Dmd-mcp ultraviolet imaging system and distortion correction method

Abstract

The present application relates to the field of spectral analysis instrument, specifically relates to a kind of DMD-MCP ultraviolet imaging system and distortion correction method.The distortion correction method is generated by lattice mask, collects MCP distortion image, extracts feature points, and is fused nonlinear polynomial modeling and neural network model to carry out joint correction, realizes two-stage processing from rough correction to fine correction.The present application fills the blank of DMD-MCP ultraviolet imaging system space mapping and geometric distortion correction, significantly improves spectral quantitative precision, adapts high-end spectrometer, can be extended to X-ray, electron imaging and other scenes, and engineering value is high.

Description

Technical Field

[0001] This invention belongs to the field of spectroscopic analysis instrument technology, and particularly relates to a DMD-MCP ultraviolet imaging system and a distortion correction method. Background Technology

[0002] In quantitative elemental analysis in key fields such as environmental monitoring, food safety, and geological exploration, atomic fluorescence spectrometers (AFS) and inductively coupled plasma atomic emission spectrometers (ICP-OES) have become indispensable core tools, providing crucial support for accurately acquiring elemental content information. However, the traditional cryogenic CCD / CMOS detectors upon which these spectrometers rely have significant technical limitations in the deep ultraviolet band: on the one hand, their quantum efficiency decays significantly, leading to a substantial decrease in the ability to capture deep ultraviolet light signals; on the other hand, their dynamic range is limited, making it difficult to meet the imaging requirements of high sensitivity and a wide signal amplitude range. These two major problems have become the core bottlenecks restricting further improvements in the overall performance of the spectrometers.

[0003] To overcome these bottlenecks, the industry has gradually focused its research on the integrated application of novel detection and optical modulation devices. Among these, microchannel plate (MCP) detectors, or single-photon counting imaging detectors, have become a research hotspot in deep ultraviolet imaging due to their single-photon-level detection sensitivity and stable operation without deep cooling. Digital micromirror devices (DMDs), on the other hand, possess high-speed, programmable spatial light modulation capabilities, enabling flexible control of the spatial distribution of optical signals and providing a new control dimension for spectral imaging systems. Theoretically, combining DMDs and MCPs can achieve high-sensitivity, wide dynamic range deep ultraviolet imaging without increasing the system's thermal load, opening up new avenues for improving spectrometer performance.

[0004] However, in practical applications, significant spatial mapping errors exist between the DMD and MCP imaging planes. Simultaneously, the MCP itself introduces nonlinear distortion. These two factors combined lead to noticeable geometric distortion in the imaging results. Because elemental spectral lines are highly concentrated in the spectral image, even sub-pixel-level spectral line position drift can severely impact the accuracy of quantitative elemental analysis. Currently, there is a technological gap in the industry regarding imaging distortion correction methods for DMD-MCP deep ultraviolet imaging systems, and existing research cannot solve the aforementioned problems of geometric distortion and insufficient quantitative accuracy.

[0005] Currently, some patents involve pixel mapping technology between DMD and other imaging devices. For example, Chinese invention patent application CN118470132A, published on August 9, 2024, entitled "A Projector Calibration Method Based on Double Checkerboard Radial Basis Interpolation," establishes the coordinate correspondence between the camera and the projector (including DMD) by combining a special projected checkerboard image with a color calibration plate. However, this method relies on the high resolution and low distortion characteristics of the CCD detector and does not consider the unique properties of MCP, such as nonlinear response, low spatial resolution, and large pixel size. Another example is Chinese invention patent application CN108088365A, published on May 29, 2018, entitled "A Precise Coordinate Mapping Method for a Phase-Based Digital Micromirror Camera," which uses a sinusoidal fringe phase matching method to achieve phase matching between DMD and CCD. It also uses CCD as the adapter and cannot cope with the distortion problem of MCP in deep ultraviolet imaging. Therefore, none of the aforementioned existing technologies can be directly applied to and effectively correct the complex distortions in the DMD-MCP deep ultraviolet imaging system. Summary of the Invention

[0006] In view of this, the present invention aims to provide a DMD-MCP ultraviolet imaging system and a distortion correction method to solve the problems of spatial mapping error and image distortion in existing DMD-MCP deep ultraviolet imaging systems.

[0007] To achieve the above objectives, the technical solution created by this invention is implemented as follows:

[0008] A distortion correction method for a DMD-MCP ultraviolet imaging system includes the following steps:

[0009] S1: Generate a regularly distributed dot matrix mask on the image plane of the digital micromirror device;

[0010] S2: Turn on the ultraviolet light source. The ultraviolet beam is modulated by the digital micromirror device and then imaged onto the image plane of the single-photon counting imaging detector by the optical system to obtain a dot matrix distortion image.

