A multispectral-based microscopic three-dimensional measurement method and device

Abstract

This invention discloses a multispectral microscopic three-dimensional measurement method and apparatus. The method includes acquiring structured light fringe images of the sample under test, performing differential operations and separation extraction to obtain a multi-channel modulation image stack; performing modulation correction and global optimization selection on the multi-channel modulation image stack to obtain relative height values; and performing convergence processing based on the relative height values ​​to generate three-dimensional topographic point cloud data. This invention acquires structured light fringe images at multiple axial positions under different exposure conditions and obtains multiple sets of modulation image stacks through differential operations. After separation extraction according to multiple spectral channels, the multiple modulation curves corresponding to each pixel are corrected and optimally fused, and the relative height values ​​are obtained through peak fitting positioning. Finally, the relative height values ​​are converged to generate three-dimensional topographic point cloud data of the sample under test. This enables complete three-dimensional topographic reconstruction of complex industrial samples with spatially non-uniform reflectance characteristics.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a multispectral microscopic three-dimensional measurement method and apparatus. Background Technology

[0002] Currently, structural imaging microscopy, with its advantages of high spatial resolution and high efficiency, is widely used for precise three-dimensional topographic measurement of functional surfaces such as semiconductor chips, integrated circuits, and micro-optical components. However, with the development of micro-nano fabrication and electronic manufacturing technologies, the functional surfaces being measured are becoming increasingly complex in terms of structural scale and material composition, and their reflection characteristics exhibit significant spatial non-uniformity. When the local reflected light intensity of the sample exceeds the dynamic range limit of the system, the fringe modulation and signal-to-noise ratio decrease significantly, severely limiting the accuracy of three-dimensional topographic measurement.

[0003] To address the aforementioned limitation in dynamic range, existing technologies primarily focus on two approaches. The first is optimizing optical detection methods by improving system structure to suppress the impact of sample reflectivity differences on signal modulation, such as employing a multi-detector differential confocal scheme. The second is implementing adaptive illumination control, dynamically adjusting projected light intensity to optimize camera exposure; typical strategies include pixel-by-pixel encoding and multi-exposure image fusion. While these methods extend the dynamic range to some extent, the multi-detector structure increases system complexity and cost, while pre-calibration and multiple-exposure strategies significantly prolong acquisition and processing time, making it difficult to meet the high efficiency and stability requirements of industrial online inspection. Furthermore, to address measurement efficiency, existing technologies have proposed various schemes to compress the acquisition process, including time-domain phase-shifting methods based on axial scanning and fast imaging methods without axial scanning. However, the former is limited by the projection speed of sinusoidal fringes and the influence of system nonlinearity, while the latter is typically applicable only to surfaces with uniform reflectivity, and the calibration and stability control of multi-camera systems are quite difficult.

[0004] Existing technologies have proposed a three-dimensional measurement method for structured microscopy based on spatially phase-shifted complementary binary fringes. By projecting two complementary binary fringe images, the minimum number of projection images required for traditional structured microscopy is reduced from three to two, significantly improving image acquisition efficiency. However, this method does not fully consider the spatial non-uniformity of reflectivity widely present on the surface of actual industrial samples. Its dynamic range is still limited by the camera's single-exposure response, and it cannot achieve complete three-dimensional morphology reconstruction of surfaces with non-uniform reflectivity caused by complex material composition or complex surface structure. Summary of the Invention

[0005] This invention provides a multispectral microscopic three-dimensional measurement method and apparatus, aiming to solve the technical problem that existing structural illumination microscopic three-dimensional measurement methods cannot achieve complete three-dimensional morphology reconstruction of complex industrial samples with spatially non-uniform reflectance characteristics.

[0006] In a first aspect, embodiments of the present invention provide a multispectral microscopic three-dimensional measurement method, comprising: The structured light stripe images of the sample under test at multiple axial positions under different exposure conditions are acquired. The structured light stripe images at adjacent axial positions are processed by differential operation to obtain a stack of multiple modulation images. The multiple modulation image stacks are separated and extracted according to multiple preset spectral channels to obtain a multi-channel modulation image stack; Modulation correction and global optimization are performed on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack to obtain the optimal modulation curve for each pixel. Peak fitting and positioning are then performed on the optimal modulation curve to obtain the relative height value of each pixel. The three-dimensional topographic point cloud data of the tested sample is generated by convergence processing based on the relative height values ​​of all the pixels.

[0007] Secondly, embodiments of the present invention provide a multispectral microscopic three-dimensional measurement device, comprising: The image acquisition unit is used to acquire structured light stripe images of the sample under test at multiple axial positions under different exposure conditions, and to perform differential operation processing on the structured light stripe images at adjacent axial positions to obtain multiple sets of modulation image stacks. The image extraction unit is used to separate and extract the multiple modulation image stacks according to multiple preset spectral channels to obtain a multi-channel modulation image stack. The image correction unit is used to perform modulation correction and global optimization selection on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack, to obtain the optimal modulation curve for each pixel, and to perform peak fitting and positioning on the optimal modulation curve to obtain the relative height value of each pixel. The image output unit is used to perform convergence processing based on the relative height values ​​of all the pixels to generate three-dimensional topographic point cloud data of the sample under test.

[0008] This invention provides a multispectral microscopic three-dimensional measurement method, comprising: acquiring structured light fringe images of a sample under test at multiple axial positions under different exposure conditions; performing differential processing on the structured light fringe images at adjacent axial positions to obtain multiple sets of modulation index image stacks; separating and extracting the multiple sets of modulation index image stacks according to multiple preset spectral channels to obtain a multi-channel modulation index image stack; performing modulation index correction and global optimization selection on multiple modulation index curves corresponding to each pixel in the multi-channel modulation index image stack to obtain the optimal modulation index curve for each pixel, and performing peak fitting and positioning on the optimal modulation index curve to obtain the relative height value of each pixel; and performing convergence processing based on the relative height values ​​of all the pixels to generate three-dimensional topographic point cloud data of the sample under test. This invention acquires structured light stripe images at multiple axial positions under different exposure conditions and obtains multiple sets of modulation image stacks through differential operations. After separation and extraction according to multiple spectral channels, the multiple modulation curves corresponding to each pixel are corrected and optimally fused, and the relative height value is obtained through peak fitting and positioning. Finally, the three-dimensional topography point cloud data of the sample under test is generated. In this way, complete three-dimensional topography reconstruction can be achieved for complex industrial samples with non-uniform reflectance characteristics.

[0009] This invention also provides a multispectral microscopic three-dimensional measurement device, which has the same beneficial effects as described above. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A schematic flowchart of a multispectral microscopic three-dimensional measurement method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram comparing the low signal-to-noise ratio column pixel modulation correction for surface reflection characteristic continuity provided in an embodiment of the present invention; Figure 3 This is a schematic diagram comparing the peak offset correction of multi-channel modulation provided in an embodiment of the present invention; Figure 4 A comparative schematic diagram of wide-field illumination image reconstruction of wafer surface cracks provided in an embodiment of the present invention; Figure 5 This is a comparative schematic diagram of wide-field illumination image reconstruction of PCB blind vias provided in an embodiment of the present invention; Figure 6This is a schematic block diagram of a multispectral microscopic three-dimensional measurement device provided in an embodiment of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0014] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0015] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0016] Please see below. Figure 1 , Figure 1 The flowchart of a multispectral microscopic three-dimensional measurement method provided in this embodiment of the invention specifically includes steps S101 to S104.

[0017] S101. Acquire structured light stripe images of the sample under test at multiple axial positions under different exposure conditions, and perform differential operation on the structured light stripe images at adjacent axial positions to obtain multiple sets of modulation image stacks. S102. The multi-modulation image stack is separated and extracted according to multiple preset spectral channels to obtain a multi-channel modulation image stack. S103. Modulation correction and global optimization selection are performed on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack to obtain the optimal modulation curve for each pixel, and peak fitting and positioning are performed on the optimal modulation curve to obtain the relative height value of each pixel. S104. Based on the relative height values ​​of all the pixels, perform convergence processing to generate three-dimensional topographic point cloud data of the sample under test.

[0018] In step S101, the sample to be tested is placed on the platform of the microscopic measurement system, and the sample is driven by an electrically controlled displacement stage to perform a stepwise axial scan along the optical axis. At each axial position, a preset structured light fringe pattern is projected onto the surface of the sample, and the fringe image is simultaneously acquired with at least two different exposure times, thereby obtaining structured light fringe images of the sample at multiple axial positions under different exposure conditions. The purpose of using different exposure conditions is to adapt to the spatial non-uniformity of the reflectivity of the sample surface, enabling the system to acquire image information of different brightness levels at the same axial position, thereby expanding the dynamic range of the acquisition system. After completing the image acquisition at all axial positions, the structured light fringe images acquired at adjacent axial positions are subjected to layer-by-layer differential operations to eliminate the interference of background intensity terms and extract the modulation information related to the surface morphology of the sample, thereby obtaining a stack of multiple modulation images.

[0019] In one embodiment, step S101 includes: Adjust the tilt corrector so that the surface of the sample under test on the tilt corrector is perpendicular to the optical axis; The sample under test is driven by an electronically controlled displacement stage to perform axial scanning according to a preset scheme. At each axial position, the same stripe pattern is projected twice by a projector. The trigger mode other than the camera performs two synchronous acquisitions at the same axial position with low exposure time and high exposure time respectively. The synchronous acquisition process is repeated at all axial positions to obtain the original image stack of different exposure times at different axial positions. The stripe patterns projected at adjacent axial positions have a π phase shift and form complementary patterns. The original image stack is preprocessed by layer-by-layer difference to eliminate the corresponding background intensity terms, resulting in multiple sets of modulation image stacks.

