A method for managing laser measurement and surface profile characteristic data

Abstract

The application discloses a kind of laser measurement and surface shape characteristic data management method, can avoid the frequency mirror image problem when multiple surface laser measurement occurs, and realize the high-precision detection of measured piece and the effective storage of surface shape data, utilize the spectral characteristic quantization analysis of regional pixel under different truncation function to realize surface shape high-precision detection, and the element of surface shape data containing surface shape characteristic is integrated management.The method proposed in the patent mainly includes two modules of sampling characteristic quantization based on regional frequency peak analysis and surface shape data efficient storage.Sampling characteristic quantization based on regional frequency peak analysis includes: effective interference area pixel selection and spectrum analysis matrix acquisition, interference peak analysis under different truncation function, spectral quantization analysis of sampling quality under different truncation function, regional interference frequency solving;Surface shape data efficient storage includes: surface shape data storage and preliminary processing, information matrix expansion and processing, surface shape blurring processing.

Description

Technical Field

[0001] This invention relates to a laser measurement and surface feature data management method, which can avoid the frequency mirroring problem that occurs during multi-surface laser measurement and achieve high-precision detection of the test piece and effective storage of surface feature data. The proposed method mainly includes two modules: sampling feature quantization based on regional frequency peak analysis and efficient storage of surface feature data. It is applied in the field of precision laser detection and data storage. Background Technology

[0002] Laser interferometry is an effective method for surface shape inspection of high-precision optical components. It uses a matrix to represent information (or elements) about the height of the measured surface, and the size of this matrix is ​​the same as the camera resolution of the measuring equipment. During laser interferometry (hereinafter referred to as laser measurement), environmental factors such as vibration, phase shifter errors, and noise from internal components of the measuring equipment can cause sampling errors or contamination in the acquired interferogram. Furthermore, due to the complexity of the actual measurement environment, these errors often overlap and are reflected in the sampled interferogram simultaneously. This means that contamination, sampling errors, or random fluctuations in the sampled light intensity signal (collectively referred to as measurement disturbance errors) coexist in the interferogram, which adversely affects the information of the interferometric light intensity signal.

[0003] The general process of laser measurement can be seen in the inventor's paper, "Lin Chang, et al. A general auto-shift minimal-step phase-shifting algorithm for arbitrary cavity length[J]. Optics and Laser in Engineering, 2022, 149: 106791," which can be divided into two main steps: interferogram acquisition and optical signal processing. When processing the interferometric intensity signal, it is often necessary to solve for the frequencies of the interferometric harmonics. However, the presence of measurement perturbation errors significantly reduces the accuracy of the interferometric harmonic frequency solution. Frequency determination generally relies on the Fast Fourier Transform, and to reduce the adverse effects of truncation and picket fence effects on the accuracy of frequency determination, a truncation function is needed to perform preliminary processing on the sampled intensity signal.

[0004] However, the problem is that when the measurement error in the actual detection environment is complex, the frequency spectrum (hereinafter referred to as spectrum) obtained by different truncation functions is different, and the impact of error on different pixels is different. The light intensity spectrum of some severely affected pixels cannot be used to solve for effective frequency information. At the same time, in the selection of truncation function, ideally, a narrow main lobe, a low side lobe peak, and fast harmonic attenuation should be selected. However, these three are contradictory. When the sampling length (truncation function length) remains unchanged, the truncation function cannot take into account the above performance. Although the inventors studied the influence of the selection of truncation function based on characteristic polynomial on laser measurement in the paper "Lin Chang, et al. Multi-surface phase-shifting algorithm using the window function fitted by the nonlinear least squares method[J]JOURNAL OF MODERNOPTICS, 2022, 69(3): 160-171", they still failed to solve the problem of selecting the optimal (or most effective) truncation function under different error conditions. Therefore, how to make the optimal selection of truncation function under actual measurement conditions is also one of the problems currently faced in the field.

[0005] Especially when the component under test is an optical plate with a high degree of parallelism between its front and rear surfaces, the interference harmonics corresponding to each surface under test will all be displayed in the acquired interferogram, making optical signal processing particularly complex. When measuring optical plate components, as the sampling frequency increases, a frequency peak mirroring effect occurs, meaning that the detected frequency peak is inaccurate, and the true frequency peak is mirrored to the other side with half the sampling frequency as the axis of symmetry, thus making it impossible to obtain accurately.

[0006] Some high-end precision parts have extremely high machining accuracy. Determining this accuracy and quantitatively evaluating the performance of these parts requires a quantitative assessment of the surface height distribution. As the foundation for this quantitative assessment and parts storage management, the storage, allocation, and management of the quantitative characteristics of these parts, as well as the characteristic data of their surface height distribution (stored in matrix form), including: the root mean square (RMS) value of the height distribution, the peak-to-valley distance (PV) between the highest and lowest points, and the peak-to-valley distance (PVQ) after removing 5% of the highest and lowest points, are crucial.

[0007] Especially for products such as optical components used in aerospace, military industry, and key scientific devices (such as inertial confinement fusion devices), the efficient management and confidentiality of surface shape and feature data of these parts are essential due to the special nature of their applications. The following further explains the fields and problems addressed by this invention.

[0008] The key characteristics of high-precision components, such as optical components, are: high cost and specific application scenarios; the surface finish quality of the components directly determines the product quality; and the surface shape data of the components is related to the core performance of high-end devices. After measuring the surface shape of components using various testing methods, matrix arrays are typically used to store the surface shape data. However, traditional methods only store the surface shape data, neglecting important characteristic data such as RMS, PV, and PVQ values, which are crucial for classifying and quantifying component processing quality. Furthermore, the simple storage of surface shape data easily leads to data leakage and increases the difficulty of digital management, which is highly detrimental to the effective and safe management of high-precision products.

[0009] Therefore, the current technical problems in the field of laser detection can be summarized as follows: (1) How to achieve high-precision harmonic frequency calculation and surface shape detection, and avoid the measurement inaccuracy caused by frequency mirroring problem; (2) How to achieve efficient and encrypted storage of surface shape data to prevent surface shape data leakage.