[0011] S3: Extract the light spot feature points from the raster-distorted image and obtain the pixel coordinates of the light spot feature points. ;

[0012] S4: Based on the spot feature points in the central region of the dot matrix distortion image, generate an ideal dot matrix distortion-free reference image; wherein, the pixel coordinates of the spot feature points in the dot matrix distortion-free reference image are... ;

[0013] S5: Perform pixel-by-pixel translation scanning of the dot matrix mask within the image plane of the digital micromirror device, repeating steps S1 to S4 to obtain multiple sets of paired data of dot matrix distorted images and dot matrix undistorted reference images.

[0014] S6: Construct the pixel coordinates of light spot feature points in the raster-distorted image based on paired data. Pixel coordinates of light spot feature points in a distortion-free dot matrix reference image The one-to-one correspondence between them forms a mapping dataset;

[0015] S7: Construct a nonlinear polynomial model using partial data from the mapping dataset, and solve for the correction coefficients in the nonlinear polynomial model by fitting the least squares method.

[0016] S8: Using the solved correction coefficients, perform pixel coordinate mapping and grayscale interpolation resampling on the remaining part of the bit-distorted image in the mapping dataset to achieve coarse correction of the bit-distorted image and generate a coarsely corrected image.

[0017] S9: Use the coordinates of the spot feature points in the coarsely corrected image as input and the coordinates of the spot feature points in the corresponding undistorted reference image as output to train the neural network model.

[0018] S10: Input the pixel coordinates of the coarsely corrected image one by one into the trained neural network model to obtain the corrected pixel coordinates. ;

[0019] S11: Calculate the corrected pixel coordinates Relative to pixel coordinates residual offset Using residual offset Residual compensation is performed on the pixel coordinates of the coarsely corrected image to achieve fine correction of the raster-distorted image.

[0020] Furthermore, in step S3, before extracting the spot feature points in the bit-distorted image, the bit-distorted image is preprocessed, specifically as follows:

[0021] The raster-distorted image is denoted as:

[0022] ;

[0023] in, For the raster-distorted image, the coordinates are The grayscale value of the pixel, For the image domain, This represents the global maximum grayscale value.

[0024] Threshold segmentation is used to segment the bit-distorted image to obtain a binary image:

[0025] ;

[0026] ;

[0027] in, The threshold set for threshold segmentation, It is a binary image.

[0028] Furthermore, in step S3, the specific process of extracting the spot feature points in the dot matrix distortion image is as follows:

[0029] In binary images Perform connected component analysis to obtain N independent connected regions. :

[0030] ;

[0031] For N connected regions Perform filtering to retain connected regions with more than 4 pixel values. As an effective area;

[0032] Calculate each effective region The zeroth and first moments:

[0033] ;

[0034] ;

[0035] ;

[0036] in, It is the zeroth moment. Let be the first moment, representing the mass distribution of the effective region in the y-direction. Let be the first moment, representing the mass distribution of the effective region in the x-direction;

[0037] Based on each valid region Calculate the zeroth and first moments for each effective region. centroid coordinates :

[0038] ;

[0039] ;

[0040] Each valid area centroid coordinates Pixel coordinates of light spot feature points in a raster-distorted image .

[0041] Furthermore, in step S4, the process of generating the dot matrix distortion-free reference image is as follows:

[0042] Using nine spot feature points in the central region of the dot matrix distortion image as a reference, an ideal dot matrix distortion-free reference image is generated in reverse according to the ideal geometric relationship; wherein, the ideal geometric relationship includes the position of the nine spot feature points and the distance between two adjacent spot feature points.

[0043] Furthermore, in step S5, the specific process of performing pixel-by-pixel translation scanning of the dot matrix mask within the digital micromirror device is as follows:

[0044] The dot matrix mask is translated pixel by pixel within the image plane of the digital micromirror device. Steps S1 to S4 are repeated for each translation, and multiple sets of paired data of dot matrix distorted images and dot matrix undistorted reference images are accumulated.

[0045] Furthermore, in step S7, the nonlinear polynomial model is a system of polynomial equations of multiple degrees:

[0046] ;

[0047] in, pixel coordinates with pixel coordinates relational functions, pixel coordinates with pixel coordinates The relationship function is given by P and Q, which are correction coefficients, and N is the degree of the polynomial equation system.

[0048] Furthermore, in step S8, the specific process of performing pixel coordinate mapping on the remaining raster-distorted images in the mapping dataset is as follows:

[0049] Establish a reference mesh based on a reference coordinate system: the reference coordinate system is defined with the x-axis pointing to the right and the y-axis pointing downwards, and the coordinate unit is 1 pixel;

[0050] Using the solved correction coefficients P and Q, the pixel coordinates of each pixel in the remaining part of the raster-distorted image in the mapping dataset are mapped to the reference grid, resulting in floating-point coordinates corresponding to each pixel coordinate. This forms a floating-point coordinate graph.