[0020] In this embodiment, the tilt corrector is adjusted so that the surface of the sample under test is perpendicular to the optical axis. Specifically, the tilt corrector is a precision mechanical device used to adjust the placement posture of the sample under test. It achieves precise alignment of the sample surface normal with the optical axis of the microscopic measurement system through a multi-degree-of-freedom angle fine-tuning mechanism. Before the measurement begins, the sample under test is placed horizontally on the tilt corrector. By adjusting the angle adjustment knob or electronic adjustment mechanism of the tilt corrector, the tilt posture of the sample under test is finely corrected so that the surface normal of the sample is parallel to the optical axis of the system, that is, to ensure that the surface of the sample is perpendicular to the optical axis. The purpose of this step is to eliminate the systematic error introduced by the tilt of the sample placement, and to ensure that the structured light fringe images acquired at each position during the subsequent axial scanning process can truly reflect the morphological features of the sample surface, thereby providing a reliable data foundation for high-precision three-dimensional morphological reconstruction.

[0021] After tilt correction, the sample under test is axially scanned according to a preset scheme using a motorized translation stage. The motorized translation stage is a high-precision linear displacement platform driven by a computer or controller, capable of moving the sample under test gradually along the optical axis at preset step sizes and speeds, thereby achieving layer-by-layer scanning of the sample at different axial positions. During the axial scanning process, the motorized translation stage sequentially positions the sample under test to each target axial position according to the preset scanning scheme. At each axial position, the same fringe pattern is projected twice onto the surface of the sample under test by a projector; that is, two consecutive projections of structured light fringes are performed at the same axial position, and the fringe patterns used in the two projections are completely identical in spatial frequency, phase, and pattern content. Simultaneously, the camera's acquisition mode is set to External Trigger Mode, meaning the camera does not automatically acquire images but initiates exposure and acquisition upon receiving an external trigger signal. In this embodiment, the projector generates a rising edge signal when it begins projecting the fringe pattern; the camera is triggered and begins exposure upon receiving this rising edge signal. At the same axial position, the camera performs two simultaneous acquisitions with different exposure times: a low exposure time of 3000 μs and a high exposure time of 10000 μs. The purpose of using different exposure times for the two acquisitions is to accommodate the spatial non-uniformity of the reflectivity of the sample surface: low-exposure acquisition avoids the loss of modulation information due to signal saturation in high-reflectivity areas, while high-exposure acquisition enhances the signal strength and signal-to-noise ratio in low-reflectivity areas. The combination of these two methods effectively expands the dynamic range of the system.

[0022] During the projection and acquisition process described above, the fringe patterns projected at adjacent axial positions exhibit a π-phase shift and form complementary patterns. When the electrically controlled displacement stage moves the sample under test from the current axial position to the next adjacent axial position, the fringe pattern projected by the projector at this adjacent axial position has a spatial phase shift of π radians relative to the fringe pattern at the previous axial position. This results in the fringe patterns projected at two adjacent axial positions being out of phase in terms of brightness distribution, forming a complementary pattern relationship. This π-phase shift complementary fringe projection strategy between adjacent axial positions effectively eliminates the influence of the background intensity term when performing differential operations on the images of adjacent axial positions. Simultaneously, it reduces the minimum number of projection frames required for traditional structural illumination microscopy from three to two, significantly improving image acquisition efficiency.

[0023] Furthermore, the electrically controlled displacement stage moves the sample under test to the next axial position according to the preset scanning scheme, and repeats the above projection and synchronous acquisition process until the scanning and acquisition of all preset axial positions are completed. At this time, the raw image stack of the sample under test at different axial positions and different exposure times is obtained. This raw image stack contains the complete sequence of structured light stripe images of the sample under test acquired under both low and high exposure time conditions throughout the entire axial scanning range.

[0024] Finally, the original image stack is preprocessed by layer-by-layer difference to eliminate the corresponding background intensity terms, resulting in multiple modulation index image stacks. Since each frame in the original image stack contains a superposition of background intensity and modulation information, directly using the original images for subsequent processing would introduce background intensity interference, affecting the accuracy of modulation index extraction. Therefore, the original image stack is subjected to layer-by-layer difference operations according to axial position, that is, the structured light fringe images acquired at adjacent axial positions are subtracted pixel-by-pixel and the absolute value is taken. Since the fringe patterns at adjacent axial positions are π-phase-shift complementary patterns, the difference operation can effectively eliminate the common background intensity components in the two frames, retaining the effective information related to the fringe modulation. After the above layer-by-layer difference processing, two sets of modulation index image stacks corresponding to low exposure time and high exposure time are obtained, i.e., multiple modulation index image stacks, providing input data for subsequent multispectral channel separation extraction and modulation index correction fusion processing.

[0025] In one embodiment, the stripe patterns projected from adjacent axial positions exhibit a π phase shift and constitute complementary patterns, including: The fringe projection mode is set to temporal phase-shifted binary fringe. The system magnification, reflectivity of each pixel coordinate, background intensity, modulation degree, initial phase, binary fringe frequency, axial shift step size, and phase shift step number are obtained respectively. The binary fringe frequency is multiplied by the pixel coordinates, and the phase shift determined by the axial movement step size and the phase shift step size is superimposed. The superposition result is then subjected to a cosine operation to obtain the fringe phase cosine value. The phase cosine values ​​of the stripes are binarized to obtain a binary stripe signal; The modulation index and the binary fringe signal are scaled and then superimposed on the background intensity. The ratio of the reflectivity to the square of the system magnification is then used to calculate complementary binary fringes. The complementary binary fringes at adjacent axial positions have a two-step phase shift, and the fringe patterns at adjacent positions are complementary.

[0026] In this embodiment, the fringe projection mode is set to Temporal Phase-Shifting Binary Fringe. Binary fringes refer to a fringe pattern where the grayscale value of each pixel only takes two discrete values: 0 and 255. Compared to traditional sinusoidal fringes, binary fringes have advantages such as high projection rate and are not affected by the gamma nonlinearity of the projection device, making them suitable for high-speed structural illumination micro-measurement scenarios. Temporal phase shift refers to the phase of the fringe pattern increasing sequentially as the measured sample moves gradually in the axial direction, thus achieving a systematic phase shift in the time dimension. After setting the fringe projection mode, the system magnification is obtained. M Pixel coordinates (x, y) reflectivity at R(x, y) Background intensity a(x, y) Adjustment system b(x, y) Initial phase φ Binary fringe frequency Axial movement step size and the number of phase shift steps N Among them, the system amplification factor M Characterizes the magnification ratio of the microscope objective on the sample under test; reflectivity. R(x, y) Describes the reflectivity of incident light at various locations on the surface of the sample under test; background intensity. a(x, y) For a uniform background light intensity component that does not contain stripe modulation information; modulation degree b(x, y) Characterizing the contrast intensity of structured light fringes; initial phase φ The initial phase offset of the stripe pattern; binary stripe frequency. Determines the spatial period of the stripe pattern in the horizontal direction; axial movement step size The distance the electrically controlled displacement stage moves between adjacent axial positions; the number of phase shift steps. N The number of steps required to complete one full 2π phase cycle. In this embodiment, the number of axial movement steps is defined. It is an integer representing the number of steps the sample moves along the axial direction.

[0027] Furthermore, the binary fringe frequency With the pixel coordinates x Perform a product operation and add the product of the axial movement step size. With the number of phase shift steps N Given a specific phase shift, a cosine operation is performed on the superposition result to obtain the fringe phase cosine value. That is, the binary fringe frequency is used as the basis for the calculation. With pixel coordinates x product As the basic spatial phase of the stripes, the number of axial movement steps i and phase shift steps N Determined phase shift As the temporal phase shift component that varies with axial position, the above two terms are superimposed and then subjected to cosine operation to obtain the stripe phase cosine value at the current axial position of the pixel coordinate. The cosine value comprehensively reflects the spatial periodic distribution characteristics of the fringe pattern and the temporal phase shift effect introduced during axial scanning. The fringe phase cosine value is then binarized to obtain a binary fringe signal. A sign function can be used. The cosine values ​​of the stripe phase are binarized: +1 is output when the cosine value is greater than or equal to zero, and -1 is output when the cosine value is less than zero. This converts the continuous cosine modulation signal into a binary stripe signal with only two discrete values. Since the binary stripe signal contains only two states, +1 and -1, they correspond to two pixel values, 255 and 0, after mapping to the image grayscale space. Therefore, the resulting binary stripe signal meets the definition requirements of binary stripes.

[0028] Finally, the aforementioned modulation system b(x, y) After scaling the binary stripe signal, it is compared with the background intensity. a(x, y) Superimpose the values, and then combine them with the reflectivity. R(x, y) With the system magnification M The ratio of the squares of the two elements is used to calculate the complementary binary fringes. The intensity distribution expression of the complementary binary fringes loaded on a digital micromirror device (DMD) is as follows: In the above expression, modulation b(x, y) Half of the binary stripe signal The product of these values ​​constitutes the fringe modulation component, which is then modulated by the background intensity using the "±" operator. a(x, y) The results are superimposed to form a stripe intensity distribution containing both positive and negative modulation polarities; then the superposition result is multiplied by the reflectivity. R(x, y) With system amplification factor MThe ratio of the squares To compensate for differences in the surface reflectance of the tested samples and the influence of the magnification of the microscope system on image intensity, complementary binary fringes were ultimately obtained. In this embodiment, the specific values ​​for each parameter are: number of phase shift steps. Binary fringe frequency System magnification Axial movement step size .

[0029] Due to the number of phase shift steps Only two phase shifts are needed to complete a full 2π phase cycle, that is, the phase shift amount in each step is... Therefore, when the sample being tested moves from its current axial position to an adjacent axial position, the phase of the projected fringes shifts by exactly π radians, causing the projected binary fringes at adjacent axial positions to be out of phase in terms of brightness distribution. Since the fringes are binary, each pixel has only two values: 0 and 255. Therefore, after the phase shift of π, the values ​​of each pixel are exactly flipped, and the two fringes at adjacent positions are complementary patterns. This characteristic of complementary binary fringes at adjacent axial positions, with two-step phase shifts and complementary fringes at adjacent positions, provides data support for subsequent layer-by-layer difference operations to eliminate background intensity terms and extract modulation information.

