrogram of the preprocessed phase sequence; taking the spectrogram corresponding to each gesture as a training sample, and pairing each training sample to obtain a twin network data set; training a twin network by using the twin network data set to obtain an RFID gesture recognition model; and matching the gesture samples to be recognized with the identified gesture samples one by one, taking the matched data as the input of the RFID gesture recognition model, and recognizing the gesture samples to be recognized by utilizing a template matching method. The method of the invention not only can greatly reduce the training samples required by the recognition model and reduce the deployment cost of the wireless perception system, but also obtains higher recognition precision and strong practicability.

Description

RFID gesture recognition method, computer device, product and storage medium
Technical Field
The invention relates to the technical field of wireless sensing, in particular to an RFID gesture recognition method, a computer device, a product and a storage medium.
Background
With the rapid development of wireless networks, wireless sensing technology has been widely applied in people's lives. The wireless sensing technology is easy to deploy and low in cost, and can help users to enjoy better life quality without special equipment. However, the wireless sensing technology often encounters a big problem in practical application: due to the multipath influence of the wireless signals, the wireless signals are reflected by different reflectors in different environments, so that the acquired wireless signals are correspondingly changed, the performance of the wireless sensing system in a new environment is greatly reduced, and the application of the wireless sensing system in actual life is seriously hindered. This is known as the cross-scene problem of wireless perception.
The mainstream solution methods for the current cross-scene problem are divided into two types, namely data training is collected again in a new scene, knowledge is migrated from an existing scene by using migration learning, and training samples needed by a target scene are reduced. However, the wireless sensing system is time-consuming and labor-consuming to collect data, and can obtain a good effect only by collecting dozens or even hundreds of gestures, so that the wireless sensing system cannot be applied in practice; and migration learning requires that data distribution of a source scene and a target scene is similar, otherwise harmful knowledge can be migrated to cause negative migration, and meanwhile, the source scene is required to have sufficient data.
CN107633227B discloses a CSI-based fine-grained gesture recognition method and system, which classify the extracted CSI features by using a machine learning method, and classify the features into corresponding stroke sequences. However, such methods based on machine learning and deep learning all require a large amount of data to train the model, so that a more accurate decision boundary can be obtained.
Disclosure of Invention
The invention aims to solve the technical problem that aiming at the defects of the prior art, an RFID gesture recognition method, a computer device, a product and a storage medium are provided, and high accuracy of gesture classification can be obtained only by a small amount of training samples of a new scene.
In order to solve the technical problems, the technical scheme adopted by the invention is as follows: an RFID gesture recognition method comprises the following steps:
s1, acquiring a phase sequence generated by the user gesture;
s2, preprocessing the phase sequence, and identifying a gesture type corresponding to the gesture sample in the phase sequence;
s3, acquiring a spectrogram of the preprocessed phase sequence;
s4, taking the spectrogram corresponding to each gesture as a training sample, and pairing each training sample to obtain a twin network data set;
s5, training a twin network by using the twin network data set to obtain an RFID gesture recognition model;
s6, pairing the gesture samples to be recognized and the identified gesture samples one by one, taking paired data as input of the RFID gesture recognition model, and recognizing the gesture samples by using a template matching method.
The invention uses the twin network to input two gesture samples simultaneously, judges the similarity between the samples, and classifies the samples by using a template matching method to realize high-precision small sample gesture recognition. According to the method, the high accuracy of gesture classification can be obtained only by a small number of training samples of new scenes.
In step S2, the specific implementation process of preprocessing the phase sequence includes: filtering the phase sequence using a Savitzky-Golay filter to filter ambient noise; and normalizing the filtered phase sequence by using a Min-Max normalization method to obtain a preprocessed phase sequence.
In step S5, the twin network includes two weight-shared neural networks. Weight sharing means that two independent neural networks share the weight of one set of network. The weight sharing has the advantages that the parameter number of the network is reduced, and meanwhile, the output generated by the two networks to the same input is the same, so that the condition of different outputs of the same input can not occur.
The two neural networks have the same structure; the neural network comprises a first convolution layer, a first maximum pooling layer, a second convolution layer, a second maximum pooling layer and a full-connection layer which are connected in sequence. The convolutional layer can extract local characteristics of the spectrogram and enlarge the receptive field of the network so as to obtain characteristics with high identification degree; the pooling layer can effectively reduce the parameter quantity of the network and accelerate the network training speed; and finally, the full-connection layer converts the two-dimensional features into one-dimensional features, so that the similarity between the features can be calculated conveniently.
