Microstructure element cross-scale ultra-precision detection system and working method

Abstract

The present application relates to a kind of microstructure element cross-scale ultra-precision detection system and working method, belong to precision measurement technical field, this system is based on the phase information acquisition full aperture splicing system of digital holography and dynamic interference to record the information of object completely, and aberration correction is carried out to entire system, the problem of measurement accuracy is solved, then using the method that compressive sensing is combined with super-resolution algorithm, remove interference chart noise speckle simultaneously, improve resolution, to achieve higher measurement accuracy.Finally, by scanning splicing method to reach the full aperture large field of view detection of millimeter level, and using dynamic interference measurement system ensures the accuracy of splicing process.Compared with the traditional optical element surface detection system, the present application has the advantages of real-time, accuracy, high precision, large measurement range, etc., while ensuring the accuracy of surface detection, improves the efficiency of absolute detection method, can meet the needs of various kinds of microstructure optical element surface fast measurement.

Description

TECHNICAL FIELD

[0001] The application relates to a microstructure element cross-scale ultra-precision detection system and a working method, in particular to a high-precision full-aperture cross-scale optical element surface shape ultra-precision detection system and a working method based on digital holography, and belongs to the technical field of precision measurement. BACKGROUND

[0002] With the development of the fields of national defense industry, aerospace and life science, the application of microstructure optical elements is becoming more and more extensive. Unlike traditional optical elements, the overall size of the microstructure optical element is small, facilitating the integration and array of the entire optical system, and the microstructure size on the element surface is also relatively small, generally in the micron level. Therefore, the machining precision of the microstructure optical element depends to a great extent on the surface shape detection technology, and the requirement for detection precision is also higher and higher.

[0003] The digital holographic interferometric measurement technology has the general characteristics of real-time, non-contact, high precision and the like of optical interference technology, and has a micron or even sub-micron transverse resolution and a sub-nanometer longitudinal resolution, and has become one of the research hotspots of topography measurement technology. Color digital holography is used to shoot an object by using a multi-wavelength (red, green and blue) light source and record the information of the object in a digital form. Compared with a monochromatic light hologram, the color digital holography can more comprehensively and completely reflect the information of the object, and can realize accurate surface topography measurement.

[0004] At present, a measurement system only relying on digital holography is difficult to achieve the high-precision detection requirement of the nanometer level. The transverse resolution of the digital holographic measurement is less than 1 um, and the longitudinal resolution is 1 nm. However, the actual resolution is far lower than the theoretical value, and is affected by speckle noise, which reduces the signal-to-noise ratio of the image, affects the three-dimensional reconstruction accuracy, and limits the application of digital holography in microstructure measurement and micro object identification and the like. Due to the limitation of the array size and the total number of pixels, the imaging field of view and the imaging resolution of the CCD (or CMOS) are always restricted to each other, which affects the measurement range of the digital holography technology.

[0005] Therefore, solving the problem of nanometer-micron-millimeter cross-scale surface topography ultra-precision fast full-aperture detection is an extremely challenging task. SUMMARY

[0006] In view of the shortcomings of the prior art, the application provides a microstructure element cross-scale ultra-precision detection system and a working method. Compared with the traditional optical element surface shape detection system, the application has the advantages of real-time, accuracy, high precision, large measurement range and the like. While ensuring the surface shape detection precision, the efficiency of the absolute detection method is improved, and the demand of the surface shape fast measurement of various microstructure optical elements can be met.

[0007] The application adopts the following technical scheme:

[0008] A microstructure element cross-scale ultra-precision detection system, comprising a red-green-blue three-color laser, a non-polarization beam splitter A, a spatial filter, a polarizer, a polarization beam splitter A, a polarization beam splitter B, a polarization beam splitter C, a polarization beam splitter D, a color polarization camera, a spiral phase plate, a filter, a right-angle mirror, a polarization camera, a PZT displacement table and a three-axis displacement table;

[0009] The red-green-blue three-color laser is modulated and combined by three non-polarization beam splitters A, expanded and filtered by a spatial filter, and then transmitted to the polarization beam splitter A through the polarizer with the polarization direction at 45° to the optical axis. The polarization beam splitter A divides the incident light into S and P two polarization states, wherein the S light component with the polarization direction perpendicular to the incident plane is reflected into the reference light path, and the reference light reflected by the microscope objective A is transmitted into the imaging light path through the polarization beam splitter A after being modulated by the 4f system composed of the 1 / 4 wave plate A and the lens group.

[0010] The other P light component with the polarization direction parallel to the incident plane is projected into the measurement light path through the polarization beam splitter A, and after being modulated by the 1 / 4 wave plate B and the lens group in the measurement light path, the light is divided by the non-polarization beam splitter B. One way is to illuminate the measured object through the measurement light path, and the light carrying the topographic information of the measured object surface is reflected back through the microscope objective B, and then enters the imaging light path through the polarization beam splitter A. The measured object is arranged on the three-axis displacement table. The other way is to become a 532nm single-wavelength light beam after passing through the filter, and then enter the dynamic interference measurement system. In the imaging light path, the reference light reflected by the microscope objective A and the object light carrying the topographic information of the measured object surface reflected by the microscope objective B are modulated by the 1 / 4 wave plate C, and then become circularly polarized light with opposite rotation directions, and then enter the color polarization camera to interfere, and the color digital holographic interference pattern is recorded by the color polarization camera.

