Fingerprint iteration imaging method based on nonlinear Lamb wave frequency dispersion effect

Abstract

The invention provides a non-linear Lamb wave frequency dispersion effect-based fingerprint iterative imaging method. The method comprises the following steps of: acquiring a time domain sound field signal and a fingerprint grayscale image when a fingertip presses a glass plate; frequency dispersion curve changes of Lamb wave fundamental waves and second harmonics are calculated, and second harmonic excitation parameters are determined; constructing a fingerprint ridge-valley image, and establishing a speed model; determining a second harmonic propagation velocity by using a group velocity frequency dispersion curve, extracting signal features, and establishing a frequency domain sound field matrix; fingerprint primary positioning is completed through secondary harmonic ray tomography, and an initial model and a sound field matrix are updated; establishing an offline training data set, and training by adopting a supervised descent method; and completing fingerprint iterative imaging by using the descending gradient matrix output by training. According to the method, the high sensitivity of the nonlinear Lamb waves to fingerprints is utilized, the method has the characteristics of zero occupation of space under a transducer screen, small data set and low imaging error, and wide-area, accurate and rapid representation of submillimeter fingerprint details can be realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biometric authentication, and particularly relates to a Lamb wave fingerprint imaging method. TECHNICAL BACKGROUND

[0002] Biometric authentication methods use unique biological features to improve security, optimize user experience, and enhance the accuracy and convenience of identity verification in the digital age. Compared with other human physiological feature authentication methods, fingerprint authentication has the advantages of stable performance and high cost performance, and accounts for more than half of the global biometric authentication market. With the widespread application of fingerprint authentication technology in various electronic devices, the technology is gradually developing towards multi-fingerprint or palmprint authentication. However, existing optical, capacitive and ultrasonic fingerprint scanning methods all rely on signal acquisition methods perpendicular to the screen and complex under-screen devices, which limit the scanning domain, and the former two are easily disturbed by environmental pollution, thereby affecting the convenience and accuracy of personal authentication.

[0003] Domestic and foreign scholars have carried out related research in the fields of plate waveguide acoustic field theory and target detection, and proposed guided wave tomography methods such as ray tomography, diffraction tomography and full waveform inversion. The above methods use the multi-mode dispersion characteristics of guided waves to provide a possibility to break through the limitations of the scanning domain. Among them, the minimum resolvable size of the ray tomography method is limited by the first Fresnel zone, and the diffraction tomography method is suitable for detecting defects with low contrast and smoothness. The patent document with the application publication number CN117825526A discloses a shape reconstruction method based on ultrasonic full waveform inversion, which considers characteristic parameters such as diffraction and multiple scattering group velocity, and the resolution reaches below the excitation wavelength. Although this method has the ability to accurately reconstruct sub-millimeter textures, it needs to calculate the Hessian matrix to find the optimal reconstruction result, resulting in extremely high computational cost, which limits its timeliness in practical applications and is not suitable for fingerprint authentication.

[0004] In recent years, intelligent fast guided wave tomography has become a research hotspot in the field of quantitative imaging. This method not only has high reliability and computational efficiency, but also makes full use of the rich characteristic parameters of the acoustic field, expands it to the high-frequency field and solves the problem of mapping the changes of the fingertip pressing guided wave acoustic field and the changes of the fingerprint morphology, which can provide a new solution for large-scale fast scanning and characterization of sub-millimeter level fingerprint details.

[0005] There is a significant difference in material parameters between the fingertip epithelium and the glass, which will cause the dispersion effect to become more severe when the fingertip contacts the glass surface. The traditional ultrasonic guided wave tomography method mainly relies on the characteristic parameters of linear guided waves such as travel time, amplitude, signal difference coefficient, and is often used for macroscopic target detection. The patent document with the application publication number CN119810876A discloses an ultrasonic guided wave fingerprint imaging and authentication method, which characterizes the fingerprint morphology through the linear acoustic field spectrum peak value change caused by the fingertip pressing the glass plate. However, the inherent unevenness of the fingerprint surface causes micro-contact between the epithelial layer and the plate surface, and this micro-contact usually has a weak effect on the acoustic field, but it may introduce errors in the fingerprint morphology and thickness reconstruction process. In comparison, the nonlinear parameters of guided waves can carry more information about the subtle changes in the structure, and should have higher accuracy in characterizing sub-millimeter textures, and can effectively suppress imaging artifacts. Therefore, it is urgent to propose an intelligent and fast guided wave tomography method that can introduce more nonlinear acoustic parameters, fully utilize acoustic field information to reduce reconstruction errors, and achieve wide-area accurate and fast characterization of sub-millimeter fingerprint details. SUMMARY

[0006] In view of the problems that the existing fingerprint scanning technology has limited scanning domain, and the traditional ultrasonic guided wave tomography method has deficiencies in resolution, timeliness and sensitivity, the present application proposes a fingerprint iterative imaging method based on nonlinear Lamb wave dispersion effect, which detects the change of Lamb wave acoustic field propagating along the plate surface through a transducer array, effectively expands the fingerprint scanning domain, uses the second harmonic characteristic for fingertip positioning in the first imaging, and uses the fundamental wave dispersion effect for quantitative reconstruction of the fingerprint morphology in the second imaging, thereby fully utilizing the Lamb wave acoustic field information, reducing the size of the fingerprint imaging data set, and characterizing sub-millimeter fingerprint details with low error and low imaging artifacts.