[0010] To overcome the aforementioned problems and avoid the frequency mirroring issue that occurs during multi-surface laser measurement, and to achieve high-precision detection of the workpiece and efficient storage of surface shape data, this patent utilizes spectral feature quantization analysis of regional pixels under different truncation functions to achieve high-precision surface shape detection, and integrates and manages elements containing surface shape features in the surface shape data. The laser measurement and surface shape feature data management method proposed in this patent mainly includes two modules: sampling feature quantization based on regional frequency peak analysis and efficient storage of surface shape data. Among them, the sampling feature quantization based on regional frequency peak analysis includes: selection of effective interference region pixels and acquisition of spectrum analysis matrix, interference peak analysis under different truncation functions, spectrum quantization analysis of sampling quality under different truncation functions, and solving for regional interference frequency; efficient storage of surface data includes: surface data storage and preliminary processing, information matrix expansion and processing, and surface blurring processing. Summary of the Invention

[0011] First, let's explain the terms involved: Optical inspection refers to the detection of the height distribution of the surface shape of a test part through optical methods. For example, an image with a certain resolution (such as 1000 pixels × 1000 pixels) is acquired by optical interferometry, and then the height of each pixel is reconstructed by the corresponding solution algorithm, thereby obtaining a matrix containing the surface shape and height of the test part (its size is also 1000 pixels × 1000 pixels). RMS, PV, and PVQ are commonly used parameters in optical inspection. For clarity, the definitions of these parameters are explained here, representing: the root mean square (RMS) value of the surface height distribution, the peak-valley distance between the highest and lowest points (PV) value, and the peak-valley distance after removing 5% of the highest points and the peak-valley distance after removing 5% of the lowest points (PVQ) value (conventional method). In addition, in this invention, the corresponding peak-valley distances for removal amounts ranging from 1% to 5% are limited. The minimum and maximum data involved in face shape storage are, respectively, the data of the lowest point in the face shape height represented by the elements in the face shape matrix and the data of the highest point in the face shape height. Measurement error: When using inspection equipment to perform optical inspection on the surface of a part, each inspection may be affected by different inspection environments (such as the movement of operators or vibration of the inspection environment). Since optical inspection technology is sensitive to external disturbances, the surface inspection results obtained after each measurement may contain different occasional measurement errors. Truncation function: When performing Fourier transform, since only sequences of finite length can be analyzed, when affected by measurement disturbances, the analyzed spectral amplitude spectrum (i.e., the spectrum) will exhibit spectral leakage. This phenomenon is manifested in the spectrum as the spectral curve that should be concentrated at a certain peak position will be scattered at the surrounding frequency positions, thus showing several "small frequency peaks", and the height of the original frequency peaks will be affected and become lower.

[0012] To avoid the frequency mirroring problem that occurs during multi-surface laser measurement and to achieve high-precision detection of the measured part and efficient storage of surface shape data, this patent utilizes spectral feature quantization analysis of regional pixels under different truncation functions to achieve high-precision surface shape detection, and integrates and manages elements containing surface shape features in the surface shape data. The laser measurement and surface shape feature data management method proposed in this patent mainly includes two modules: sampling feature quantization based on regional frequency peak analysis and efficient storage of surface shape data. Among them, the sampling feature quantization based on regional frequency peak analysis includes: selection of effective interference region pixels and acquisition of spectrum analysis matrix, interference peak analysis under different truncation functions, spectrum quantization analysis of sampling quality under different truncation functions, and solving for regional interference frequency; efficient storage of surface data includes: surface data storage and preliminary processing, information matrix expansion and processing, and surface blurring processing.

[0013] Preferably, the effective interference region pixel selection and spectrum analysis matrix acquisition in the sampling feature quantization based on regional frequency peak analysis are as follows: Select an analysis region IZ in the interferogram. The size of IZ should not be less than 100×100 pixels and should be located in the effective interference region. Select all pixels point by point within IZ and perform Fast Fourier Transform on the interference light intensity of the selected pixels at different sampling frame numbers. Take the right half of the obtained spectrum result and select it as the spectrum analysis object. Store the spectrum one by one in the preliminary spectrum analysis matrix. In the preliminary spectrum analysis matrix, the horizontal coordinate is the frequency point number and the vertical coordinate is the number of the selected pixel. After obtaining the preliminary spectrum analysis matrix, since the horizontal coordinate is the frequency point number and cannot represent the frequency component information, the horizontal coordinate is transformed. With n as the horizontal coordinate symbol, K as the total number of sampling frames, and N as the phase shift interval of the signal with the minimum optical path difference in the interference harmonics corresponding to each surface, the horizontal frequency coordinate after the coordinate transformation is (nN) / (2π 2 Let F0 be the relative frequency abscissa, and store F0 separately in a vector. Then, plot the spectrum with F0 as the abscissa and the spectrum analysis object after amplitude normalization as the ordinate, and perform spectrum cutting to obtain the spectrum cutting matrix MF. The spectrum cutting is as follows: extract the spectrum with the relative frequency abscissa between (1+M) / 2 and N / 2, where M is the ratio between the distance of the test object in front of the interferometer reference mirror and the optical thickness, and the spectrum corresponding to the endpoint of the range (1+M) / 2 to N / 2 must be included when extracting the relative frequency abscissa. Preferably, the interference peak analysis under different truncation functions in the sampling feature quantization based on regional peak analysis specifically involves: multiplying the spectrum of each row in the spectrum cutting matrix MF by the selected set of truncation functions, denoted as the truncated spectrum cutting matrix MF_s. The subscript s of MF_s corresponds to the selected set of truncation functions, including: Hanning truncation function, Hamming truncation function, and Blackman truncation function. Then, MF_s is analyzed: first, the number of main peaks is analyzed. If, for the Hanning truncation function, Hamming truncation function, and Blackman truncation function, the spectrum of each row in the obtained MF_s satisfies the following condition: there are two frequency peaks with the largest amplitudes... If the difference in the horizontal coordinates between the two frequency peaks with the largest amplitudes is between F0=0.8 and 1.2, then the pixel corresponding to the spectrum in that row satisfies the three-peak sampling condition. If the number of rows satisfying the three-peak sampling condition in all rows of MF_s is not less than 75% of the total number of rows, then the sampling is considered valid, and each has its own independent peak. The spectrum quantization score is recorded as 1. Otherwise, it is recorded as 0. If the spectrum quantization score is 1, the following calculation and analysis will continue. If the spectrum quantization score is 0, the sampling frequency will be increased to acquire the interferogram, and the effective interference region pixel selection and spectrum analysis matrix acquisition, and the interference peak analysis under different truncation functions will be performed again until the spectrum quantization score is 1. Preferably, the spectral quantization analysis of sampling quality under different cutoff functions in the sampling feature quantization based on regional peak analysis is specifically performed as follows: the spectrum is quantized into MF_s of 1 and the following calculations are performed: the amplitudes of the two main peaks in MF_s corresponding to the Hanning cutoff function, Hamming cutoff function, and Blackman cutoff function are calculated respectively. The amplitudes of the two peaks from left to right in the spectrum corresponding to each row in each MF_s are recorded as DLP1 and DLP2 respectively. If DLP1 is greater than DLP2 for the spectrum of that row, the peak height quantization score PGD is recorded as 1; otherwise, it is recorded as 0. If more than 60% of the spectrum in different rows of each MF_s has PGD=1, the current sampling is determined to be normal sampling and subsequent calculations can continue. Otherwise, the interferogram is determined to be abnormal sampling and frequency-increased until it is determined to be normal sampling. Preferably, the solution of regional interference frequencies in the sampling feature quantization based on regional peak analysis is specifically as follows: The F0 corresponding to the first and second peak values ​​in the MF_s of the three truncation functions (Hanning, Hamming, and Blackman) determined to be normal sampling is searched row by row. The average value of F0 at the positions of the first and second peak values ​​in each row of the current truncation function is calculated and denoted as F1_s and F2_s. Then, the average results of the first and second peak values ​​corresponding to the three truncation functions are calculated as harmonic frequencies F1 and F2. Based on these, initial phase demodulation, unwrapping, de-tilting, and surface reconstruction are performed. The obtained front and rear surface shape results are stored in the surface shape data storage matrix and denoted as SP1 and SP2, respectively.