[0030] In step S102, since the structured light fringe images acquired by the system contain mixed information from multiple spectral response bands, in order to fully utilize the differentiated response characteristics of different wavelength light sources on the surface of the sample under test, the multiple modulation image stacks obtained in step S101 need to be separated and extracted according to multiple preset spectral channels. Based on the spectral distribution characteristics of the system light source and the multispectral response channel configuration of the camera sensor, the pixel information in each modulation image is split and classified according to the corresponding spectral channel, and the modulation image under each spectral channel is extracted separately, thus obtaining a multi-channel modulation image stack. This multi-channel separation processing enables subsequent steps to perform targeted correction and fusion based on the independent modulation information of each channel, laying the foundation for improving measurement accuracy and dynamic range.

[0031] In one embodiment, step S102 includes: Two sets of original image stacks are obtained, and adjacent layer difference operations are performed on the two sets of original image stacks to obtain the modulation image stack; For the modulation images in the modulation image stack, the intensity values ​​of adjacent original images at the same pixel coordinates are subtracted and the absolute value is taken to obtain the modulation value at the corresponding pixel coordinates. The system intensity point spread function is obtained, and the modulation value is convolved with the system intensity point spread function to obtain the modified modulation image at each pixel coordinate of each layer.

[0032] In this embodiment, after axial scanning and simultaneous dual-exposure acquisition in step S101, two sets of original image stacks corresponding to the low-exposure time and the high-exposure time are obtained respectively. Since the system uses a color camera for image acquisition, each frame of the original image contains red... (Red, R) ,green (Green, G) ,blue (Blue, B) The spectral response information of the three pixel channels. Therefore, the images in the two sets of original image stacks are separated and extracted according to the red, green, and blue three pixel channels respectively, resulting in six sets of modulation image stacks, which are respectively the red channel under high exposure. Green Channel under High Exposure Blue channel under high exposure Red channel under low exposure Green Channel under Low Exposure and the blue channel under low exposure ,in p The pixel coordinates are represented. The six sets of modulation image stacks mentioned above correspond to different combinations of exposure conditions and different spectral channels, enabling subsequent processing to fully utilize the differentiated response characteristics of each channel under different exposure conditions, providing multi-dimensional modulation information for achieving high dynamic range measurement.

[0033] Furthermore, assume that the original image stack contains a total of i The layer performs adjacent-layer difference operations on the original image stack of each channel to obtain the corresponding layer for each channel. i- A stack of modulation images with 1 layer. For the modulation image of the z-th layer... B (x, y, z) The calculation method is to take the first... z +1 floor and the z The original image of the layer has the same pixel coordinates. (x, y) The strength values ​​at each point are subtracted and their absolute values ​​are taken, i.e.: Substituting the complementary binary fringe intensity distribution expression established in step S101 into the above formula, since the fringe patterns projected from adjacent axial positions have a π phase shift and constitute complementary patterns, the difference operation can effectively eliminate the common background intensity term in the two frames of images. a(x, y) Retain relevant information related to the stripe tone system. After simplification, the first... z The modulation plot of a layer can be represented as: In the above expression, R(x, y) The reflectance at each pixel coordinate on the surface of the sample being tested. M This is the system amplification factor. b(x, y) In order to adjust the system, For binary fringe frequencies,N For the number of phase shift steps, i This represents the number of steps of axial movement. For symbolic functions, the symbol " "Indicates convolution operation, Let be the system intensity point spread function (IPSF). This expression indicates that the modulation image... B(x, y, z) The modulation spatial distribution characteristics of the sample surface at the current axial position are accurately reflected by the convolution operation of the stripe modulation component after scaling by reflectivity and system magnification with the system intensity point spread function.

[0034] Finally, the system intensity point spread function The intensity response distribution characteristics of a microscopic imaging system to a point light source are described, characterizing the influence of diffraction and defocusing effects of the system's optical elements on image quality. In this embodiment, the system's coherence transfer function... h(x, y, z) The specific form is as follows: in, z The defocus distance represents the offset of the current axial position of the sample being measured relative to the focal plane of the system. λ The wavelength of the lighting source; α The half-aperture angle of a microscope objective describes the objective's maximum ability to converge light. w The system aperture size defines the spatial range of the effective imaging area. i The unit is the imaginary number. In this embodiment, the specific value of the above parameter can be: half-aperture angle. System aperture size System strength point spread function This is the coherent transfer function mentioned above. h(x, y, z) The square of the modulus value, in its physical sense, quantifies the intensity distribution of a point light source after passing through a microscopic imaging system. In the calculation of the modulation image, the convolution operation between the fringe modulation component and the system intensity point spread function reflects the optical transmission effect of the microscopic imaging system on the structured light fringe signal. This ensures that the modulation values ​​at each pixel coordinate in each layer accurately reflect the actual influence of the system's optical characteristics on the spatial distribution of modulation, thus obtaining the modulation image at each pixel coordinate in each layer corrected by the system's optical transmission effect—the corrected modulation image. This corrected modulation image provides more accurate input data for subsequent correction and global optimization selection of multi-channel modulation curves.

[0035] In step S103, after acquiring the multi-channel modulation image stack, modulation correction processing is performed on multiple modulation curves corresponding to each pixel along the axial direction to eliminate modulation deviations caused by differences in system optical characteristics and noise interference. Based on this, a global optimization selection is performed on the corrected multiple modulation curves, that is, by comprehensively comparing the modulation response intensity of each spectral channel at the pixel, the modulation curve with the best signal quality is selected as the optimal modulation curve for that pixel. Then, peak fitting and positioning are performed on the optimal modulation curve. By extracting data points near the peak and performing curve fitting, the peak position of the fitted curve is used to determine the focal position of the pixel, and the axial coordinates corresponding to the focal position are used as the relative height value of the pixel.

[0036] In one embodiment, step S103 includes: The modulation of the low signal-to-noise ratio column pixels in the multi-channel modulation image stack is corrected based on the continuity of surface reflection characteristics to obtain the corrected multi-channel modulation image stack. For each pixel, multiple modulation curves corresponding to the pixel are extracted from the corrected multi-channel modulation image stack, and low-pass filtering is performed on the multiple modulation curves to obtain filtered multiple modulation curves. Multi-channel modulation peak offset correction is performed on the filtered multiple modulation curves to obtain offset-corrected multiple modulation curves; Compare the peak values ​​of the multiple modulation curves after offset correction, and take the curve with the largest peak value as the optimal modulation curve for the pixel. Data points near the peak of the optimal modulation curve are extracted and fitted with Gaussian curves. The horizontal coordinate of the peak of the fitted curve is used as the relative height value of the pixel.

[0037] In this embodiment, in the stack of six modulation images obtained after spectral channel separation and extraction in step S102, due to the spatial periodic distribution characteristics of the structured light stripes, some pixel columns in the stripe pattern are located precisely at the node positions of the stripe brightness transition region, that is, pixel columns where the stripe phase is an odd multiple of half the period. The stripe modulation values ​​at these pixel columns are close to zero, resulting in a significantly lower signal-to-noise ratio (SNR) than other pixel columns, forming low SNR pixel columns. The modulation values ​​of low SNR pixel columns are severely affected by noise interference, and if not corrected, it will adversely affect the subsequent optimal fusion of modulation curves and peak fitting and positioning. Given that the reflectivity characteristics of the sample surface are generally continuous in space, meaning that the changes in reflectivity and modulation intensity between adjacent pixels have a smooth transition, an average interpolation can be performed using the modulation intensity values ​​of the two adjacent columns of pixels in the same row corresponding to the low signal-to-noise ratio (SNR) column pixels. This interpolated value then replaces the original modulation intensity value of the low SNR column pixels, thus effectively correcting the modulation intensity of the low SNR column pixels. After performing this correction operation on each of the six modulation intensity image stacks, a corrected multi-channel modulation intensity image stack is obtained.

[0038] Furthermore, after correcting the modulation of low signal-to-noise ratio (SNR) pixels, for each pixel on the surface of the sample under test, the modulation values ​​of each layer along the axial direction are extracted from the stack of the six corrected modulation images. These values ​​are then arranged in axial order to form six modulation curves corresponding to that pixel. These six modulation curves correspond to the red channel under high exposure. Green Channel under High Exposure Blue channel under high exposure Red channel under low exposure Green Channel under Low Exposure and the blue channel under low exposure The modulation index of the six channels varies with axial position. Due to system noise and environmental interference during actual acquisition, high-frequency noise components are superimposed on the modulation index curves, affecting the accuracy of subsequent peak localization. Therefore, low-pass filtering is required to remove noise from the six modulation index curves. Specifically, a two-dimensional Fourier transform is performed on the six modulation index curves to transform them from the spatial domain to the frequency domain, obtaining the spectrum corresponding to each modulation index curve. A one-dimensional rectangular window function is applied to the central region of the frequency domain for filtering. The width of this rectangular window function is set to one-tenth of the total spectrum width and is centrally symmetrically distributed in the central region of the spectrum to retain the effective low-frequency signal components and suppress high-frequency noise components. An inverse two-dimensional Fourier transform is performed on the filtered spectrum, and the modulus value is taken to recover the filtered modulation index distribution, thus extracting the six filtered modulation index curves. After the above low-pass filtering process, the high-frequency noise components of each modulation curve are effectively suppressed, and the curve shape is smoother, providing a data basis for subsequent peak offset correction and peak fitting positioning.