[0007] In order to achieve the above purpose, the technical scheme of the present application is as follows: a fingerprint iterative imaging method based on nonlinear Lamb wave dispersion effect, comprising the following steps:

[0008] Step 1: Use a wide-area ultrasonic transducer and a fingerprint acquisition system to acquire time-domain acoustic field signals and fingerprint grayscale images when the fingertip at any position presses the glass plate;

[0009] Step 2: Calculate and draw the variation of the Lamb wave fundamental wave and the second harmonic dispersion curve at the junction of the fingertip epithelium and the glass material using the global matrix method, and determine the working frequency f w of the second harmonic excitation point and the transducer;

[0010] Step three: the optical sensor collected fingerprint grayscale image is processed into a fingerprint ridge valley image by image processing, and is mapped to the phase velocity variation interval of the Lamb wave target mode at the adhesion position of the fingerprint epidermis and the glass material, and a fingerprint velocity model is established according to the processed ridge valley image;

[0011] Step four: the group velocity dispersion curve is used to determine the wave packet propagation velocity of the Lamb wave second harmonic, the time domain signal of the target mode is locked and extracted, the energy flow amplitude between each two channels is extracted one by one through fast Fourier transform and Gaussian filter smoothing processing, and a frequency domain sound field matrix is established;

[0012] Step five: the fingerprint is positioned and reconstructed by the Lamb wave second harmonic ray tomography method, the first fingerprint imaging result is obtained, the sound field matrix is updated by the finite difference method, and the initial model and the frequency domain sound field matrix are provided for the subsequent second fingerprint imaging method;

[0013] Step six: through steps three to five, the velocity model of the Lamb wave fundamental wave target mode, the updated initial model and the frequency domain sound field matrix of the off-line training data set example are obtained, the off-line training data set of the second fingerprint imaging method is established, and the descending gradient matrix is obtained through off-line training by the supervised descent method;

[0014] Step seven: through steps three to five, the velocity model of the fundamental wave target mode, the updated initial model and the frequency domain sound field matrix of the test example are obtained, and the second fingerprint imaging result is obtained by using the descending gradient matrix obtained in step six.

[0015] Preferably, the fingerprint wide-area acquisition ultrasonic transducer is composed of an n-channel transducer array composed of n piezoelectric ceramic chips, the piezoelectric ceramic chips adopt PZT-5H material with high center frequency, and the electrical interconnection of each channel element is formed by positive and negative copper foils; the negative copper foil is a continuous whole, and provides an acoustic field waveguide and a ground wire for each channel element; the number of positive copper foils is the same as the number of channels, and is divided into four positive copper foil bundles according to the number and is led out in four directions;

[0016] The fingerprint acquisition system is composed of a fingerprint wide-area acquisition ultrasonic transducer, an optical sensor, an ultrasonic instrument and a signal transmission circuit; the fingerprint wide-area acquisition ultrasonic transducer, the ultrasonic instrument and the signal transmission circuit are used to acquire the time domain sound field signal when the fingertip presses the glass plate, the fingerprint wide-area acquisition ultrasonic transducer acquires the signal in the working mode of single-element excitation and full-array reception; the optical sensor is used to acquire the fingerprint grayscale image when the fingertip presses the glass plate.

[0017] Preferably, the change of the fundamental wave and the second harmonic wave dispersion curve of the fingerprint epidermis at the adhesion position with the glass material is related to the obvious acoustic field scattering caused by the pressing of the fingerprint, and the change of the dispersion curve is obtained by taking the thickness, density, Young's modulus and Poisson's ratio of the glass material and the fingerprint epidermis material and the height difference between the fingerprint ridge and valley as inputs, and calculating the dispersion curves of the fundamental wave and the second harmonic wave of the fingerprint epidermis before and after adhesion with the glass material by using the global matrix method.