[0051] Furthermore, in step S8, the floating-point coordinate image is resampled by grayscale interpolation using bilinear interpolation or cubic spline interpolation to generate a coarsely corrected image aligned with the reference grid.

[0052] The specific process of performing grayscale interpolation resampling on a floating-point coordinate image using bilinear interpolation is as follows:

[0053] Select any floating-point coordinate in the floating-point coordinate graph Take the floating-point coordinates The four neighboring floating-point coordinates , , , ;in, , , , , floor It is a rounding function;

[0054] Based on the four neighboring floating-point coordinates , , , Calculate floating-point coordinates grayscale value :

[0055] ;

[0056] in, , ;

[0057] Repeat the first two steps to calculate the grayscale value of each floating-point coordinate in the floating-point coordinate image, thereby resampling the floating-point coordinate image;

[0058] The specific process of grayscale interpolation and resampling of a floating-point coordinate image using cubic spline interpolation is as follows:

[0059] First, in integer pixel positions For each node, the grayscale value of each row or column in the floating-point coordinate graph. Establish a system of cubic spline equations, and obtain the second derivative coefficients of each node by solving the tridiagonal matrix; for the internal nodes... satisfy:

[0060] ;

[0061] in, The interval length is... The coefficients of the second derivative to be solved are given; natural boundary conditions are used. , By solving the corresponding tridiagonal matrix, the second derivative coefficients of each node are obtained using the chasing method.

[0062] Secondly, use the second derivative coefficients of each node. Assemble a piecewise cubic polynomial in the interval The polynomial on is:

[0063] ;

[0064] Then, based on the piecewise cubic polynomial for any floating-point coordinate... Perform four multiplication-addition operations and use Horner's rule to calculate the grayscale value of the floating-point coordinates:

[0065] ;

[0066] Finally, the grayscale value of each floating-point coordinate in the floating-point coordinate image is calculated row by row or column by column to achieve resampling of the floating-point coordinate image.

[0067] Furthermore, the dot matrix mask can be a standard two-dimensional dot matrix mask or a pseudo-random dot matrix mask;

[0068] The standard two-dimensional dot matrix mask is generated as follows:

[0069] Using the center pixel of the image plane of the digital micromirror device as a reference, a regularly distributed grid pattern is generated across the entire image plane of the digital micromirror device at a fixed step distance; the corresponding micromirrors in the digital micromirror device are set to the on state according to the grid pattern, and the remaining micromirrors in the digital micromirror device are set to the off state, thus generating a standard two-dimensional dot matrix mask on the image plane of the digital micromirror device.

[0070] The pseudo-random dot matrix mask is generated as follows:

[0071] A random dot matrix pattern is generated using a pseudo-random number generator; the corresponding micromirrors in the digital micromirror device are set to the on state according to the random dot matrix pattern, and the remaining micromirrors in the digital micromirror device are set to the off state, and finally a pseudo-random dot matrix mask is generated on the image plane of the digital micromirror device.

[0072] A DMD-MCP ultraviolet imaging system, comprising:

[0073] Ultraviolet light source, used to provide ultraviolet light beams;

[0074] Digital micromirror devices are used to load lattice masks and modulate ultraviolet beams;

[0075] An optical system used to image the reflected light spot modulated by a digital micromirror device;

[0076] A single-photon counting imaging detector is used to receive reflected light spots modulated by a digital micromirror device and acquire lattice-distorted images;

[0077] The image processing module is used to implement steps S3 to S11 in the above method.

[0078] Compared with the prior art, the present invention can achieve the following beneficial effects:

[0079] 1. Achieved high-precision correction for deep ultraviolet imaging: This invention proposes for the first time a systematic correction scheme specifically designed to solve the problems of spatial mapping error, nonlinearity, and geometric distortion in DMD-MCP deep ultraviolet imaging systems. Through the joint correction of nonlinear polynomials and neural network models, it can effectively compensate for the distortion unique to MCP and achieve sub-pixel-level mapping accuracy, fundamentally solving the technical bottleneck of decreased accuracy of spectral quantitative analysis caused by image distortion in the deep ultraviolet band.

[0080] 2. Balancing Stability and High Accuracy of the Calibration Model: This invention creatively employs a two-stage strategy combining coarse and fine calibration. A nonlinear polynomial model is used for initial geometric transformation, ensuring the model's stability and generalization ability. Then, a neural network model is used for residual learning to compensate for complex nonlinear errors that the polynomial model cannot fit. This method, integrating traditional geometric modeling with data-driven machine learning, achieves high accuracy that is difficult to attain with a single method while ensuring algorithm robustness.