[0014] Preferably, the efficient storage and preliminary processing of surface data in the aforementioned surface data storage specifically includes: taking surface data storage matrices SP1 and SP2 as analysis objects, and denoting the current surface data storage matrix as SP, with the size of the SP matrix denoted as XM×YM; calculating the root mean square (RMS) and peak-to-valley (PV) values ​​of SP; calculating the maximum value of SP matrix elements after removing the maximum value of elements in the matrix with an m% ratio, and the minimum value of SP matrix elements after removing the minimum value of elements in the matrix with an m% ratio, and then subtracting the two values ​​and denoting them as PVQ_m, where the variable m in m% and PVQ_m is selected as five integers from 1 to 5; if PVQ_m satisfies (PVQ_m-PV) / PV>3, then the current surface quality is determined to be poor; if PVQ_m satisfies 3≥(PVQ_m-PV) / PV>1.5, then the current surface quality is determined to be slightly poor; if PVQ_m satisfies 1.5≥(PVQ_m-PV) / PV>1.5, then the current surface quality is determined to be slightly poor. If Q_m - PV) / PV > 0.5, the current surface quality is judged as good; if both PV and Q_m satisfy 0.5 ≥ (PVQ_m - PV) / PV, the current surface quality is judged as excellent. Furthermore, the surface quality is quantified and its feature information encrypted using the quantitative index SQ: the absolute value of the difference between the surface RMS value and the PV value is used as the quantification standard. The specific quantification method is as follows: if the current surface quality is poor, then SQ = |RMS - PV| / K1 (|| represents taking the absolute value). If the face quality is slightly poor, then SQ = |RMS-PV| / K2. If the face quality is good, then SQ = |RMS-PV| / K3. If the face quality is excellent, then SQ = |RMS-PV| / K4. The calculation result of SQ is rounded to two decimal places. The values ​​of K1 to K4 are set according to the user's encryption requirements. It is required that the SQ calculation results corresponding to K1 to K4 remain distinct from each other after being rounded to two decimal places, and the values ​​of K1 to K4 are generally set to integers.

[0015] Preferably, the information matrix expansion and processing in the efficient storage of surface data specifically includes: while keeping the original matrix elements and their positions unchanged in the surface data storage matrix SP, adding a row with all elements being zero in SP, referred to as the added row, extending the matrix size to (XM+1)×YM, and then arranging the following surface parameters from left to right in the added row: PV, RMS, PVQ5, PVQ3, PVQ2, PVQ1, SQ, RS, EX, EXG, MH. The surface data storage matrix after the added row is denoted as SFMA. The parameters in the added row have the following meanings: RS is a parameter representing the distribution form of the surface shape of the high-precision component, EX is the outlier removal operator, and MH is the fuzzification processing operand.

[0016] Preferably, the square face refers to a face shape height data in each element of the face shape data storage matrix before adding rows, denoted as RS=R; the circular face refers to a face shape height data in which not every element of the face shape data storage matrix is ​​a face shape height data in each element before adding rows. In the face shape data storage matrix of the circular face, the face shape height data is centered at the matrix's center point (XM / 2, YM / 2), the face data distribution conforms to the characteristics of a circle, and the radius of the circle is less than XM / 2, denoted as RS=S. Before adding rows, the face shape data storage matrix is ​​a square matrix with equal rows and columns. Storing circular faces in a square matrix cannot fill all matrix elements. To distinguish between valid and invalid face shape regions, the element values ​​of the regions outside the circle in the circular face shape data matrix are all set to EM, and EM should be different from any element value in the face shape data to distinguish between valid and invalid face shape regions.

[0017] Preferably, EX is the outlier removal operator. If EX=1, the outlier removal operation is performed; otherwise, it is not performed, and EX=0 is recorded. EXG is the outlier detection parameter during the outlier removal operation. During the outlier removal operation (when EX=1), elements in the effective area of ​​the face matrix that are greater than the face RMS value EXG times are identified as outliers. The element identified as an outlier is replaced with the average value of the effective face area elements of the four elements surrounding it (top, bottom, left, and right). If there are three or two effective face area elements surrounding the outlier, the outlier is replaced with the average value of the three or two adjacent effective face area elements. If there is only one effective face area element surrounding the outlier, the outlier removal operation is not performed even if EX=1. Preferably, the surface fuzzification processing in the efficient storage of surface data specifically includes: during fuzzification processing, there are two selectable methods. The first method is to swap the elements of even-numbered rows in SP if MH=2, and the second method is to swap the elements of odd-numbered rows in SP if MH=1. For both swapping methods, elements that have already participated in the swapping will no longer participate in the swapping.