[0039] Due to the axial chromatic aberration effect of microscope objectives—that is, light of different wavelengths has different focal lengths after refraction through the microscope objective—the focal plane positions of the red, green, and blue spectral channels differ along the axial direction. This difference in focal plane position is reflected in the modulation curves, manifesting as a shift in the peak positions of the modulation curves of different spectral channels along the axial direction. Without correction, accurate peak comparison and optimal fusion of the modulation curves between different channels cannot be performed under a unified axial coordinate system. Therefore, the numerical aperture (NA) of the microscope objective and the reference wavelengths corresponding to each set of spectral channels are obtained. For each set of spectral channels, the difference between a preset wavelength (usually a reference wavelength of a benchmark channel) and the reference wavelength of the channel to be corrected is calculated. This difference is then divided by the square of the numerical aperture to obtain the focal length deviation of each channel. This focal length deviation is the axial position shift of the focal plane relative to the reference channel's focal plane caused by the axial chromatic aberration effect. Based on the focal length deviation of each channel, the modulation curves of the corresponding channels in the six filtered modulation curves are axially offset and corrected. That is, the modulation curves of each channel are shifted along the axial direction by the corresponding focal length deviation, so that the modulation curves of all channels are aligned to a unified axial coordinate reference, resulting in six modulation curves after offset correction.

[0040] Furthermore, for each pixel, six modulation curves are obtained after offset correction across all effective channels, where the effective channels include a high-exposure red channel, a high-exposure green channel, a high-exposure blue channel, a low-exposure red channel, a low-exposure green channel, and a low-exposure blue channel. The peak value of the modulation curve corresponding to each effective channel is extracted, and the peak values ​​of all effective channels are compared to determine the effective channel with the largest peak value. The modulation curve corresponding to the effective channel with the largest peak value is taken as the optimal modulation curve for that pixel. The principle behind selecting the modulation curve with the largest peak value as the optimal modulation curve is that a larger peak value indicates a stronger fringe modulation signal and a higher signal-to-noise ratio at the current pixel, thus the peak position of its modulation curve can more accurately reflect the true focus position of the pixel. By selecting the best among the six channels, pixels in different reflective characteristic regions can each choose the channel with the best signal quality for height positioning, thereby achieving reliable extraction of the relative height value of each pixel under high dynamic range conditions.

[0041] Finally, Gaussian fitting is performed on the data points near the peak of the optimal modulation curve, and the abscissa of the peak of the fitted curve is used as the relative height value of the pixel. After determining the optimal modulation curve for each pixel, numerical analysis is performed on the optimal modulation curve to locate the position of the maximum modulation value, and several data points near this position are extracted as data points near the peak. Subsequently, Gaussian fitting is performed on the data points near the peak, that is, the least squares curve is fitted to the extracted data points using a Gaussian function to obtain a continuous Gaussian fitted curve. Since the Gaussian function has good symmetry and single-peak characteristics, its fitting result can determine the peak position of the modulation curve with sub-pixel accuracy. The focal position of the pixel is located using the abscissa of the peak of the fitted Gaussian curve, that is, the axial coordinate corresponding to the maximum value of the fringe modulation at the pixel, and the axial coordinate corresponding to the focal position is used as the relative height value of the pixel. By repeating the above step S103 process on all pixels on the surface of the sample, the relative height values ​​of all pixels can be obtained.

[0042] Combination Figure 2 As shown, in one embodiment, the step of correcting the modulation of low signal-to-noise ratio column pixels in the multi-channel modulation image stack based on the continuity of surface reflection characteristics to obtain a corrected multi-channel modulation image stack includes: Obtain the corresponding stripe phase in the multi-channel modulation image stack, and take the pixel column where the stripe phase is an odd multiple of half the period as the low signal-to-noise ratio column pixel; For each of the low signal-to-noise ratio column pixels, the modulation values ​​of the two adjacent columns of pixels in the corresponding row are obtained, and the average interpolation calculation is performed on the modulation values ​​of the two adjacent columns of pixels to obtain the interpolated and corrected modulation values ​​of the low signal-to-noise ratio column pixels. The interpolated modulation value is replaced with the original modulation value of the corresponding low signal-to-noise ratio column pixel in the multi-channel modulation image stack to obtain the corrected multi-channel modulation image stack.

[0043] In this embodiment, as Figure 2 As shown on the left, in the spatial distribution of structured light fringes, the fringe phase at each pixel coordinate is determined by the binary fringe frequency. Pixel coordinates x , number of axial movement steps i and phase shift steps N This is decided jointly. Regarding pixel coordinates... (x, y) The corresponding fringe phase When the phase value satisfies the condition: When k is an odd number, this pixel column is located precisely at a node position in the transition region between light and dark areas of the binary fringe pattern. At these node positions, the positive and negative half-cycles of the fringe signal switch, causing the fringe modulation value of this pixel column to approach zero, thus making the signal-to-noise ratio (SNR) at this pixel column significantly lower than that of a normal pixel column. Figure 2 As shown in the schematic diagram on the left, the binary fringe pattern exhibits a rectangular wave distribution with alternating bright and dark areas. At pixel columns where the fringe phase meets the aforementioned conditions, the fringe signal intensity attenuates sharply, forming dark areas with near-zero modulation. These pixel columns with fringe phases that are odd multiples of half the period are labeled as low signal-to-noise ratio (SNR) columns. Since the modulation values ​​at low SNR columns are mainly composed of noise components and lack effective topographic modulation information, directly involving them in subsequent modulation curve analysis and peak fitting localization will severely affect the accuracy of extracting the relative height value at these pixel columns. Figure 2 As shown on the right, in the modulation curve distribution of the six spectral channels, the modulation curve of the low signal-to-noise ratio column pixels exhibits significant noise fluctuation characteristics. Its envelope shape is significantly different from the smooth modulation curve of the adjacent normal pixel column, further confirming the necessity of correcting the low signal-to-noise ratio column pixels.

[0044] Based on the physical assumption that the surface reflectance characteristics of the tested sample are spatially continuous—that is, the reflectance and modulation changes between adjacent pixels exhibit a smooth transition—the true modulation value at a low signal-to-noise ratio pixel column can be approximately estimated from the modulation value of its adjacent normal pixel column. For coordinates... (x, y) For the low signal-to-noise ratio column pixels, obtain the left adjacent column pixels of their corresponding row. (x- 1 , y)and the adjacent column of pixels on the right (x+ 1 , y) The moderating value m(x- 1 , y) and m(x+ 1 , y) The interpolated modulation value of the low signal-to-noise ratio column pixels is calculated according to the following mean interpolation formula: in, m̃(x, y) Indicates coordinates as (x, y) The modulation values ​​of the pixels after interpolation correction. m(x- 1 , y) and m(x+ 1 , y) These are the modulation values ​​of the pixels in the two adjacent columns of the same row. This mean interpolation method utilizes the prior knowledge of the continuity of surface reflection characteristics, estimating the modulation value of the low signal-to-noise ratio column pixels by the arithmetic mean of the modulation values ​​of the normal pixel columns on both sides. This achieves a reasonable reconstruction of the modulation of the low signal-to-noise ratio column pixels with low computational complexity. It should be noted that after the above interpolation correction, the axial grayscale curve of the low signal-to-noise ratio column pixels becomes: in, a For pixels (x, y) Background light intensity at that location R For pixels (x, y) Reflectivity at that location M The system magnification is denoted by . This expression indicates that the axial grayscale curve of the low signal-to-noise ratio (SNR) column pixels after interpolation correction contains only the DC component determined by the background light intensity, reflectivity, and system magnification. Its fringe modulation information has been obtained from the adjacent normal pixel column through interpolation and replaced the original noise-dominated modulation value, thereby eliminating the adverse effects of noise interference at the low SNR column pixels on the modulation curve shape.

[0045] Finally, the above-described identification and interpolation correction operations for low signal-to-noise ratio (SNR) columns of pixels are performed on each of the six modulation index image stacks, and the corrected modulation index value at each low SNR column pixel is calculated by mean interpolation. m̃(x, y) The original modulation values ​​at corresponding positions in the multi-channel modulation image stack are replaced one by one to obtain the corrected multi-channel modulation image stack. After correction based on the continuity of surface reflection characteristics, the noise fluctuations of the modulation curves of each channel are effectively suppressed, and the curve shape tends to be smooth and continuous. This provides a higher quality modulation data foundation for subsequent steps such as low-pass filtering, multi-channel modulation peak shift correction, and optimal fusion and Gaussian fitting localization.

[0046] In one embodiment, the step of performing low-pass filtering on the multiple modulation index curves to obtain filtered multiple modulation index curves includes: Perform a one-dimensional Fourier transform on each of the modulation curves to obtain the spectrum corresponding to each modulation curve; After center-shifting the spectrum, a one-dimensional rectangular window function is applied to the central region of the spectrum for filtering to obtain the filtered spectrum; wherein the width of the one-dimensional rectangular window function is one-tenth of the total width of the spectrum, and it is centrally symmetrically distributed in the central region of the spectrum; A one-dimensional inverse Fourier transform is performed on the filtered spectrum and the modulus value is taken to obtain the filtered modulation distribution, so as to extract multiple modulation curves after filtering.

[0047] In this embodiment, after the low signal-to-noise ratio pixel correction based on the continuity of surface reflectivity in the aforementioned steps, although the modulation curves of each pixel in the six spectral channels have eliminated the outlier interference caused by the fringe node positions, they still inevitably have high-frequency noise components introduced by factors such as system electronic noise, environmental vibration, light source fluctuations, and sensor quantization errors. These high-frequency noise components cause the modulation curves to exhibit irregular rapid oscillations, seriously affecting the accurate positioning of subsequent peak positions. To effectively remove the above-mentioned high-frequency noise and retain the low-frequency effective signal components reflecting the surface morphology information of the sample under test, the modulation curves need to be low-pass filtered. In this embodiment, the one-dimensional modulation distribution data composed of multiple modulation curves corresponding to each pixel is used as input, and a one-dimensional Fourier transform is performed on it to transform the modulation information from the spatial domain to the frequency domain. After the one-dimensional Fourier transform, the low-frequency components in the modulation distribution are concentrated in the central region of the spectrum, while the high-frequency noise components are dispersed in the edge region of the spectrum, thereby achieving effective separation of effective signals and noise in the frequency domain space, providing a basis for subsequent frequency domain filtering operations.