[0011] In the dynamic interference measurement system, the light is divided by the polarization beam splitter B into the measurement light path and the reference light path. The light in the measurement light path is divided by the polarization beam splitter C, one way is incident to the PZT displacement table, and after reflection, it interferes with the reference light through the polarization beam splitter D. The other way of light in the reference light path is modulated into right-handed vortex light by the spiral phase plate, and then forms a petal-shaped interference pattern after interference with the linearly polarized light beam, i.e. the reference light, through the right-angle mirror, which is recorded by the polarization camera.

[0012] A working method of the above-mentioned microstructure element cross-scale ultra-precision detection system, comprising the following steps:

[0013] (1) Collecting holographic interferogram: the reference light reflected back by microscopic objective A and the object light carrying the surface topography information of the object to be measured reflected back by microscopic objective B interfere, and the color digital holographic interferogram is recorded by the color polarization camera to obtain three red, green and blue color four holographic interferograms, wherein one red and one blue color are collected, two green colors are collected, and one sub-image is recovered by four-step phase shift method (four-step phase shift needs to collect four images, according to the principle of color digital holography, one image is collected for each of the three colors, and the fourth image can be collected again with a random color, here the green color is selected); Figure One

[0014] (2) Correcting system error: after obtaining three holographic interferograms containing the phase of the test object and the system aberration, that is, after collecting the first hologram (the hologram can be the superposition of three primary colors, that is, white light), the three-axis displacement table is moved orthogonally in the vertical plane of the optical axis twice and the holograms are captured respectively to obtain the system phase aberration. Subtract the calculated system phase aberration from the sub-image obtained in step (1) to obtain a holographic sub-image without aberration;

[0015] (3) Image enhancement and surface reconstruction: the holographic sub-image obtained in step (2) is subjected to image enhancement to obtain a high-quality, low-noise and high-resolution sub-image. The sub-image is the image of the measured element after system aberration correction and image enhancement, that is, the surface shape of the measured element. The surface shape of the entire element is obtained by splicing multiple sub-images;

[0016] (4) Full-aperture image splicing: move the three-axis displacement table and repeat steps (1)-(3) to collect multiple sub-images. The full-aperture large field of view image is obtained by splicing the multiple sub-images.

[0017] Preferably, when moving the three-axis displacement table in step (4), first, the displacement amount of the three-axis displacement table along the parallel optical axis direction is roughly determined according to the overlap area requirement and the single imaging area size, and then the displacement amount of the three-axis displacement table is accurately controlled by the dynamic interference measurement system.

[0018] Each time the three-axis displacement table moves a position, the three-axis displacement table is controlled to move scanning along the vertical optical axis direction and a sub-image is collected, wherein the single-wavelength vortex light dynamic interference system is used to control the displacement amount of the three-axis displacement table;

[0019] Preferably, the process of accurately controlling the displacement amount of the three-axis displacement table by the dynamic interference measurement system is as follows:

[0020] The PZT displacement table introduces phase shift, according to the four-step phase shift principle, the introduced phase shift amounts are 0, π / 2, π and 3π / 2 respectively. The PZT displacement table needs to move vertically three times, and each time the displacement amount is λ / 8. The light intensity of the four-phase conjugate vortex light interferograms captured by the polarization camera in real time is:

[0021] ​

[0022] Where I1, I2, I3, I4 are the light intensity of the four images, a is the background light intensity, b is the modulation amplitude of the fringes, and the spiral wave front of the vortex light at any point (x, y) can be obtained by simultaneously solving the four equations of formula (1). The vortex light phase line rotates uniformly around the wave vector k, and the rotation angle is proportional to the optical path change. In the medium of air, the rotation angle of the phase line is θ, and the propagation distance h and the rotation angle satisfy a simple linear relationship: h = θ x λ / 2π. Therefore, the micro-displacement value of the three-axis displacement table can be indirectly measured by detecting the phase line of the vortex light beam. In this way, the displacement of the three-axis displacement table is controlled, the displacement table of the measured element is moved for scanning, the acquisition of multiple sub-images is realized, and the detection of a large field of view and full aperture is realized by splicing the acquired sub-images.

[0023] Preferably, the orthogonal movement of the three-axis displacement table in step (2) means that in the vertical plane of the optical axis, the first movement is performed to capture the hologram, and then the second movement is performed to capture the hologram again. The second movement is orthogonal to the first movement, and the second movement is based on the original position (the displacement amount of the first and second movements can be any value).

