Method and system for calculating parallelism of detector

Abstract

The invention discloses a method and system for calculating the parallelism of a detector, and belongs to the technical field of optical detection. The method comprises the following steps: acquiring at least four positioning points on the lower surface of the detector, at least three reference points on the upper surface of the detector and full-pixel point cloud data of the upper surface of the detector, and performing de-weighting and coordinate calibration; fitting a lower surface plane; performing rough fitting based on the upper surface reference point, and generating a prior weight reflecting the physical credibility according to the distance from each pixel to a rough plane; in fine fitting of the upper surface, iteratively fusing the prior weight and a Huber dynamic weight based on residual adaptive calculation, and solving a fine plane parameter through a weighted robust least square algorithm; and finally calculating an upper and lower surface normal vector included angle to obtain parallelism. According to the method, outlier interference is cooperatively suppressed through a dual weight mechanism, the global trend and the local precision are considered, the anti-interference performance and the precision of parallelism calculation are remarkably improved, and the method is particularly suitable for precise adjustment of a high-resolution optical detector.

Description

Technical Field

[0001] This application relates to the field of optical detection technology, and more specifically to a method and system for calculating the parallelism of a detector. Background Technology

[0002] The demand for "large field of view, high sensitivity, and high resolution" detection is increasingly urgent in fields such as astronomical surveys, high-resolution remote sensing, medical imaging, and high-energy physics. However, traditional single-chip detectors are limited by material growth, manufacturing processes, and physical properties, making it difficult to achieve ultra-large arrays or ultra-high resolution. To overcome these limitations, stitching technology has emerged. The surface flatness of the stitched detector is a core indicator determining its performance, directly affecting detection accuracy, imaging quality, and system reliability. Calculating the parallelism of the upper and lower surfaces is a crucial step in solving the surface flatness problem of stitched detectors. Its importance lies in supporting accuracy control and performance assurance in multiple dimensions. Parallelism calculation provides a quantitative basis for precise calibration. During the stitching and assembly of optical detectors, the support structure can be adjusted or micro-displacement compensation can be used to avoid flatness loss of control due to implicit parallelism defects. Especially for high-resolution detectors such as 4k×4k, the parallelism error needs to be controlled within the second range to meet the requirements of precision optical measurement. The core of parallelism calculation is to solve for the angle between the normal vectors of the two planes by fitting the three-dimensional planes of the upper and lower surfaces of the detector. Current mainstream technologies are limited by the characteristics of point cloud data and the defects of fitting algorithms, making it difficult to simultaneously achieve both anti-interference capability and fitting accuracy.

[0003] In existing technologies, traditional weighted least squares fitting assigns fixed weights to points at different locations in the point cloud to improve fitting accuracy in key regions, and then solves the plane by minimizing the sum of squared weighted residuals. Huber robust fitting, on the other hand, addresses outliers in the point cloud by dynamically adjusting weights using the Huber piecewise loss function. When the residuals are less than or equal to a threshold, squared loss is used; when the residuals are greater than the threshold, linear loss is used, which reduces the weights of outliers. The plane parameters are then optimized iteratively using weighted least squares until convergence.

[0004] However, existing technologies use fixed weighted least squares weights, which cannot dynamically handle outliers. If a pixel becomes an outlier due to sudden noise, its high weight will still lead to fitting bias, making it difficult to meet high accuracy requirements. Huber fitting only dynamically adjusts the weights through residuals, without incorporating prior information from the detector pixels, resulting in insufficient weight discrimination and lower-than-expected accuracy of the fitted plane.

[0005] Therefore, there is an urgent need for a method and system for calculating detector parallelism to solve the technical problems of outlier interference and low accuracy of fitting plane in existing technologies. Summary of the Invention

[0006] The purpose of this application is to provide a method and system for calculating detector parallelism, which can solve at least one of the aforementioned technical problems. The specific solution is as follows: A method for calculating detector parallelism includes the following steps: Acquire the coordinates of at least four positioning points on the surface below the detector, the coordinates of at least three reference points on the upper surface, and the coordinates of the point cloud of all pixels on the upper surface; The coordinates of the reference points on the upper surface and the coordinates of all the pixel point clouds on the upper surface are subjected to duplicate data removal and coordinate calibration. Based on the positioning points, obtain the planar parameters of the lower surface; Based on the reference point, obtain the parameters of the rough plane of the upper surface; Calculate the distance from each pixel to the coarse plane and obtain the prior weights; Based on the prior weights and Huber dynamic weights calculated from the residuals, the data of the upper surface pixel point cloud after removing duplicate data and calibrating the coordinates are iteratively fitted to obtain the parameters of the fine plane of the upper surface. The angle between the upper and lower surfaces of the detector is calculated based on the planar parameters of the lower surface and the fine plane parameters of the upper surface.