The loss function L (W) of the twin network is:
Figure BDA0003627412350000021
where W is the twin network parameter, N is the logarithm of the input,
Figure BDA0003627412350000022
is the n-th input of the input,
Figure BDA0003627412350000023
and
Figure BDA0003627412350000024
is any two gesture samples, Y n Is composed of
Figure BDA0003627412350000025
And
Figure BDA0003627412350000026
whether the same type of label is 0 or not, and if not, the same type of label is 1; d (W, S) n ) Euclidean distance for two input features; m is a boundary constraint on the heterogeneous sample distance such that the heterogeneous sample distance is constrained to be within m. The advantage of this loss function is: when the input sample pairs belong to the same class (i.e. Y) n 0), if
Figure BDA0003627412350000027
And
Figure BDA0003627412350000028
if the similarity is large, the loss function is also large, and vice versa; if not, then
Figure BDA0003627412350000029
And
Figure BDA00036274123500000210
the similarity between them is large and the loss function is small, and vice versa. Therefore, the network is optimized through the loss function, the similarity of the same type of samples can be improved, the similarity of different types of samples can be reduced, and the classification accuracy is improved.
As an inventive concept, the present invention also provides a computer arrangement comprising a memory, a processor and a computer program stored on the memory; the processor executes the computer program to implement the steps of the method of the present invention.
As an inventive concept, the present invention also provides a computer program product comprising computer programs/instructions; which when executed by a processor implement the steps of the method of the present invention.
As an inventive concept, the present invention also provides a computer-readable storage medium having stored thereon a computer program/instructions; which when executed by a processor, perform the steps of the method of the invention.
Compared with the prior art, the invention has the following beneficial effects: according to the RFID gesture recognition method based on small sample learning, the phase value caused by the user gesture is obtained through the RFID equipment, the time domain feature vector is expanded into the time-frequency domain through short-time Fourier transform, the user gesture is recognized through the twin network, and therefore high-accuracy RFID gesture recognition under the condition of small samples is achieved. The method of the invention not only can greatly reduce the training samples required by the recognition model and reduce the deployment cost of the wireless perception system, but also obtains higher recognition precision and strong practicability.
Drawings
FIG. 1 is a flow chart of a method according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a twin network structure employed in a method according to an embodiment of the present invention;
FIG. 3 illustrates a gesture recognized by a method according to an embodiment of the present invention;
FIG. 4 is a graph showing the comparison of the results of the method of the present invention with 4 conventional methods.
Detailed Description
The method of the embodiment of the invention comprises the following steps:
s1, acquiring a phase sequence generated by a user gesture by using commercial RFID equipment.
S2, denoising and normalizing the phase sequence obtained in the step 1, and identifying the gesture type.
And S3, converting the phase sequence obtained in the step 2 by using short-time Fourier transform to obtain a corresponding spectrogram.
And S4, adopting the spectrogram corresponding to each gesture obtained in the step 3 as a training sample, and pairing the training samples one by one to form a data pair so as to obtain a data set of the twin network. And inputting the obtained data pair into a network for training to obtain the RFID gesture recognition model.
And S5, matching the gesture data to be recognized with the identified gesture data one by one, inputting the matched gesture data into the RFID gesture recognition model obtained in the step S4, and recognizing the gesture data by using a template matching method.
The step S1 of obtaining the phase sequence generated by the user gesture includes collecting the phase data generated by the user gesture through the english jack Speedway R420 reader, the circular polarized Laird S9028PCL antenna, and the Monza AZ-8654 passive RFID tag.
Step S2, denoising and normalizing the phase sequence obtained in step 1, specifically comprising: filtering the original phase data acquired in the step S1 by using a Savitzky-Golay filter, and filtering environmental noise; and normalizing the filtered phase sequence by using a Min-Max normalization method, so that signal fluctuation caused by the gesture is more obvious.
And S3, converting the denoised phase data obtained in the step S2 into a spectrogram by using short-time Fourier transform, and expanding the data from a time domain to a time-frequency domain, so that the feature dimensionality is increased. Meanwhile, the twin network input is generated by utilizing a full-array-based data set generation strategy, specifically, each gesture training sample is paired with other gesture training samples one by one to form a binary set, and the binary set forms a twin network input set.
And S4, pairing each training sample one by adopting the spectrogram obtained in the step S3, specifically pairing each gesture training sample with other gesture training samples one by one to form a binary group, wherein the binary group forms a twin network input set. And inputting the data into a network for training to obtain the RFID gesture recognition model.