[0018] Compared with the prior art, the present invention has the following obvious and prominent substantive features and significant advantages: (1) By selecting effective interference region pixels and obtaining the spectrum analysis matrix, the spectrum corresponding to the interference light intensity of a certain region pixel can be accurately obtained, thus avoiding the influence of measurement error when performing spectrum analysis on the light intensity of a certain pixel; (2) By quantitatively analyzing the peak distribution characteristics of the pixel points in a certain region under different truncation functions, and performing upsampling operations accordingly, the frequency mirroring problem can be avoided, and the harmonic spectrum can be solved with high precision and the surface shape can be accurately reconstructed. (3) By adding the calculation and display of each parameter in the row of the surface matrix, the efficiency of component data management is improved, and multi-dimensional information interpretation can be performed; (4) It realizes the digital storage of high-precision product surface shape data and feature data, providing a solid data foundation for quantitative evaluation and product classification; (5) Through a variety of optional fuzzing methods, the confidential storage of high-precision product data is achieved, which can effectively prevent the leakage of important product information. Attached Figure Description

[0019] The present invention will be further described below with reference to the accompanying drawings and examples. Figure 1 This is a schematic diagram of frequency mirroring; Figure 2 A schematic diagram showing the effective and ineffective interference regions; Figure 3 This is a schematic diagram of coordinate transformation; Figure 4 A schematic diagram of the spectrum matrix for which a cutoff function is applied; Figure 5 This is a schematic diagram of undersampling; Figure 6 A schematic diagram of MF_s; Figure 7 This is a schematic diagram for harmonic spectrum calculation; Figure 8 This is a three-dimensional schematic diagram of the surface shape distribution of the component under test; Figure 9 Schematic diagrams showing two forms of the tested component: square and circular. Figure 10 This is a diagram illustrating outlier removal. Figure 11 This is a schematic diagram of the blurring process; Figure 12 This is a flowchart illustrating the implementation of the present invention. Detailed Implementation

[0020] To avoid the frequency mirroring problem that occurs during multi-surface laser measurement and to achieve high-precision detection of the measured part and efficient storage of surface shape data, this patent utilizes spectral feature quantization analysis of regional pixels under different truncation functions to achieve high-precision surface shape detection, and integrates and manages elements containing surface shape features in the surface shape data. The laser measurement and surface shape feature data management method proposed in this patent mainly includes two modules: sampling feature quantization based on regional frequency peak analysis and efficient storage of surface shape data. The sampling feature quantization based on regional peak analysis includes: selection of effective interference region pixels and acquisition of the spectrum analysis matrix, interference peak analysis under different truncation functions, spectrum quantization analysis of sampling quality under different truncation functions, and solving for regional interference frequencies; efficient storage of surface data includes: surface data storage and preliminary processing, information matrix expansion and processing, and surface blurring processing. Preferred embodiments of the present invention are described in detail below with reference to the accompanying drawings: Example 1: The frequency peak mirroring effect when measuring optical flat panel elements is shown in the attached figure. Figure 1 As shown, the original spectrum cannot display effective frequency information at this time. This problem can be avoided by increasing the sampling frequency and re-sampling the interferogram.

[0021] The specific steps for selecting effective interference region pixels and obtaining the spectrum analysis matrix in the sampling feature quantization based on regional peak analysis are as follows: Select an analysis region IZ in the interferogram. The size of IZ should not be less than 100×100 pixels and should be located in the effective interference region (as shown in the attached figure). Figure 2 As shown in the figure, all pixels within the IZ region are selected point by point, and the interference light intensity of the selected pixels at different sampling frame numbers is directly subjected to Fast Fourier Transform. The right half of the obtained spectrum result is then selected as the spectrum analysis object, and the spectrum is stored one by one in the preliminary spectrum analysis matrix. In the preliminary spectrum analysis matrix, the horizontal axis is the frequency point index, and the vertical axis is the index of the selected pixel. After obtaining the preliminary spectrum analysis matrix, since the horizontal axis is only the frequency point index and cannot represent the frequency component information, a coordinate transformation is performed on the horizontal axis, with n as the horizontal axis symbol (which can be seen from the attached figure). Figure 3 It can be seen that n is actually the frequency index of the clipped spectrum matrix, K is the total number of sampling frames, and N is the phase shift interval of the signal with the minimum optical path difference in the interference harmonics corresponding to each surface. Then, the transverse frequency coordinate after coordinate transformation is (nN) / (2π). 2Let F0 be the relative frequency abscissa, and store F0 separately in a vector. Then, plot the spectrum with F0 as the abscissa and the spectrum analysis object after amplitude normalization as the ordinate, and perform spectrum cutting to obtain the spectrum cutting matrix MF. The spectrum cutting is as follows: extract the spectrum with the relative frequency abscissa between (1+M) / 2 and N / 2, where M is the ratio between the distance of the test object in front of the interferometer reference mirror and the optical thickness, and the spectrum corresponding to the endpoint of the range (1+M) / 2 to N / 2 must be included when extracting the relative frequency abscissa. To further explain the above process: the purpose of selecting the effective interference region is to only select the part where interference occurs, as shown in the attached figure. Figure 2 As shown, apart from the interference region, the surface shape information of the test piece cannot be solved in the invalid interference region of the entire image. Directly performing a fast Fourier transform on the interference light intensity is equivalent to applying a rectangular truncation function to the signal; after coordinate transformation, the horizontal coordinate of the spectrum changes from 1 to K / 2 (since half of the spectrum is selected, the number of effective frequency points in this spectrum is half of the ordinal number of the interferogram) to 0~N. The sampling interval N is also the sampling frequency in interferometry. N can determine the phase shift value of the harmonic signal corresponding to the test piece with the minimum interference cavity length as N / (2π). The result after coordinate transformation is denoted as the relative frequency horizontal coordinate F0. Using it as the horizontal axis and the normalized frequency point height as the vertical axis, a relative frequency spectrum diagram can be obtained. This diagram can more intuitively show the relative relationship between each harmonic frequency. The purpose of extracting the spectrum cutting matrix MF is: since one of the conditions for determining frequency mirroring is that the amplitude height of the second peak in the spectrum is lower than that of the third peak, it is only necessary to analyze the height difference between these two peaks. The first peak is not necessary for determining frequency mirroring. The normalization process described above refers to dividing each element of a sequence by the value of the largest element in the sequence, thus ensuring that all element values ​​are between 0 and 1. This process is more conducive to analyzing the magnitude relationship of sequence element values. A spectrum is plotted with the frequency index on the horizontal axis and the normalized amplitude value on the vertical axis, as shown in the attached figure. Figure 3 The left figure shows the results before and after the coordinate transformation, as shown in the attached figure. Figure 3 As shown in the figure on the right.