[0048] Furthermore, a one-dimensional rectangular window function is an ideal low-pass filter with a clear cutoff boundary in the frequency domain. It has a gain of 1 in the passband and a gain of 0 in the stopband, effectively preserving frequency components within the passband while completely suppressing them within the stopband. In this embodiment, the width of the one-dimensional rectangular window function is set to one-tenth of the total spectrum width, and it is centrally symmetrically distributed in the central region of the spectrum. This parameter setting is based on the fact that the modulation curve changes corresponding to the surface morphology information of the tested sample are usually low-frequency signals, with their effective frequency components mainly concentrated in a narrow frequency band near the center of the spectrum, while the high-frequency components corresponding to system noise and random interference are distributed in high-frequency regions far from the center of the spectrum. Setting the width of the one-dimensional rectangular window function to one-tenth of the total spectrum width effectively preserves the low-frequency modulation information related to the sample morphology while effectively suppressing most high-frequency noise components, achieving a good balance between filtering effect and information preservation. The symmetrical distribution of the one-dimensional rectangular window function in the central region of the spectrum ensures the symmetry of the filtering operation in the frequency domain, avoiding the introduction of additional phase distortion or spectral distortion due to the asymmetrical configuration of the filter. By multiplying the one-dimensional rectangular window function element by element with the spectral data, the high-frequency components outside the central region of the spectrum can be set to zero, retaining only the low-frequency effective components within the central region, thus obtaining the filtered spectrum.

[0049] Finally, the spectral data filtered by the one-dimensional rectangular window function is transformed back from the frequency domain to the spatial domain. Since the inverse transform result may contain tiny imaginary components due to numerical computation precision, the modulus of the inverse transform result is taken to obtain the real-valued modulation distribution data. After the above one-dimensional inverse Fourier transform and modulus taking, the filtered modulation distribution is obtained. From this filtered modulation distribution, the modulation values ​​of each layer along the axial direction are extracted according to the coordinates of each pixel point, resulting in multiple filtered modulation curves. After the above low-pass filtering, the high-frequency noise components in each modulation curve are effectively removed, the curve shape is smoother and more continuous, and the modulation peak position is more clearly distinguishable, providing high-quality modulation data input for multi-channel modulation peak offset correction and global optimization selection in subsequent steps.

[0050] Combination Figure 3 As shown, in one embodiment, the step of performing multi-channel modulation peak offset correction on the filtered multiple modulation curves to obtain offset-corrected multiple modulation curves includes: Obtain the numerical aperture of the microscope objective and the reference wavelength corresponding to each spectral channel; For each spectral channel, the difference between the preset wavelength and the reference wavelength of the channel to be corrected is calculated, and the difference is divided by the square of the numerical aperture to obtain the focal length deviation of each channel. Based on the focal length deviation of each channel, the modulation curve of the corresponding channel in the filtered multiple modulation curves is axially offset to obtain the offset-corrected multiple modulation curves.

[0051] In this embodiment, due to the axial chromatic aberration effect of the microscope objective, light of different wavelengths has different focal lengths after refraction by the objective, resulting in differences in the focal plane positions of the red, green, and blue spectral channels in the axial direction. For example... Figure 3 As shown on the left, before correction, the peak positions of the modulation curves for each spectral channel exhibit a significant mutual offset along the axial direction z. The peaks of curves of different colors are not aligned to the same axial coordinate. This peak offset is caused by the axial chromatic aberration effect of the microscope objective. To quantitatively correct the above peak offset, it is necessary to obtain the numerical aperture (NA) of the microscope objective and the reference wavelength corresponding to each group of spectral channels. In this embodiment, the objective used in the experiment is an OLYMPUS 20x semi-apochromatic objective, whose numerical aperture... The reference wavelengths corresponding to each spectral channel are as follows: Red channel reference wavelength Green channel reference wavelength Blue channel reference wavelength = 479.99nm. The above reference wavelength is the center wavelength with the highest response sensitivity of each spectral channel, which is determined by the spectral response characteristics of the camera sensor and the spectral distribution of the system light source.

[0052] Then, the green pixel channel is selected as the reference channel, using the reference wavelength of the green channel. As a preset wavelength, focal length deviations were calculated for both the red and blue channels. Focal length deviations for each channel. The calculation formula is: in, The preset wavelength, i.e., the reference wavelength of the reference channel. ; The reference wavelength for the channel to be corrected; NA This is the numerical aperture of the microscope objective. The formula is derived from the dispersion theory of lenses in geometric optics. Its physical meaning is that light of different wavelengths has different focal lengths after refraction through the microscope objective. The difference in focal length between two wavelengths is directly proportional to the wavelength difference and inversely proportional to the square of the numerical aperture. Substituting the above parameters into the formula, the specific values ​​of the focal length deviation between each channel can be obtained: the focal length deviation between the red channel and the blue channel. Focal length deviation between the red and green channels Focal length deviation between the green channel and the blue channel The aforementioned focal length deviation quantifies the relative position shift of the focal plane in each spectral channel due to axial chromatic aberration, providing a theoretical basis for subsequent axial shift correction.

[0053] Furthermore, in actual measurement, the focal length deviation is also slightly affected by the optical properties of the sample surface material. Therefore, there is a certain residual deviation between the focal length deviation value calculated solely based on the theoretical formula and the actual measurement result. Thus, a systematic, small correction factor is introduced. This is to compensate for the residual deviation between theoretical and experimental results. Correction amount The determination is based on a rigorous pre-calibration process: under the same system conditions, multiple calibration samples of known similar materials are measured using each channel. The measured axial offset between channels is compared with the theoretical value calculated by the above theoretical formula, and the statistical average of the differences is the result. Extensive calibration experiments have confirmed that, for a fixed imaging system and samples of similar materials, the axial offset relationship between channels is stable and repeatable, which ensures... Reliability is a constant quantity in the system.

[0054] Finally, using the green pixel channel as the reference channel and the peak position of its modulation curve as the axial reference, the modulation curves of the red and blue channels are corrected for axial offset. For non-reference channels (red and blue channels), their height is corrected using the following formula: in, For channel c The original height value, For channel c The corrected height value, k A scaling factor related to objective lens optical design. The reference wavelength for the reference channel. The reference wavelength for the channel to be corrected. NA Numerical aperture, This represents the systemic minor correction amount corresponding to this channel. In this embodiment, the specific values ​​of each parameter are: proportional coefficient. Red channel correction amount Blue channel correction amount By using the above correction formula, the modulation curves of each non-reference channel are shifted axially by the sum of the focal length deviation and the system correction, so that the modulation curves of all spectral channels are aligned to a unified axial coordinate system based on the green channel. For example... Figure 3As shown on the right, after multi-channel modulation peak shift correction, the peak positions of the modulation curves of each spectral channel are accurately aligned to the same axial coordinate, eliminating the peak shift between channels caused by axial chromatic aberration. The envelope shape of each channel curve is clearly distinguishable, and the consistency of peak positions is significantly improved. After the above shift correction process, multiple modulation curves with shift correction are obtained, providing a data basis for accurate comparison of the peak values ​​of each channel's modulation curve and selection of the optimal modulation curve in subsequent steps.

[0055] In one embodiment, comparing the peak values ​​of the multiple modulation curves after offset correction and selecting the curve with the largest peak value as the optimal modulation curve for the pixel includes: For each pixel, obtain multiple modulation curves of the pixel after offset correction under all effective channels; wherein, the effective channels include high exposure red channel, high exposure green channel, high exposure blue channel, low exposure red channel, low exposure green channel and low exposure blue channel; The peak value of the modulation curve corresponding to each effective channel is extracted, and the peak values ​​of all effective channels are compared to determine the effective channel with the largest peak value. The modulation curve corresponding to the effective channel with the largest peak value is taken as the optimal modulation curve for the corresponding pixel.

[0056] In this embodiment, after the aforementioned steps of low signal-to-noise ratio pixel correction, low-pass filtering, and multi-channel modulation peak shift correction, the modulation curves of each pixel in the six spectral channels have achieved noise suppression and axial coordinate alignment, thus meeting the conditions for inter-channel comparison and optimal selection under a unified axial reference. This applies to each pixel on the surface of the sample under test. p The modulation curves of the pixel after offset correction are obtained in all effective channels. The effective channels are... C It is a collection of six spectral response channels, specifically including: a high-exposure red channel. (HR) High-exposure green channel (HG) High-exposure blue channel (HB) Low-exposure red channel (LR) Low-exposure green channel (LG) and low-exposure blue channel (LB) The six effective channels described above consist of a combination of two exposure conditions (high exposure and low exposure) and three spectral channels (red, green, and blue), covering all modulation information acquired by the system at different exposure times and spectral response bands on the surface of the sample under test. Each effective channel corresponds to a modulation curve after offset correction. ,in C Indicates a valid channel identifier. pRepresents pixel coordinates. z This curve represents the axial position coordinates and describes the pixel point. p In effective channels C The modulation value varies with axial position. z The changing relationship.

[0057] Then, based on the positive correlation between signal strength and signal-to-noise ratio (SNR), that is, the larger the peak value of the modulation curve, the stronger the fringe modulation signal acquired by that channel at the current pixel, the higher its SNR, and the more accurate the indication of the true focus position of that pixel. Therefore, for each pixel... p The modulation curves of each of the six effective channels were analyzed. Numerical analysis was performed to extract the curves along the axial direction. z The maximum modulation value, i.e., the peak value of each modulation curve. Subsequently, the peak values ​​extracted from the six effective channels are compared one by one to determine the effective channel with the largest peak value. The optimal fusion process can be expressed as: in, Representing pixels p The optimal modulation curve, C Represents all valid channels (i.e.) HR, H- G, HB, LR, LG, LB The set of ) For channel C At pixel p The adjustment curve at the location, For channel C At pixel p The `arg max` operation is used to select the modulation curve corresponding to the channel that maximizes the peak value from all effective channels. The core principle of this optimal fusion strategy is that for areas with high reflectivity on the sample surface, images acquired under high exposure conditions may experience signal saturation, suppressing the peak value of the modulation curve in the high-exposure channel. In this case, the low-exposure channel can provide unsaturated effective modulation information, with a larger peak value and higher signal-to-noise ratio. Conversely, for areas with low reflectivity, the signal strength acquired under low exposure conditions is insufficient, resulting in a larger peak value in the modulation curve of the high-exposure channel, providing more reliable focus position information. Furthermore, different spectral channels exhibit different reflection response characteristics on different material surfaces; some materials may show stronger modulation responses under specific spectral channels. By selecting the modulation curve with the largest peak value from the six effective channels, the system can adaptively select the channel with the best signal quality for each pixel, fully utilizing the complementary advantages of multiple exposure conditions and multiple spectral channels to effectively expand the system's dynamic range.

[0058] Finally, based on the comparison results above, for each pixel... p The modulation curve corresponding to the peak maximum effective channel determined through the optimal fusion process. This serves as the optimal modulation curve for that pixel. This optimal modulation curve has the highest peak value among the modulation curves of the six effective channels, representing the modulation response with the best signal quality and highest signal-to-noise ratio at that pixel. Its peak position most accurately reflects the true focus position of that pixel. By performing the above global optimization selection on all pixels on the surface of the test sample, each pixel obtained its own optimal modulation curve, thus enabling pixels in different regions on the complex industrial sample surface with non-uniform reflectivity to adaptively select the modulation information under the best combination of channels and exposure conditions. This provides high-quality input data for subsequent Gaussian fitting of data points near the peak and accurate positioning of relative height values.

[0059] In one embodiment, the step of extracting data points near the peak of the optimal modulation curve and performing Gaussian fitting, using the abscissa of the peak of the fitted curve as the relative height value of the pixel, includes: Obtain the axial grayscale curve corresponding to each pixel in the modulation image, and perform differential operation decoding on the axial grayscale curves of adjacent axial positions to obtain the axial modulation curve corresponding to each pixel. Numerical analysis is performed on the optimal modulation curve based on the axial modulation curve to extract data points near the peak value at the location of the maximum modulation value. Gaussian fitting is performed on the data points near the peak, and the focal position of the pixel is located by the horizontal coordinate of the peak of the fitted curve. The axial coordinate corresponding to the focal position is used as the relative height value of the pixel.

[0060] In this embodiment, based on the principle of determining the relative height of an object point by locating the peak value of the modulation curve, for any pixel in the modulation image... Its axial grayscale curve This describes the image intensity value at that pixel as a function of axial position. z The relationship between the changes. Combined with the aforementioned intensity distribution model of complementary binary fringes, this axial grayscale curve... It can be represented as: Where R is a pixel Reflectivity at that location M The system amplification factor is... a For background intensity, b In order to adjust the system, For binary fringe frequencies, For the initial phase, The expression is a sign function. It shows that the axial grayscale curve is composed of a superposition of a background intensity term and a fringe modulation term, where the fringe modulation term carries effective information related to the surface morphology of the sample being tested. Based on the aforementioned method of solving the modulation scheme using time-domain phase shift, the finite difference method is used to analyze the axial grayscale curve. Decoding involves performing a difference operation on the grayscale values ​​at adjacent axial positions to obtain the axial modulation curve of the pixel. Specifically, regarding axial position The regulation value of the place It passes through two adjacent axial positions and The difference between the gray values ​​at each point and the absolute value are obtained as follows: in, Let "intensity point spread function" be the system intensity function, denoted by "". "This represents the convolution operation. The first term in the above expression is the residual component of the background intensity term after difference, and the second term is the effective signal component containing fringe modulation information. Since the fringe patterns projected from adjacent axial positions are complementary patterns with π phase shifts, the difference operation can effectively suppress the influence of the background intensity term, thus improving the axial modulation curve." It is mainly dominated by stripe modulation signal components.

[0061] Furthermore, regarding the aforementioned axial modulation curve Theoretical analysis can reveal the physical laws governing the variation of modulation intensity with defocus distance. An approximate expansion of the axial modulation intensity curve yields: Based on the Stokes approximation, further simplification yields: in, This is an approximate expression for the optical transfer function, with the defocus distance normalized. Normalized fringe frequency This is a first-order Bessel function of the first kind. The above theoretical analysis shows that when the normalized defocus distance... u When the modulation index is 0, that is, when the surface of the sample being measured is exactly at the focal plane of the system, the modulation index reaches its maximum value; as the defocus distance increases, the modulation index gradually decreases. By performing numerical analysis on the modulation index of each pixel and finding the axial position corresponding to the maximum modulation index, the focal position of that pixel can be located, and this value can be used as the relative height value of that pixel.

[0062] Furthermore, after determining the optimal modulation curve for each pixel through global optimization in the aforementioned steps, a numerical traversal analysis is performed along the axial direction on the optimal modulation curve to locate the axial position where the modulation reaches its maximum value. Since the actual acquired modulation curve is a discrete data point sequence, the axial resolution of the discrete sampling is limited by the axial movement step size of the electronically controlled displacement stage. Directly using discrete maximum points as peak positions only yields axial accuracy at the step level, which is insufficient for high-precision 3D topography measurement. Therefore, after locating the discrete maximum point, several data points near this location are extracted as peak proximity data points for subsequent sub-pixel-level high-precision peak positioning. The extraction range for peak proximity data points is typically selected by choosing several data points on both sides of the discrete maximum point to ensure that the extracted data points fully cover the rising and falling segments of the modulation curve peak region, providing sufficient effective constraint information for subsequent Gaussian fitting.

[0063] Finally, since the modulation curve near its peak approximates a Gaussian function distribution—that is, the modulation value decreases symmetrically in a bell shape around the peak—leaving the data points near the peak using a Gaussian function for least-squares curve fitting can determine the continuous peak positions of the modulation curve with sub-pixel accuracy. The Gaussian fitting process constructs a Gaussian function model, iteratively solving for parameters such as the peak position, peak amplitude, and half-maximum width of the Gaussian function with the optimization objective of minimizing the sum of squared residuals between the fitted curve and the actual data points. After fitting, the abscissa of the peak of the resulting Gaussian curve is the precise axial coordinate corresponding to the maximum modulation value at that pixel. The physical meaning of this axial coordinate is the axial position of the sample surface at that pixel when it is exactly located on the system's focal plane, i.e., the focal position of that pixel. The axial coordinate corresponding to this focal position is used as the relative height value of the pixel, which characterizes the axial height information of the sample surface at that pixel with sub-pixel accuracy. By performing the above-mentioned peak-near data point extraction and Gaussian fitting processing on all pixels on the surface of the sample, the relative height values ​​of all pixels can be obtained, providing complete and high-precision height information for the subsequent step S104 to aggregate and generate the three-dimensional topography point cloud data of the sample.

[0064] In step S104, after calculating the relative height values ​​of all pixels, the one-dimensional planar coordinates of each pixel are correlated and converged with their corresponding relative height values ​​to construct the three-dimensional spatial coordinate information of each measurement point on the surface of the sample under test. By integrating the three-dimensional coordinate data of all pixels, complete point cloud data characterizing the three-dimensional morphology of the sample surface can be generated. This point cloud data can comprehensively reflect the microstructural features and height distribution information of the sample surface and can serve as the basic data for subsequent applications such as three-dimensional morphology analysis, surface quality assessment, and defect detection.

[0065] Combination Figure 4 and Figure 5 As shown, in order to verify the actual effect of the multispectral microscopic three-dimensional measurement method provided in the embodiments of the present invention, representative industrial samples were selected for experimental testing, specifically including wafer surface crack detection experiment and PCB blind via aspect ratio detection experiment.

[0066] Please see Figure 4 , Figure 4 This is a schematic diagram of the experimental results for detecting surface cracks on wafers provided in an embodiment of the present invention. Surface cracks on wafers are typically caused by mechanical stress or process defects. They exhibit significant tilt angles on both sides of the wafer, and the local reflectivity varies considerably due to differences in the coating material. They are typical test objects with significant spatial non-uniformity in reflectivity, making detection challenging. In this experiment, the cracked region Roi 1 was selected for three-dimensional morphology reconstruction, as shown... Figure 4 As shown in (a). Figure 4 (d) Shows the two single-channel independent reconstruction results (Ⅰ, Ⅱ) with the highest reconstruction rate among the six effective channels, and the reconstruction result (Ⅲ) of the method of this invention, with a comparison of the corresponding point cloud quantity and reconstruction rate. Figure 4As shown in (b), region A corresponds to the high reflectivity area of ​​the sample surface. In this region, the blue channel pixels are overexposed, and the reflected light intensity exceeds the upper limit of the system's dynamic range. This causes the fringe modulation of this channel in region A to be lower than the preset threshold, making effective peak fitting and positioning impossible, resulting in the failure of 3D morphology reconstruction. Region B corresponds to the low reflectivity area of ​​the sample surface. In this region, the image brightness of the green channel is low, especially for low signal-to-noise ratio pixels, which are located at nodes where the fringe phase is an odd multiple of half the period. The modulation value is further reduced, resulting in a lack of longitudinal modulation in the reconstruction result, and the surface exhibits abnormal texture. The above analysis shows that single-channel reconstruction methods all result in missing point cloud data in some areas, making it impossible to achieve complete 3D morphology reconstruction of the sample surface. This invention, by acquiring structured light fringe images at multiple axial positions under different exposure conditions, and then separating and extracting multiple spectral channels, corrects and globally optimizes the multiple modulation curves of each pixel, effectively expanding the system's dynamic range, obtaining higher point cloud density, and achieving a reconstruction rate significantly better than any single-channel reconstruction result, essentially restoring the complete 3D morphology of the cracked area. Crack cross-sectional profile as shown Figure 4 As shown in (c), the crack height difference measured after plane fitting is 4.41 μm, which accurately reflects the height difference information between the two planes on both sides of the crack.

[0067] Please see Figure 5 , Figure 5 This is a schematic diagram illustrating the experimental results of PCB blind via aspect ratio testing provided in an embodiment of the present invention. PCBs (Printed Circuit Boards) have numerous via structures distributed on their surface, mainly including through-holes and blind vias. Blind vias connect the surface layer and the inner layer, and are only visible from one side. When processing defects exist, they can easily cause signal transmission or power supply failures. The metal layer at the top of a blind via and the substrate at the bottom of the via exhibit significant differences in reflectivity, and the ratio of via depth to diameter is large. This makes it a typical test object with significant spatial non-uniformity of reflective characteristics and a complex structure on a macroscopic scale. In this experiment, a blind via sample with a nominal diameter of 0.1 mm and a depth of approximately 2 mm was selected for testing. Figure 5 As shown in (a), the applicability of the method of the present invention in the measurement of macroscopic complex structures is verified. In the independent reconstruction of the six effective channels, the high-exposure blue channel (HB) and the low-exposure green channel (LG) showed the highest reconstruction rates, and their reconstruction results are shown below. Figure 5As shown in Figures II and III of (c), it can be seen that due to the extremely high reflectivity of the metal layer at the top of the blind aperture, severe overexposure occurs in the high-exposure blue channel, significantly reducing the signal-to-noise ratio of the stripe modulation and resulting in poor reconstruction quality in the top region. Conversely, the reflectivity at the bottom of the blind aperture is low, leading to insufficient signal strength in the low-exposure green channel, making it difficult to accurately measure the bottom depth and resulting in a lack of three-dimensional topographic information. In contrast, this invention globally optimizes the modulation curves of each pixel in the six effective channels, adaptively selecting the modulation information of the low-exposure channel for the high-reflectivity region at the top of the aperture to avoid overexposure, and adaptively selecting the modulation information of the high-exposure channel for the low-reflectivity region at the bottom of the aperture to enhance signal strength. This effectively balances the exposure difference between the top and bottom of the aperture, achieving complete reconstruction of the three-dimensional topography of the blind aperture, as shown in Figures II and III. Figure 4 As shown in (d)Ⅰ, the blind hole depth was measured to be approximately 1.932 mm and the diameter to be approximately 0.105 mm, with relative errors of 3.4% and 4.7% respectively compared to the nominal dimensions. This verifies the adaptability and reliability of the method of the present invention for the three-dimensional morphology measurement of large-scale complex structural samples.

[0068] Combination Figure 6 As shown, Figure 6 A schematic block diagram of a multispectral-based three-dimensional microscopic measurement device 600 provided in this embodiment of the invention, comprising: The image acquisition unit 601 is used to acquire structured light stripe images of the sample under test at multiple axial positions under different exposure conditions, and to perform differential operation processing on the structured light stripe images at adjacent axial positions to obtain multiple sets of modulation image stacks. Image extraction unit 602 is used to separate and extract the multiple modulation image stacks according to multiple preset spectral channels to obtain a multi-channel modulation image stack. The image correction unit 603 is used to perform modulation correction and global optimization selection on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack, to obtain the optimal modulation curve for each pixel, and to perform peak fitting and positioning on the optimal modulation curve to obtain the relative height value of each pixel. The image output unit 604 is used to perform convergence processing based on the relative height values ​​of all the pixels to generate three-dimensional topographic point cloud data of the sample under test.

[0069] In this embodiment, the image acquisition unit 601 acquires structured light stripe images of the sample under test at multiple axial positions under different exposure conditions, performs differential processing on the structured light stripe images at adjacent axial positions to obtain multiple sets of modulation index image stacks; the image extraction unit 602 separates and extracts the multiple sets of modulation index image stacks according to multiple preset spectral channels to obtain a multi-channel modulation index image stack; the image correction unit 603 performs modulation index correction and global optimization selection on multiple modulation index curves corresponding to each pixel in the multi-channel modulation index image stack to obtain the optimal modulation index curve for each pixel, and performs peak fitting and positioning on the optimal modulation index curve to obtain the relative height value of each pixel; the image output unit 604 performs convergence processing based on the relative height values ​​of all the pixels to generate three-dimensional topographic point cloud data of the sample under test.

[0070] In one embodiment, the image acquisition unit 601 is specifically used for: Adjust the tilt corrector so that the surface of the sample under test on the tilt corrector is perpendicular to the optical axis; The sample under test is driven by an electronically controlled displacement stage to perform axial scanning according to a preset scheme. At each axial position, the same stripe pattern is projected twice by a projector. The trigger mode other than the camera performs two synchronous acquisitions at the same axial position with low exposure time and high exposure time respectively. The synchronous acquisition process is repeated at all axial positions to obtain the original image stack of different exposure times at different axial positions. The stripe patterns projected at adjacent axial positions have a π phase shift and form complementary patterns. The original image stack is preprocessed by layer-by-layer difference to eliminate the corresponding background intensity terms, resulting in multiple sets of modulation image stacks.

[0071] In one embodiment, the image acquisition unit 601 is further configured to: The fringe projection mode is set to temporal phase-shifted binary fringe. The system magnification, reflectivity of each pixel coordinate, background intensity, modulation degree, initial phase, binary fringe frequency, axial shift step size, and phase shift step number are obtained respectively. The binary fringe frequency is multiplied by the pixel coordinates, and the phase shift determined by the axial movement step size and the phase shift step size is superimposed. The superposition result is then subjected to a cosine operation to obtain the fringe phase cosine value. The phase cosine values ​​of the stripes are binarized to obtain a binary stripe signal; The modulation index and the binary fringe signal are scaled and then superimposed on the background intensity. The ratio of the reflectivity to the square of the system magnification is then used to calculate complementary binary fringes. The complementary binary fringes at adjacent axial positions have a two-step phase shift, and the fringe patterns at adjacent positions are complementary.

[0072] In one embodiment, the image extraction unit 602 is specifically used for: Two sets of original image stacks are obtained, and adjacent layer difference operations are performed on the two sets of original image stacks to obtain the modulation image stack; For the modulation images in the modulation image stack, the intensity values ​​of adjacent original images at the same pixel coordinates are subtracted and the absolute value is taken to obtain the modulation value at the corresponding pixel coordinates. The system intensity point spread function is obtained, and the modulation value is convolved with the system intensity point spread function to obtain the modified modulation image at each pixel coordinate of each layer.

[0073] In one embodiment, the image correction unit 603 is specifically used for: The modulation of the low signal-to-noise ratio column pixels in the multi-channel modulation image stack is corrected based on the continuity of surface reflection characteristics to obtain the corrected multi-channel modulation image stack. For each pixel, multiple modulation curves corresponding to the pixel are extracted from the corrected multi-channel modulation image stack, and low-pass filtering is performed on the multiple modulation curves to obtain filtered multiple modulation curves. Multi-channel modulation peak offset correction is performed on the filtered multiple modulation curves to obtain offset-corrected multiple modulation curves; Compare the peak values ​​of the multiple modulation curves after offset correction, and take the curve with the largest peak value as the optimal modulation curve for the pixel. Data points near the peak of the optimal modulation curve are extracted and fitted with Gaussian curves. The horizontal coordinate of the peak of the fitted curve is used as the relative height value of the pixel.

[0074] In one embodiment, the image correction unit 603 is further specifically used for: Obtain the corresponding stripe phase in the multi-channel modulation image stack, and take the pixel column where the stripe phase is an odd multiple of half the period as the low signal-to-noise ratio column pixel; For each of the low signal-to-noise ratio column pixels, the modulation values ​​of the two adjacent columns of pixels in the corresponding row are obtained, and the average interpolation calculation is performed on the modulation values ​​of the two adjacent columns of pixels to obtain the interpolated and corrected modulation values ​​of the low signal-to-noise ratio column pixels. The interpolated modulation value is replaced with the original modulation value of the corresponding low signal-to-noise ratio column pixel in the multi-channel modulation image stack to obtain the corrected multi-channel modulation image stack.

[0075] In one embodiment, the image correction unit 603 is further specifically used for: Perform a one-dimensional Fourier transform on each of the modulation curves to obtain the spectrum corresponding to each modulation curve; After center-shifting the spectrum, a one-dimensional rectangular window function is applied to the central region of the spectrum for filtering to obtain the filtered spectrum; wherein the width of the one-dimensional rectangular window function is one-tenth of the total width of the spectrum, and it is centrally symmetrically distributed in the central region of the spectrum; A one-dimensional inverse Fourier transform is performed on the filtered spectrum and the modulus value is taken to obtain the filtered modulation distribution, so as to extract multiple modulation curves after filtering.

[0076] In one embodiment, the image correction unit 603 is further specifically used for: Obtain the numerical aperture of the microscope objective and the reference wavelength corresponding to each spectral channel; For each spectral channel, the difference between the preset wavelength and the reference wavelength of the channel to be corrected is calculated, and the difference is divided by the square of the numerical aperture to obtain the focal length deviation of each channel. Based on the focal length deviation of each channel, the modulation curve of the corresponding channel in the filtered multiple modulation curves is axially offset to obtain the offset-corrected multiple modulation curves.

[0077] In one embodiment, the image correction unit 603 is further specifically used for: For each pixel, obtain multiple modulation curves of the pixel after offset correction under all effective channels; wherein, the effective channels include high exposure red channel, high exposure green channel, high exposure blue channel, low exposure red channel, low exposure green channel and low exposure blue channel; The peak value of the modulation curve corresponding to each effective channel is extracted, and the peak values ​​of all effective channels are compared to determine the effective channel with the largest peak value. The modulation curve corresponding to the effective channel with the largest peak value is taken as the optimal modulation curve for the corresponding pixel.

[0078] In one embodiment, the image correction unit 603 is further specifically used for: Obtain the axial grayscale curve corresponding to each pixel in the modulation image, and perform differential operation decoding on the axial grayscale curves of adjacent axial positions to obtain the axial modulation curve corresponding to each pixel. Numerical analysis is performed on the optimal modulation curve based on the axial modulation curve to extract data points near the peak value at the location of the maximum modulation value. Gaussian fitting is performed on the data points near the peak, and the focal position of the pixel is located by the horizontal coordinate of the peak of the fitted curve. The axial coordinate corresponding to the focal position is used as the relative height value of the pixel.