In step S5, the RFID recognition model obtained in step S4 is adopted, the gesture sample to be recognized and the identified gesture sample are paired one by one, and input to the RFID recognition model to obtain a corresponding feature vector, the euclidean distance between the gesture sample to be recognized and the identified gesture sample feature vector is calculated, and the sample to be recognized is classified into a category corresponding to the identified gesture sample with the shortest distance, that is, the gesture data is recognized by using a template matching method.
The twin network is a network formed by two weight-sharing subneural networks. The two sub-networks may be identical or different in structure, and a twin network composed of different sub-networks is also called a pseudo-twin network for processing two inputs with larger difference. The invention adopts two twin networks with the same sub-network structure, and specifically comprises two-dimensional convolution layers, three full-connection layers and one maximum pooling layer. The loss function trained by the invention uses a comparison loss function, the loss function can make the intra-class distance of the same type of samples smaller and the inter-class distance between different samples larger, and the comparison function is specifically as follows:
Figure BDA0003627412350000041
wherein W is a twin network parameter and N isThe number of the pairs is input and,
Figure BDA0003627412350000042
is the nth input. Y is n Is composed of
Figure BDA0003627412350000043
And
Figure BDA0003627412350000044
and whether the same type of label is 0 or not, and otherwise, the same type of label is 1. D (W, S) n ) Is the euclidean distance of the features of the two inputs.
FIG. 1 shows a schematic process flow diagram of the process of the present invention. The RFID gesture recognition method based on small sample learning provided by the invention comprises the following steps:
s1, acquiring a phase sequence generated by a user gesture by using commercial RFID equipment, and specifically collecting phase data generated by the user gesture through an Injejun Speedway R420 reader, a circularly polarized Laird S9028PCL antenna and a Monza AZ-8654 passive RFID tag.
S2, denoising and normalizing the phase sequence obtained in the step 1, and identifying the gesture type. The method comprises the following specific steps: filtering the original phase data acquired in the step S1 by using a Savitzky-Golay filter, and filtering environmental noise; and normalizing the filtered phase sequence by using a Min-Max normalization method, so that signal fluctuation caused by the gesture is more obvious. And finally, manually marking a gesture label on the acquired data for identification.
And S3, converting the phase sequence obtained in the step S2 by using short-time Fourier transform to obtain a corresponding spectrogram.
And S4, matching each training sample one by adopting the spectrogram obtained in the step 3 to obtain a data set of the twin network. And inputting the data into a network for training to obtain the RFID gesture recognition model.
S5, matching the gesture data to be recognized with the identified gesture data one by one, inputting the matched gesture data to the RFID gesture recognition model obtained in the step S4, and recognizing the gesture data by using a template matching method.
The invention is described in more detail below with reference to a practical example:
first, a user makes a gesture in an environment where the invention is deployed, and phase data generated by the user gesture is collected through an English Jack Speedway R420 reader, a circularly polarized Laird S9028PCL antenna and a Monza AZ-8654 passive RFID tag. The present invention is intended to recognize 18 common gestures, as shown in FIG. 3:
then, filtering the acquired original phase data by using a Savitzky-Golay filter, and normalizing the phase sequence to a [0,1] interval by using a Min-Max normalization method, so that fluctuation generated by the user gesture is more obvious.
The processed phase values are then converted into spectrograms using a short-time fourier transform. Setting a window function traversal phase sequence at a specific position, and carrying out Fourier transform on the phase sequence in the window function; and after traversing one window function, moving the window function backwards, and continuously performing Fourier transform on the phase sequence in the window function until the complete phase sequence is traversed to obtain a final spectrogram.
And then, pairing each training sample with other samples to obtain an input set of the twin network, and sending the input set into the twin network for training by using the contrast loss.
And finally, pairing the sample to be tested with the training sample, and inputting the obtained data set into the trained twin network to obtain a corresponding feature vector. And calculating Euclidean distances between the feature vectors of the samples to be detected and the feature vectors of the training samples, and classifying the samples to be detected into the class with the shortest average distance.
Aiming at the RFID gesture recognition method based on small sample learning, the invention finally achieves 88% of recognition accuracy rate under the condition of one training sample, and is superior to other gesture recognition methods. See figure 4. Compared with the traditional method for training the model by re-collecting data, the method has the advantages that the accuracy rate is increased by 45 percentage points; compared with the gesture recognition method based on transfer learning, which is commonly used in recent years, the accuracy of the method is improved by 34 percent.
According to the practical cases, the method can effectively reduce the amount of training samples required by the wireless sensing system, and meanwhile, the system can achieve higher identification precision, and is high in reliability, good in practicability and effectiveness.