[0018] The working frequency f of the fingerprint wide-area acquisition ultrasonic transducer w The same frequency value is used, and the determination method is as follows: through global matrix method wave structure analysis, a Lamb wave mode with energy flow mainly concentrated in the normal direction is selected as the fundamental wave mode for exciting the target mode of the second harmonic wave;

[0019] Preferably, the method for establishing the fingerprint velocity model is as follows: the high dynamic range fingerprint gray scale image collected by the optical sensor is adjusted to a low dynamic range image by using an image processing method, the connected region of the fingerprint detail edge is removed, the gray scale change caused by the pressing force is eliminated, and a fingerprint ridge-valley image is obtained; the phase velocity descending interval [c d2 ,c u2 ] of the target second harmonic wave at the second harmonic wave excitation point in the dispersion curve is determined, the gray scale interval of the fingerprint ridge-valley gray scale image is mapped to the phase velocity interval, and a second harmonic wave velocity model is obtained; the fundamental wave mode with the shortest wavelength at the frequency f w is taken as the target mode, and its phase velocity descending interval [c d1 ,c u1 ] is determined, the gray scale interval of the fingerprint ridge-valley gray scale image is mapped to the phase velocity interval, and a fundamental wave velocity model is obtained.

[0020] The second harmonic wave velocity model is used for comparison and verification of the fingerprint morphology characterization effect; the fundamental wave velocity model is used as the velocity model in the subsequent second fingerprint imaging, which takes into account the advantages of the nonlinear Lamb wave characteristics while taking into account the linear characteristics and resolution.

[0021] Preferably, the method for establishing the second harmonic wave frequency domain acoustic field matrix is as follows: at the frequency f w , the wave packet propagation velocity of the second harmonic wave is determined according to the group velocity dispersion curve, the time domain signal of the target mode is locked and extracted by using a window function; the energy flow amplitude between each two channels is extracted one by one by using fast Fourier transform and Gaussian filter smoothing processing, thereby generating an n x n frequency domain acoustic field matrix with one dimension related to the number n of channels of the fingerprint wide-area acquisition ultrasonic transducer.

[0022] Preferably, the first fingerprint imaging method is: establishing a baseline frequency domain sound field matrix and a fingerprint pressing frequency domain sound field matrix, calculating the peak drop value between each two channels, using an elliptical algorithm to calculate the signal difference coefficient distribution, and obtaining a fingerprint positioning reconstruction result; through signal difference coefficient threshold control, extracting the fingerprint pressing part and mapping to the phase velocity drop interval [c d1 ,c u1 ] of the fundamental wave target mode, thereby providing an initial model that changes with the fingerprint pressing state for the second fingerprint imaging method.

[0023] Preferably, the sound field matrix updating method is: using a frequency domain finite difference method based on a linear elastic model to calculate the frequency domain sound field matrix corresponding to the initial model, and replacing the second harmonic sound field matrix in the first fingerprint imaging, and then using the data set offline training based on the supervised descent method and the online inversion of the test example of the second fingerprint imaging.

[0024] Preferably, the second fingerprint imaging method is: after obtaining the descent gradient matrix in the data set offline training based on the supervised descent method, establishing a fingerprint velocity model, establishing a second harmonic frequency domain sound field matrix, first fingerprint imaging and sound field matrix updating, obtaining the velocity model of the fundamental wave target mode, the updated initial model and the frequency domain sound field matrix of the test example, and finally obtaining the online inversion result.

[0025] Compared with the prior art, the present application has the following advantages: the unique advantages of Lamb wave propagation on the surface or inside of the waveguide structure are fully utilized, and the change rule of the Lamb wave sound field characteristics after the fingerprint ridge is attached to the glass plate is realized. The transducer array realizes wide-area characterization of the fingerprint form on the waveguide structure. The imaging method with two iterations takes into account the high sensitivity of nonlinear Lamb waves to the inherent unevenness of the fingerprint surface, and the discrimination accuracy of the fundamental Lamb wave dispersion effect on the fingerprint epidermis and the glass plate. The sound field change caused by the fingerprint pressing on the glass plate is fully utilized, and the information utilization rate of the data set example is greatly improved. In addition, by moving most of the calculation to the offline training stage, the time efficiency limitation of the traditional full waveform inversion method is effectively broken. The iterative imaging result shows that the imaging artifacts and image noise are significantly reduced, and only a ring-shaped band caused by the deviation of the fingerprint profile in two imaging exists. Since the ring-shaped band is located in the external area of the quantitative reconstruction of the fingerprint form, the subsequent image processing and fingerprint authentication process can thus reduce the calculation pressure.

[0026] The fingerprint iterative imaging method based on the nonlinear Lamb wave dispersion effect has the characteristics of zero space occupation under the transducer screen, small offline training data set and low imaging error. The method can realize wide-area, accurate and rapid characterization of sub-millimeter fingerprint details, which not only opens up a new path for large-screen identity authentication of diversified consumer electronics and public security equipment, but also provides an efficient solution for full-screen multi-fingerprint and palmprint authentication. BRIEF DESCRIPTION OF DRAWINGS

[0027] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0028] Figure 1 Flowchart of the present application.

[0029] Figure 2 Fingerprint collection operation schematic diagram of the embodiment of the present application.

[0030] Figure 3 Fingerprint collection transducer system of the embodiment of the present application.