[0024] Preferably, in step (2), the system aberration phase parameters of the Chebyshev polynomials are obtained from the three holographic interference images, wherein the system aberration phase is represented by a series of Chebyshev polynomials:

[0025]

[0026] Where T'1(x, y), T'2(x, y), T'3(x, y) represent the phase parameters corresponding to the three holographic interference images, R(x, y) and O(x, y) are the system aberration phase and the test object phase respectively, a i is the coefficient of the Chebyshev polynomial, C i (x, y) is the term of the Chebyshev polynomial, n is the number of terms of the Chebyshev polynomial, Δx is the displacement amount of the three-axis displacement table in the x direction, and Δy is the displacement amount of the three-axis displacement table in the y direction.

[0027] The differential operation of formula (2) obtains the phase difference data ΔT1 and ΔT2 of the system aberration phase before and after the two displacements:

[0028]

[0029] ΔC ix (x, y) and ΔC iy (x, y) are the differentials of the Chebyshev polynomials in the x and y directions respectively. Formula (3) is written in matrix form and solved to obtain the coefficients a iThen, the system aberration R(x,y) is obtained;

[0030] The advantage of this absolute calibration method is that it can effectively protect the mid-to-high frequency information of the object phase, and can eliminate various low-order and high-order aberrations even for very complex aberrations.

[0031] Preferably, the image enhancement in step (3) includes a compressed sensing algorithm and a super-resolution algorithm. The compressed sensing algorithm and the super-resolution algorithm process the interference image obtained by the color polarization camera. The compressed sensing algorithm uses compressed sensing to perform sparse representation of the noisy image, and then uses the reconstruction algorithm to recover the original image to achieve the purpose of denoising. When sampling the interference image, the compressed sensing algorithm selects a sampling rate of 50% to ensure that the complete spectral information is preserved.

[0032] Furthermore, the image compressed sensing algorithm includes three aspects: ① using a sparse basis ψ to sparsely represent the image signal x as x = ψa, where a is the sparse coefficient; ② designing an observation matrix Φ ~ N(0, 1 / M) that contains rich information from the original image in the measurement information to obtain the observation value y, y = Φa; ③ reconstructing the sparse coefficients a' from the observation value y by solving the underdetermined problem, and reconstructing the signal x = ψa' with high probability. Therefore, image reconstruction is the process of solving the underdetermined system of equations, which is the core of the CS algorithm.

[0033] Preferably, a typical convex optimization algorithm—Basis Pursuit (BP)—is chosen during reconstruction. The core of the Basis Pursuit algorithm is to transform the problem of finding the minimum L0 norm into the problem of finding the minimum L1 norm, as shown in formula (4):

[0034] min||s||1 (4)

[0035] Where s is the input signal, and then by variable substitution, equation (4) is transformed into a more mature linear programming problem. Specifically, μ and ν are two non-negative vectors, μ = [μ i ] N×1 ≥0, ν=[ν i ] N×1 ≥0, N is a positive integer, replace equation (4) with the linear optimization problem shown in equation (5):

[0036]

[0037] If another Θ = [A, -A], Equation (5) can then be rewritten as

[0038] minc T z stΘz=y (6)

[0039] Then, the original-scale radiometric method can be used to solve equation (3): First, a feasible solution z is obtained. (0) Make it satisfy Θz=y, and then follow the path that makes the objective function f=c T The function value of z moves in the direction of descent (i.e., the negative gradient direction of the objective function, which is the -c direction), gradually approaching the optimal solution.

[0040] Preferably, the super-resolution algorithm uses the FSRCNN network framework, which is implemented in Python. Specifically, each input to the network consists of two pairs of images, with a training stride of 2500 steps per epoch and 25 rounds of iterative training. In each iteration, all samples are learned once. The network consists of six steps, corresponding to Conv1, Conv2, Conv3, Conv4, deconvolution, and supplementation, which correspond to the direct connection channels from the input to the output image: feature extraction, shrinkage, mapping, expansion, deconvolution, and supplementation. The input is a low-resolution interferogram, which is processed by Conv1 (5×5, 64 convolution kernels). After obtaining the feature map of the low-resolution image, Conv2 (1×1, 12 convolutional kernels, 56 channels) is used to reduce the feature dimension of the resolution and save computational cost. Conv3, the mapping step, contains M=4 layers, each with 12 3×3 convolutional kernels. Then, Con4 (1×1, 56 convolutional kernels, 12 channels) is used to expand the feature dimension to obtain a high-resolution image. The deconvolution layer (9×9, 1 convolutional kernel, 56 channels) is used to sample and aggregate the previous features and output a high-resolution result. The final supplement layer adds low-frequency constraints to the processing result of the neural network convolutional layer to prevent distortion caused by the loss of basic image information.

[0041] In this invention, four-step phase shifting can reconstruct surface information from interference fringes. Image enhancement can further process the image obtained from the four-step phase shifting, with the aim of reconstructing a clear image from a blurred image.

