A device and method for detecting the surface shape of a meter-class planar element by eliminating intermediate frequency coherent noise

Abstract

The application discloses a meter-level plane element surface shape detection device and method for eliminating intermediate frequency coherent noise, wherein the detection method comprises the following steps: (1) adjusting the angle of a large-aperture element clamped by an electric five-dimensional adjusting frame to obtain an interference pattern with clear focusing and good fringe contrast; (2) adjusting the voltage of a laser controller to change the wavelength of a tunable laser, thereby realizing wavelength phase shifting, and using a measurement camera to collect a group of phase-shifted interference patterns in time sharing mode; (3) sequentially performing aperture selection, random step phase shifting algorithm demodulation surface shape, neural network method unwrapping phase and aberration fitting removal tilt on the phase-shifted interference patterns to obtain a surface shape diagram phi (x, y), and removing system errors from the surface shape diagram phi (x, y); and (4) processing by using a convolutional neural network to eliminate intermediate frequency coherent noise to obtain phi dn (x, y). According to the application, the intermediate frequency surface shape of the meter-level plane element can be effectively and highly accurately measured.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of optical precision detection, and particularly relates to a device and method for detecting the surface shape of a meter-level plane element by eliminating middle-frequency coherent noise. BACKGROUND

[0002] Residual errors in the manufacture of optical elements will directly affect the quality of laser output, and in particular, the middle-frequency band (1 / 12-1 / 160 of the aperture) manufacturing error of the optical element will reduce the brightness of the central bright spot, widen the bright spot width, and is an important reason for causing nonlinear self-focusing in a high-power laser system, which seriously affects the performance of the system.

[0003] Large-aperture optical elements with an aperture of more than 1 m are referred to as meter-level optical elements, and the production cost of the meter-level optical elements is high, the aperture is large, and the middle-frequency manufacturing precision of the surface shape is very high. In order to ensure that the middle-frequency surface shape manufacturing error of the optical element can meet the requirements, high-precision measurement needs to be performed on the optical element during the entire manufacturing process. The method of interference detection is usually used for high-precision measurement, but in the interference detection, many system factors will affect the precision of the interference measurement, and the existence of coherent noise has a greater influence on the interference measurement result.

[0004] Most interference detection systems select a laser light source with good coherence as an illumination light source, and due to some defects or multiple reflections in the elements, a large amount of coherent fringe noise exists in the interference image when the light passes through numerous optical elements. Most of such coherent noise is Newton's ring or a bullseye, and the frequency range is in the first frequency band (PSD1) of the power spectral density, and the period is 1 / 12-1 / 160 of the full aperture, which overlaps with the frequency band that needs to be detected with high precision.

[0005] Although the commonly used frequency domain filtering method can weaken these coherent noise rings, the key surface shape information in the full frequency band is often lost.

[0006] Therefore, an effective and high-precision interference detection algorithm is needed to remove the middle-frequency coherent noise, so as to meet the extreme detection requirements of the large-aperture optical element. SUMMARY

[0007] The application provides a device and method for detecting the surface shape of a meter-level plane element by eliminating middle-frequency coherent noise, and high-precision middle-frequency surface shape measurement of the meter-level plane element can be realized.

[0008] A device for detecting the surface shape of a meter-level plane element by eliminating middle-frequency coherent noise, comprising an interference detection system and a computer processing module.

[0009] The interference detection system is installed on the air floating optical platform, and comprises a tunable laser, a polarizer, a beam expander, a spatial filter, a mirror, a beam splitter prism, an imaging lens, a measurement camera, a small-aperture collimator, a small-aperture negative lens, a first folding mirror, a second folding mirror, a large-aperture collimator, a transmission standard flat crystal TF and a reflection standard flat crystal RF.

[0010] The laser emitted by the tunable laser obtains linearly polarized light after passing through the polarizer, and then passes through the beam expander, the spatial filter, the mirror, the beam splitter prism and the small-aperture collimator to exit as parallel light; the parallel light is reflected by the small-aperture negative lens, and is folded twice by the first folding mirror and the second folding mirror, and then passes through the large-aperture collimator to exit as parallel light after beam expansion; when the parallel light after beam expansion passes through the transmission standard flat crystal TF, part of the light is reflected back to the beam splitter prism as interference reference light; the other part of the light is transmitted through the mirror to be measured and is reflected by the reflection standard flat crystal RF to return to the beam splitter prism as interference detection light; the interference reference light and the interference detection light are reflected by the beam splitter prism and then pass through the imaging lens to interfere at the image plane of the measurement camera to obtain an interference pattern.