[0031] Figure 4 The fingerprint velocity model establishment method based on Lamb wave second harmonic proposed by the present application, wherein (a) and (b) are enlarged diagrams in the dashed box in the phase velocity (c p ) dispersion curve diagram describing the sound velocity drop.

[0032] Figure 5 The obtaining schematic diagram of the Lamb wave second harmonic frequency domain sound field matrix proposed by the present application, wherein (a) is an enlarged diagram in the dashed box in the group velocity (c g ) dispersion curve diagram describing the sound propagation velocity.

[0033] Figure 6 The first imaging result of one fingertip in the embodiment of the present application.

[0034] Figure 7 The initial model update result and the sound field matrix update result of the embodiment shown in Figure 6 .

[0035] Figure 8 The second imaging result and the enlarged diagram of the fingerprint morphology reconstruction of the embodiment shown in Figure 6 .

[0036] Figure 9 The second imaging result and the enlarged diagram of the fingerprint morphology reconstruction of the embodiment shown in Figure 6Magnified view of the fundamental wave based FIT imaging results and the fingerprint morphology reconstruction of the shown embodiment, the set of results used for comparison with Figure 8 the results shown.

[0037] In the figure, 1 is a transducer array, 2 is a glass plate, 3 represents a fingertip pressing position, 4 represents sound field scattering, and 5 is a negative copper foil. DETAILED DESCRIPTION

[0038] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of the present application.

[0039] As Figure 1 shown, a fingerprint iterative imaging method based on nonlinear Lamb wave dispersion effect aims to improve the sensitivity of the Lamb wave tomography method to the micro-contact between the fingerprint and the glass plate, improve the utilization rate of the training set examples of the intelligent fast guided wave tomography method, and reduce the fingerprint reconstruction error. Relying on a fingerprint collection ultrasonic transducer with a scanning domain of φ100 mm and a fingerprint collection system developed in the invention patent with application number 202411880989.X as the verification device of the imaging method, the change of the Lamb wave second harmonic dispersion effect caused by the fingertip pressing on the glass plate is used to develop the frequency domain sound field matrix extraction and fingerprint velocity model establishment method, an offline training data set for fingerprint iterative imaging is established, and an iterative imaging method of cascaded positioning reconstruction and quantitative reconstruction is proposed. The method can realize high sensitivity and low artifact fingerprint wide-area fast imaging of the test example fingerprint. Among them, the Lamb wave second harmonic ray tomography method is used for fingerprint positioning (first fingerprint imaging), which provides an initial model for the subsequent second fingerprint imaging. On this basis, the sound field matrix is updated by finite difference forward modeling, and the descent gradient matrix output by offline training using supervised descent method is used for the second fingerprint imaging (i.e. fingerprint morphology quantitative), so as to realize sub-millimeter fingerprint detail reconstruction. The specific steps are as follows:

[0040] Step one: use the designed fingerprint wide-area collection ultrasonic transducer and fingerprint collection system to collect the time-domain sound field signals and fingerprint grayscale images when the fingertip presses the glass plate at any position.

[0041] As Figure 2As shown, when the fingertip presses against the glass plate 2, the mechanical properties of the fingertip epithelium are closer to those of a fluid, exhibiting a significant difference from the rigid glass material. This results in noticeable sound field scattering 4 and dispersion effects at the fingertip pressing point 3, where the Lamb wave propagating along the glass surface is located. The transducer array 1 employs a single-element sequential excitation and simultaneous reception mode, and the acquired time-domain signal contains rich sound field information, providing a crucial data foundation for sound field matrix establishment and fingerprint reconstruction.

[0042] like Figure 3 As shown, the fingerprint wide-area acquisition ultrasonic transducer used in this embodiment has a scanning area of ​​φ100mm. Its transducer array 2 consists of 128 piezoelectric ceramics (model: PZT-5H) with a center frequency of 5MHz. Each element is a φ2mm circular wafer, which is fixed between positive and negative copper foils with conductive silver paste. The negative copper foil 5 also serves as a sound field waveguide. It is fixed to the glass plate 2 with epoxy resin adhesive and led outwards through 4 pins to form 4 ground wires. There is no electrical interconnection between the positive copper foils. Its 128 channels are divided into 4 positive copper foil bundles and led outwards. Each positive copper foil bundle is flexibly fixed with a polydimethylsiloxane (PDMS) film and forms an interface with a ground wire, which is then connected to the ultrasonic instrument (model: Verasonics multi-channel ultrasonic system). An insulating layer is also attached to the bottom of each positive copper foil bundle. In addition, an optical sensor (model: EPSON V600) is installed below the glass plate 2 to collect grayscale images of the fingerprint pattern when the fingertip presses on the glass plate. This data is used for offline training of the intelligent fast guided wave tomography method and provides a velocity model for calibration of the acoustic field matrix in the simulation. The fingerprint acquisition ultrasonic transducer, optical sensor, ultrasonic instrument, and its signal transmission circuit together constitute the fingerprint acquisition system.