[0042] This invention employs a full-aperture stitching system based on RGB three-color coaxial reflective digital holography and dynamic interference to acquire phase information and completely record object information. Aberration correction is applied to the entire system, solving the accuracy problem of measurement. Then, a method combining compressed sensing and super-resolution algorithms is used to remove noise and speckle in the interferogram while improving resolution, thus achieving higher measurement accuracy. Finally, a scanning stitching method is used to achieve millimeter-level full-aperture large-field-of-view detection, and a single-wavelength vortex light dynamic interference system is used to ensure the accuracy of the stitching process. Compared with traditional optical element surface shape detection systems, this invention has advantages such as real-time performance, accuracy, high precision, and a large measurement range. While ensuring surface shape detection accuracy, it improves the efficiency of absolute detection methods and can meet the needs of rapid surface shape measurement for various types of microstructure optical elements.

[0043] For any details not covered in this invention, please refer to the prior art.

[0044] The beneficial effects of this invention are as follows:

[0045] This invention offers measurement accuracy down to the nanometer and micrometer scales, and measurement aperture down to the millimeter scale, representing a cross-scale ultra-precision detection system for microstructured components. This invention is the first to combine digital holography with compressed sensing, super-resolution algorithms, and vortex light dynamic interferometry displacement measurement. Compared to traditional optical component surface shape detection methods, this invention can simultaneously achieve nanometer-level high-precision measurement and millimeter-level full-aperture large-field-of-view measurement. While ensuring surface shape detection accuracy, it improves the efficiency and range of the detection method. The method is simple to implement, highly accurate, fast, low-cost, and highly practical, and can meet the needs of cross-scale surface shape detection for microstructured optical components. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the microstructure element cross-scale ultra-precision detection system of the present invention;

[0047] Figure 2 This is a flowchart illustrating the working method of the microstructure component cross-scale ultra-precision detection system of the present invention;

[0048] Figure 3 A vortex light interferogram was captured by a polarization camera;

[0049] Figure 4 These are holograms of pollen samples captured by a color polarization camera, where (a), (b), and (c) are three holograms of the sample, respectively.

[0050] Figure 5 Hologram of pollen sample after aberration correction;

[0051] Among them, 1-Red, green and blue three-color laser, 2-Unpolarized beam splitter A, 3-Spatial filter, 4-Polarizer, 5-Polarized beam splitter A, 6-Polarized beam splitter B, 7-Polarized beam splitter C, 8-Polarized beam splitter D, 9-Color polarized camera, 10-Spiral phase plate, 11-Filter, 12-Right angle mirror, 13-Polarized camera, 14-PZT displacement stage, 15-Triaxial displacement stage, 16-1 / 4 wave plate A, 17-Lens group, 18-Microscopic objective A, 19-1 / 4 wave plate B, 20-Unpolarized beam splitter B, 21-Object under test, 22-Microscopic objective B, 23-1 / 4 wave plate C. Detailed implementation method:

[0052] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments. However, this description is not limited thereto. All aspects not described in detail in the present invention are based on conventional techniques in the field.

[0053] Example 1:

[0054] A multi-scale ultra-precision detection system for microstructured components, such as Figure 1 As shown, it includes a red, green and blue laser 1, an unpolarized beam splitter A 2, a spatial filter 3, a polarizer 4, a polarizing beam splitter A 5, a polarizing beam splitter B 6, a polarizing beam splitter C 7, a polarizing beam splitter D 8, a color polarizing camera 9, a spiral phase plate 10, a filter 11, a right-angle mirror 12, a polarizing camera 13, a PZT displacement stage 14, and a triaxial displacement stage 15;

[0055] The red, green, and blue lasers 1 are modulated by three non-polarizing beam splitters A2 and then combined. The beam is expanded and filtered by a spatial filter 3, and then transmitted to a polarizing beam splitter A5 after passing through a polarizer 4 whose polarization direction is at 45° to the optical axis. The polarizing beam splitter A5 splits the incident light into two polarization states, S and P. The S component, whose polarization direction is perpendicular to the incident plane, is reflected into the reference optical path. In the reference optical path, it is modulated by a 4f system composed of a quarter-wave plate A16 and a lens group 17. The 4f system is a conventional filtering system. The reference light reflected back by the microscope objective A18 passes through the polarizing beam splitter A5 and enters the imaging optical path.

[0056] Another P-component, with its polarization direction parallel to the incident plane, is projected into the measurement optical path via polarization beam splitter A5. In the measurement optical path, after being modulated by a quarter-wave plate B19 and a lens group, the light is split by a non-polarization beam splitter B20. One beam passes through the measurement optical path and illuminates the object under test 21. The light carrying the morphological information of the surface of the object under test is reflected back by the microscope objective B22 and enters the imaging optical path via polarization beam splitter A5. The object under test 21 is set on a three-axis displacement stage 15. The other beam passes through filter 11 and becomes a single-wavelength beam of 532nm, which enters the dynamic interferometry system. In the imaging optical path, the reference light reflected back by the microscope objective A18 and the object light carrying the morphological information of the surface of the object under test reflected back by the microscope objective B22 are modulated by a quarter-wave plate C23 and become circularly polarized light with orthogonal rotation. The light is then incident on the color polarization camera 9 and interferes. The color polarization camera 9 records a color digital holographic interferogram.