[0011] The computer processing module comprises a hardware control module, an image acquisition module and an interference pattern data analysis processing module; the hardware control module is connected with the controller of the tunable laser, the wavelength of the tunable laser is changed by adjusting the controller voltage of the tunable laser, so that the wavelength is shifted, and the wavelength is shifted; the image acquisition module is connected with the measurement camera, and after a group of wavelength-shifted interference patterns are obtained, the data is transmitted to the interference pattern data analysis processing module for analysis.

[0012] Further, the wavelength of the tunable laser is λ0=632.8nm, the wavelength is changed by adjusting the controller voltage of the tunable laser to modulate the interference phase, so that the wavelength is shifted; wherein the phase change amount and the wavelength change amount Δλ satisfy h is the length of the interference cavity.

[0013] A meter-level plane element surface shape detection method for eliminating intermediate frequency coherent noise, using the meter-level plane element surface shape detection device, the specific process is as follows:

[0014] Step 1, the angles of the large-aperture collimator, the transmission standard flat crystal TF, the mirror to be measured and the reflection standard flat crystal RF clamped are adjusted by the electric five-dimensional adjusting frame to obtain an interference pattern with clear focus and good fringe contrast;

[0015] Step 2, the wavelength of the tunable laser is changed by adjusting the controller voltage of the tunable laser, so that the wavelength is shifted, a group of shifted interference patterns are obtained by using the measurement camera to collect in time;

[0016] Step 3, aperture selection, random step phase shifting algorithm demodulation of the surface shape, neural network method unwrapping phase, aberration fitting to remove tilt are sequentially performed on the phase shifting interferogram to obtain the surface shape φ(x,y), and the system error is removed from the surface shape φ(x,y);

[0017] Step 4, further processing is performed by using the neural network to eliminate the intermediate frequency coherent noise to obtain the final surface shape φ dn (x,y).

[0018] In step (3), in the process of aberration fitting to remove tilt, Zernike polynomials are used for aberration fitting to remove tilt for regular circular apertures, and Schmidt orthogonalization is used for aberration fitting to remove tilt for irregular apertures.

[0019] The specific process of step (4) is as follows:

[0020] (4-1) Construct a simulation data set;

[0021] (4-2) Divide the data set into a training set, a validation set and a test set; select a PDU convolutional neural network, use the training set to train the convolutional neural network, use the validation set to verify the network learning effect, and use the test set to evaluate the accuracy of the trained and verified network;

[0022] (4-3) Pre-process the surface shape φ(x,y) to obtain a low-frequency surface shape φ l (x,y), an intermediate-frequency surface shape φ m (x,y) and a high-frequency surface shape φ h (x,y);

[0023] (4-4) Input the pre-processed intermediate-frequency surface shape φ m (x,y) into the trained PDU convolutional neural network to obtain a surface shape result φ' mt (x,y) with removed coherent noise and a separated noise n(x,y);

[0024] (4-5) Correct the surface shape result φ' mt (x,y) with removed coherent noise by using the separated noise n(x,y) to obtain the final denoising result φ' m (x,y) of the intermediate-frequency surface shape φ m (x,y);

[0025] (4-6) Add the low-frequency surface shape φ l (x,y) and the processed intermediate-frequency surface shape φ' m (x,y) to obtain the final surface shape φ dn (x,y) = φ l (x,y) + φ' m (x,y).

[0026] The specific process of step (4-1) is as follows:

[0027] (4-1-1) generating a surface shape W(x, y) containing high, medium and low frequency bands using a random surface shape generating function;

[0028] (4-1-2) performing an error function filtering on the surface shape W(x, y) in a first frequency band PSD1 to obtain a medium frequency true value surface shape W r (x, y), and the error function filtering is as follows:

[0029]

[0030] In the formula, f1 and f2 are the set PSD1 frequency range;

[0031] (4-1-3) randomly generating a noise center point N with coordinates (x0, y0), and the annular coherent noise radius is R, so that the phase of one annular coherent noise is:

[0032]

[0033] In the formula, r is the number of pixels to the noise center point N, and dx is the actual size of each pixel point;

[0034] (4-1-4) adding the simulation medium frequency true value surface shape W r (x, y) and the coherent noise phase n to obtain a simulation medium frequency surface shape containing noise Where k is the number of annular coherent noises.