[0043] Step 2: Calculate and plot the dispersion curves of the Lamb fundamental and second harmonic waves at the junction of the fingertip epithelium and the glass material using the global matrix method, and determine the second harmonic excitation point and the transducer's operating frequency f. w .

[0044] In this embodiment, Schott B270 soda-lime glass was selected as the glass material, with a thickness d of 0.55 mm and a density ρ of 2.56 g / cm³. 3 The Young's modulus E is 71.10 GPa, and the Poisson's ratio ν is 0.22. The material parameters of the fingertip epithelium are: d = 1.5 mm, ρ = 1.02 g / cm³. 3 E = 2.19 GPa, ν = 0.4, and the height difference Δh between the ridge and valley is 0.25 mm. The dispersion curves of the fundamental frequency, second harmonic phase velocity, and group velocity calculated and plotted using the global matrix method are shown below. Figure 4 and Figure 5In the dispersion curve, the fundamental mode is marked by capital letters, the symmetric mode (S mode) and the anti-symmetric mode (A mode) are sequentially marked from low frequency to high frequency, for example: A0, A1, S0, S1, etc.; the second harmonic mode is marked by lowercase letters, since the second harmonic with energy flow accumulation effect is all symmetric mode, so the mark is only s0, s1, s2, etc.; the second harmonic is excited by the fundamental wave, and the excitation point is identified in the mode of "fundamental mode-second harmonic mode", for example: S0-s0 mode pair, S1-s2 mode pair, etc.

[0045] Through the global matrix method wave structure analysis, it can be obtained that in the thin plate material at high working frequency, the energy flow of the S1 mode is mainly concentrated in the normal direction, and the S1-s2 mode pair excited by the second harmonic has higher energy flow transmission efficiency. Therefore, the s2 mode is selected as the Lamb wave second harmonic target mode in this test example, and the S1 mode is the fundamental wave mode for exciting the s2 mode. As shown in Figure 4 , the frequency corresponding to the S1-s2 mode pair is 5.06 MHz, which will be used as the excitation frequency of the second harmonic s2 mode and the working frequency of this embodiment, that is, f w = 5.06 MHz.

[0046] Step three: process the fingerprint ridge-valley image obtained by the optical sensor into a fingerprint ridge-valley image through image processing, and map it to the change interval of the Lamb wave phase velocity at the adhesion position of the fingerprint epidermis and the glass material, and establish a fingerprint velocity model.

[0047] As shown in Figure 4 , the image collected by the optical sensor through multiple exposures is a high dynamic range (HDR) image, and the fingerprint details in such an image are extremely complex, and it is difficult to directly process it into a velocity model. Therefore, a bilateral filtering method is used to adjust the image to a low dynamic range (LDR), which effectively compresses the contrast while ensuring the image quality. Further, the spatial filtering method is used to remove the connected region of the fingerprint detail edge, simplify the fingerprint detail, and eliminate the gray scale change caused by the pressing force, so as to obtain a ridge-valley gray scale image representing only the fingerprint features.

[0048] In this embodiment, the phase velocity (c p ) value of the second harmonic excitation point S1-s2 mode pair of the glass material is 7.14 mm / μs, and after the fingerprint epidermis is adhered to the glass material, the c p value drops to 6.74 mm / μs, that is, the phase velocity drop interval [c d2 , c u2 ] is [6.74, 7.14] mm / μs. Next, the gray scale interval of the processed fingerprint ridge-valley gray scale image is mapped to the c pThe interval is determined, thereby establishing a second harmonic velocity model describing the fingerprint morphology, which is only used for comparison and verification of the fingerprint morphology characterization effect.

[0049] On the other hand, among the numerous modes at the operating frequency of 5.06 MHz, the c p value of the fundamental mode A0 is the smallest and has the shortest wavelength. Therefore, the mode is selected as the target mode of the Lamb wave fundamental. In the second fingerprint imaging later, the present application will determine the c p interval of the A0 mode and establish a velocity model (the phase velocity falling interval [c d1 , c u1 ] is [2.78, 2.97] mm / μs). The coexistence of the two models can effectively take advantage of the linear and nonlinear Lamb wave characteristics, further improving the accuracy of fingerprint imaging.

[0050] Step four: According to the group velocity dispersion curve, the wave packet propagation velocity of the second harmonic s2 mode is determined, the time domain signal of the mode is locked and extracted, and the energy flow amplitude between each two channels is extracted one by one through fast Fourier transform and Gaussian filter smoothing processing, thereby establishing a frequency domain sound field matrix.