[0022] The interferometric peak analysis under different truncation functions in the sampling feature quantization based on regional peak analysis is specifically as follows: the spectrum of each row in the spectrum cutting matrix MF is multiplied by the selected set of truncation functions, denoted as the truncated spectrum cutting matrix MF_s. The subscript s of MF_s corresponds to the selected set of truncation functions, including: Hanning truncation function, Hamming truncation function, and Blackman truncation function (as shown in the attached figure). Figure 4 As shown, where Figure 4 The left figure is a schematic diagram of the MF_Hanning cutoff window under a certain measurement condition. Figure 4The right figure shows the spectrum of a row in the MF_Hanning truncation window matrix. Then, the MF_s is analyzed: First, the number of main peaks is analyzed. For the Hanning truncation function, Hamming truncation function, and Blackman truncation function, if the spectrum of each row in the obtained MF_s satisfies the following: there are two frequency peaks with the largest amplitude and the difference between the horizontal coordinates of these two frequency peaks with the largest amplitude is between F0=0.8 and 1.2, then the pixel corresponding to the spectrum of that row satisfies three-peak sampling. And the number of three-peak sampling in all rows of MF_s is not less than 75% of the total number of rows. Then, the sampling is considered to be valid, and there are independent peaks. The spectrum quantization is recorded as 1. Otherwise, it is recorded as 0. If the spectrum quantization is 1, the following calculation and analysis are continued. If the spectrum quantization is 0, the sampling frequency is increased to acquire the interferogram, and the effective interference region pixel selection and spectrum analysis matrix acquisition, and the interference peak analysis under different truncation functions are repeated until the spectrum quantization is 1. To further explain the above process: In the sampled spectrum, due to the influence of measurement error, each spectrum contains peaks of various sizes. However, the three peaks with the largest amplitudes correspond to the thickness variation signal of the measured optical plate (generated from the self-interference of the front and rear surfaces of the measured part), the front surface signal, and the rear surface signal, respectively. When the above characteristics are not met, a certain harmonic is undersampled, as shown in the attached figure. Figure 5 As shown, the harmonics of the rear surface cannot be reconstructed at this time. Furthermore, in the above analysis, the selected spectrum only includes the frequency peaks of the front and rear surfaces (thickness change signals generally exist independently during measurement and are not easily affected, therefore they are not extracted). The existence of these two frequency peaks with the largest amplitudes is a preliminary determination that the sampling is valid. The difference in the horizontal axis is considered because, under the influence of errors, spectral leakage may cause what was originally an independent peak to become two peaks, which is detrimental to high-precision measurement. Therefore, sampling is considered valid only when each peak is independently distributed and has a certain difference between them (the specified frequency difference is to separate the frequencies without causing inter-spectral interference). Setting the spectral quantization score is a necessary operation for convenient implementation on hardware devices, providing a direct data reference for determining whether to perform frequency upsampling. As mentioned above, the subscript 's' of MF_s corresponds to the selected truncation function group, including: Hanning truncation function (e.g., MF_Hanning truncation function), Hamming truncation function, and Blackman truncation function, as shown in the attached figure. Figure 6 The results shown are for each MF_s in a set of measurements. It can be seen that the spectrum in each set of results is different.

[0023] The spectral quantization analysis of sampling quality under different cutoff functions in the sampling feature quantization based on regional peak analysis is specifically as follows: The spectrum is quantized into MF_s of 1 and the following calculations are performed: The amplitudes of the two main peaks in MF_s corresponding to the Hanning cutoff function, Hamming cutoff function, and Blackman cutoff function are calculated respectively. The amplitudes of the two peaks from left to right in the spectrum of each row in each MF_s are recorded as DLP1 and DLP2 respectively. If DLP1 is greater than DLP2 for the spectrum of that row, the peak height quantization division PGD is recorded as 1; otherwise, it is recorded as 0. If more than 60% of the spectrum in different rows (i.e., different pixels) of each MF_s has PGD=1, the current sampling is determined to be normal sampling and subsequent calculations can continue. Otherwise, the interferogram is determined to be abnormal sampling and frequency-incremented until it is determined to be normal sampling. Further explanation of the above process: Due to the complexity of various error factors in the actual measurement environment, different cutoff functions have different comprehensive effects on the spectrum (for example, the Hanning cutoff function may be more robust to linear errors but less robust to nonlinear errors, but the proportion of linear and nonlinear errors in the actual measurement environment is unknown). In the above operation, the purpose of multiplying the spectrum of each row in the spectrum cutting matrix MF with the selected cutoff function group is to observe the quality of the spectrum under the action of different cutoff functions, and thus optimize the cutoff function through the quantification of spectrum quality. The purpose of determining that the current sampling is normal sampling when PGD=1 exists in more than 60% of the spectrum is that some pixels are affected by measurement errors and disturbances, and their peak heights may not necessarily show the characteristic that the second peak is greater than the third peak. However, not all pixels will produce this effect. Therefore, judging by the peak height characteristics of the spectrum of most pixels in a region is an effective method. At the same time, the interferogram determined to be abnormal sampling can be re-sampled until the peak height distribution characteristics meet the measurement requirements. Otherwise, the spectrum mirroring problem when there are complex measurement errors is difficult to solve. The peak height quantization (PGD) setting is to provide a data reference basis for implementation on hardware devices.

[0024] The specific process for solving the regional interference frequency in the sampling feature quantization based on regional peak analysis is as follows: The F0 values ​​corresponding to the first and second peak values ​​in the MF_s of the three truncation functions (Hanning, Hamming, and Blackman, which are determined to be normal sampling) are searched row by row. The average value of F0 at the positions of the first and second peak values ​​in each row of the current truncation function is calculated and denoted as F1_s and F2_s. Then, the average results of the first and second peak values ​​corresponding to the above three truncation functions are calculated as harmonic frequencies F1 and F2 (as shown in the attached figure). Figure 7As shown in the figure, initial phase demodulation, unwrapping, de-tilting and surface reconstruction are performed accordingly. The obtained front and back surface surface shapes are stored in the surface shape data storage matrix and are denoted as SP1 and SP2, respectively.

[0025] Further explanation of the above process: Due to the influence of measurement errors, the harmonic frequencies obtained from the spectrum corresponding to each truncation function may be different. Furthermore, since the measurement errors experienced by each pixel are also different, they may exhibit different characteristics in the spectrum. Therefore, the method used here, which averages the spectral peaks corresponding to different truncation functions and different pixels, can better avoid the problem of inaccurate harmonic frequency calculation caused by certain independent factors. The specific calculation process of F1 and F2 is as follows: First, the first and second peak values ​​of each row in MF_s corresponding to a certain truncation function are calculated and averaged. Then, the above calculation is performed for each truncation function, and the two peak values ​​corresponding to different truncation functions are averaged again. Meanwhile, the processing method proposed in this patent mainly targets spectrum processing and does not limit the further processing methods after obtaining the harmonic frequencies. The reason is that the most important step in multi-surface laser interferometry is to locate the harmonic frequency characteristics of each component signal. Phase demodulation, unwrapping, de-tilting, and wavefront reconstruction (based on the linear relationship between the initial phase and the wavefront height) after obtaining the harmonic frequencies are not the core issues addressed in this patent. As an example, phase demodulation can be performed using the least squares fitting method proposed in the paper "Chang Lin, Yan Ketao, et al. Multi-surface interferometry of transparent test objects considering secondary reflection fringes [J]. Optics and Precision Engineering, 2020, 28 (12): 2605-2613".

