CN115700595B - Identity recognition method and device based on radio frequency fingerprint deep learning - Google Patents
Identity recognition method and device based on radio frequency fingerprint deep learning Download PDFInfo
- Publication number
- CN115700595B CN115700595B CN202211389328.8A CN202211389328A CN115700595B CN 115700595 B CN115700595 B CN 115700595B CN 202211389328 A CN202211389328 A CN 202211389328A CN 115700595 B CN115700595 B CN 115700595B
- Authority
- CN
- China
- Prior art keywords
- signal
- identity
- detection area
- radio frequency
- value
- 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
Classifications
-
- 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
- Collating Specific Patterns (AREA)
Abstract
The invention relates to the technical field of identity recognition, and particularly discloses an identity recognition method and device based on radio frequency fingerprint deep learning, wherein the method comprises the following steps: s1: an identity detection area is arranged, and a wireless signal sensor is arranged in the identity detection area; s2: the received signal strength output by the signal receiver is collected, and compares the collected received signal strength with a reference value, judging whether the identity detection area passes by a person or not; if so, reconstructing a shadow fading image caused by a human body through the received signal intensity, and within the time of passing the identity detection area a segment of shadow fading image continuous frame data is formed, and the human body identification result is obtained by using the human body identification result as input to identify in the long-term and short-term memory neural network model. The invention can utilize the penetrability of the wireless signal, does not have a recognition blind area, does not need to pre-install equipment or a sensor on the measured object, and greatly improves the recognition efficiency.
Description
Technical Field
The invention belongs to identity in the technical field of identification, in particular to an identity recognition method and device based on radio frequency fingerprint deep learning.
Background
At present, many ways for realizing identity recognition exist, the traditional identity recognition method is to install equipment or a sensor on a detected object, and then the equipment or the sensor is sensed by sensing equipment to realize identity recognition, and the mode needs to install the equipment or the sensor on the detected object in advance, so that the method is extremely inconvenient; the other is that identity recognition can be realized based on image analysis, for example, people or objects in an image can be recognized through visual analysis based on the image, but the acquisition path of the image is generally through a camera, but the camera is easy to have a visual blind area in the process of acquiring the image, so that the image is easy to be influenced by the visual blind area during detection based on the image analysis, and the problem of low recognition efficiency is caused.
Disclosure of Invention
The invention aims to solve the defects and the shortcomings of the prior art, and provides an identity recognition method and device based on radio frequency fingerprint deep learning, which can utilize the penetrability of wireless signals, have no recognition dead zone, and simultaneously do not need to pre-install equipment or sensors on a detected object, thereby greatly improving the recognition efficiency.
Firstly, the invention provides a method for identifying identity based on radio frequency fingerprint deep learning for solving the problems, which comprises the following steps:
s1: an identity detection area is arranged, a wireless signal sensor is arranged in the identity detection area, the wireless signal sensor comprises a signal transmitter and a signal receiver, and the signal receiver is used for receiving a radio frequency signal sent by the signal transmitter and passing through the identity detection area and outputting the intensity of the received signal;
s2: collecting the intensity of the received signal output by the signal receiver, and comparing the collected intensity of the received signal with a reference value to judge whether personnel pass through the identity detection area; if so, reconstructing a shadow fading image caused by a human body through the received signal intensity, forming a section of continuous frame data of the shadow fading image in the time when a person passes through an identity detection area, and inputting the continuous frame data into a long-period memory neural network model for identity recognition to obtain a human body identity recognition result;
the step S2 further includes preprocessing the acquired received signal strength, including the following steps: obtaining an original signal according to the received signal strength received by the signal receiver; filtering the original signal to obtain a filtered signal; cleaning and filtering the filtering signal to obtain a test signal;
reconstructing the shadow fading image caused by the human body through the received signal strength comprises recording a test signal when a person passes through the identity recognition area, and obtaining a detection value; determining an attenuation value according to the difference value between the detection value and the reference value, wherein the attenuation value is the radio frequency fingerprint of the person; performing image reconstruction based on the radio frequency fingerprint to obtain image characteristics; carrying out convolution processing on the image characteristics to obtain convolution characteristics; flattening the convolution characteristic to obtain an input characteristic;
the reference value is the received signal strength output by the signal receiver when the identity detection area is not occupied.
Compared with the prior art, the method has the advantages that the identity of the tested person is identified by adopting the wireless signal, the problem of an identification blind area is avoided by utilizing the penetrability of the wireless signal, and the shadow fading image of the tested person is reconstructed by the wireless signal, so that the method is not influenced by ambient light, the identity can be accurately identified even in an environment with poor light, meanwhile, the radio frequency fingerprint of the tested person is identified by utilizing the long-short-period memory neural network model, and the identification accuracy and the identification efficiency can be effectively improved; secondly, identity recognition of the detected personnel is realized by utilizing wireless signals, and the power source of the wireless signal sensor is commercial power, so that the problem that the wearable detection method cannot work due to the fact that a device battery is not conductive can be effectively solved.
As a further improvement of the present invention, the comparing the collected received signal strength with a reference value in the step S2 to determine whether the identity detection area has a person or not includes the steps of:
obtaining a reference value according to the test signal;
obtaining a monitoring value according to the test signal;
obtaining a comparison value according to the difference value between the monitoring value and the reference value;
and comparing the comparison value with a preset threshold value, and if the comparison value is larger than the preset threshold value, judging that personnel pass through the current identity detection area.
As a further improvement of the invention, the step S2 of identifying the identity in the long-term and short-term memory neural network model by taking the identity as input, and obtaining the identification result comprises the following steps:
performing feature extraction processing on the input features to obtain output features;
inputting the output characteristics into a first fully-connected network to obtain processed characteristics;
and inputting the processed characteristics to a second full-connection layer to obtain a recognition result.
As a further improvement of the present invention, the feature extraction processing is performed on the input feature, and the obtaining of the output feature includes the following steps:
carrying out forgetting gate information extraction processing on the input characteristics to obtain forgetting gate information;
carrying out input gate information extraction processing on the input features to obtain input gate information;
carrying out extraction processing on the output gate information of the input features to obtain output gate information;
candidate state information extraction processing is carried out on the input features, and candidate state information is obtained;
performing information updating processing based on the candidate state information, the forget gate information and the input gate information obtained in the steps to obtain updated information;
performing feature output processing based on the updated information and the output gate information to obtain output features
As a further development of the invention, the first fully connected network comprises a plurality of first fully connected layers and a plurality of batch normalization layers, which are arranged between two adjacent first fully connected layers.
As a further development of the invention, the activation function in the second fully connected layer is a softmax activation function.
As a further improvement of the invention, said filtering the original signal comprises: and filtering the original signal by adopting a Gaussian convolution smoothing algorithm.
As a further improvement of the invention, the training method of the long-term and short-term memory neural network model comprises the following steps:
acquiring radio frequency fingerprint information and identities of different time sequences of different personnel, and respectively constructing a sample database by taking the radio frequency fingerprint information and the identities as characteristics and labels;
dividing a sample database into a training set and a testing set;
and training the long-term and short-term memory neural network model through the training set and the testing set.
In order to achieve another object of the present invention, the present invention further provides an identification device based on rf fingerprint deep learning corresponding to the above method, which includes:
the identity detection area establishment module is used for setting an identity detection area, a wireless signal sensor is arranged in the identity detection area, the wireless signal sensor comprises a signal transmitter and a signal receiver, and the signal receiver is used for receiving radio frequency signals sent by the signal transmitter and passing through the identity detection area and outputting received signal strength;
an identity recognition module: the identity recognition module is used for collecting the intensity of the received signal output by the signal receiver and comparing the collected intensity of the received signal with a reference value to judge whether personnel pass through the identity detection area; if so, reconstructing a shadow fading image caused by a human body through the received signal intensity, forming a section of continuous frame data of the shadow fading image in the time when a person passes through an identity detection area, and inputting the continuous frame data into a long-period memory neural network model for identity recognition to obtain a human body identity recognition result;
the method also comprises the step of preprocessing the acquired received signal strength, and comprises the following steps: obtaining an original signal according to the received signal strength received by the signal receiver; filtering the original signal to obtain a filtered signal; cleaning and filtering the filtering signal to obtain a test signal;
reconstructing the shadow fading image caused by the human body through the received signal strength comprises recording a test signal when a person passes through the identity recognition area, and obtaining a detection value; determining an attenuation value according to the difference value between the detection value and the reference value, wherein the attenuation value is the radio frequency fingerprint of the person; performing image reconstruction based on the radio frequency fingerprint to obtain image characteristics; carrying out convolution processing on the image characteristics to obtain convolution characteristics; flattening the convolution characteristic to obtain an input characteristic;
the reference value is the intensity of the received signal output by the signal receiver when the identity detection area is not occupied.
Compared with the prior art, the invention realizes the identity recognition of the tested person by adopting the wireless signal, avoids the problem of a recognition blind area by utilizing the penetrability of the wireless signal, reconstructs the shadow fading image of the tested person by the wireless signal, is not influenced by the ambient light, can accurately realize the identity recognition even in the environment with poorer light, and can effectively improve the recognition accuracy and the recognition efficiency by utilizing the long-short-period memory neural network model to recognize the radio frequency fingerprint of the tested person; secondly, identity recognition of the detected personnel is realized by utilizing wireless signals, and the power source of the wireless signal sensor is commercial power, so that the problem that the wearable detection method cannot work due to the fact that a device battery is not conductive can be effectively solved.
For a better understanding and implementation, the present invention is described in detail below with reference to the drawings.
Drawings
The drawings are for a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:
FIG. 1 is a flow chart of an identification method based on radio frequency fingerprint deep learning in an embodiment of the invention;
fig. 2a is a characteristic of a shadow fading image of a person under test 1 at one moment in an embodiment of the present invention;
fig. 2b is a characteristic of a shadow fading image of the person under test 1 at another moment in an embodiment of the present invention;
fig. 3a is a characteristic of a shadow fading image of one moment of the person under test 2 in an embodiment of the present invention;
fig. 3b is a shadow fading image feature of the person under test 2 at another moment in an embodiment of the present invention;
FIG. 4 is a schematic diagram of test signal data according to an embodiment of the present invention;
FIG. 5 is a diagram of a specific long and short term memory network architecture in an embodiment of the invention;
FIG. 6 is a block diagram of an identification device based on RF fingerprint deep learning in an embodiment of the present invention;
in the figure: 100. an identity detection area establishing module; 101. a radio frequency signal sensor; 200. identification recognition and (5) a module.
The achievement of the objects, functional features and advantages of the present invention will be further described with reference to the accompanying drawings, in conjunction with the embodiments.
Detailed Description
Examples
The traditional identity recognition method needs to install equipment or a sensor on a detected object, and is very inconvenient to operate; the identity recognition scheme based on image analysis is characterized in that the problem of recognition dead zone is avoided by utilizing the penetrability of a wireless signal, the image of the detected object is reconstructed by utilizing the wireless signal, the influence of ambient light is avoided, the identity recognition can be accurately realized even in an environment with poor light, and meanwhile, the radio frequency fingerprint of the detected object is recognized by utilizing a long-term memory neural network model, so that the recognition accuracy and the recognition efficiency can be effectively improved; secondly, identity recognition of the detected personnel is realized by utilizing wireless signals, and the power source of the wireless signal sensor is commercial power, so that the problem that the wearable detection method cannot work due to the fact that a device battery is not conductive can be effectively solved. The identification method based on the radio frequency fingerprint deep learning is specifically described below.
Referring to fig. 1, the identification method based on the radio frequency fingerprint deep learning of the present invention includes the following steps:
step S1: an identity detection area is arranged, a wireless signal sensor is arranged in the identity detection area, the wireless signal sensor comprises a signal transmitter and a signal receiver, and the signal receiver is used for receiving a radio frequency signal sent by the signal transmitter and passing through the identity detection area and outputting received signal strength (Received Signal Strength, RSS);
s2: collecting the intensity of the received signal output by the signal receiver, and comparing the collected intensity of the received signal with a reference value to judge whether personnel pass through the identity detection area; if so, reconstructing a shadow fading image caused by a human body through the received signal intensity, forming a section of continuous frame data of the shadow fading image in the time when a person passes through an identity detection area, and inputting the continuous frame data into a long-period memory neural network model for identity recognition to obtain a human body identity recognition result;
the reference value is the received signal strength output by the signal receiver when the identity detection area is not occupied.
In this embodiment, the rf signal sensors 101 are exemplarily used as signal sensors, the rf signal sensors 101 are mounted on a door frame, the plurality of rf signal sensors 101 form a sensor array, and an area surrounded by the sensor array is an identification area, where each rf signal sensor 101 can transmit signals and receive signals, each rf signal sensor 101 can at least receive signals sent by another rf signal sensor 101, the rf signal sensor 101 is a WiFi or ZigBee device, and the signal strength sent by the rf signal sensor needs to reach a level that enables the signals to pass through a person to be identified.
The step S2 further includes preprocessing the acquired radio frequency signal, including the following steps:
step S201: obtaining an original signal according to the received signal strength received by the signal receiver;
step S202: filtering the original signal to obtain a filtered signal;
step S203: and cleaning and filtering the filtered signals to obtain test signals.
In this embodiment, the received signal strength of the radio frequency signal sensor 101 is obtained as an original signal, the original signal is filtered by a filter, a smooth linear filter is exemplarily used as the filter of the scheme, the original signal is filtered by a gaussian convolution smoothing algorithm in this embodiment to obtain a filtered signal, and the filtered signal is filtered and cleaned by an upper computer to obtain a test signal; referring to fig. 4, a test signal acquired by a person in the process of passing through the identity detection area is shown in fig. 4, the acquired test signal data is an array of t×n×m, and each cell in the array represents a shadow fading density value of an nth row and an mth column of pixel block of the person to be tested in the identity detection area at time t in the process of passing through the identity detection area.
The step S2 of comparing the collected received signal strength with a reference value to determine whether the person passes through the identity detection area includes the following steps:
step S204: obtaining a reference value according to the test signal;
step S205: obtaining a monitoring value according to the test signal;
step S206: obtaining a comparison value according to the difference value between the monitoring value and the reference value;
step S207: and comparing the comparison value with a preset threshold value, and if the comparison value is larger than the preset threshold value, judging that personnel pass through the current identity detection area.
In this embodiment, the test signal data value y is recorded when no person under test passes through the identity detection area 0 And using the value as a reference value to continuously monitor the intensity of the test signal obtained from the current identity monitoring area to obtain a monitoring value y q And the obtained monitoring value y q And a reference value y 0 The specific calculation method is to take the monitoring value y q And a reference value y 0 Difference y between w And the difference y is w As the comparison value, the comparison value y w The calculation formula of (2) is as follows:
y w ＝y q -y 0 (1)
the obtained comparison value y w Comparing with a preset threshold value, when comparing the value y w If the identity detection area is larger than the preset threshold value, judging that the personnel pass through the current identity detection area, otherwise, judging that the personnel pass through the current identity detection area; the rf signal sensor 101 in the sensor array works by adopting a frequency division multiplexing method, that is, the rf signal sensor 101 adopts different frequencies to transmit signals, adopts a low-frequency transmit signal when the monitored value is smaller than a preset threshold value, and adopts a high-frequency transmit signal when the monitored value is larger than the preset threshold value.
The reconstructing the shadow fading image caused by the human body through the received signal intensity in the step S2 includes the following steps:
step S208: recording a test signal when a person passes through the identity recognition area to obtain a detection value;
step S209: determining an attenuation value according to the difference value between the detection value and the reference value, wherein the attenuation value is the radio frequency fingerprint of the person;
step S210: performing image reconstruction based on the radio frequency fingerprint to obtain image characteristics;
step S211: carrying out convolution processing on the image characteristics to obtain convolution characteristics;
step S212: and flattening the convolution characteristic to obtain an input characteristic.
If the person passes through the identity detection area, recording a wireless signal data value y of the time period when the person passes through the identity recognition area t And takes this value as the detection value y t Based on the reference value y 0 Detection value y t To obtain a wireless signal strength attenuation value y of the person in the time period passing through the identification area, the attenuationThe subtracted value y is the radio frequency fingerprint of the person, and the specific calculation formula of y is as follows:
y＝y t -y 0 (2)
referring to fig. 2a, fig. 2b, fig. 3a and fig. 3b, in this embodiment, when a person passes through the identity detection area, the radio frequency fingerprint of the person can be recorded, the shadow fading image is reconstructed by a back projection imaging mode based on the radio frequency fingerprint to obtain image features, wherein fig. 2a and fig. 2b are image features of the person 1 to be detected at different moments in the process of passing through the identity detection area, fig. 3a and fig. 3b are image features of the person 2 to be detected at different moments in the process of passing through the identity detection area, it can be found by observation that the image features formed by the person 1 to be detected and the person 2 to be detected are different, the height of the person 1 to be detected is higher than that of the person 2 to be detected, and the width of the body is also different, so that we can identify the identity of the person based on the different image features of each person; and then carrying out convolution processing on the image features to obtain convolution features, wherein the convolution features are two-dimensional features, so that the convolution features are required to be leveled to be converted into one-dimensional features, and further one-dimensional input features are obtained.
In the step S2, the identity recognition is performed on the long-term and short-term memory neural network model by using the long-term and short-term memory neural network model as input, and the recognition result is obtained, including the following steps:
step S213: performing feature extraction processing on the input features to obtain output features;
step S214: the output characteristics are input to a first fully connected network, obtaining processed features;
step S215: and inputting the processed characteristics to a second full-connection layer to obtain a recognition result.
In this embodiment, the neural network data generally exist independently, and the input of the former data and the input of the latter data are not always related, but because the person passes through the identification area and is a process, the detection values of the person at different moments in the process of passing through the identification area can be obtained during the process, that is, the detection values originate from the same person to be detected, so that the one-dimensional input features at different moments are not independent; processing the one-dimensional input features through a long-short-term memory network to obtain output features, inputting the output features into a first full-connection network, performing nonlinear transformation and addition processing on the output features by a first weight connection network to obtain processed features, inputting the processed features into a second full-connection layer, identifying and classifying the processed features by the second full-connection layer, and outputting an identification result, namely determining personnel identity information; the activation function in the second full-connection layer adopts a softmax activation function, and the aim of classification and identification is achieved through the softmax activation function.
Wherein, in the step S213, the feature extraction process is performed on the input feature, and the obtaining the output feature includes the following steps:
s2131: carrying out forgetting gate information extraction processing on the input characteristics to obtain forgetting gate information;
s2132: carrying out input gate information extraction processing on the input features to obtain input gate information;
s2133: carrying out extraction processing on the output gate information of the input features to obtain output gate information;
s2134 the method comprises the following steps: candidate state information extraction processing is carried out on the input features, and candidate state information is obtained;
s2135: performing information updating processing based on the candidate state information, the forget gate information and the input gate information obtained in the steps to obtain updated information;
s2136: and (3) performing feature output processing based on the updated information and the output gate information acquired in the step (S2133) to acquire output features.
In this embodiment, nonlinear spatial transformation is performed on the input features to obtain forgetting door information f t The specific calculation process is as follows:
f t ＝σ(W f h t-1 +W f X t +b f ) (3)
nonlinear transformation processing is carried out on the input characteristics to obtainObtain input gate information i t The specific calculation process is as follows:
i t ＝σ(W i h t-1 +W i X t +b i ) (4)
the input characteristics are subjected to output gate information extraction processing to obtain output gate information o t The specific calculation process is as follows:
o t ＝σ(W o h t-1 +W o X t +b o ) (5)
the input characteristics are subjected to nonlinear transformation processing to obtain candidate state information C', and the specific calculation process is as follows:
C′＝tanh(W c h t-1 +W c X t +b i ) (6)
information update processing is performed based on the candidate state information, the forget gate information and the input gate information to obtain updated information C t The specific calculation process is as follows:
C t ＝f t *C t-1 +i t *C′ (7)
based on the output gate information and the update information, performing feature output processing to obtain hidden variable h t The hidden variable is an output characteristic, and the specific calculation process is as follows:
h t ＝o t *ReLU(C t ) (8)
wherein X is t Representing the input characteristics at time t, b f ，b i ，b o ，b c Bias vectors respectively representing forget gate, input gate, output gate and time status, W f ，W i ，W o ，W c Respectively representing the weight of forgetting gate, input gate, output gate and time status, C t Representing the state at time t, C t-1 The state at time t-1 is represented, sigma is a sigmoid activation function, and the calculation formula of the sigmoid activation function is as follows:feature selection of input features at past t-time for better representationThe conventional sigmoid activation function is adopted, and the value of the sigmoid activation function is mapped between 0 and 1.
Wherein the first fully connected network in step S214 includes a plurality of first fully connected layers and a plurality of batch normalization layers (Batch Normalization), the batch normalization layer is arranged between two adjacent first full connection layers.
The training method of the long-term and short-term memory neural network model comprises the following steps:
acquiring radio frequency fingerprint information and identities of different time sequences of different personnel, and respectively constructing a sample database by taking the radio frequency fingerprint information and the identities as characteristics and labels;
sample database at 4:1 into training sets and test sets;
and training the long-term and short-term memory neural network model through the training set and the testing set.
In this embodiment, the activation function of the first full connection layer adopts a ReLU activation function for the output feature h t The feature information of the past moment is saved, and the 0-1 mapping is not needed, so that the scheme selects a ReLU activation function, and the calculation formula of the ReLU activation function is as follows:
f(x)＝max(0，x) (9)
compared with the traditional tanh activation function, the ReLU activation function avoids the power function calculation, so that the speed of a model can be increased, in addition, the tanh activation function has the problem of gradient disappearance when the number of layers of the model is increased, and the ReLU activation function can effectively alleviate the problem.
The fully connected layers are also connected by a ReLU activation function, the calculation rate of the model can be effectively improved, and the condition of gradient disappearance can be prevented.
However, as the model deepens, the ReLU activation function also prevents the problem of gradient vanishing, so the invention introduces a new structure, namely a residual block. The residual block can realize jump connection by directly taking the output of the upper layer as the input without weighting and summing, and the specific realization formula is as follows:
y＝F(x)+x (10)
as shown in the formula (10), even if the gradient of F (x) is 0 and the derivative result of x is 1, there is no case where the gradient disappears, so the introduction of the residual block can effectively solve the problem of model gradient disappearance, and even if the gradient of F (x) is 0, the performance of the model is not weakened due to the increase of the number of layers.
The optimizer is critical to training of model parameters, and although the Adam optimizer converges quickly, it is prone to causing non-convergence or missing locally optimal solutions. Therefore, the invention uses the Adam optimizer to update parameters in the first half of the rounds and uses the SGD optimizer to adjust the parameters in the second half of the rounds.
In order to map the output of each layer of network in a certain range, the invention adds a batch normalization layer (Batch Normalization) after the ReLU activation function, so that the output of each layer of network is mapped in a certain spatial range, and the training speed of the network is accelerated.
Referring to fig. 5, in this embodiment, a total of 80 pieces of data are generated for identification and classification, a radio frequency fingerprint is used as input, a deep learning network is built, a specific network architecture is shown in fig. 5, and finally, the classification accuracy in a test set is as high as 93.75%,16 pieces of test data only have one classification error, only the test sample of one to-be-tested person 1 is wrongly classified into a to-be-tested person 2, and the identification results of the rest test samples are correct.
Referring to fig. 6, correspondingly, the invention also provides an identification device based on rf fingerprint deep learning corresponding to the identification method based on rf fingerprint deep learning, which comprises an identification detection area establishment module 100 and an identification module 200,
the identity detection area establishing module 100 is configured to set an identity detection area, wherein a wireless signal sensor is arranged in the identity detection area, the wireless signal sensor comprises a signal transmitter and a signal receiver, and the signal receiver is configured to receive and output a radio frequency signal sent by the signal transmitter and penetrating through the identity detection area;
the identity recognition module 200 is configured to collect the radio frequency signal received by the signal receiver, and compare the collected radio frequency signal with a reference value to determine whether a person passes through the identity detection area; if so, reconstructing a shadow fading image of the human body through a radio frequency signal, forming a section of continuous frame data of the continuous shadow fading image in the time when a person passes through an identity detection area, inputting the continuous frame data into a long-period and short-period memory neural network model for identity recognition, and obtaining a recognition result;
the reference value is a radio frequency signal value received by the signal receiver when the identity detection area is not occupied.
In this embodiment, a radio frequency signal sensor 101 is used as a wireless signal sensor, the radio frequency signal sensor 101 is mounted on a door frame, a plurality of radio frequency signal sensors 101 form a sensor array, an area surrounded by the sensor array is an identification area, each radio frequency signal sensor 101 can transmit signals and receive signals, each radio frequency signal sensor 101 can at least receive signals transmitted by another radio frequency signal sensor 101, the radio frequency signal sensor 101 is a ZigBee or WiFi signal transmitter, the signal strength transmitted by the radio frequency signal sensor is required to reach a level that enables the signals to pass through a person to be identified, the highest mounting height of the radio frequency signal sensor 101 is higher than the height of the person to be identified, and the lowest mounting height is lower than the knee height of the person to be identified; in this embodiment, a smoothing linear filter is used to implement filtering processing on the test signal, and a gaussian convolution smoothing algorithm is used to implement filtering processing on the test signal.
The present invention is not limited to the above-described embodiments, but, if various modifications or variations of the present invention are not departing from the spirit and scope of the present invention, the present invention is intended to include such modifications and variations as fall within the scope of the claims and the equivalents thereof.
Claims (8)
1. The identity recognition method based on the radio frequency fingerprint deep learning is characterized by comprising the following steps of:
s1: an identity detection area is arranged, a wireless signal sensor is arranged in the identity detection area, the wireless signal sensor comprises a signal transmitter and a signal receiver, and the signal receiver is used for receiving a radio frequency signal sent by the signal transmitter and passing through the identity detection area and outputting the intensity of the received signal;
s2: collecting the intensity of the received signal output by the signal receiver, comparing the collected intensity of the received signal with a reference value, judging whether the identity detection area passes by a person or not; if so, reconstructing a shadow fading image caused by a human body through the received signal intensity, forming a section of continuous frame data of the shadow fading image in the time when a person passes through an identity detection area, and inputting the continuous frame data into a long-period memory neural network model for identity recognition to obtain a human body identity recognition result;
the step S2 further includes preprocessing the acquired received signal strength, including the following steps: obtaining an original signal according to the received signal strength received by the signal receiver; filtering the original signal to obtain a filtered signal; cleaning and filtering the filtering signal to obtain a test signal;
reconstructing the shadow fading image caused by the human body through the received signal strength comprises recording a test signal when a person passes through the identity recognition area, and obtaining a detection value; determining an attenuation value according to the difference value between the detection value and the reference value, wherein the attenuation value is the radio frequency fingerprint of the person; performing image reconstruction based on the radio frequency fingerprint to obtain image characteristics; carrying out convolution processing on the image characteristics to obtain convolution characteristics; flattening the convolution characteristic to obtain an input characteristic;
the reference value is the received signal strength output by the signal receiver when the identity detection area is not occupied.
2. The method for identifying an identity based on deep learning of radio frequency fingerprints according to claim 1, wherein comparing the collected received signal strength with a reference value in the step S2 to determine whether a person passes through the identity detection area comprises the following steps:
obtaining a reference value according to the test signal;
monitoring the test signal intensity of the current identity area to obtain a monitoring value;
obtaining a comparison value according to the difference value between the monitoring value and the reference value;
and comparing the comparison value with a preset threshold value, and if the comparison value is larger than the preset threshold value, judging that personnel pass through the current identity detection area.
3. The identification method based on the deep learning of the radio frequency fingerprint according to claim 1, wherein the step S2 of identifying the identity in the long-term and short-term memory neural network model by using the identification result as an input comprises the following steps:
performing feature extraction processing on the input features to obtain output features;
inputting the output characteristics into a first fully-connected network to obtain processed characteristics;
and inputting the processed characteristics to a second full-connection layer to obtain a recognition result.
4. The identification method based on the deep learning of the radio frequency fingerprints according to claim 3, wherein, the step of extracting the input features to obtain output features comprises the following steps:
carrying out forgetting gate information extraction processing on the input characteristics to obtain forgetting gate information;
carrying out input gate information extraction processing on the input features to obtain input gate information;
carrying out extraction processing on the output gate information of the input features to obtain output gate information;
candidate state information extraction processing is carried out on the input features, and candidate state information is obtained;
performing information updating processing based on the candidate state information, the forget gate information and the input gate information obtained in the steps to obtain updated information;
and carrying out feature output processing based on the updated information and the output gate information to obtain output features.
5. The identification method based on the radio frequency fingerprint deep learning as claimed in claim 3, wherein the identification method is characterized in that: the first full-connection network comprises a plurality of first full-connection layers and a plurality of batch normalization layers, wherein the batch normalization layers are arranged between two adjacent first full-connection layers.
6. The identification method based on the radio frequency fingerprint deep learning as claimed in claim 3, wherein the identification method is characterized in that: the activation function in the second fully-connected layer is a softmax activation function.
7. The method for identifying an identity based on deep learning of radio frequency fingerprints according to claim 1, wherein the filtering the original signal includes: and filtering the original signal by adopting a Gaussian convolution smoothing algorithm.
8. An identification device based on radio frequency fingerprint deep learning, which is characterized by comprising:
the identity detection area establishment module is used for setting an identity detection area, a wireless signal sensor is arranged in the identity detection area, the wireless signal sensor comprises a signal transmitter and a signal receiver, and the signal receiver is used for receiving radio frequency signals sent by the signal transmitter and passing through the identity detection area and outputting received signal strength;
an identity recognition module: the identity recognition module is used for collecting the intensity of the received signal output by the signal receiver and comparing the collected intensity of the received signal with a reference value to judge whether personnel pass through the identity detection area; if so, reconstructing a shadow fading image caused by a human body through the received signal intensity, forming a section of continuous frame data of the shadow fading image in the time when a person passes through an identity detection area, and inputting the continuous frame data into a long-period memory neural network model for identity recognition to obtain a human body identity recognition result;
the method also comprises the step of preprocessing the acquired received signal strength, and comprises the following steps: obtaining an original signal according to the received signal strength received by the signal receiver; filtering the original signal to obtain a filtered signal; cleaning and filtering the filtering signal to obtain a test signal;
reconstructing the shadow fading image caused by the human body through the received signal strength comprises recording a test signal when a person passes through the identity recognition area, and obtaining a detection value; determining an attenuation value according to the difference value between the detection value and the reference value, wherein the attenuation value is the radio frequency fingerprint of the person; performing image reconstruction based on the radio frequency fingerprint to obtain image characteristics; carrying out convolution processing on the image characteristics to obtain convolution characteristics; flattening the convolution characteristic to obtain an input characteristic; the reference value is the intensity of the received signal output by the signal receiver when the identity detection area is not occupied.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211389328.8A CN115700595B (en) | 2022-11-08 | 2022-11-08 | Identity recognition method and device based on radio frequency fingerprint deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211389328.8A CN115700595B (en) | 2022-11-08 | 2022-11-08 | Identity recognition method and device based on radio frequency fingerprint deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115700595A CN115700595A (en) | 2023-02-07 |
| CN115700595B true CN115700595B (en) | 2023-07-28 |
Family
ID=85121081
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211389328.8A Active CN115700595B (en) | 2022-11-08 | 2022-11-08 | Identity recognition method and device based on radio frequency fingerprint deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115700595B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116541693B (en) * | 2023-07-07 | 2023-09-12 | 河北省水利工程局集团有限公司 | Open construction environment personnel detection method and working personnel on-duty management system |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288018A (en) * | 2019-06-24 | 2019-09-27 | 桂林电子科技大学 | A kind of WiFi personal identification method merging deep learning model |
| CN112398553A (en) * | 2020-11-03 | 2021-02-23 | 上海电机学院 | Communication radiation source individual identification method based on differential equipotential sphere diagram |
Family Cites Families (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10497242B2 (en) * | 2016-12-15 | 2019-12-03 | David H. Williams | Systems and methods for monitoring for and preempting pre-identified restriction violation-related behavior(s) of persons under restriction |
| US10373165B2 (en) * | 2017-09-25 | 2019-08-06 | Capital One Services, Llc | Automated sensor-based customer identification and authorization systems within a physical environment |
| JP2021135619A (en) * | 2020-02-25 | 2021-09-13 | 日本放送協会 | Situation identification device, situation learning device, and program |
| CN111884974B (en) * | 2020-07-16 | 2022-08-30 | 东南大学 | Broadband signal radio frequency fingerprint feature extraction method based on frequency spectrum decomposition |
| CN115294615A (en) * | 2022-08-24 | 2022-11-04 | 南京理工大学 | Radio frequency fingerprint identification method based on machine learning |
-
2022
- 2022-11-08 CN CN202211389328.8A patent/CN115700595B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110288018A (en) * | 2019-06-24 | 2019-09-27 | 桂林电子科技大学 | A kind of WiFi personal identification method merging deep learning model |
| CN112398553A (en) * | 2020-11-03 | 2021-02-23 | 上海电机学院 | Communication radiation source individual identification method based on differential equipotential sphere diagram |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115700595A (en) | 2023-02-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112587153B (en) | End-to-end non-contact atrial fibrillation automatic detection system and method based on vPPG signal | |
| CN110403600B (en) | Intelligent analysis method and system for paroxysmal atrial fibrillation based on difference time scatter diagram | |
| CN112418649B (en) | Building stair pedestrian flow estimation system based on multi-dimensional MEMS inertial sensor | |
| CN111556453A (en) | Multi-scene indoor action recognition method based on channel state information and BilSTM | |
| CN115700595B (en) | Identity recognition method and device based on radio frequency fingerprint deep learning | |
| CN111698258B (en) | WiFi-based environmental intrusion detection method and system | |
| Li et al. | An improved sparse representation over learned dictionary method for seizure detection | |
| CN116552306B (en) | Monitoring system and method for direct current pile | |
| Zhang et al. | Roughness-length-based characteristic analysis of intracranial EEG and epileptic seizure prediction | |
| CN111901028B (en) | Human body behavior identification method based on CSI (channel State information) on multiple antennas | |
| CN117598700B (en) | Intelligent blood oxygen saturation detection system and method | |
| CN115886840A (en) | Epilepsy prediction method based on domain-pair-resistant multi-level deep convolution feature fusion network | |
| CN114595725A (en) | Electroencephalogram signal classification method based on addition network and supervised contrast learning | |
| CN113069117A (en) | Electroencephalogram emotion recognition method and system based on time convolution neural network | |
| CN116977807A (en) | Multi-sensor fusion-based intelligent monitoring system and method for refrigerator | |
| CN116524380A (en) | Target detection method based on brain-computer signal fusion | |
| CN110200624A (en) | Based on convolutional neural networks-Recognition with Recurrent Neural Network-support vector machines mixed model disease identification algorithm | |
| CN115581435A (en) | Sleep monitoring method and device based on multiple sensors | |
| CN112883905B (en) | Human behavior recognition method based on coarse-grained time-frequency features and multi-layer fusion learning | |
| CN115067878A (en) | EEGNet-based resting state electroencephalogram consciousness disorder classification method and system | |
| CN110547772B (en) | Individual age prediction method based on brain signal complexity | |
| CN111612746B (en) | Dynamic detection method for functional brain network central node based on graph theory | |
| CN110647870B (en) | Method for calculating approximate entropy of resting state fMRI data based on sliding window | |
| CN110363096B (en) | Brain time signal processing method based on hidden Markov model | |
| CN115173971B (en) | Broadband signal real-time detection method based on frequency spectrum data |
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 |