[0057] In the dynamic interferometric measurement system, the light is split by polarization beam splitter B6 into a measurement optical path and a reference optical path. In the measurement optical path, the light is split by polarization beam splitter C7. One beam is incident on the PZT displacement stage 14, and after reflection, it passes through polarization beam splitter D8 and interferes with the reference light. The other beam in the reference optical path is modulated into a right-handed vortex beam by spiral phase plate 10. After passing through right-angle mirror 12, it interferes with another linearly polarized beam, i.e., the reference light, to form a petal-shaped interference pattern, which is recorded by polarization camera 13.

[0058] Example 2:

[0059] A working method for a multi-scale ultra-precision detection system for microstructured components, such as Figure 2 As shown, it includes the following steps:

[0060] (1) Acquiring holographic interferograms: The reference light reflected back by microscope objective A18 interferes with the object light reflected back by microscope objective B22, which carries information about the surface morphology of the object under test. The color polarization camera 9 records the color digital holographic interferograms, obtaining four holographic interferograms of red, green, and blue. One image each of red and blue is acquired, and two images of green are acquired. A sub-image is then recovered using a four-step phase-shifting method (four-step phase-shifting requires acquiring four images; according to the principle of color digital holography, the three colors interfere). Figure One For the first image, one image is captured for each color. For the fourth image, another image can be captured for a random color (green was selected here).

[0061] (2) Correcting system errors: After obtaining three holographic interferograms containing the phase of the test object and the system aberration, that is, after acquiring the first hologram (which can be the superposition of three primary colors of light, i.e. white light), the three-axis displacement stage is orthogonally moved twice on the vertical plane of the optical axis and the holograms are captured respectively to obtain the system phase aberration. The sub-image obtained in step (1) is subtracted from the calculated system phase aberration to obtain an aberration-free holographic sub-image.

[0062] (3) Image enhancement and surface reconstruction: The aberration-free holographic sub-image obtained in step (2) is enhanced to obtain a high-quality, low-noise, and high-resolution sub-image. This sub-image is the image after system aberration correction and image enhancement, which is the surface shape of the component under test. The surface shape of the entire component is obtained by stitching together multiple sub-images.

[0063] (4) Full aperture image stitching: Move the three-axis displacement stage 15 and repeat steps (1)-(3) to acquire multiple sub-images. Stitch the acquired multiple sub-images together to obtain a full aperture large field of view image.

[0064] Example 3:

[0065] A working method of a microstructure component cross-scale ultra-precision detection system is described in Example 2. The difference is that when moving the three-axis displacement stage 15 in step (4), the displacement of the three-axis displacement stage along the parallel optical axis is first roughly determined according to the overlap area requirements and the size of the single imaging area. Then, the displacement of the three-axis displacement stage is precisely controlled by the dynamic interferometric measurement system.

[0066] Each time the three-axis displacement stage moves to a position, it controls the three-axis displacement stage to move along the vertical optical axis to scan and acquire a sub-image. A single-wavelength vortex light dynamic interferometry system is used to control the displacement of the three-axis displacement stage.

[0067] Furthermore, the process of precisely controlling the displacement of the triaxial displacement stage using a dynamic interferometric measurement system is as follows:

[0068] The PZT displacement stage introduces a phase shift. Based on the four-step phase shift principle, the introduced phase shifts are 0, π / 2, π, and 3π / 2, respectively. The PZT displacement stage needs to move longitudinally three times, with each shift being λ / 8. Four phase-shifted conjugate vortex interferograms are captured in real-time by a polarization camera (e.g., Figure 3 The light intensity (as shown) is:

[0069]

[0070] Where I1, I2, I3, and I4 are the light intensities of the four images, a is the background light intensity, and b is the modulation amplitude of the stripes. By solving the four equations of formula (1), the spiral wavefront of the vortex light at any point (x,y) can be obtained. The equiphase line of the vortex light rotates uniformly around the wave vector k. The rotation angle is proportional to the change in optical path it produces. The medium is air, and the rotation angle of the equiphase line is θ. The propagation distance h and the rotation angle satisfy a simple linear relationship: h = θ × λ / 2π. Therefore, as long as the equiphase line of the vortex beam can be detected, the micro-displacement value of the three-axis displacement stage can be indirectly measured. By controlling the displacement of the three-axis displacement stage in this way, the displacement stage of the measured element is moved to scan, and multiple sub-images are acquired. The acquired sub-images are stitched together to achieve the detection of a large field of view and full aperture.

[0071] Example 4:

[0072] The working method of a microstructure component cross-scale ultra-precision detection system is as described in Example 2. The difference is that the orthogonal movement of the three-axis displacement stage 15 in step (2) means: on the vertical plane of the optical axis, it moves once to capture a hologram, and then moves a second time to capture a hologram again. The second movement is orthogonal to the first movement direction, and the second movement is based on the original position (the displacement of the first and second movements can be any value).