[0026] Example 2: After obtaining accurate surface shape data through the sampling feature quantization module based on regional peak analysis, efficient surface shape data storage is performed. This mainly includes three steps: surface shape data storage and preliminary processing, information matrix expansion and processing, and surface fuzzification processing. Specifically: The efficient storage and preliminary processing of surface data in the aforementioned surface data includes: taking surface data storage matrices SP1 and SP2 as analysis objects respectively, and denoting the current surface data storage matrix as the analysis object as SP, with the size of the SP matrix denoted as XM×YM (as shown in the attached figure). Figure 8The diagram shows the surface distribution of an image with a resolution of 600×600. The root mean square (RMS) and peak-to-valley (PV) values ​​of the SP are calculated. The maximum value of the SP matrix element after removing the maximum value of elements in the matrix representing m% of the SP, and the minimum value of the SP matrix element after removing the minimum value of elements in the matrix representing m% of the SP, are calculated. The difference between these two values ​​is denoted as PVQ_m. The variable m in m% and PVQ_m is selected from five integers from 1 to 5. If PVQ_m satisfies (PVQ_m-PV) / PV>3, the current surface quality is considered poor. If PVQ_m satisfies 3≥(PVQ_m-PV) / PV>1.5, the current surface quality is considered slightly poor. If PVQ_m satisfies 1.5≥(PVQ_m-PV) / PV>0.5, the current surface quality is considered good. If PVQ_m satisfies 0.5, the current surface quality is considered good. If 5 ≥ (PVQ_m - PV) / PV, the current surface quality is determined to be excellent. Furthermore, the surface quality is quantified and its feature information encrypted using the quantitative index SQ. The absolute value of the difference between the surface RMS value and the PV value is used as the quantification standard. The specific quantification method is as follows: If the current surface quality is poor, then SQ = |RMS - PV| / K1 (|| represents taking the absolute value); if the current surface quality is slightly poor, then SQ = |RMS - PV| / K2; if the current surface quality is good, then SQ = |RMS - PV| / K3; if the current surface quality is excellent, then SQ = |RMS - PV| / K4. The calculation result of SQ is retained to two decimal places. The values ​​of K1 to K4 are set according to the user's encryption requirements. It is required that the SQ calculation results corresponding to K1 to K4 remain distinct after retaining two decimal places, and the values ​​of K1 to K4 are generally set to integers.

[0027] To further explain the above process: Removing m% and then calculating the peak-valley values ​​is an optional processing method in the field. Its purpose is that when some pixels are affected by measurement disturbances or dust in the instrument, the surface height of that point is unreliable. Therefore, removing a certain percentage of the maximum or minimum values ​​can more fully observe the data distribution characteristics. Specifically: the surface data peak-valley values ​​after removing 1% of the data with the largest absolute value and 1% of the data with the smallest absolute value (denoted as PVQ_1), the surface data peak-valley values ​​after removing 2% of the data with the largest absolute value and 2% of the data with the smallest absolute value (denoted as PVQ_2), the surface data peak-valley values ​​after removing 3% of the data with the largest absolute value and 3% of the data with the smallest absolute value (denoted as PVQ_3), and the surface data peak-valley values ​​after removing 5% of the data with the largest absolute value and 5% of the data with the smallest absolute value (denoted as PVQ_5). For example, if the matrix is ​​[-10 1 1 1; 2 2 2 2; 5 5 5 10], with 12 elements, after removing the maximum value (10) and the minimum value (-10) by 1 / 12, the peak-to-valley value in the matrix is ​​4 (because 5-1=4). It can be seen that -10 and 10 are clearly abnormal values ​​compared to other matrix elements. The purpose of setting K1 to K4 is to prevent the direct leakage of surface shape data. In other words, if the RMS or PV data of the surface shape is stored directly, once this data is leaked, the core parameter information that directly reflects the manufacturing performance of the tested parts will be leaked. The settings of K1 to K4 and SQ can initially encrypt the surface shape feature information that reflects processing performance and processing capability.

[0028] The information matrix expansion and processing in the efficient storage of surface data specifically includes: while keeping the original matrix elements and their positions unchanged in the surface data storage matrix SP, adding a row with all elements being zero (hereinafter referred to as the added row) to SP, extending the matrix size to (XM+1)×YM, and then arranging the following surface parameters from left to right in the added row: PV, RMS, PVQ5, PVQ3, PVQ2, PVQ1, SQ, RS, EX, EXG, MH. The surface data storage matrix after the added row is denoted as SFMA. The parameters in the added row have the following meanings: RS is a parameter representing the distribution form of the surface shape of the high-precision component, EX is the outlier removal operator, and MH is the fuzzification processing operand.

[0029] Further explanation of the above process: In the process described above, "while maintaining the original matrix elements and their positions in the surface data storage matrix SP, a row with all zero elements is added to SP (hereinafter referred to as the added row), extending the matrix size to (XM+1)×YM." Clearly, the added row should be located above or below the original matrix. This patent does not specify a particular position here; users can choose according to their needs. RS in the surface parameters represents the distribution of the surface shape of the high-precision component, as shown in the attached diagram. Figure 9 As shown, there are two types of face shapes: circular and square. The following explanation uses the circular face shape as an example: The difference between circular and square face shapes is mainly related to the image features processed during optical inspection. Under certain conditions, the effective test area of ​​an optical inspection system is circular, but the acquired image is square. Therefore, many invalid areas exist outside the circular area. These invalid areas cannot reflect the high-precision features of the face shape (the face shape can be represented by an image with uneven color distribution, as shown in the attached figure). Figure 9 As shown, the distribution of height values ​​in the entire image can be seen by using the value corresponding to a reference "color bar". This has no impact on the measurement results, therefore, the effective measurement area in the center circle needs to be cropped. This processing method will result in the image shown in the attached figure. Figure 8 The form shown on the left is called a circular face. (See attached diagram.) Figure 10 As shown, when performing outlier removal, it can effectively handle intermediate outliers based on the distribution characteristics of surrounding elements.

[0030] The term "square face" refers to a face shape data matrix where, before adding rows, every element is a face shape height. Therefore, RS = R. The term "circular face" refers to a face shape data matrix where, before adding rows, not every element is a face shape height. In a circular face shape data matrix, the face shape height is centered at the matrix's center point (XM / 2, YM / 2), and the face data distribution conforms to the characteristics of a circle (the radius of the circle is less than XM / 2). Therefore, RS = S. Before adding rows, the face shape data matrix is ​​a square matrix with equal rows and columns. Storing circular faces in a square matrix cannot fill all matrix elements. To distinguish between valid and invalid face shape regions, the element values ​​outside the circle in the circular face shape data matrix are all set to EM. Furthermore, EM should be different from any element value in the face shape data to differentiate between valid and invalid face shape regions.

[0031] Note: When RS = R or S, simply specify the letter variable in the corresponding position. For example, [1 3 0.2 .... S....].

[0032] Where EX is the outlier removal operator. If EX=1, the outlier removal operation is performed; otherwise, it is not performed, and EX=0 is recorded. EXG is the anomaly detection parameter during the outlier removal operation. During the outlier removal operation (when EX=1), elements in the valid region of the face matrix that are greater than the face RMS value EXG times are identified as outliers. The element identified as an outlier is replaced with the average value of the valid face region elements of the four elements directly above, below, to the left, and to the right of it. If there are three or two valid face region elements around the outlier, the outlier is replaced with the average value of the three or two adjacent valid face region elements. If there is only one valid face region element around the outlier, the outlier removal operation is not performed even if EX=1. The surface fuzzification process in the efficient storage of surface data specifically includes: during fuzzification, there are two selectable methods. The first method is to swap the elements of even-numbered rows in SP if MH=2, and the second method is to swap the elements of odd-numbered rows in SP if MH=1. For both swapping methods, elements that have already participated in the swapping will no longer participate in the swapping. To further explain the above process: the purpose of blurring is to further ensure that the face data is not easily leaked and interpreted. Once the face data is leaked, it cannot be directly interpreted because the positions of the face height elements have been changed. Taking the first blurring method as an example, the elements in the even-numbered rows of SP are swapped, such as swapping the 2nd and 4th elements in SP, swapping the 6th and 8th elements in SP, and so on. As shown in the attached figure... Figure 11 As shown, after blurring, the original arrangement of the face elements changes. This can be seen from the difference in face shapes before and after the swap in the figure. While the blurred face shapes can largely retain their original features, they cannot accurately reveal the face shape information. The flowchart of this patent implementation is shown in the attached figure. Figure 12 As shown.

[0033] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. The complete flowchart of the present invention is shown in the accompanying drawings. Figure 12 As shown, the present invention is not limited to the above embodiments. Various changes can be made according to the purpose of the invention. Any changes, modifications, substitutions, combinations or simplifications made in accordance with the spirit and principle of the technical solution of the present invention shall be equivalent substitutions. As long as they meet the purpose of the invention and do not deviate from the technical principle and inventive concept of the invention, they shall fall within the protection scope of the present invention.

Claims

1. A method for laser measurement and surface feature data management, characterized in that: It includes two modules: sampling feature quantization based on regional peak analysis and efficient storage of surface data. The sampling feature quantization based on regional peak analysis includes: selection of effective interference region pixels and acquisition of spectrum analysis matrix, interference peak analysis under different truncation functions, spectrum quantization analysis of sampling quality under different truncation functions, and solving for regional interference frequency; efficient storage of surface data includes: surface data storage and preliminary processing, information matrix expansion and processing, and surface blurring processing; The specific steps for selecting effective interference region pixels and obtaining the spectrum analysis matrix in the sampling feature quantization based on regional frequency peak analysis are as follows: Select an analysis region IZ in the interferogram. The size of IZ should not be less than 100×100 pixels and should be located in the effective interference region. Select all pixels point by point from IZ and perform Fast Fourier Transform on the interference light intensity of the selected pixels at different sampling frame numbers. Take the right half of the obtained spectrum result and select it as the spectrum analysis object. Store the spectrum one by one in the preliminary spectrum analysis matrix. In the preliminary spectrum analysis matrix, the horizontal coordinate is the frequency point number and the vertical coordinate is the number of the selected pixel. After obtaining the preliminary spectrum analysis matrix, since the horizontal coordinate is the frequency point number and cannot represent the frequency component information, the horizontal coordinate is transformed. Let n be the horizontal coordinate symbol, K be the total number of sampling frames, and N be the phase shift interval of the signal with the minimum optical path difference in the interference harmonics corresponding to each surface. Then, the horizontal frequency coordinate after the coordinate transformation is (nN) / (2π 2 Let F0 be the relative frequency abscissa, and store F0 separately in a vector. Then, plot the spectrum with F0 as the abscissa and the spectrum analysis object after amplitude normalization as the ordinate, and perform spectrum cutting to obtain the spectrum cutting matrix MF. The spectrum cutting is as follows: extract the spectrum with the relative frequency abscissa between (1+M) / 2 and N / 2, where M is the ratio between the distance of the test object in front of the interferometer reference mirror and the optical thickness, and the spectrum corresponding to the endpoint of the range (1+M) / 2 to N / 2 must be included when extracting the relative frequency abscissa.

2. The laser measurement and surface feature data management method according to claim 1, characterized in that: The interferometric peak analysis under different truncation functions in the sampling feature quantization based on regional peak analysis includes: multiplying the spectrum of each row in the spectrum cutting matrix MF by the selected set of truncation functions, denoted as the truncated spectrum cutting matrix MF_s. The subscript s of MF_s corresponds to the selected set of truncation functions, including: Hanning truncation function, Hamming truncation function, and Blackman truncation function. Then, MF_s is analyzed: first, the number of main peaks is analyzed. If, for the Hanning truncation function, Hamming truncation function, and Blackman truncation function, the spectrum of each row in the obtained MF_s satisfies: there are two frequency peaks with the largest amplitudes and these two... If the difference in the horizontal coordinates between the frequency peaks with the largest amplitude is between F0 = 0.8 and 1.2, then the corresponding pixel in that row of spectrum satisfies the three-peak sampling condition. If the number of rows satisfying the three-peak sampling condition in all rows of MF_s is not less than 75% of the total number of rows, then the sampling is considered valid, and each has its own independent peak. The spectrum quantization score is recorded as 1. Otherwise, it is recorded as 0. If the spectrum quantization score is 1, the following calculation and analysis are continued. If the spectrum quantization score is 0, the sampling frequency is increased to acquire the interferogram, and the effective interference region pixel selection and spectrum analysis matrix acquisition, and the interference peak analysis under different truncation functions are repeated until the spectrum quantization score is 1.