[0081] 3. Improved the practicality and reliability of the entire DMD-MCP deep ultraviolet imaging system: A mapping dataset is constructed through high-density dot matrix sliding scans, and a reference benchmark is generated using the low-distortion region at the image center. The method is highly self-contained and does not rely on external precision calibration equipment. The entire process is clear, computationally efficient, and easily integrated and deployed into existing DMD-MCP deep ultraviolet imaging systems via software upgrades, greatly promoting the practical application of this technology in high-end spectroscopic instruments.

[0082] 4. Excellent versatility and scalability: The correction framework proposed in this invention is not limited to specific DMD or MCP models. Its core concept is applicable to any imaging system with complex spatial mapping and nonlinear distortion. By replacing different mask patterns (such as pseudo-random dot matrix masks) or selecting different interpolation algorithms and neural network models, it can flexibly adapt to different hardware platforms and performance requirements. It can be widely used in various high-end scientific instrument fields such as X-ray imaging, electron imaging, ion imaging, and multi-channel spectrometers, with broad application prospects. Attached Figure Description

[0083] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments and descriptions of the invention are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings:

[0084] Figure 1 A schematic diagram of the structure of the DMD-MCP ultraviolet imaging system described in the embodiments of the present invention;

[0085] Figure 2A schematic flowchart of the distortion correction method for the DMD-MCP ultraviolet imaging system described in the embodiments of the present invention;

[0086] Figure 3 A schematic diagram of the dot matrix distortion image described in the embodiment of the present invention;

[0087] Figure 4 A schematic diagram of a dot matrix distortion-free reference image as described in an embodiment of the present invention;

[0088] Figure 5 This is a schematic diagram of a dot matrix fine-correction image as described in an embodiment of the present invention.

[0089] Explanation of reference numerals in the attached figures:

[0090] 1. Ultraviolet light source; 2. Digital micromirror device; 3. Optical system; 4. Single-photon counting imaging detector; 5. Image processing module. Detailed Implementation

[0091] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0092] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0093] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation on this invention. Furthermore, the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first," "second," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0094] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "assembly," "connection," and "joining" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0095] The invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0096] Optical axis error is a key factor limiting the measurement accuracy of reflective phase deflection systems. It mainly stems from the low-frequency information aliasing caused by the system's optical axis error and the approximation of the imaging model, making it difficult to effectively distinguish between the system's optical axis error and the measured surface shape. To address this, this invention proposes a calibration method that integrates a pinhole imaging model and mirror multi-pose geometric constraints. This method explicitly extracts the system's optical axis error while calibrating the system's geometric extrinsic parameters between the camera and the display screen, achieving decoupling and separation of the error during the calibration stage.

[0097] like Figure 1 As shown, this invention provides a DMD-MCP ultraviolet imaging system, including an ultraviolet light source 1, a digital micromirror device 2, an optical system 3, a single-photon counting imaging detector 4, and an image processing module 5. The ultraviolet light source 1 provides a stable, parallel ultraviolet beam; the digital micromirror device 2 is used to load a dot matrix mask to modulate the ultraviolet beam; the optical system 3 is used to image the reflected light spot modulated by the digital micromirror device 2; the single-photon counting imaging detector 4 receives the reflected light spot modulated by the digital micromirror device 2 and acquires the dot matrix distortion image; and the image processing module 5 performs image processing, neural network model training, and coarse and fine image correction.

[0098] like Figure 2 As shown, this invention provides a distortion correction method for a DMD-MCP ultraviolet imaging system, comprising the following steps:

[0099] S1: Generate a regularly distributed dot matrix mask on the image plane of the digital micromirror device.

[0100] A lattice mask is a special binary (black and white) optical pattern loaded and displayed by a digital micromirror device. As a spatial calibration tool, a lattice mask contains a series of micromirrors (i.e., light spots) sparsely distributed according to specific rules or algorithms, each in an open (flipped) state, forming a discrete array in space. Each light spot in the lattice mask is a flipped-open micromirror.

[0101] The dot matrix mask can use a standard two-dimensional dot matrix mask or a pseudo-random dot matrix mask. Compared with the standard two-dimensional dot matrix mask, the pseudo-random dot matrix mask has better noise resistance and robustness.

[0102] The standard two-dimensional dot matrix mask is generated as follows:

[0103] Using the center pixel of the image plane of the digital micromirror device as a reference, a regularly distributed grid pattern is generated across the entire image plane of the digital micromirror device at a fixed step size (e.g., 29 pixels). The corresponding micromirrors in the digital micromirror device are set to the on state according to the grid pattern, and the remaining micromirrors in the digital micromirror device are set to the off state. Finally, a standard two-dimensional dot matrix mask is generated on the image plane of the digital micromirror device.