[0051] As shown in Figure 5 , the time domain signals between the two transducers on the diameter are shown. At the operating frequency of 5.06 MHz, the dispersion phenomenon is significant, and the guided wave mode is obviously increased. Therefore, the present embodiment sets a trapezoidal short window with a total length of 5 μs and a rising and falling edge of 1 μs, which is used to extract the time domain signal of the s2 mode with a group velocity (c g ) value of 2.73 mm / μs. The frequency spectrum after fast Fourier transform shows that after the fingertip is pressed, the spectral peak value decreases from 0.15 to 0.054, with a decrease of 64.00%. After further Gaussian filter smoothing processing, the peak value is updated from 8.4×10 -2 to 3.2×10 -2 , with a decrease of 61.90%. This significant change can be attributed to two aspects: first, the adhesion of the ridge to the glass plate causes a significant decrease in c p , which causes the mismatch of the second harmonic excitation point, thereby causing the energy flow of the excited s2 mode to decrease; second, the adhesion of the ridge to the glass plate causes scattering of the sound field, which is manifested as a decrease in the spectral intensity collected by the transducer. The superposition of the above two factors makes the Lamb wave second harmonic highly sensitive to the overall structural changes caused by the fingertip pressing.

[0052] In the transducer array, the time-domain signal between every two channels is processed using the method described above, extracting the spectral peak after Gaussian filtering and smoothing, thereby generating a frequency-domain acoustic field matrix with dimension n×n. Since the number of channels n in the fingerprint wide-area acquisition ultrasonic transducer is 128, the dimension of this matrix is ​​128×128, and it has a one-to-one correspondence with the fingerprint velocity model. To avoid non-numerical problems caused by the presence of zero values ​​during the offline training process of the second imaging, the acoustic field energy flow amplitude is mapped to the interval [1,2].

[0053] In addition, the pure black grayscale image of the optical sensor when it is unloaded and the sound field collected by the fingerprint acquisition ultrasonic transducer when it is unloaded together constitute the baseline signal for obtaining the velocity model difference and the frequency domain sound field difference.

[0054] Step 5: Fingerprint localization and reconstruction are performed using Lamb wave second harmonic ray tomography to achieve the first fingerprint imaging, providing an initial model for the subsequent second fingerprint imaging method, and the acoustic field matrix is ​​updated using finite difference.

[0055] like Figure 6 As shown, after obtaining the baseline frequency domain acoustic field matrix and the fingertip pressure frequency domain acoustic field matrix, the peak drop value between every two channels is calculated, and an elliptic algorithm is used for imaging, forming a Lamb wave second harmonic ray tomography method. This allows for the preliminary acquisition of the fingertip pressure localization reconstruction result, thus completing the first fingerprint imaging. This result is represented as a localization probability distribution map, used to describe the distribution of the signal difference coefficient (SDC) in variable form.

[0056] In this embodiment, the higher 20% of the SDC value range is extracted to achieve SDC threshold control, thereby extracting the fingertip pressing portion. Subsequently, the SDC value range is mapped to... Figure 4 The fundamental A0 mode c shown p The interval is used to generate a velocity model that describes the fingertip pressure, providing an initial model for the second fingerprint imaging method that varies with the fingertip pressure state, such as... Figure 7 As shown.

[0057] Furthermore, using this initial model and the finite-difference forward modeling method, the sound field matrix is ​​updated numerically, and the results are presented in... Figure 7 In this model, the amplitude of the sound field energy flow is also mapped to the interval [1,2]. Considering the time constraints of fingerprint imaging, this embodiment adopts a frequency domain finite difference method based on a linear elastic model. It should be noted that using this method means that the second harmonic of the Lamb wave generated by the coupling between the sound field and the structure will no longer exist at the second harmonic position. Therefore, in the second imaging, the initial model and velocity model corresponding to the A0 mode should be used, and the phase velocity decrease interval is [2.78,2.97] mm / μs.

[0058] Step six: Establish the offline training dataset of the second fingerprint imaging method, and obtain the descent gradient matrix by offline training with the supervised descent method.

[0059] In this embodiment, the left thumb is selected as the registered fingertip, and is pressed at an arbitrary position in the scanning domain of the transducer device. Through steps three to five, the A0 mode velocity model, the updated initial model, and the frequency domain sound field matrix of the offline training dataset example are obtained. In each example, the three have a one-to-one correspondence.

[0060] The offline training is performed on a HP workstation equipped with 32 Intel Xeon Gold 6230 CPUs, and the running environment is MATLAB R2020a. Using the supervised descent method, the nonlinear mapping relationship between the sound field change and the velocity model change caused by the fingertip pressing the glass plate is learned, and the formula is:

[0061]

[0062] In the formula, R is the descent gradient matrix. Δs = s0-s1 represents the sound field change, s0 and s1 are the frequency domain sound field matrix extracted in step four and the sound field matrix updated in step five, respectively. m represents the fingerprint model described by the slowness matrix, and its relationship with the fingerprint velocity model c p is: Therefore, Δm = m0-m1 is the velocity model change, and m0 and m1 are the initial model updated in step five and the A0 mode fingerprint velocity model established in step three, respectively. The supervised descent method is essentially a fitting process, which obtains R by minimizing the sound field mismatch function between m0 and m1.