[0035] In step (4-2), the proportions of the training set, the validation set and the test set are 3:1:1.

[0036] In step (4-3), the process of pre-processing the surface shape map φ(x, y) is as follows:

[0037] The surface shape map φ(x, y) is subjected to four-image topology to obtain φ1(x, y), so that the surface shape edge is continuous in space; the spatial frequency spectrum F(x, y) is obtained by Fourier transform on the topological surface shape φ1(x, y); the spatial frequency spectrum F(x, y) is divided into low frequency F l (x, y), medium frequency F m (x, y) and high frequency F h (x, y) according to PSD1, and the inverse Fourier transform and topology solution operation are performed on them in sequence to obtain surface shapes in different frequency bands: low frequency φ l (x, y), medium frequency φ m (x, y) and high frequency φ h (x, y).

[0038] In step (4-5), the correction process is φ m (x,y) = φ mt (x,y) + t n(x,y), where t is a correction coefficient, and RMS(φ m (x,y)) is minimized.

[0039] Compared with other phase denoising methods, the present application has the following advantages:

[0040] 1. The present application can eliminate the mid-frequency coherent noise in the surface profile of a meter-level planar element, and effectively reduce the PSD1 value of the surface profile of the meter-level planar element.

[0041] 2. The present application only processes the mid-frequency band of the surface profile, and does not lose the original low-frequency and high-frequency information of the surface profile.

[0042] 3. The processing result of the neural network of the present application contains separated noise, which can be used for secondary correction of the result of the first output network, and further eliminate the mid-frequency coherent noise. BRIEF DESCRIPTION OF DRAWINGS

[0043] Figure 1 Fig. 1 is a schematic diagram of a meter-level planar element surface profile detection device for eliminating mid-frequency coherent noise according to the present application;

[0044] Figure 2 Fig. 2 is a neural network structure diagram in an embodiment of the present application;

[0045] Figure 3 Fig. 3 is a surface profile diagram obtained after steps 1-3 in an embodiment of the present application;

[0046] Figure 4 Fig. 4 is a denoised surface profile diagram obtained after step 4; Figure 3

[0047] Fig. 5 is a mid-frequency (PSD1 band) surface profile of φ Figure 5 Figure 3 Fig. 6 is a mid-frequency (PSD1 band) surface profile obtained after neural network processing of φ

[0048] Figure 6 Figure 5 Fig. 7 is a separated noise obtained after neural network processing of φ

[0049] Figure 7 Fig. 8 is a surface profile obtained after secondary correction of the separated noise of φ Figure 5 DETAILED DESCRIPTION

[0050] The present application will be further described in detail below in combination with the drawings and embodiments, and it should be noted that the following embodiments are intended to facilitate the understanding of the present application, and do not limit the present application in any way.

[0051] For example, Figure 1 ​​​As shown, a device for detecting the surface shape of a meter-level planar element for eliminating intermediate frequency coherent noise includes an interference detection system and a computer processing module.

[0052] The entire interferometric detection system is fixed on an air-floating vibration-isolated optical platform 1, and includes a tunable laser 2, a polarizer 3, a beam expander 4, a spatial filter 5, a reflector 6, a beam splitter 7, an imaging lens 8, a measuring camera 9, a small-aperture collimating objective lens group 10, a small-aperture negative lens 11, a first deflecting reflector 12, a second deflecting reflector 13, a large-aperture collimating objective lens 14, a transmission standard flat crystal TF15, and a reflection standard flat crystal RF17.

[0053] Among them, the laser emitted by the tunable laser 2 obtains linear polarization after passing through the polarizer 3, and then passes through the beam expander 4, the spatial filter 5, the reflector 6, the beam splitter prism 7, and the small-aperture collimating objective lens group 10 in sequence to emit parallel light; after the parallel light is emitted, it passes through the small-aperture negative lens 11, passes through the first folding reflector 12 and the second folding reflector 13 to achieve two turns, and then passes through the large-aperture collimating objective lens 14 to emit the expanded parallel light; when the expanded parallel light passes through the transmission standard flat crystal TF15, a part of the light will be reflected and narrowed back to the beam splitter prism 7 as interference reference light; the other part of the light is transmitted as measurement light through the mirror to be measured 16 and then reflected by the reflection standard flat crystal RF17, and returns to the beam splitter prism 7 along the original path as interference detection light; the interference reference light and the interference detection light are reflected by the beam splitter prism 7 and pass through the imaging lens 8, and interfere at the image plane of the measuring camera 9 to obtain an interference pattern.