[0104] The pseudo-random dot matrix mask is generated as follows:

[0105] The image size, number of pixels, and minimum spacing between pixels are set, and a random dot matrix pattern is generated using a pseudo-random number generator. The corresponding micromirrors in the digital micromirror device are set to the on state according to the random dot matrix pattern, and the remaining micromirrors in the digital micromirror device are set to the off state. Finally, a pseudo-random dot matrix mask is generated on the image surface of the digital micromirror device.

[0106] S2: The ultraviolet light source is turned on. The ultraviolet beam is modulated by a digital micromirror device and then imaged onto the image plane of a single-photon counting imaging detector by an optical system, resulting in a dot-matrix image, such as... Figure 3 As shown.

[0107] S3: Extract the light spot feature points from the raster-distorted image and obtain the pixel coordinates of the light spot feature points. .

[0108] Before extracting the spot feature points from the raster-distorted image, the raster-distorted image is preprocessed, specifically as follows:

[0109] The raster-distorted image is denoted as:

[0110] ;

[0111] in, For the raster-distorted image, the coordinates are The grayscale value of the pixel, For the image domain, This represents the global maximum grayscale value.

[0112] To avoid interference from background noise, a threshold segmentation method is used to segment the raster-distorted image, resulting in a binary image.

[0113] ;

[0114] ;

[0115] in, The threshold set for threshold segmentation, It is a binary image.

[0116] The specific process for extracting spot feature points from a raster-distorted image is as follows:

[0117] In binary images Perform connected component analysis to obtain N independent connected regions. :

[0118] ;

[0119] For N connected regions Perform filtering to retain connected regions with more than 4 pixel values. As an effective area;

[0120] Calculate each effective region The zeroth and first moments:

[0121] ;

[0122] ;

[0123] ;

[0124] in, It is the zeroth moment. Let be the first moment, representing the mass distribution of the effective region in the y-direction. Let be the first moment, representing the mass distribution of the effective region in the x-direction;

[0125] Based on each valid region Calculate the zeroth and first moments for each effective region. centroid coordinates :

[0126] ;

[0127] ;

[0128] Each valid area centroid coordinates Pixel coordinates of light spot feature points in a raster-distorted image .

[0129] S4: Based on the spot feature points in the central region of the dot matrix distortion image, generate an ideal dot matrix distortion-free reference image; wherein, the pixel coordinates of the spot feature points in the dot matrix distortion-free reference image are... .

[0130] The process of generating a raster-based distortion-free reference image is as follows:

[0131] Since the distortion at the center of the bitmap image is negligible and can be considered zero distortion, an ideal bitmap-free reference image is generated by using the nine spot feature points in the central region of the bitmap image as a reference, according to ideal geometric relationships. Figure 4 As shown. The ideal geometric relationship includes the positions of the nine light spot feature points and the distance between two adjacent light spot feature points.

[0132] S5: Perform pixel-by-pixel translation scanning of the dot matrix mask within the image plane of the digital micromirror device, repeating steps S1 to S4 to obtain multiple sets of paired data of dot matrix distorted images and dot matrix undistorted reference images.

[0133] The specific process of performing pixel-by-pixel translation scanning of the dot matrix mask within the digital micromirror device is as follows:

[0134] The dot matrix mask is translated pixel by pixel within the image plane of the digital micromirror device. Steps S1 to S4 are repeated for each translation, and multiple sets of paired data of dot matrix distorted images and dot matrix undistorted reference images are accumulated.

[0135] For example: acquire paired data of 30 sets of dot matrix distorted images and dot matrix undistorted reference images. Currently, the pixel coordinates displayed on the image plane of the digital micromirror device are (0, 0), (0, 30), (0, 60), the next round displays (1, 0), (1, 30), (1, 60), until the final display is (29, 0), (29, 30), (29, 60), completing 30 translation scans.

[0136] S6: Construct the pixel coordinates of light spot feature points in the raster-distorted image based on paired data. Pixel coordinates of light spot feature points in a distortion-free dot matrix reference image The one-to-one correspondence between them forms a mapping dataset.

[0137] The method constructs a mapping dataset through high-density dot matrix sliding scans and generates a reference benchmark using the low-distortion region at the image center. It is highly self-contained and requires no external precision calibration equipment. The entire process is clear, computationally efficient, and easily integrated and deployed in existing DMD-MCP deep ultraviolet imaging systems via software upgrades, greatly advancing the practical application of this technology in high-end spectroscopic instruments.

[0138] S7: Construct a nonlinear polynomial model using a portion of the data in the mapping dataset, and solve for the correction coefficients in the nonlinear polynomial model by fitting the data using the least squares method.

[0139] The nonlinear polynomial model is a system of polynomial equations of degree:

[0140] ;

[0141] in, pixel coordinates with pixel coordinates relational functions, pixel coordinates with pixel coordinates The relationship function is given by P and Q, which are correction coefficients, and N is the degree of the polynomial equation system.