[0063] Since the iterative imaging method utilizes the nonlinear Lamb wave characteristics to realize fingertip positioning in the first imaging, and utilizes the fundamental frequency Lamb wave dispersion effect to quantitatively reconstruct the fingerprint morphology in the second imaging, the sound field change caused by the fingertip pressing the glass plate is fully utilized. Therefore, this embodiment only uses the 512 examples obtained by pressing the registered fingertip at an arbitrary position in the scanning domain to establish the offline training dataset. Ultimately, the fingerprint iterative imaging method proposed in this application belongs to a small sample guided wave quantitative imaging method.

[0064] Step seven: Use the descent gradient matrix to perform the second fingerprint imaging to quantitatively reconstruct the fingerprint morphology of the test example with low error and low imaging artifacts, and obtain a fingerprint iterative imaging method based on the nonlinear Lamb wave dispersion effect.

[0065] Specifically, after obtaining the descending gradient matrix R through step six, the online inversion of the fingerprint pressing test example is started. Steps three to five are repeated to obtain the A0 mode velocity model of the test example, the updated initial model and the frequency domain sound field matrix, and input them into the second imaging online inversion formula:

[0066]

[0067] In the formula, M test is the slowness form of the test example imaging result, is the updated initial model of the test example, Δs test is the sound field change of the test example.

[0068] Finally, the relationship is used to convert M test into the velocity model form to obtain the fingerprint second imaging result, as shown in Figure 8 As a comparison, Figure 9 shows the result of directly using the Lamb wave A0 mode for FIT imaging. In terms of reconstruction root mean square error, the fingerprint iterative imaging method is 9.53x10 -3 , lower than the control group of 10.50x10 -3 . This difference is mainly due to the fact that the iterative imaging method has less imaging artifacts and image noise, and only one annular band caused by the deviation of the two imaging fingerprint profiles. Since the annular band is not located in the internal area of the quantitative reconstruction of the fingerprint morphology, it has little effect on subsequent image post-processing and fingerprint authentication. In summary, the fingerprint iterative imaging method proposed in the present application can effectively reduce the reconstruction error and realize the wide-area accurate and fast characterization of sub-millimeter fingerprint details.

[0069] The above only describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method of fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect, characterized in that, The steps include: Step one: using the fingerprint wide-area collection ultrasonic transducer and the fingerprint collection system to collect the time-domain acoustic field signal and the fingerprint gray image when the fingertip presses the glass plate at any position; Step two: Calculate and draw the dispersion curve of the fundamental wave and the second harmonic wave of the Lamb wave at the junction of the fingertip epithelium and the glass material using the global matrix method, and determine the second harmonic excitation point and the working frequency f of the transducer w ; Step three: processing the fingerprint gray image collected by the optical sensor into a fingerprint ridge-valley image through image processing, and mapping it to the velocity variation interval of the Lamb wave target mode at the adhesion position of the fingertip epithelium and the glass material, and establishing a fingerprint velocity model according to the processed ridge-valley image; Step four: using the group velocity dispersion curve to determine the wave packet propagation velocity of the Lamb wave second harmonic, locking and extracting the time-domain signal of the target mode, and through fast Fourier transform and Gaussian filter smoothing processing, the energy flow amplitude between each two channels is extracted one by one to establish a frequency domain acoustic field matrix; Step five: using the Lamb wave second harmonic ray tomography method to locate and reconstruct the fingerprint to obtain the first fingerprint imaging result, and using the finite difference to update the acoustic field matrix to provide an initial model and a frequency domain acoustic field matrix for the subsequent second fingerprint imaging method; Step six: through steps three to five, the velocity model of the Lamb wave fundamental wave target mode, the updated initial model and the frequency domain acoustic field matrix of the off-line training data set example are obtained, and the off-line training data set of the second fingerprint imaging method is established, and the descending gradient matrix is obtained through off-line training by the supervised descent method; Step seven: through steps three to five, the velocity model of the Lamb wave fundamental wave target mode, the updated initial model and the frequency domain acoustic field matrix of the test example are obtained, and the second fingerprint imaging result is obtained by using the descending gradient matrix obtained in step six.