[0054] In this embodiment, the small-aperture collimating objective lens group 10 is 4 inches, the small-aperture negative lens 11 is 4 inches, and the large-aperture collimating objective lens 14 is 32 inches.

[0055] The computer processing module includes a hardware control module, an image acquisition module, and an interferogram data analysis and processing module. The hardware control module is connected to the tunable laser 2 and adjusts its voltage to change its wavelength, thereby achieving wavelength phase shifting. The image acquisition module is connected to the measurement camera 9. After generating a set of wavelength phase-shifted interferograms, the data is transmitted to the interferogram data analysis and processing module for analysis.

[0056] The above device is used to perform meter-level planar component surface detection to eliminate intermediate frequency coherent noise, including the following steps:

[0057] Step 1: Use the electric five-dimensional adjustment frame to adjust the angles of the large-aperture collimating objective lens 14, the transmission standard flat crystal TF15, the test mirror 16 and the reflection standard flat crystal RF17 to obtain an interference pattern with clear focus and good fringe contrast.

[0058] Step 2, by adjusting the tunable laser 2 voltage, so that the wavelength of the tunable laser 2 changes, thereby realizing wavelength phase shifting, a set of phase-shifted interferograms is obtained by using the measurement camera 9 to collect in time.

[0059] Here the interference cavity length is 800 mm, according to the formula It is calculated that one step of π / 2 phase shift requires a wavelength change of Δλ = 6.26 × 10 4 nm, and the number of collected interferograms is 5.

[0060] Step 3, select an aperture for the phase-shifted interferogram, as much as possible to include the complete interferogram, use the random step phase shifting algorithm to demodulate the surface shape, and use the advanced iteration algorithm (AIA), which uses the phase of each pixel point on each interferogram to iteratively calculate the step size and the phase of the interferogram based on the least squares method. After obtaining the demodulated surface shape, use the neural network method to unwrap the phase. For regular circular apertures, Zernike polynomials are used for aberration fitting to remove tilt, and for irregular apertures, Schmidt orthogonalization is used for aberration fitting to remove tilt, to obtain the surface shape φ(x,y). The calibrated cavity error is used to remove the system error of the surface shape φ(x,y).

[0061] Step 4, use the neural network to process and eliminate the intermediate frequency coherent noise to obtain φ dn (x,y). The specific process is as follows:

[0062] (4-1) Construct a simulation data set.

[0063] (4-2) Divide the data set into training set, validation set and test set; select PDU (Phase Denoise Unet) convolutional neural network, use the training set to train the convolutional neural network, use the validation set to verify the network learning effect, and use the test set to evaluate the accuracy of the trained and verified network.

[0064] The structure of the PDU convolutional neural network is as shown in Figure 2

[0065] The main body of the network is composed of six layers of U-Net, and DenseNet is used to transition between each layer. U-Net extracts features by scaling the interferogram multiple times, and as the number of U-Net layers increases, the number of extracted features increases, which helps to extract the features of the phase. At the same time, the U-Net layer and the layer are expanded by the DenseNet, which expands the network depth and convolution time, so that the network can extract higher-level and abstract features. In addition, U-Net and DenseNet use a large number of skip connection operations, which ensures the transmission of gradients, improves the stability of the network, and shortens the training time.

[0066] ​The PDU convolutional neural network weight is obtained by training using the simulation data set in step (4-1).

[0067] (4-3) The surface shape is pre-processed to obtain low-frequency surface shape l (x,y), medium-frequency surface shape m (x,y) and high-frequency surface shape h (x,y).

[0068] (4-4) The pre-processed medium-frequency surface shape m (x,y) is input into the trained PDU convolutional neural network to obtain the surface shape result mt (x,y) and the separated noise n(x,y).

[0069] (4-5) The surface shape result mt (x,y) is corrected using the separated noise n(x,y) to obtain the final denoising result m (x,y) of the medium-frequency surface shape m (x,y).

[0070] (4-6) The low-frequency surface shape l (x,y) and the processed medium-frequency surface shape m (x,y) are added to obtain the final surface shape dn (x,y) = φ l (x,y) + φ' m (x,y).