[0142] S8: Using the solved correction coefficients, perform pixel coordinate mapping and grayscale interpolation resampling on the remaining part of the bit-distorted image in the mapping dataset to achieve coarse correction of the bit-distorted image and generate a coarsely corrected image.

[0143] The specific process of performing pixel coordinate mapping on the remaining part of the raster-distorted image in the mapping dataset using the solved correction coefficients P and Q is as follows:

[0144] A reference grid is established based on a reference coordinate system: The reference coordinate system is defined with the x-axis pointing to the right and the y-axis pointing downwards, with each coordinate unit being 1 pixel, and the origin located at the top-left pixel (0,0). The reference coordinate system is the coordinate framework that defines the pixel position, while the reference grid is a regular array of pixels within this reference coordinate system; it is a concrete pixel structure of the reference coordinate system.

[0145] Using the solved correction coefficients P and Q, the pixel coordinates of each pixel in the remaining part of the raster-distorted image in the mapping dataset are mapped to the reference grid, resulting in floating-point coordinates corresponding to each pixel coordinate. This forms a floating-point coordinate graph.

[0146] The floating-point coordinate image is resampled by grayscale interpolation using bilinear interpolation or cubic spline interpolation to generate a coarsely corrected image aligned with the reference grid.

[0147] For bilinear interpolation:

[0148] The pixel coordinates (x, y) of each pixel in the remaining raster-distorted images in the mapping dataset (excluding those used to construct the nonlinear polynomial model) are mapped to the reference grid to obtain floating-point positions (x′, y′). Then, the gray levels of the four neighborhoods are sampled with distance weighting using bilinear interpolation to generate a coarsely corrected image aligned with the reference grid, providing an initial mapping for subsequent fine correction.

[0149] The specific process of bilinear interpolation is as follows:

[0150] First, select any floating-point coordinate in the floating-point coordinate graph. Take the floating-point coordinates The four neighboring floating-point coordinates , , , ;in, , , , , floor This is the floor function.

[0151] Secondly, based on the four neighboring floating-point coordinates , , , Calculate floating-point coordinates grayscale value :

[0152] ;

[0153] in, , .

[0154] Finally, repeat the first two steps to calculate the grayscale value of each floating-point coordinate in the floating-point coordinate image, thereby resampling the floating-point coordinate image.

[0155] For cubic spline interpolation:

[0156] First, the two-dimensional image is decomposed into two independent one-dimensional data chains, rows and columns. Then, a "node-grayscale" mapping relationship is established for each data chain. Subsequently, a cubic polynomial segment is constructed between adjacent nodes, and these segments are required to maintain the continuity of function values, first derivatives and second derivatives at the nodes, thereby forming a globally smooth curve.

[0157] The specific process of cubic spline interpolation is as follows:

[0158] First, in integer pixel positions For each node, the grayscale value of each row or column in the floating-point coordinate graph. Establish a system of cubic spline equations, and obtain the second derivative coefficients of each node by solving the tridiagonal matrix; for the internal nodes... satisfy:

[0159] ;

[0160] in, The interval length is... The coefficients of the second derivative to be solved are given; natural boundary conditions are used. , By solving the corresponding tridiagonal matrix, the second derivative coefficients of each node are obtained using the chasing method.

[0161] Secondly, use the second derivative coefficients of each node. Assemble a piecewise cubic polynomial in the interval The polynomial on is:

[0162] ;

[0163] Then, based on the piecewise cubic polynomial for any floating-point coordinate... Perform four multiplication-addition operations and use Horner's rule to calculate the grayscale value of the floating-point coordinates:

[0164] ;

[0165] Finally, the grayscale value of each floating-point coordinate in the floating-point coordinate image is calculated row by row or column by column to achieve resampling of the floating-point coordinate image.

[0166] A dimension-separating strategy is employed for 2D images: first, horizontal interpolation is performed on each row to obtain an intermediate image; then, vertical interpolation is performed on each column of the intermediate image, ultimately completing full image resampling. Compared to bilinear interpolation, cubic spline interpolation improves edge smoothness.

[0167] S9: Use the coordinates of the spot feature points in the coarsely corrected image as input and the coordinates of the spot feature points in the corresponding undistorted reference image as output to train the neural network model.

[0168] When training the neural network model, the spot feature points of the coarsely corrected image are used as the basis, and the training set, test set and validation set are randomized.

[0169] The neural network model uses an MLP neural network, which has a 6-layer fully connected structure (input layer → 256-neuron hidden layer → 1024-neuron hidden layer → 2048-neuron hidden layer → 1024-neuron hidden layer → 256-neuron hidden layer → output layer). The activation function is ReLU, and the loss function is mean squared error (MSE). Backpropagation training is performed using the Adam optimizer until the validation set loss converges.

[0170] Neural network models are not limited to MLP neural networks; BP neural networks, RBF networks, or lightweight CNN networks can also be used to adapt to different hardware platforms.