Claims (8)

1. An RFID gesture recognition method is characterized by comprising the following steps:
s1, acquiring a phase sequence generated by the user gesture;
s2, preprocessing the phase sequence and identifying the gesture type;
s3, acquiring a spectrogram of the preprocessed phase sequence;
s4, taking the spectrogram corresponding to each gesture as a training sample, and pairing each training sample to obtain a twin network data set;
s5, training a twin network by using the twin network data set to obtain an RFID gesture recognition model;
s6, pairing the gesture samples to be recognized and the identified training samples one by one, taking the paired data as the input of the RFID gesture recognition model, and recognizing the gesture samples by using a template matching method.
2. The RFID gesture recognition method according to claim 1, wherein in step S2, the specific implementation process of preprocessing the phase sequence includes: filtering the phase sequence using a Savitzky-Golay filter to filter ambient noise; and normalizing the filtered phase sequence by using a Min-Max normalization method to obtain a preprocessed phase sequence.
3. The RFID gesture recognition method according to claim 1, wherein in step S5, the twin network comprises two weight-shared neural networks.
4. The RFID gesture recognition method of claim 3, wherein the two neural networks are identical in structure; the neural network comprises a first convolution layer, a first maximum pooling layer, a second convolution layer, a second maximum pooling layer and a full-connection layer which are connected in sequence.
5. The RFID gesture recognition method according to claim 1, wherein the loss function l (w) of the twin network is:
Figure FDA0003627412340000011
where W is the twin network parameter, N is the logarithm of the input,
Figure FDA0003627412340000012
is the n-th input to the computer,
Figure FDA0003627412340000013
and
Figure FDA0003627412340000014
is any two gesture samples, Y n Is composed of
Figure FDA0003627412340000015
And
Figure FDA0003627412340000016
whether the same type of label is 0 or not, and if not, the same type of label is 1; d (W, S) n ) Euclidean distance for two input features; m is a boundary constraint for different types of gesture sample distances such that different types of sample distances are constrained to be within m.
6. A computer apparatus comprising a memory, a processor and a computer program stored on the memory; characterized in that the processor executes the computer program to implement the steps of the method according to one of claims 1 to 5.
7. A computer program product comprising a computer program/instructions; characterized in that the computer program/instructions, when executed by a processor, performs the steps of the method according to one of claims 1 to 5.
8. A computer readable storage medium having stored thereon a computer program/instructions; characterized in that the computer program/instructions, when executed by a processor, implement the steps of the method of one of claims 1 to 5.
CN202210480019.5A 2022-05-05 2022-05-05 RFID gesture recognition method, computer device, product and storage medium Pending CN115100733A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210480019.5A CN115100733A (en) 2022-05-05 2022-05-05 RFID gesture recognition method, computer device, product and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210480019.5A CN115100733A (en) 2022-05-05 2022-05-05 RFID gesture recognition method, computer device, product and storage medium