[0071] Figures 3 to 7 The application is applied to eliminate the results of the medium-frequency coherent noise of the plane element surface shape detection. Among them Figure 3 is the surface shape diagram φ(x,y) obtained in step 3, Figure 4 is the surface shape diagram φ dm (x,y) obtained by using the neural network to process and eliminate the medium-frequency coherent noise in step 4, Figure 5 is the medium-frequency φ m (x,y) of the surface shape diagram φ(x,y), Figure 6 is the neural network processing medium-frequency result φ' m (x,y), Figure 7 is the separated noise n(x,y). Among them, the PV value of the surface shape φ(x,y) is 382.85 nm, the RMS value is 72.21 nm, and the equivalent PSD1 value of φ m (x,y) is 4.77 nm. The PV value of the surface shape φ dm (x,y) after the neural network processing is 355.90 nm, the RMS value is 75.18 nm, and the equivalent PSD1 value of φ'm The RMS value of (x, y) is 2.81 nm. Through the processing of the neural network, the equivalent PSD1 value of the experimental surface shape can be reduced from 4.77 nm to 2.81 nm, and the medium-frequency coherent noise can be effectively separated.

[0072] The above embodiments describe the technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the present application. Any modification, supplement and equivalent replacement made within the principle range of the present application shall be included in the protection scope of the present application.

Claims

1. A device for detecting the surface shape of a micrometer order planar element by eliminating intermediate frequency coherent noise, characterized by, The system comprises an interference detection system and a computer processing module. The interference detection system is installed on an air-floating optical platform (1) and comprises a tunable laser (2), a polarizer (3), a beam expander (4), a spatial filter (5), a mirror (6), a beam splitter prism (7), an imaging lens (8), a measurement camera (9), a small-aperture collimator lens group (10), a small-aperture negative lens (11), a first folding mirror (12), a second folding mirror (13), a large-aperture collimator lens (14), a transmission standard flat TF (15) and a reflection standard flat RF (17). The laser emitted by the tunable laser (2) is linearly polarized after passing through the polarizer (3), and then passes through the beam expander (4), the spatial filter (5), the mirror (6), the beam splitter prism (7) and the small-aperture collimator lens group (10) in sequence to exit as parallel light. The parallel light is passed through the small-aperture negative lens (11), the first folding mirror (12) and the second folding mirror (13) to realize two turns, and then passed through the large-aperture collimator lens (14) to exit as expanded parallel light. When the expanded parallel light passes through the transmission standard flat TF (15), part of the light is reflected back to the beam splitter prism (7) as interference reference light, and the other part of the light is transmitted through the mirror to be measured (16) as measurement light, and then reflected by the reflection standard flat RF (17) to return to the beam splitter prism (7) as interference detection light. The interference reference light and the interference detection light are reflected by the beam splitter prism (7) and then passed through the imaging lens (8) to interfere at the image plane of the measurement camera (9) to obtain an interference pattern. The computer processing module comprises a hardware control module, an image acquisition module and an interference pattern data analysis processing module. The hardware control module is connected with the controller of the tunable laser (2), and the wavelength of the tunable laser (2) is changed by adjusting the controller voltage of the tunable laser (2) to realize wavelength phase shifting. The image acquisition module is connected with the measurement camera (9), and after a set of wavelength phase shifting interference patterns are obtained, the data is transmitted to the interference pattern data analysis processing module for analysis.

2. The apparatus for removing intermediate frequency coherent noise from a meter-class planar element surface profile measurement according to claim 1, wherein The wavelength of the tunable laser (2) is λ0=632.8nm, the wavelength is changed by adjusting the controller voltage of the tunable laser (2) to realize the phase modulation of interference, and the wavelength is shifted; wherein the phase change amount And the wavelength change amount Δλ satisfies h is the length of the interference cavity.