[0171] S10: Input the pixel coordinates of the coarsely corrected image one by one into the trained neural network model to obtain the corrected pixel coordinates. .

[0172] S11: Calculate the corrected pixel coordinates Relative to pixel coordinates residual offset Using residual offset Residual compensation is performed on the pixel coordinates of the coarsely corrected image to achieve fine correction of the raster-distorted image. The final finely corrected raster image is as follows: Figure 5 As shown.

[0173] This invention proposes for the first time a systematic correction scheme specifically designed to solve the spatial mapping error, nonlinearity, and geometric distortion problems in DMD-MCP deep ultraviolet imaging systems. Through the joint correction of nonlinear polynomials and neural network models, it can effectively compensate for the distortion unique to MCP, achieve sub-pixel-level mapping accuracy, and fundamentally solve the problem of decreased accuracy of spectral quantitative analysis caused by image distortion in the deep ultraviolet band.

[0174] This invention creatively employs a two-stage strategy combining coarse and fine correction. A nonlinear polynomial model is used for initial geometric transformation, ensuring the model's stability and generalization ability. Then, a neural network model is used for residual learning to compensate for complex nonlinear errors that the polynomial model cannot fit. This method, which integrates traditional geometric modeling with data-driven machine learning, achieves high accuracy that is difficult to attain with a single method while ensuring algorithm robustness.

[0175] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0176] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A distortion correction method for a DMD-MCP ultraviolet imaging system, characterized in that, Includes the following steps: S1: Generate a regularly distributed dot matrix mask on the image plane of the digital micromirror device; S2: Turn on the ultraviolet light source. The ultraviolet beam is modulated by the digital micromirror device and then imaged onto the image plane of the single-photon counting imaging detector by the optical system to obtain a dot matrix distortion image. S3: Extract the light spot feature points from the raster-distorted image and obtain the pixel coordinates of the light spot feature points. ; S4: Based on the spot feature points in the central region of the dot matrix distortion image, generate an ideal dot matrix distortion-free reference image; wherein, the pixel coordinates of the spot feature points in the dot matrix distortion-free reference image are... ; S5: Perform pixel-by-pixel translation scanning of the dot matrix mask within the image plane of the digital micromirror device, repeating steps S1 to S4 to obtain multiple sets of paired data of dot matrix distorted images and dot matrix undistorted reference images. S6: Construct the pixel coordinates of light spot feature points in the raster-distorted image based on paired data. Pixel coordinates of light spot feature points in a distortion-free dot matrix reference image The one-to-one correspondence between them forms a mapping dataset; S7: Construct a nonlinear polynomial model using partial data from the mapping dataset, and solve for the correction coefficients in the nonlinear polynomial model by fitting the least squares method. S8: Using the solved correction coefficients, perform pixel coordinate mapping and grayscale interpolation resampling on the remaining part of the bit-distorted image in the mapping dataset to achieve coarse correction of the bit-distorted image and generate a coarsely corrected image. S9: Use the coordinates of the spot feature points in the coarsely corrected image as input and the coordinates of the spot feature points in the corresponding undistorted reference image as output to train the neural network model. S10: Input the pixel coordinates of the coarsely corrected image one by one into the trained neural network model to obtain the corrected pixel coordinates. ; S11: Calculate the corrected pixel coordinates Relative to pixel coordinates residual offset Using residual offset Residual compensation is performed on the pixel coordinates of the coarsely corrected image to achieve fine correction of the raster-distorted image.

2. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 1, characterized in that, In step S3, before extracting the spot feature points in the bit-distorted image, the bit-distorted image is preprocessed, specifically as follows: The raster-distorted image is denoted as: ； in, For the raster-distorted image, the coordinates are The grayscale value of the pixel, For the image domain, This represents the global maximum grayscale value. Threshold segmentation is used to segment the bit-distorted image to obtain a binary image: ； ； in, The threshold set for threshold segmentation, It is a binary image.

3. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 2, characterized in that, In step S3, the specific process of extracting the spot feature points in the dot matrix distortion image is as follows: In binary images Perform connected component analysis to obtain N independent connected regions. : ； For N connected regions Perform filtering to retain connected regions with more than 4 pixel values. As an effective area; Calculate each effective region The zeroth and first moments: ； ； ； in, It is the zeroth moment. Let be the first moment, representing the mass distribution of the effective region in the y-direction. Let be the first moment, representing the mass distribution of the effective region in the x-direction; Based on each valid region Calculate the zeroth and first moments for each effective region. centroid coordinates : ； ； Each valid area centroid coordinates Pixel coordinates of light spot feature points in a raster-distorted image .

4. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 1, characterized in that, In step S4, the process of generating the dot matrix distortion-free reference image is as follows: Using nine spot feature points in the central region of the dot matrix distortion image as a reference, an ideal dot matrix distortion-free reference image is generated in reverse according to the ideal geometric relationship; wherein, the ideal geometric relationship includes the position of the nine spot feature points and the distance between two adjacent spot feature points.

5. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 1, characterized in that, In step S5, the specific process of performing pixel-by-pixel translation scanning of the dot matrix mask within the digital micromirror device is as follows: The dot matrix mask is translated pixel by pixel within the image plane of the digital micromirror device. Steps S1 to S4 are repeated for each translation, and multiple sets of paired data of dot matrix distorted images and dot matrix undistorted reference images are accumulated.

6. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 1, characterized in that, In step S7, the nonlinear polynomial model is a system of polynomial equations of multiple degree: ； in, pixel coordinates With pixel coordinates relational functions, pixel coordinates with pixel coordinates The relationship function is given by P and Q, which are correction coefficients, and N is the degree of the polynomial equation system.

7. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 1, characterized in that, In step S8, the specific process of performing pixel coordinate mapping on the remaining raster-distorted images in the mapping dataset is as follows: Establish a reference mesh based on a reference coordinate system: the reference coordinate system is defined with the x-axis pointing to the right and the y-axis pointing downwards, and the coordinate unit is 1 pixel; Using the solved correction coefficients P and Q, the pixel coordinates of each pixel in the remaining part of the raster-distorted image in the mapping dataset are mapped to the reference grid, resulting in floating-point coordinates corresponding to each pixel coordinate. This forms a floating-point coordinate graph.

8. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 7, characterized in that, In step S8, the floating-point coordinate image is resampled by grayscale interpolation using bilinear interpolation or cubic spline interpolation to generate a coarsely corrected image aligned with the reference grid. The specific process of performing grayscale interpolation resampling on a floating-point coordinate image using bilinear interpolation is as follows: Select any floating-point coordinate in the floating-point coordinate graph Take the floating-point coordinates The four neighboring floating-point coordinates , , , ;in, , , , , floor It is a rounding function; Based on the four neighboring floating-point coordinates , , , Calculate floating-point coordinates grayscale value : ； in, , ; Repeat the first two steps to calculate the grayscale value of each floating-point coordinate in the floating-point coordinate image, thereby resampling the floating-point coordinate image; The specific process of grayscale interpolation and resampling of a floating-point coordinate image using cubic spline interpolation is as follows: First, in integer pixel positions For each node, the grayscale value of each row or column in the floating-point coordinate graph. Establish a system of cubic spline equations, and obtain the second derivative coefficients of each node by solving the tridiagonal matrix; for the internal nodes... satisfy: ； in, The interval length is... The coefficients of the second derivative to be solved are given; natural boundary conditions are used. , By solving the corresponding tridiagonal matrix, the second derivative coefficients of each node are obtained using the chasing method. Secondly, use the second derivative coefficients of each node. Assemble a piecewise cubic polynomial in the interval The polynomial on is: ； Then, based on the piecewise cubic polynomial for any floating-point coordinate... Perform four multiplication-addition operations and use Horner's rule to calculate the grayscale value of the floating-point coordinates: ； Finally, the grayscale value of each floating-point coordinate in the floating-point coordinate image is calculated row by row or column by column to achieve resampling of the floating-point coordinate image.

9. The distortion correction method for the DMD-MCP ultraviolet imaging system according to claim 1, characterized in that, The dot matrix mask can be a standard two-dimensional dot matrix mask or a pseudo-random dot matrix mask; The standard two-dimensional dot matrix mask is generated as follows: Using the center pixel of the image plane of the digital micromirror device as a reference, a regularly distributed grid pattern is generated across the entire image plane of the digital micromirror device at a fixed step distance; the corresponding micromirrors in the digital micromirror device are set to the on state according to the grid pattern, and the remaining micromirrors in the digital micromirror device are set to the off state, thus generating a standard two-dimensional dot matrix mask on the image plane of the digital micromirror device. The pseudo-random dot matrix mask is generated as follows: Random dot matrix patterns are generated using a pseudo-random number generator. According to the random dot matrix pattern, the corresponding micromirrors in the digital micromirror device are set to the on state, and the remaining micromirrors in the digital micromirror device are set to the off state, and finally a pseudo-random dot matrix mask is generated on the image plane of the digital micromirror device.

10. A DMD-MCP ultraviolet imaging system, comprising: Ultraviolet light source, used to provide ultraviolet light beams; Digital micromirror devices are used to load lattice masks and modulate ultraviolet beams; An optical system used to image the reflected light spot modulated by a digital micromirror device; A single-photon counting imaging detector is used to receive reflected light spots modulated by a digital micromirror device and acquire lattice-distorted images; The ultraviolet imaging system is characterized in that it further includes an image processing module for implementing steps S3 to S11 in the method of any one of claims 1 to 9.

Images