[0079] Since the embodiments of the apparatus and the embodiments of the method correspond to each other, please refer to the description of the embodiments of the method for the embodiments of the apparatus, which will not be repeated here.

[0080] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed, can perform the steps provided in the above embodiments. The storage medium may include various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0081] This invention also provides a computer device, which may include a memory and a processor. The memory stores a computer program, and when the processor calls the computer program in the memory, it can implement the steps provided in the above embodiments. Of course, the computer device may also include various network interfaces, a power supply, a graphics card, etc., to utilize the graphics card's performance to operate the model, such as for inference and training.

[0082] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to in the method section. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

[0083] It should also be noted that, in this specification, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

Claims

1. A microscopic three-dimensional measurement method based on multispectral imaging, characterized in that, include: The structured light stripe images of the sample under test at multiple axial positions under different exposure conditions are acquired. The structured light stripe images at adjacent axial positions are processed by differential operation to obtain a stack of multiple modulation images. The multiple modulation image stacks are separated and extracted according to multiple preset spectral channels to obtain a multi-channel modulation image stack; Modulation correction and global optimization selection are performed on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack to obtain the optimal modulation curve for each pixel. Peak fitting and positioning are performed on the optimal modulation curve to obtain the relative height value of each pixel. The three-dimensional topographic point cloud data of the tested sample is generated by convergence processing based on the relative height values ​​of all the pixels.

2. The multispectral-based three-dimensional microscopic measurement method according to claim 1, characterized in that, The process involves performing modulation correction and global optimization selection on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack to obtain the optimal modulation curve for each pixel, and then performing peak fitting and localization on the optimal modulation curve to obtain the relative height value of each pixel, including: The modulation of the low signal-to-noise ratio column pixels in the multi-channel modulation image stack is corrected based on the continuity of surface reflection characteristics to obtain the corrected multi-channel modulation image stack. For each pixel, multiple modulation curves corresponding to the pixel are extracted from the corrected multi-channel modulation image stack, and low-pass filtering is performed on the multiple modulation curves to obtain filtered multiple modulation curves. Multi-channel modulation peak offset correction is performed on the filtered multiple modulation curves to obtain offset-corrected multiple modulation curves; Compare the peak values ​​of the multiple modulation curves after offset correction, and take the curve with the largest peak value as the optimal modulation curve for the pixel. Data points near the peak of the optimal modulation curve are extracted and fitted with Gaussian curves. The horizontal coordinate of the peak of the fitted curve is used as the relative height value of the pixel.

3. The multispectral-based three-dimensional microscopic measurement method according to claim 2, characterized in that, The step of correcting the modulation of low signal-to-noise ratio column pixels in the multi-channel modulation image stack based on the continuity of surface reflection characteristics to obtain a corrected multi-channel modulation image stack includes: Obtain the corresponding stripe phase in the multi-channel modulation image stack, and take the pixel column where the stripe phase is an odd multiple of half the period as the low signal-to-noise ratio column pixel; For each of the low signal-to-noise ratio column pixels, the modulation values ​​of the two adjacent columns of pixels in the corresponding row are obtained, and the average interpolation calculation is performed on the modulation values ​​of the two adjacent columns of pixels to obtain the interpolated and corrected modulation values ​​of the low signal-to-noise ratio column pixels. The interpolated modulation value is replaced with the original modulation value of the corresponding low signal-to-noise ratio column pixel in the multi-channel modulation image stack to obtain the corrected multi-channel modulation image stack.

4. The multispectral-based three-dimensional microscopic measurement method according to claim 2, characterized in that, The low-pass filtering process performed on the multiple modulation index curves yields multiple filtered modulation index curves, including: Perform a one-dimensional Fourier transform on each of the modulation curves to obtain the spectrum corresponding to each modulation curve; After center-shifting the spectrum, a one-dimensional rectangular window function is applied to the central region of the spectrum for filtering to obtain the filtered spectrum; wherein the width of the one-dimensional rectangular window function is one-tenth of the total width of the spectrum, and it is centrally symmetrically distributed in the central region of the spectrum; A one-dimensional inverse Fourier transform is performed on the filtered spectrum and the modulus value is taken to obtain the filtered modulation distribution, so as to extract multiple modulation curves after filtering.

5. The multispectral-based three-dimensional microscopic measurement method according to claim 2, characterized in that, The step of performing multi-channel modulation peak offset correction on the filtered multiple modulation curves to obtain offset-corrected multiple modulation curves includes: Obtain the numerical aperture of the microscope objective and the reference wavelength corresponding to each spectral channel; For each spectral channel, the difference between the preset wavelength and the reference wavelength of the channel to be corrected is calculated, and the difference is divided by the square of the numerical aperture to obtain the focal length deviation of each channel. Based on the focal length deviation of each channel, the modulation curve of the corresponding channel in the filtered multiple modulation curves is axially offset to obtain the offset-corrected multiple modulation curves.

6. The multispectral-based three-dimensional microscopic measurement method according to claim 2, characterized in that, The step of comparing the peak values ​​of multiple modulation curves after offset correction and selecting the curve with the largest peak value as the optimal modulation curve for the pixel includes: For each pixel, obtain multiple modulation curves of the pixel after offset correction under all effective channels; wherein, the effective channels include high exposure red channel, high exposure green channel, high exposure blue channel, low exposure red channel, low exposure green channel and low exposure blue channel; The peak value of the modulation curve corresponding to each effective channel is extracted, and the peak values ​​of all effective channels are compared to determine the effective channel with the largest peak value. The modulation curve corresponding to the effective channel with the largest peak value is taken as the optimal modulation curve for the corresponding pixel.

7. The multispectral-based three-dimensional microscopic measurement method according to claim 2, characterized in that, The step of extracting data points near the peak of the optimal modulation curve and performing Gaussian fitting, using the horizontal coordinate of the peak of the fitted curve as the relative height value of the pixel, includes: Obtain the axial grayscale curve corresponding to each pixel in the modulation image, and perform differential operation decoding on the axial grayscale curves of adjacent axial positions to obtain the axial modulation curve corresponding to each pixel. Numerical analysis is performed on the optimal modulation curve based on the axial modulation curve to extract data points near the peak value at the location of the maximum modulation value. Gaussian fitting is performed on the data points near the peak, and the focal position of the pixel is located by the horizontal coordinate of the peak of the fitted curve. The axial coordinate corresponding to the focal position is used as the relative height value of the pixel.

8. The multispectral-based three-dimensional microscopic measurement method according to claim 1, characterized in that, The process involves acquiring structured light fringe images of the sample under test at multiple axial positions under different exposure conditions, performing differential processing on the structured light fringe images at adjacent axial positions to obtain a stack of multiple modulation index images, including: Adjust the tilt corrector so that the surface of the sample under test on the tilt corrector is perpendicular to the optical axis; The sample under test is driven by an electronically controlled displacement stage to perform axial scanning according to a preset scheme. At each axial position, the same stripe pattern is projected twice by a projector. The trigger mode other than the camera performs two synchronous acquisitions at the same axial position with low exposure time and high exposure time respectively. The synchronous acquisition process is repeated at all axial positions to obtain the original image stack of different exposure times at different axial positions. The stripe patterns projected at adjacent axial positions have a π phase shift and form complementary patterns. The original image stack is preprocessed by layer-by-layer difference to eliminate the corresponding background intensity terms, resulting in multiple sets of modulation image stacks.

9. The multispectral-based three-dimensional microscopic measurement method according to claim 8, characterized in that, The stripe patterns projected from adjacent axial positions exhibit a π phase shift and constitute complementary patterns, including: The fringe projection mode is set to temporal phase-shifted binary fringe. The system magnification, reflectivity of each pixel coordinate, background intensity, modulation degree, initial phase, binary fringe frequency, axial shift step size, and phase shift step number are obtained respectively. The binary fringe frequency is multiplied by the pixel coordinates, and the phase shift determined by the axial movement step size and the phase shift step size is superimposed. The superposition result is then subjected to a cosine operation to obtain the fringe phase cosine value. The phase cosine values ​​of the stripes are binarized to obtain a binary stripe signal; The modulation index and the binary fringe signal are scaled and then superimposed on the background intensity. The ratio of the reflectivity to the square of the system magnification is then used to calculate complementary binary fringes. The complementary binary fringes at adjacent axial positions have a two-step phase shift, and the fringe patterns at adjacent positions are complementary.

10. The multispectral-based three-dimensional microscopic measurement method according to claim 1, characterized in that, The step of separating and extracting the multi-channel modulation image stack according to multiple preset spectral channels to obtain a multi-channel modulation image stack includes: Two sets of original image stacks are obtained, and adjacent layer difference operations are performed on the two sets of original image stacks to obtain the modulation image stack; For the modulation images in the modulation image stack, the intensity values ​​of adjacent original images at the same pixel coordinates are subtracted and the absolute value is taken to obtain the modulation value at the corresponding pixel coordinates. The system intensity point spread function is obtained, and the modulation value is convolved with the system intensity point spread function to obtain the modified modulation image at each pixel coordinate of each layer.

11. A multispectral microscopic three-dimensional measurement device, characterized in that, include: The image acquisition unit is used to acquire structured light stripe images of the sample under test at multiple axial positions under different exposure conditions, and to perform differential operation processing on the structured light stripe images at adjacent axial positions to obtain multiple sets of modulation image stacks. The image extraction unit is used to separate and extract the multiple modulation image stacks according to multiple preset spectral channels to obtain a multi-channel modulation image stack. The image correction unit is used to perform modulation correction and global optimization selection on multiple modulation curves corresponding to each pixel in the multi-channel modulation image stack, to obtain the optimal modulation curve for each pixel, and to perform peak fitting and positioning on the optimal modulation curve to obtain the relative height value of each pixel. The image output unit is used to perform convergence processing based on the relative height values ​​of all the pixels to generate three-dimensional topographic point cloud data of the sample under test.

Images