3. A surface detection method for meter-level planar components that eliminates intermediate frequency coherent noise, characterized in that: The meter-level plane element surface shape detection device of claim 1 or 2 is used in the following specific process: Step 1: The angles of the large-aperture collimator lens (14), the transmission standard flat TF (15), the mirror to be measured (16) and the reflection standard flat RF (17) are adjusted by the electric five-dimensional adjustment frame to obtain an interference pattern with clear focus and good fringe contrast. Step 2: The wavelength of the tunable laser (2) is changed by adjusting the controller voltage of the tunable laser (2) to realize wavelength phase shifting, and a set of phase shifting interference patterns are obtained by using the measurement camera (9) to collect in time. Step 3: The phase shifting interference patterns are sequentially subjected to aperture selection, random step phase shifting algorithm demodulation of surface shape, neural network method unwrapping phase, aberration fitting to remove tilt, to obtain a surface shape pattern φ(x, y), and the surface shape pattern φ(x, y) is subjected to system error removal. Step 4, further processing with a convolutional neural network to eliminate mid- frequency coherent noise to obtain the final surface shape φ dn (x,y).

4. The method according to claim 3, wherein In step (3), in the process of aberration fitting to remove tilt, Zernike polynomials are used for regular circular apertures, and Schmidt orthogonalization is used for irregular apertures.

5. The method of claim 3, wherein the method is characterized by: The specific process of step (4) is as follows: (4-1) Construct a simulation data set; (4-2) Divide the data set into a training set, a validation set, and a test set; select a PDU convolutional neural network, use the training set to train the convolutional neural network, use the validation set to verify the learning effect of the network, and use the test set to evaluate the accuracy of the trained and validated network; (4-3) pre-processing the surface shape φ(x, y) to obtain a low-frequency surface shape φ l (x, y), a medium-frequency surface shape φ m (x, y), and a high-frequency surface shape φ h (x, y); (4-4) The pre-processed intermediate frequency surface shape φ m (x,y) input the trained PDU convolutional neural network to obtain the surface shape result φ' after removing coherent noise mt (x,y) and the separated noise n(x,y); (4-5) The surface shape result φ' of removing coherent noise is obtained by using the separated noise n(x, y) mt (x, y) to obtain the intermediate frequency surface shape φ m (x, y) The final de-noising result φ' of (x, y) is obtained m (x, y) (4-6) The low frequency surface shape φ l (x,y) and the processed mid frequency surface shape φ' m (x,y) are added to obtain the final surface shape φ dn (x,y) = φ l (x,y) + φ' m (x,y).

6. The method according to claim 5, wherein The specific process of step (4-1) is as follows: (4-1-1) Use a random surface shape generation function to generate a surface shape W(x, y) containing high, medium, and low frequency bands; (4-1-2) The power spectrum density first frequency band PSD1 error function filtering is performed on the surface shape W(x, y) to obtain the middle frequency true value surface shape W r (x, y), and the error function filtering is: In the formula, f1 and f2 are the set PSD1 frequency range; (4-1-3) Randomly generate a noise center point N with coordinates (x0, y0), and the circular ring coherent noise radius is R, so the phase of a circular ring coherent noise is: In the formula, r is the number of pixels to the noise center point N, and dx is the actual size of each pixel. (4-1-4) The simulated intermediate frequency true value surface shape W r (x, y) and the coherent noise phase n are added to obtain the simulated intermediate frequency surface shape with noise where k is the number of coherent noise circles.

7. The method of claim 5, wherein the method is a meter-class planar element surface profile detection method for eliminating intermediate frequency coherent noise, characterized in that, In step (4-2), the proportions of the training set, the validation set, and the test set are 3:1:

1.

8. The method of claim 5, wherein the method is a meter-class planar element surface profile detection method for eliminating intermediate frequency coherent noise, characterized in that, In step (4-3), the process of pre-processing the surface shape map φ(x, y) is as follows: Perform four-image topology on the face graph φ(x,y) to obtain φ1(x,y), so that the face edge is continuous in space; perform Fourier transform on the face graph φ1(x,y) after topology to obtain the spatial spectrum F(x,y); divide the spatial spectrum F(x,y) into the following frequency domains according to PSD1: low frequency F l (x,y), intermediate frequency F m (x,y) and high frequency F h (x, y), and then perform inverse Fourier transform and topological solution operations to obtain the surface shapes of different frequency bands: low frequency φ l (x,y), intermediate frequency φ m (x,y) and high frequency φ h (x,y).

9. The method of claim 5, wherein the method is a meter-class planar element surface profile detection method for eliminating intermediate frequency coherent noise, characterized in that, In step (4-5), the correction process is φ' m (x,y) = φ' mt (x,y) + t n(x,y), where t is a correction coefficient, satisfying that RMS(φ' m (x,y)) is minimized.

Images