3. The laser measurement and surface feature data management method according to claim 2, characterized in that: The spectral quantization analysis of sampling quality under different cutoff functions in the sampling feature quantization based on regional peak analysis includes: dividing the spectrum into MF_s of 1 and performing the following calculations: calculating the amplitudes of the two main peaks in MF_s corresponding to the Hanning cutoff function, Hamming cutoff function, and Blackman cutoff function respectively. The amplitudes of the two peaks from left to right in each row of each MF_s are denoted as DLP1 and DLP2 respectively. If DLP1 is greater than DLP2 for the spectrum of that row, the peak height quantization score PGD is recorded as 1; otherwise, it is recorded as 0. If more than 60% of the spectrum in different rows of each MF_s has PGD=1, the current sampling is determined to be normal sampling and subsequent calculations can continue. Otherwise, the interferogram is determined to be abnormal sampling and frequency-incremented until it is determined to be normal sampling.

4. The laser measurement and surface feature data management method according to claim 3, characterized in that: The solution of regional interference frequencies in the sampling feature quantization based on regional peak analysis includes: performing a row-by-row search on the F0 corresponding to the first and second peak values ​​in the MF_s of the three truncation functions (Hanning, Hamming, and Blackman) that are determined to be normal sampling; and calculating the average value of F0 at the positions of the first and second peak values ​​in each row of the current truncation function, denoted as F1_s and F2_s. Then, calculating the average result of the first and second peak values ​​corresponding to the above three truncation functions as harmonic frequencies F1 and F2, and performing initial phase demodulation, unwrapping, de-tilting, and surface reconstruction accordingly. The obtained front and rear surface shape results are stored in the surface shape data storage matrix and denoted as SP1 and SP2, respectively.

5. The laser measurement and surface feature data management method according to claim 4, characterized in that: The efficient storage and preliminary processing of surface data in the surface data storage method includes: taking surface data storage matrices SP1 and SP2 as analysis objects, and denoting the current surface data storage matrix as SP, with the size of the SP matrix denoted as XM×YM. The root mean square (RMS) and peak-to-valley (PV) values ​​of SP are calculated. Then, the maximum value of the SP matrix element after removing the maximum value of elements in the matrix with an m% percentage reduction, and the minimum value of the SP matrix element after removing the minimum value of elements in the matrix with an m% percentage reduction, are calculated, and the difference between the two is denoted as PVQ_m. The variable m in m% and PVQ_m is selected as one of five integers from 1 to 5. If PVQ_m satisfies (PVQ_m-PV) / PV>3, the current surface quality is determined to be poor. If PVQ_m satisfies 3≥(PVQ_m-PV) / PV>1.5, the current surface quality is determined to be slightly poor. If PVQ_m satisfies 1.5≥(PVQ_m-PV) / PV>1.5, the current surface quality is determined to be slightly poor. If PV / PV > 0.5, the current surface quality is judged as good; if PVQ_m both satisfy 0.5 ≥ (PVQ_m - PV) / PV, the current surface quality is judged as excellent. Furthermore, the surface quality is quantified and its feature information encrypted using the quantitative index SQ: the absolute value of the difference between the surface RMS value and the PV value is used as the quantification standard. The specific quantification method is as follows: if the current surface quality is poor, then SQ = |RMS - PV| / K1 (|| represents taking the absolute value); if the current surface quality is poor... If the face quality is slightly poor, then SQ = |RMS-PV| / K2. If the current face quality is good, then SQ = |RMS-PV| / K3. If the current face quality is excellent, then SQ = |RMS-PV| / K4. The calculation result of SQ is rounded to two decimal places. The values ​​of K1 to K4 are set according to the user's encryption requirements. It is required that the SQ calculation results corresponding to K1 to K4 remain distinct from each other after rounding to two decimal places, and the values ​​of K1 to K4 are generally set to integers.

6. The laser measurement and surface feature data management method according to claim 5, characterized in that: The information matrix expansion and processing in the efficient storage of surface data includes: while keeping the original matrix elements and their positions in the surface data storage matrix SP unchanged, adding a row with all elements being zero in SP, referred to as the added row, and extending the matrix size to (XM+1)×YM. Then, in the added row, the following surface parameters are arranged from left to right: PV, RMS, PVQ5, PVQ3, PVQ2, PVQ1, SQ, RS, EX, EXG, MH. The surface data storage matrix after the added row is denoted as SFMA. The parameters in the added row have the following meanings: RS is a parameter representing the distribution form of the surface shape of the high-precision component, EX is the outlier removal operator, and MH is the fuzzification processing operand. A square face refers to a face where, before adding rows, every element in the face data storage matrix is ​​face height data, denoted as RS=R. A circular face refers to a face where, before adding rows, not every element in the face data storage matrix is ​​face height data. In a circular face, the face height data is centered at the matrix's center point (XM / 2, YM / 2), and the face data distribution conforms to the characteristics of a circle with a radius less than XM / 2, denoted as RS=S. Before adding rows, the face data storage matrix is ​​a square matrix with equal rows and columns. Storing circular faces in a square matrix cannot fill all matrix elements. To distinguish between valid and invalid face regions, the element values ​​outside the circle in the circular face data matrix are all set to EM, and EM should be different from any element value in the face data to differentiate between valid and invalid face regions. Where EX is the outlier removal operator. If EX=1, the outlier removal operation is performed; otherwise, it is not performed, and EX=0 is recorded. EXG is the anomaly detection parameter during the outlier removal operation. During the outlier removal operation, elements in the valid region of the face matrix that are greater than EXG times the face RMS value are identified as outliers. The element identified as an outlier is replaced with the average value of the valid face region elements of its four surrounding elements (top, bottom, left, and right). If there are three or two valid face region elements surrounding the outlier, the outlier is replaced with the average value of the three or two adjacent valid face region elements. If there is only one valid face region element surrounding the outlier, the outlier removal operation is not performed even if EX=1. There are two options for fuzzification: the first is to swap the elements in even-numbered rows of SP if MH=2, and the second is to swap the elements in odd-numbered rows of SP if MH=1. For both swapping methods, elements that have already been swapped will not be swapped again.

Images