Publications (1)

Publication Number Publication Date
CN115100733A true CN115100733A (en) 2022-09-23

Family

ID=83287807

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210480019.5A Pending CN115100733A (en) 2022-05-05 2022-05-05 RFID gesture recognition method, computer device, product and storage medium

Country Status (1)

Country Link
CN (1) CN115100733A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117292404A (en) * 2023-10-13 2023-12-26 哈尔滨工业大学 High-precision gesture data identification method, electronic equipment and storage medium

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117292404A (en) * 2023-10-13 2023-12-26 哈尔滨工业大学 High-precision gesture data identification method, electronic equipment and storage medium
CN117292404B (en) * 2023-10-13 2024-04-19 哈尔滨工业大学 High-precision gesture data identification method, electronic equipment and storage medium

Similar Documents

Publication Publication Date Title
CN113378632B (en) Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method
WO2018019194A1 (en) Image recognition method, terminal, and nonvolatile storage medium
Ommer et al. Multi-scale object detection by clustering lines
Hammer et al. A seismic‐event spotting system for volcano fast‐response systems
CN111582320B (en) Dynamic individual identification method based on semi-supervised learning
CN107844751B (en) Method for classifying hyperspectral remote sensing images of guide filtering long and short memory neural network
CN111401145B (en) Visible light iris recognition method based on deep learning and DS evidence theory
CN105760891A (en) Chinese character verification code recognition method
CN111698258B (en) WiFi-based environmental intrusion detection method and system
CN112990282B (en) Classification method and device for fine-granularity small sample images
Priyankara et al. Computer assisted plant identification system for Android
CN114239657A (en) Time sequence signal identification method based on complex value interference neural network
CN115294615A (en) Radio frequency fingerprint identification method based on machine learning
CN114354194B (en) Rolling bearing fault diagnosis method based on full convolution self-encoder and optimized support vector machine
CN115100733A (en) RFID gesture recognition method, computer device, product and storage medium
CN111639697A (en) Hyperspectral image classification method based on non-repeated sampling and prototype network
Siméoni et al. Unsupervised object discovery for instance recognition
CN106951924B (en) Seismic coherence body image fault automatic identification method and system based on AdaBoost algorithm
CN111914922B (en) Hyperspectral image classification method based on local convolution and cavity convolution
Hewitt et al. Shape-only features for plant leaf identification
CN112163540A (en) Gesture recognition method based on WiFi
CN114676739B (en) Method for detecting and identifying time sequence action of wireless signal based on fast-RCNN
Chen et al. Spectral-spatial classification based on affinity scoring for hyperspectral imagery
CN112560629B (en) Smart phone key detection method based on 1D-CNN
CN114580476A (en) Unmanned aerial vehicle signal identification model construction method and corresponding identification method and system

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination