A method and device for identifying identity by time-frequency domain cross-correlation

Abstract

The present invention relates to a time-frequency domain cross-correlation identity recognition method and device. The present invention is implemented in two stages; the first stage is to input the infrared information model of the target object, and perform cross-correlation calculations on the target object information collected multiple times and the input information model respectively; the second stage is to collect the infrared signal of the approaching object and perform time-frequency domain correlation calculations with the input information template, input the calculation result into the trained support vector machine for identity recognition, and output the recognition result. Compared with the prior art, the pyroelectric infrared sensor is used to actively capture the infrared information radiated by the limbs of the approaching object. The pyroelectric signal is used as the recognition source to prevent the possibility of being imitated, and the price is low, the power consumption is low, and the sensitivity is high. While recognizing the infrared information of the limbs and the macroscopic overall appearance, it can also be recognized in microscopic details, which is convenient and fast, with high accuracy, and achieves identity recognition with a high safety factor.

Description

Technical Field

[0001] The present invention relates to an identity recognition method and device, and in particular to an identity recognition method and device based on time-frequency domain cross-correlation. Background Art

[0002] With the development of social economy and science and technology, information security has been paid more and more attention, and identity recognition technology has become a hot topic for research in major universities and research institutions at home and abroad. At present, the commonly used methods for identity recognition are: using video imaging equipment or fingerprint sensors for recognition. However, the recognition technology based on video imaging equipment has the defects of high cost, easy imitation, and great environmental interference; although the recognition technology based on fingerprint sensors is low-cost, fingerprint-type biometrics are not only easy to imitate, but also because fingerprint sensors use contact information collection methods, which increases the risk of spreading germs. Compared with the above two recognition methods, pyroelectric infrared sensors not only have the characteristics of non-contact, but also have the advantages of low cost and information sources that are not easy to imitate. They have been widely used in the field of body recognition, but still need to be developed in the field of identity recognition.

[0003] CN210244387 discloses a "face recognition and comparison device and access control system", which uses a pyroelectric sensor to sense the approach of a human body, uses an image sensor to collect image information, and then performs face recognition on the collected image. This method relies on face images for identity recognition, the method is reliable, and does not require direct contact with the detection device, but this device is easily affected by light and can easily lead to misjudgment, and face information can be easily imitated, which can easily cause economic losses to users.

[0004] CN110516635 A discloses a "face recognition and comparison device, access control system and method", which detects infrared radiation of the human body through an infrared pyroelectric sensor, and detects the brightness of the environment through a photosensitive sensor. If the brightness of the environment is lower than the set brightness when detecting the infrared radiation of the human body, a fill light is used to fill in the light so that the image sensor can collect face image information that meets the quality requirements; the face image information is then used to perform face recognition, and if the face recognition is successful, the door lock is unlocked. This method relies on face image recognition, and does not require direct contact with the detection device. The device can automatically detect the brightness of the environment to adjust the light, which improves the accuracy of recognition. However, this device is expensive and fails to reduce the risk of face information being imitated.

[0005] CN 211149551 U discloses an "access control system for intelligent automobile passages", wherein a touch screen is provided at the front end of an outer shell, a face recognition camera and a fingerprint recognition chip are installed above the touch screen, a human infrared pyroelectric sensor is provided at the lower end of the outer shell, and a control circuit board is provided in the inner cavity of the outer shell. The structure principle is simple, and rapid fingerprint recognition can be achieved, thereby improving the access control efficiency in automobile passages. However, the touch structure is unhygienic and there is a risk of spreading pathogens, and fingerprint-like biometrics as an identification source are easily imitated. Summary of the invention

[0006] The purpose of the present invention is to provide a time-frequency domain cross-correlation identity recognition method to address the deficiencies of the above-mentioned prior art;

[0007] Another object of the present invention is to provide a device for an identity recognition method based on time-frequency domain cross-correlation.

[0008] According to the seventh edition of the Modern Chinese Dictionary, the explanation of the Commercial Press in September 2016: "identification" means "to distinguish according to characteristics and make judgments in order to find or identify a certain object." In the identity recognition method described in this application, the task of "identification" is achieved by the correlation recognition and denoising algorithm, that is, distinguishing according to characteristics; the task of "recognition" is undertaken by the classifier designed by the support vector machine, that is, making judgments to find or identify a certain object.

[0009] Idea of ​​the invention: The method of the invention is implemented through two stages; the first stage is to input the infrared information model of the target object, perform cross-correlation calculations on the target object information collected multiple times and the information model that has been entered, and use the time-frequency domain cross-correlation coefficient matrix obtained as the feature vector to train the support vector machine (SVM); the second stage is to collect infrared signals close to the object and perform time-frequency domain correlation calculations on the information template that has been entered, input the calculation results into the trained support vector machine (SVM) for identity recognition, and output the recognition results.

[0010] The objective of the present invention is achieved through the following technical solutions:

[0011] A time-frequency domain cross-correlation identity recognition method, comprising: A. inputting a pyroelectric information model of a target object and training a vector machine, B. collecting pyroelectric information of a nearby object and distinguishing a classifier;

[0012] The target object pyroelectric information model entry and support vector machine training include:

[0013] a. Collect multi-channel signals from the target object’s fingers;

[0014] b. Processing the collected signals to obtain an information model;

[0015] c. Collect infrared information of target objects multiple times and calculate the time domain correlation coefficient of the information model;

[0016] d. Select 1-20 effective frequency points by performing principal component analysis on the power spectrum;

[0017] e. Input the time-frequency domain mutual correlation coefficient as a feature vector into the classifier for training;

[0018] The pyroelectric information collection and classifier identification of the approaching object include:

[0019] Ⅰ. Acquisition of pyroelectric information of approaching objects;

[0020] Ⅱ. Process the collected signals;

[0021] III. Calculation of the time-frequency domain mutual correlation coefficient between the close object information and the template information;

[0022] IV. Classifier designed based on support vector machine is used for discrimination.

[0023] A device for time-frequency domain cross-correlation identity recognition method is composed of an X-th group of pyroelectric sensor arrays connected to a control module via an X-th group of multiplexers and an X-th group of conditioning circuits, a first group of pyroelectric sensor arrays connected to a control module via a first group of multiplexers and a first group of conditioning circuits, a control module connected to an unlocking module via a drive circuit and a relay, a controller module connected to a memory, and a power supply module connected to an information acquisition circuit, a controller, a drive circuit, a relay and an unlocking module respectively.

[0024] The pyroelectric sensor array is a partially hollow cylinder with one end open. The partially hollow means that 2 / 3 of the cylinder is a hollow section. The hollow section is divided into four longitudinal cylindrical surfaces at equal angles at the center of the circle. Two or more circular through holes distributed in an array are arranged at equal intervals along the cylindrical surface. The through hole diameter Ф>8mm, the through hole interval is the same as 12mm, the cylinder is 150mm long, two-thirds is hollow, the diameter is 46mm, and the wall thickness is>1mm. The through holes are spaced at the same interval to facilitate the installation, adjustment and positioning of the pyroelectric probe.

[0025] Beneficial effects: Compared with the prior art, the present invention uses pyroelectric infrared sensors to actively capture infrared information radiated by the limbs of the approaching object. The pyroelectric signal is used as an identification source to reduce the possibility of being imitated, and it is low-priced, consumes less power, and has high sensitivity. It has unique advantages in human body recognition. At the same time, the identity recognition mode that combines the cross-correlation recognition noise reduction algorithm with the support vector machine can recognize microscopic details while recognizing the infrared information of the limbs and the macroscopic overall appearance. It is convenient and fast, with high accuracy, and realizes identity recognition with a high safety factor. The identity recognition device using a pyroelectric sensor has a simple structure and is easy to install. Its non-contact recognition reduces the risk of pathogen infection. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 A flow chart of a method for identification by time-frequency domain cross-correlation;

[0027] Figure 2 A circuit structure block diagram of a time-frequency domain cross-correlation identity recognition method and device;

[0028] Figure 3 A circuit diagram of a method and device for identifying an identity by time-frequency domain cross-correlation;

[0029] Figure 4 For attachment Figure 2 The structure diagram of the pyroelectric sensor array in FIG.

[0030] (a) 3D structure of pyroelectric sensor array

[0031] (a-1) Cross-sectional view of the three-dimensional structure of the pyroelectric sensor array

[0032] (b) Cross-sectional view of the pyroelectric sensor array structure

[0033] Figure 5 Power spectra of different human bodies;

[0034] Figure 6 Support vector machine accuracy plot. DETAILED DESCRIPTION

[0035] The present invention is further described in detail below with reference to the accompanying drawings and embodiments.

[0036] The method of the present invention includes two stages: the first stage is the input of the target object's pyroelectric information template and the training of the support vector machine; the second stage is the collection of the pyroelectric information of the approaching object and the final judgment of the classifier. Among them: the main work of the first stage is to input the infrared information template of the target object, and collect the information of the target object for multiple times to perform cross-correlation calculations with the information template that has been input, and the time-frequency domain cross-correlation coefficient matrix is ​​obtained as the feature vector to train the support vector machine (SVM); the main work of the second stage is to collect the infrared signal of the approaching object and perform time-frequency domain correlation calculations with the information template that has been input, and the calculation results are input into the trained support vector machine (SVM) for identity recognition, and the final recognition result is output.

[0037] A time-frequency domain cross-correlation identity recognition method, comprising: A. inputting a pyroelectric information model of a target object and training a vector machine, B. collecting pyroelectric information of a nearby object and distinguishing a classifier;

[0038] The target object pyroelectric information model entry and support vector machine training include:

[0039] a. Collect multi-channel signals from the target object’s fingers;

[0040] b. Processing the collected signals to obtain an information model;

[0041] c. Collect infrared information of target objects multiple times and calculate the time domain correlation coefficient of the information model;

[0042] d. Select 1-20 effective frequency points by performing principal component analysis on the power spectrum;

[0043] e. Input the time-frequency domain mutual correlation coefficient as a feature vector into the classifier for training;

[0044] The pyroelectric information collection and classifier identification of the approaching object include:

[0045] Ⅰ. Acquisition of pyroelectric information of approaching objects;

[0046] Ⅱ. Process the collected signals;

[0047] III. Calculation of the time-frequency domain mutual correlation coefficient between the close object information and the template information;

[0048] IV. Classifier designed based on support vector machine is used for discrimination.

[0049] A device for time-frequency domain cross-correlation identity recognition method, such as Figure 2 As shown, the Xth group of pyroelectric sensor arrays are connected to the control via the Xth group of multiplexers and the Xth group of conditioning circuits, the first group of pyroelectric sensor arrays are connected to the control module via the first group of multiplexers and the first group of conditioning circuits, the control module is connected to the unlocking module via the drive circuit and the relay, the controller module is connected to the memory, and the power supply module is respectively connected to the information acquisition circuit, the controller, the drive circuit, the relay and the unlocking module.

[0050] The information acquisition circuit is responsible for collecting infrared information of the human body. The collected information is processed by the controller and stored in the memory. The controller directly issues commands to the drive circuit, and the drive circuit drives the unlocking module through the relay. The power supply circuit supplies power to the information acquisition circuit, controller, drive circuit, relay and unlocking module respectively.

[0051] The information acquisition circuit includes a pyroelectric sensor array, a multiplexer, a signal conditioning circuit and an A / D converter. After the signal is collected by the pyroelectric sensor array, it is processed by the signal conditioning circuit through the multiplexer and then transmitted to the A / D converter for AD conversion, wherein the A / D converter is integrated inside the controller.

[0052] The controller can be any one of a single chip microcomputer, a DSP, an FPGA, and a CPU, or can identify whether the approaching object is the target object based on the correlation recognition noise reduction algorithm and the classifier. The control device controls the information acquisition circuit to extract parallel multi-channel signals from the target object's finger, and adds the extracted signals into the infrared information library of the memory after processing; the control device controls the information acquisition circuit to extract parallel multi-channel signals from the approaching object's finger, and processes the extracted signals and hands them over to the control device for the next step of identity recognition.

[0053] Pyroelectric sensor arrays, such as Figure 4 As shown, the cylinder is 150mm long, two-thirds hollow, 46mm in diameter, and >1mm thick. It is a partially hollow cylinder with one end open. The partially hollow means that 2 / 3 of the cylinder is a hollow section. The hollow section is divided into four longitudinal cylindrical surfaces at equal angles at the center of the circle. More than two circular through holes are arranged in an array at equal intervals along the cylindrical surface. The through hole diameter Ф>8mm, and the through hole interval is the same as 12mm, which is convenient for the installation, adjustment and positioning of the pyroelectric probe.

[0054] 1. Input of target object pyroelectric information template and training of support vector machine

[0055] The present invention uses a pyroelectric sensor to detect an approaching object to obtain a time domain model of a human body thermal radiation signal, and converts the thermal radiation signal into an electrical signal and inputs it into a control device for processing.

[0056] In the information entry mode, the target object places the finger into the detection area of ​​the sensor array. When the number of pyroelectric sensors is M*N, the signal array collected by the pyroelectric sensor is as follows:

[0057]

[0058] There are M*N array elements in this array, and the signal collected by each pyroelectric is:

[0059] y ij (t) = x ij (t)+n ij (t) (1)

[0060] In formula (1), y ij (t) represents the noisy signal collected by the sensor, x ij (t) indicates the signal without noise source, n ij (t) represents the noise signal, where i = 1, 2, ..., M, j = 1, 2, ..., N. The signal collected by the sensor is then subjected to noise reduction processing:

[0061]

[0062] In formula (2) Represents the signal after noise reduction processing.

[0063] Taking M=1, N=3 as an example, the signals collected by the three pyroelectric sensors are:

[0064] y 11 (t) = x 11 (t)+n 11 (t) (3)

[0065] y 12 (t) = x 12 (t)+n 12 (t) (4)

[0066] y 13 (t) = x 13 (t)+n 13 (t) (5)

[0067] In formula (3), formula (4) and formula (5), y 11 (t), y 12 (t), y 13 (t) represents the pyroelectric signals collected by three infrared sensors on the same target, x 11 (t), x 12 (t), x 13 (t) represents the signal collected by three sensors in an ideal situation, n 11 (t), n 12 (t), n 13 (t) represent the noise signals affecting the three sensors respectively.

[0068] Perform noise reduction on the signals collected by the sensor:

[0069]

[0070] In formula (6), y 11 (t), y 12 (t), y 13 (t) represents the pyroelectric signals collected by three infrared sensors on the same target, It represents the signal after noise reduction, and the processed time domain signal y(t) is used as the information template of the target object.

[0071] The pyroelectric information of the finger of the same target object is collected multiple times. After noise reduction processing, multiple pyroelectric information samples of the target object are obtained, and then the time-frequency domain mutual correlation coefficients are calculated with the recorded information templates respectively, so as to obtain the time-frequency domain correlation coefficient array. Among them, the frequency domain information uses the power spectrum as the characteristic spectrum, and a certain number of effective frequency points are selected through certain rules or algorithms. For example, the principal component analysis of the power spectrum is performed to extract the effective frequency points, and then the frequency domain mutual correlation coefficient matrix is ​​obtained according to the effective frequency points.

[0072] The calculation formula of the array time domain cross-correlation coefficient is as follows:

[0073]

[0074] In formula (7), y k and g k They represent the k-bit values ​​of the feature vectors of the discrete information template y[k] and the discrete sample signal g[k], respectively. and Respectively represent y k and g k , r is the time domain cross-correlation coefficient, where both the discrete information template y[k] and the discrete sample signal g[k] contain n points.

[0075] The calculation formula of the array frequency domain mutual correlation coefficient based on the effective frequency point is as follows:

[0076]

[0077] In formula (8), Py(f) , Pg(f) are the auto-power spectra of the discrete information template y[k] and the discrete sample signal g[k], respectively. yg (f) is the cross power spectrum of the discrete information template y[k] and the discrete sample signal g[k], C yg (f) is the frequency domain cross-correlation coefficient, and n is the number of effective frequency points selected.

[0078] Taking m samples and n effective frequency points as an example, the mixed correlation coefficient matrix obtained after correlation recognition and denoising of the close object information is:

[0079]

[0080] In formula (9), r ij is the time domain mutual correlation coefficient, C ij is the frequency domain cross-correlation coefficient, where i = 1, 2, ..., m, j = 1, 2, ..., n.

[0081] The support vector machine (SVM) is used to design the classifier with the radial basis function kernel (RBF) as the kernel function. The time-frequency domain cross-correlation coefficient array is used as the feature vector. The collected samples are cross-correlated in the time-frequency domain one by one to obtain a large number of feature vectors. These feature vectors are used as training samples to train the classifier, so that the classifier can successfully distinguish target objects from non-target objects.

[0082] 2. Collection of pyroelectric information of approaching objects and final judgment of classifier

[0083] In normal working mode, when an approaching object puts its finger into the detection area of ​​the pyroelectric sensor array, the infrared thermal radiation is converted into an electrical signal by the pyroelectric sensor. After being processed by the control device, the time domain model of the approaching object's pyroelectric information is obtained. After the information is denoised, the denoised signal is cross-correlated with the time domain template of the recorded target object to obtain a time domain cross-correlation coefficient array. The power spectrum is then used as the feature spectrum to extract the effective frequency points specified when training the classifier. The extracted power spectrum matrix is ​​cross-correlated with the power spectrum matrix of the target object to obtain a frequency domain cross-correlation coefficient matrix.

[0084] The mixed correlation coefficient matrix obtained after correlation recognition and denoising of the infrared signal of the approaching object is as follows:

[0085] [r c1 K c n ](10)

[0086] Among them, r is the time domain cross-correlation coefficient, c k is the frequency domain cross-correlation coefficient, where k = 1, 2, ..., n.

[0087] This mixed correlation coefficient matrix is ​​input as a feature vector into the trained classifier for final identity determination: if the approaching object is the target object, the identification is completed; if the approaching object is not the target object, the system will return to the initial detection state.

[0088] To verify the solution, firstly, the pyroelectric sensor array is used to collect signals from the target object's finger to obtain the target object's signal array, and then the template signal is obtained by denoising the signal array by the mean. Secondly, the pyroelectric signal of the same target object's finger is collected multiple times. After denoising, multiple pyroelectric signal samples of the target object are obtained, and then the time-frequency domain cross-correlation coefficients are calculated with the recorded template signal respectively, so as to obtain the time-frequency domain correlation coefficient array. When the approaching object puts its finger into the detection area of ​​the pyroelectric sensor array, the pyroelectric signal of the approaching object will be obtained. After the signal is denoised by the mean, the denoised signal is cross-correlated with the recorded template signal of the target object in the time domain, and the frequency domain signal uses the power spectrum as the characteristic spectrum (the power spectrum of the two is shown in Figure 2). Figure 5 ) After extracting the effective frequency points specified in advance, perform frequency domain cross-correlation to obtain the frequency domain cross-correlation coefficient matrix. Finally, the support vector machine (SVM) is used to design the classifier with the radial basis kernel function as the kernel function, and the mixed correlation coefficient matrix is ​​used as the feature vector. The collected samples are cross-correlated in the time and frequency domain one by one to obtain a large number of feature vectors. The obtained feature vectors are used as training samples to train the classifier, which can make the classifier accuracy as high as 93%. See Figure 6 It has been verified that this embodiment can accurately distinguish target objects and non-target objects.

[0089] The pyroelectric sensor device is in the shape of a partially hollow semi-open cylinder. The total length of the identification device is 10-15 cm, and two-thirds of it is hollow to allow fingers to be inserted. The inner diameter of the opening is Y cm (Y>1 cm), and the wall thickness is Z cm (Z>0.1 cm). M groups of mounting through holes are arranged in an array on all four sides of the hollow part. Each group of circular through holes with a diameter of W cm (W>0.8 cm) are arranged in a ring. The through holes are spaced at the same distance to facilitate the installation, adjustment and positioning of the pyroelectric probe.

[0090] The identification device is 15 cm long, two-thirds of which is hollow, allowing fingers to be inserted. The inner diameter of the opening is 4.6 cm and the wall thickness is 0.2 cm. M groups of mounting through holes are arranged in an array on all four sides of the hollow part, and each group of circular through holes with a diameter of 1.2 cm are arranged in a ring.

[0091] The pyroelectric sensor device is partially hollow and half open cylindrical, such as Figure 4 , as shown in the internal and external structure diagram of the pyroelectric probe. The total length is 150mm, two-thirds of which are hollow, allowing fingers to be inserted. The inner diameter of the opening is 46mm, and the wall thickness is less than 2mm. More than two groups of array-type distributed installation through holes are arranged on the four sides of the hollow part. Each group of circular through holes with a diameter of 12mm are arranged in an annular manner. The through holes are spaced at the same interval to facilitate the installation, adjustment and positioning of the pyroelectric probe.

[0092] The pyroelectric probe internal and external structure identification device is Xcm long, two-thirds of which are hollow to allow fingers to be inserted. The inner diameter of the opening is Ycm (Y>1cm) and the wall thickness is Zcm (Z>0.1cm). M groups of mounting through holes are arranged in an array on all four sides of the hollow part. Each group of circular through holes with a diameter of Wcm (W>0.8cm) are arranged in a ring. The through holes are spaced at the same distance to facilitate the installation, adjustment and positioning of the pyroelectric probe.

[0093] The total length of the identification device is X = 15 cm, the inner diameter of the opening is Y = 4.6 cm, the barrel wall thickness is Z = 0.2 cm, and the number of circular through holes is 3*4 (M = 3, N = 4). Figure 4 a, 4b illustrate the pyroelectric probe installation device.

[0094] Figure 4 a is the overall diagram of the device, Figure 4 b is a half-section view and a dimension view of the half-section view of the loading device.

[0095] See also Figure 4 There are 3*4 through-type installation holes A on a, and the pyroelectric probe is glued to the cylinder wall with 502 glue for installation.

Claims

1. A time-frequency domain cross-correlation identity recognition method, characterized in that: include: A. Target object pyroelectric information model input and vector machine training, B. Approaching object pyroelectric information collection and classifier discrimination; The target object pyroelectric information model entry and support vector machine training include: a. Collect multi-channel signals from the target object’s fingers; b. Processing the collected signals to obtain an information model; c. Collect infrared information of target objects multiple times and calculate the time domain correlation coefficient of the information model; d. Select 1-20 effective frequency points by performing principal component analysis on the power spectrum; e. Input the time-frequency domain mutual correlation coefficient as a feature vector into the classifier for training; Specifically, in the information entry mode, the target object places the finger into the detection area of ​​the sensor array. When the number of pyroelectric sensors is M*N, the signal array collected by the pyroelectric sensor is as follows: There are M*N array elements in this array, and the signal collected by each pyroelectric is: y ij (t)=x ij (t)+n ij (t) (1) In formula (1), y ij (t) represents the noisy signal collected by the sensor, x ij (t) indicates the signal without noise source, n ij (t) represents the noise signal, where i = 1, 2, ..., M, j = 1, 2, ..., N; the signal collected by the sensor is then subjected to noise reduction processing: In formula (2) represents the signal after noise reduction processing; Using the processed time domain signal as the information model of the target object; The pyroelectric information of the finger of the same target object is collected multiple times. After noise reduction processing, multiple pyroelectric information samples of the target object are obtained, and then the time-frequency domain mutual correlation coefficients are calculated with the recorded information model respectively, so as to obtain the time-frequency domain correlation coefficient array; The calculation formula of the array time domain cross-correlation coefficient is as follows: In formula (7), y k and g k They represent the k-bit values ​​of the feature vectors of the discrete information template y[k] and the discrete sample signal g[k], respectively. and Respectively represent y k and g k The mean value of , r is the time domain cross-correlation coefficient, where both the discrete information template y[k] and the discrete sample signal g[k] contain n points; The calculation formula of the array frequency domain mutual correlation coefficient based on the effective frequency point is as follows: In formula (8), P y (f), P g (f) are the auto-power spectra of the discrete information template y[k] and the discrete sample signal g[k], P yg (f) is the cross power spectrum of the discrete information template y[k] and the discrete sample signal g[k], C yg (f) is the frequency domain mutual correlation coefficient, n is the number of effective frequency points selected; With m samples and n effective frequency points, the mixed correlation coefficient matrix obtained after correlation recognition and denoising of the close object information is: In formula (9), r ij is the time domain mutual correlation coefficient, C ij is the frequency domain cross-correlation coefficient, where i = 1, 2, ..., m, j = 1, 2, ..., n; The support vector machine is used to design the classifier with the radial basis kernel function as the kernel function, and the time-frequency domain cross-correlation coefficient array is used as the feature vector. The collected samples are cross-correlated in the time-frequency domain one by one to obtain a large number of feature vectors. The obtained feature vectors are used as training samples to train the classifier, so that the classifier can successfully distinguish the target object and the non-target object. The pyroelectric information collection and classifier identification of the approaching object include: Ⅰ. Acquisition of pyroelectric information of approaching objects; Ⅱ. Process the collected signals; III. Calculation of the time-frequency domain mutual correlation coefficient between the object information and the information model; IV. Classifier designed based on support vector machine for discrimination; Among them, the mixed correlation coefficient matrix obtained after correlation recognition and noise reduction of the infrared signal of the approaching object is as follows: [r c1 K c n ] (10) Among them, r is the time domain cross-correlation coefficient, c k is the frequency domain mutual correlation coefficient, where k = 1, 2, ..., n. This mixed correlation coefficient matrix is ​​input into the trained classifier as a feature vector for final identity determination: if the approaching object is the target object, the recognition is completed; if the approaching object is not the target object, the system will return to the initial detection state.

2. A device for implementing the time-frequency domain cross-correlation identity recognition method according to claim 1, characterized in that: The X-th group of pyroelectric sensor arrays are connected to the controller via the X-th group of multiplexers and the X-th group of conditioning circuits, the first group of pyroelectric sensor arrays are connected to the controller via the first group of multiplexers and the first group of conditioning circuits, the controller is connected to the unlocking module via the drive circuit and the relay, the controller is connected to the memory, and the power module is respectively connected to the information acquisition circuit, the controller, the drive circuit, the relay and the unlocking module; wherein the pyroelectric sensor array is in the shape of a partially hollow cylinder with one end open, the cylinder is 150mm long, two-thirds hollow, the cylinder diameter is 46mm, the wall thickness is>1mm, the said partial hollow means that 2 / 3 of the cylinder is a hollow section, the hollow section is divided into four longitudinal cylindrical surfaces at equal angles at the center of the circle, and two or more circular through holes distributed in an array are arranged at equal intervals along the cylindrical surface, the through hole diameter Ф>8mm, and the through hole interval is the same as 12mm; The controller controls the information acquisition circuit to extract parallel multi-channel signals from the target object's finger, and adds the extracted signals into the infrared information library of the memory after processing; the controller controls the information acquisition circuit to extract parallel multi-channel signals from the approaching object's finger, and processes the extracted signals and hands them over to the controller for the next step of identity recognition.

Images