[0073] In step (2), the system aberration phase parameters of Chebyshev polynomials are obtained from the three holographic interferograms, where the system aberration phase is represented by a series of Chebyshev polynomials:

[0074]

[0075] Where T′1(x,y), T′2(x,y), and T′3(x,y) represent the phase parameters corresponding to the three holographic interferograms, and R(x,y) and O(x,y) are the system aberration phase and the test object phase, respectively. i Let C be the coefficients of the Chebyshev polynomial. i (x,y) are terms of the Chebyshev polynomial, n is the number of terms in the Chebyshev polynomial, Δx is the displacement of the three-axis displacement stage in the x direction, and Δy is the displacement of the three-axis displacement stage in the y direction.

[0076] Performing a difference operation on formula (2) yields the phase difference data ΔT1 and ΔT2 of the system aberration phase before and after the two displacements:

[0077]

[0078] ΔC ix (x,y) and ΔC iy(x, y) are the differentials of the Chebyshev polynomial in the x and y directions, respectively. By writing equation (3) in matrix form and solving it, we can obtain the coefficients a of the Chebyshev polynomial. i Then, the system aberration R(x,y) is obtained;

[0079] The advantage of this absolute calibration method is that it can effectively protect the mid-to-high frequency information of the object phase, and can eliminate various low-order and high-order aberrations even for very complex aberrations.

[0080] Example 5:

[0081] A working method of a microstructure component cross-scale ultra-precision detection system is as described in Example 2. The difference is that the image enhancement in step (3) includes a compressed sensing algorithm and a super-resolution algorithm. The compressed sensing algorithm and the super-resolution algorithm process the interference image obtained by the color polarization camera. The compressed sensing algorithm uses compressed sensing to perform sparse representation of the noisy image, and then uses the reconstruction algorithm to recover the original image to achieve the purpose of noise reduction. When sampling the interference image, the compressed sensing algorithm selects a sampling rate of 50% to ensure that the complete spectral information is preserved.

[0082] Furthermore, the image compressed sensing algorithm includes three aspects: ① using a sparse basis ψ to sparsely represent the image signal x as x = ψa, where a is the sparse coefficient; ② designing an observation matrix Φ ~ N(0, 1 / M) that contains rich information from the original image in the measurement information to obtain the observation value y, y = Φa; ③ reconstructing the sparse coefficients a' from the observation value y by solving the underdetermined problem, and reconstructing the signal x = ψa' with high probability. Therefore, image reconstruction is the process of solving the underdetermined system of equations, which is the core of the CS algorithm.

[0083] During reconstruction, a typical convex optimization algorithm—Basis Pursuit (BP)—is chosen. The core of the Basis Pursuit algorithm is to transform the problem of finding the minimum L0 norm into the problem of finding the minimum L1 norm, as shown in Equation (4):

[0084] min||s||1 (4)

[0085] Where s is the input signal, and then by variable substitution, equation (4) is transformed into a more mature linear programming problem. Specifically, μ and ν are two non-negative vectors, μ = [μ i ] N×1 ≥0, ν=[ν i ] N×1 ≥0, N is a positive integer, replace equation (4) with the linear optimization problem shown in equation (5):

[0086]

[0087] If another Θ = [A, -A], Equation (5) can then be rewritten as

[0088] minc T z stΘz=y (6)

[0089] Then, the original-scale radiometric method can be used to solve equation (3): First, a feasible solution z is obtained. (0) Make it satisfy Θz=y, and then follow the path that makes the objective function f=c T The function value of z moves in the direction of descent (i.e., the negative gradient direction of the objective function, which is the -c direction), gradually approaching the optimal solution.

[0090] The super-resolution algorithm uses the FSRCNN network framework, which is implemented in Python. Specifically, each input to the network consists of two pairs of images, with a training stride of 2500 steps per epoch and 25 iterations. In each iteration, all samples are learned once. The network consists of six steps, corresponding to Conv1, Conv2, Conv3, Conv4, deconvolution, and supplementation, which correspond to the direct connection channels from the input to the output image: feature extraction, shrinkage, mapping, expansion, deconvolution, and supplementation. The input is a low-resolution interferogram, which is processed by Conv1 (5×5, 64 convolution kernels, 1...). After the first layer (M×1, 12 convolutional kernels, 56 channels) is used to obtain the feature map of the low-resolution image, Conv2 (1×1, 12 convolutional kernels, 56 channels) is used to reduce the feature dimension of the resolution and save computational cost. Conv3, the mapping step, contains M=4 layers, each with 12 3×3 convolutional kernels. Then, Con4 (1×1, 56 convolutional kernels, 12 channels) is used to expand the feature dimension to obtain a high-resolution image. The deconvolution layer (9×9, 1 convolutional kernel, 56 channels) is used to sample and aggregate the previous features to output a high-resolution result. The final supplement layer adds low-frequency constraints to the processing result of the neural network convolutional layer to prevent distortion caused by the loss of basic image information.

[0091] like Figure 4 , 5 As shown, the horizontal and vertical axes represent the spatial position within the measurement field of view, in pixels. Figure 4 These are three holograms of a pollen sample captured by a color polarization camera 9. Figure 5 To correct for aberrations in the pollen sample hologram, Figure 5 The image eliminates the residual error of the system and is closer to reality.

[0092] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A cross-scale ultra-precision detection system for microstructured components, characterized in that, It includes red, green and blue lasers, unpolarized beam splitter A, spatial filter, polarizer, polarized beam splitter A, polarized beam splitter B, polarized beam splitter C, polarized beam splitter D, color polarization camera, spiral phase plate, filter, right-angle mirror, polarization camera, PZT displacement stage and triaxial displacement stage. The red, green, and blue lasers are modulated by three non-polarizing beam splitters A and then combined. The beam is expanded and filtered by a spatial filter, and then transmitted to polarizing beam splitter A after passing through a polarizer with a polarization direction at 45° to the optical axis. Polarizing beam splitter A splits the incident light into two polarization states, S and P. The S component, whose polarization direction is perpendicular to the incident plane, is reflected into the reference optical path. In the reference optical path, it is modulated by a 4f system composed of a quarter-wave plate A and a lens group. The reference light reflected back by the microscope objective A passes through polarizing beam splitter A and enters the imaging optical path. Another P-component, with its polarization direction parallel to the incident plane, is projected into the measurement optical path via polarization beam splitter A. In the measurement optical path, after being modulated by a quarter-wave plate B and a lens group, the light is split by non-polarization beam splitter B. One beam passes through the measurement optical path and illuminates the object under test. The light carrying the morphological information of the object's surface is reflected back by the microscope objective B and enters the imaging optical path via polarization beam splitter A. The object under test is set on a three-axis displacement stage. The other beam passes through a filter and becomes a single-wavelength beam of 532nm, which enters the dynamic interferometry system. In the imaging optical path, the reference light reflected back by the microscope objective A and the object light carrying the morphological information of the object's surface reflected back by the microscope objective B are modulated by a quarter-wave plate C and become circularly polarized light with orthogonal rotation. This light is then incident on the color polarization camera and interferes. The color polarization camera records a color digital holographic interferogram. In the dynamic interferometry system, the light is split by polarization beam splitter B into a measurement optical path and a reference optical path. In the measurement optical path, the light is split by polarization beam splitter C. One beam is incident on the PZT displacement stage, and after reflection, it passes through polarization beam splitter D and interferes with the reference light. The other beam is modulated into a right-hand vortex beam by a spiral phase plate in the reference optical path. After passing through a right-angle mirror, it interferes with another linearly polarized beam, i.e., the reference light, to form a petal-shaped interference pattern, which is recorded by a polarization camera.

2. A method for operating the microstructure component cross-scale ultra-precision detection system as described in claim 1, characterized in that, Includes the following steps: (1) Acquiring holographic interferograms: The reference light reflected back by the microscope objective A and the object light reflected back by the microscope objective B, which carries the surface topography information of the object to be measured, interfere with each other. The color polarization camera records the color digital holographic interferograms, and four holographic interferograms in red, green and blue are obtained. One red and one blue are acquired, and two green are acquired. A sub-image is recovered by the four-step phase shift method. (2) Correcting system errors: After obtaining three holographic interferograms containing the phase of the test object and the system aberration, that is, after acquiring the first hologram, the three-axis displacement stage is orthogonally moved twice on the vertical plane of the optical axis and the holograms are captured respectively to obtain the system phase aberration. The sub-image obtained in step (1) is subtracted from the calculated system phase aberration to obtain an aberration-free holographic sub-image. (3) Image enhancement and surface reconstruction: The aberration-free holographic sub-image obtained in step (2) is enhanced to obtain a high-quality, low-noise, and high-resolution sub-image; (4) Full aperture image stitching: Move the three-axis displacement stage and repeat steps (1)-(3) to acquire multiple sub-images. Stitch the acquired multiple sub-images together to obtain a full aperture large field of view image.

3. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 2, characterized in that, In step (4), when moving the three-axis displacement stage, the displacement of the three-axis displacement stage along the direction parallel to the optical axis is roughly determined first according to the requirements of the overlapping area and the size of the single imaging area. Then, the displacement of the three-axis displacement stage is precisely controlled by the dynamic interferometric measurement system.

4. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 3, characterized in that, The process of precisely controlling the displacement of a triaxial displacement stage using a dynamic interferometry system is as follows: A phase shift is introduced by moving the PZT displacement stage. Based on the four-step phase shift principle, the introduced phase shifts are 0, π / 2, π, and 3π / 2, respectively. The PZT displacement stage moves longitudinally three times, with each shift being λ / 8. The light intensities of the four phase-shifted conjugate vortex interferograms captured in real time by the polarization camera are: Where I1, I2, I3, and I4 are the light intensities of the four images, a is the background light intensity, and b is the modulation amplitude of the stripes. By solving the four equations of formula (1), the spiral wavefront of the vortex light at any point (x,y) can be obtained. The equiphase line of the vortex light rotates uniformly around the wave vector k. The magnitude of the rotation angle is proportional to the change in optical path it produces. The medium is air, and the rotation angle of the equiphase line is θ. The propagation distance h and the rotation angle satisfy a simple linear relationship: h = θ × λ / 2π. Therefore, as long as the equiphase line of the vortex beam can be detected, the micro-displacement value of the three-axis displacement stage can be indirectly measured.

5. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 2, characterized in that, In step (2), the orthogonal movement of the three-axis displacement stage refers to: moving once on the plane perpendicular to the optical axis to capture the hologram, and then moving a second time to capture the hologram again. The second movement is orthogonal to the first movement, and the second movement is based on the original position.

6. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 5, characterized in that, In step (2), the system aberration phase parameters of Chebyshev polynomials are obtained from the three holographic interferograms, where the system aberration phase is represented by a series of Chebyshev polynomials: Where T′1(x,y), T′2(x,y), and T′3(x,y) represent the phase parameters corresponding to the three holographic interferograms, and R(x,y) and O(x,y) are the system aberration phase and the test object phase, respectively. i Let C be the coefficients of the Chebyshev polynomial. i (x,y) are terms of the Chebyshev polynomial, n is the number of terms in the Chebyshev polynomial, Δx is the displacement of the three-axis displacement stage in the x direction, and Δy is the displacement of the three-axis displacement stage in the y direction. Performing a difference operation on formula (2) yields the phase difference data ΔT1 and ΔT2 of the system aberration phase before and after the two displacements: ΔC ix (x,y) and ΔC iy (x, y) are the differentials of the Chebyshev polynomial in the x and y directions, respectively. By writing equation (3) in matrix form and solving it, we can obtain the coefficients a of the Chebyshev polynomial. i Then, the system aberration R(x,y) is obtained.

7. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 2, characterized in that, The image enhancement in step (3) includes a compressed sensing algorithm and a super-resolution algorithm. The compressed sensing algorithm uses compressed sensing to perform sparse representation on the noisy image, and then uses a reconstruction algorithm to recover the original image in order to achieve the purpose of denoising. Furthermore, the image compressed sensing algorithm includes three aspects: ① using a sparse basis ψ to sparsely represent the image signal x as x = ψa, where a is the sparse coefficient; ② designing an observation matrix Φ ~ N(0, 1 / M) that contains rich information from the original image in the measurement information to obtain the observation value y, y = Φa; ③ reconstructing the sparse coefficients a' from the observation value y by solving the underdetermined problem, and reconstructing the signal x = ψa' with high probability.

8. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 7, characterized in that, During reconstruction, a typical convex optimization algorithm—the basis pursuit algorithm—is chosen. The core of the basis pursuit algorithm is to transform the problem of finding the minimum L0 norm into the problem of finding the minimum L1 norm, as shown in formula (4): min||s||1 (4) where s is the input signal. Then, by variable substitution, equation (4) is transformed into a linear programming problem. Specifically, μ and ν are two non-negative vectors, μ = [μ i ] N×1 ≥0, ν=[ν i ] N×1 ≥0, N is a positive integer, replace equation (4) with the linear optimization problem shown in equation (5): If another Θ = [A, -A], Then equation (5) can be rewritten as min c T z s.t.Θz=y (6) Then, the original-scale radiometric method can be used to solve equation (3): First, a feasible solution z is obtained. (0) Make it satisfy Θz=y, and then follow the path that makes the objective function f=c T The function value of z shifts in the direction of decrease, gradually approaching the optimal solution.

9. The working method of the microstructure component cross-scale ultra-precision detection system according to claim 8, characterized in that, The super-resolution algorithm uses the FSRCNN network framework, which is implemented in Python. Specifically, each input to the network consists of two pairs of images, with a training stride of 2500 steps per epoch. It employs 25 epochs of iterative training, in which all samples are learned once. The network comprises six steps: Conv1, Conv2, Conv3, Conv4, deconvolution, and supplementation, corresponding to the direct connection channels from the input to the output images. The process involves feature extraction, shrinking, mapping, expansion, deconvolution, and supplementation. The input is a low-resolution interferogram. After Conv1, a feature map of the low-resolution image is obtained. Conv2 is used to reduce the feature dimension of the resolution, saving computational costs. Conv3, the mapping step, contains M=4 layers, each with 12 3×3 convolutional kernels. Then, Conv4 is used to expand the feature dimension to obtain a high-resolution image. The deconvolution layer is used to sample and aggregate the features, outputting a high-resolution result. The final supplement layer adds low-frequency constraints to the processing results of the neural network convolutional layers to prevent distortion caused by the loss of basic image information.

Images