2. The method of claim 1, wherein the fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect is characterized by, The fingerprint wide-area collection ultrasonic transducer is composed of an n-channel transducer array (1) composed of n piezoelectric ceramic chips, the piezoelectric ceramic chips adopt PZT-5H material with high center frequency, and the electrical interconnection of each channel element is formed by positive and negative copper foils; the negative copper foil (5) is a continuous whole, which provides an acoustic field waveguide and a ground wire for each channel element; the number of positive copper foils is the same as the number of channels, and is divided into four positive copper foil bundles according to the number and is led out in four directions; The fingerprint collection system is composed of a fingerprint wide-area collection ultrasonic transducer, an optical sensor, an ultrasonic instrument and a signal transmission circuit; the fingerprint wide-area collection ultrasonic transducer, the ultrasonic instrument and the signal transmission circuit are used to collect the time-domain acoustic field signal when the fingertip presses the glass plate, the fingerprint wide-area collection ultrasonic transducer collects the signal in the working mode of single-element excitation and full-element reception; the optical sensor is used to collect the fingerprint gray image when the fingertip presses the glass plate.

3. The method of claim 1, wherein the fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect is characterized by, The change of the Lamb wave fundamental wave and second harmonic dispersion curve at the adhesion position of the fingertip epithelium and the glass material is related to the obvious acoustic field scattering (4) at the pressing position (3) of the fingertip, and the dispersion curve change is obtained by inputting the thickness, density, Young's modulus and Poisson's ratio of the glass material and the fingertip epithelium material and the height difference between the fingerprint ridge and valley, and using the global matrix method to calculate the Lamb wave fundamental wave and second harmonic dispersion curve before and after the adhesion of the fingertip epithelium and the glass material; The second harmonic excitation point and the working frequency f of the fingerprint wide-area collection ultrasonic transducer w With the same frequency value, the determination method is: through global matrix method wave structure analysis, the Lamb wave mode with the main energy flow in the normal direction is selected as the base wave mode for exciting the second harmonic target mode.

4. The method of claim 1, wherein the fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect is characterized by, The method for establishing the fingerprint speed model is: using an image processing method to adjust a high dynamic range fingerprint grayscale image collected by an optical sensor into a low dynamic range image, remove the connected area of the fingerprint detail edge, eliminate the grayscale change caused by the pressing strength, and obtain a fingerprint ridge valley image; determining the phase speed descending interval [c d2 ,c u2 ] of the target second harmonic at the second harmonic excitation point in the dispersion curve, mapping the grayscale interval of the fingerprint ridge valley grayscale image to the phase speed interval, and obtaining a second harmonic speed model; taking the fundamental mode with the shortest wavelength at the frequency f w as the target mode, determining the phase speed descending interval [c d1 ,c u1 ] thereof, mapping the grayscale interval of the fingerprint ridge valley grayscale image to the phase speed interval, and obtaining a fundamental speed model. The second harmonic velocity model is used for checking and verifying the fingerprint morphology characterization effect; the fundamental wave velocity model is used as a velocity model in subsequent second fingerprint imaging, while giving play to the advantages of nonlinear Lamb wave characteristics and taking into account linear characteristics and resolution.

5. The method of claim 1, wherein the fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect is characterized by, The method for establishing the second harmonic frequency domain sound field matrix is: at frequency f w The wave packet propagation speed of the second harmonic is determined according to the group velocity dispersion curve, the time domain signal of the target mode is locked and extracted by using a window function; the energy flow amplitude between each two channels is extracted one by one by using fast Fourier transform and Gaussian filter smoothing processing, so as to generate an n*n frequency domain sound field matrix with the dimension related to the channel number n of the fingerprint wide area acquisition ultrasonic transducer.

6. The method of claim 1, wherein the fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect is characterized by, The first fingerprint imaging method is: establishing a baseline frequency domain sound field matrix and a fingertip pressing frequency domain sound field matrix, calculating the peak drop value between each two channels, using an elliptical algorithm to calculate the signal difference coefficient distribution, and obtaining a fingerprint positioning reconstruction result; through signal difference coefficient threshold control, extracting the fingertip pressing part and mapping to the phase velocity drop interval [c d1 ,c u1 ] of the fundamental wave target mode, thereby providing an initial model that changes with the fingertip pressing state for the second fingerprint imaging method.

7. The method of claim 1, wherein the method is based on the dispersion effect of nonlinear Lamb waves. The sound field matrix updating method is: using a frequency domain finite difference method based on a linear elastic model to calculate a frequency domain sound field matrix corresponding to an initial model, and replacing a second harmonic sound field matrix in first fingerprint imaging, and then being used for data set offline training based on a supervised descent method and online inversion of a test example in second fingerprint imaging.

8. The method of claim 1, wherein the fingerprint iterative imaging based on the nonlinear Lamb wave dispersion effect is characterized by, The second fingerprint imaging method is: after obtaining a descent gradient matrix in data set offline training based on a supervised descent method, a fingerprint velocity model is established, a second harmonic frequency domain sound field matrix is established, first fingerprint imaging and sound field matrix updating are performed, a velocity model of a fundamental wave target mode of a test example, an updated initial model and a frequency domain sound field matrix are obtained, and finally online inversion results are obtained.

Images