[0007] Furthermore, the step of removing duplicate data from the coordinates of the reference points on the upper surface and the coordinates of all the pixel point clouds on the upper surface includes: using a neighborhood deduplication method based on dual thresholds of x and y coordinates to remove duplicate data.

[0008] Furthermore, the coordinate calibration includes: using the positioning point of the lower surface as a global reference, performing coordinate calibration on the reference point of the upper surface and the data of all pixel point clouds on the upper surface.

[0009] Further, the calculation of the distance from each pixel to the coarse plane and the acquisition of prior weights include: Calculate the vertical distance from each pixel to the coarse plane; Based on the inverse relationship of the distance, the normalized prior weights are generated.

[0010] Furthermore, the parameters for obtaining the fine plane of the upper surface include: Using the parameters of the coarse plane as initial values, calculate the residuals of each pixel in the current iteration; Calculate Huber's dynamic weights based on the residuals; The fusion weight is obtained by multiplying the prior weights by the Huber dynamic weights; Based on the fusion weights, the planar parameters of the upper surface are iteratively updated using the weighted least squares method until the convergence condition is met, thereby obtaining the parameters of the fine plane of the upper surface.

[0011] Furthermore, the convergence conditions include: core convergence conditions and auxiliary stopping conditions. The core convergence condition is that the change in the plane normal vector between two adjacent iterations is less than a preset tolerance. The auxiliary stopping condition is reaching a preset maximum number of iterations.

[0012] Furthermore, the method for obtaining the parameters of the coarse plane of the upper surface and the method for obtaining the plane parameters of the lower surface are: to solve using the least squares algorithm.

[0013] This application also provides a system for calculating detector parallelism, comprising: The data acquisition module is used to acquire the three-dimensional coordinates of at least four positioning points on the lower surface of the detector, the three-dimensional coordinates of at least three reference points on the upper surface, and the three-dimensional coordinates of the point cloud of all pixels on the upper surface. The coarse plane fitting module is used to fit the parameters of the coarse plane of the upper surface using the least squares method based on at least three reference points of the upper surface. The prior weight generation module is used to generate prior weights that reflect the physical credibility of a pixel based on the distance from the pixel to the coarse plane. The weighted robust fitting module is used to fuse the prior weights and Huber dynamic weights based on residual calculation, and to perform weighted robust least squares iterative fitting on all pixel point clouds of the upper surface to obtain the parameters of the fine plane of the upper surface. The parallelism calculation module is used to calculate the parallelism between the upper and lower surfaces of the detector based on the parameters of the fine plane of the upper surface and the plane parameters of the lower surface.

[0014] Furthermore, the weighted robust fitting module includes: The Huber dynamic weight calculation submodule is used to dynamically calculate Huber dynamic weights based on the current iteration residual using the Huber piecewise loss function. The weight fusion submodule is used to multiply the prior weights by the Huber dynamic weights to obtain the fused weights; The iterative fitting submodule is used to iteratively update the parameters of the fine plane of the upper surface using the weighted least squares method based on the fusion weights until the convergence condition is met.

[0015] Furthermore, the data acquisition module includes: The deduplication submodule is used to perform deduplication processing on all pixel point clouds on the upper surface based on the xy coordinate dual threshold neighborhood deduplication method. The coordinate calibration submodule is used to perform coordinate calibration on the reference points of the upper surface and the data of all pixel point clouds on the upper surface based on the positioning points of the lower surface.

[0016] Compared with the prior art, the above-described solutions of this application have at least the following beneficial effects: 1. This application discloses a method and system for calculating detector parallelism. By fusing prior weights reflecting the physical reliability of pixels with Huber dynamic weights based on iterative residual adaptive adjustment, a final fused weight is constructed, achieving dual suppression of outliers. Outliers have low prior weights due to their distance from the coarse plane and low Huber dynamic weights due to their large residuals, thus greatly weakening their influence. This effectively avoids the bias caused by fixed weights in traditional weighted least squares and the problem of insufficient weight discrimination in single Huber fitting due to lack of prior information, thereby significantly improving the robustness and accuracy of plane fitting.

[0017] 2. This application discloses a method and system for calculating detector parallelism. Prior weights provide an objective physical basis for weight allocation, anchoring the overall morphological trend of the detector; Huber dynamic weights calibrate local detail deviations during the iteration process. The fusion of these two methods results in high weights for pixels with flat regions and small residuals, dominating the fitting; while pixels at edges or with slight deviations have moderate weights, retaining their contribution. This collaborative mechanism ensures that the fitted plane reflects both the global trend and accurately captures local flatness characteristics, achieving an optimal balance between global and local aspects.

[0018] 3. The detector parallelism calculation method and system of this application, the judgment threshold in Huber dynamic weight is not a fixed value, but is dynamically calculated based on the absolute median difference (MAD) of the residual in each iteration, so that the threshold can adapt to the error distribution fluctuation of the detector pixel point cloud at different fitting stages, ensuring the scientific nature and real-time nature of the outlier judgment standard, and further enhancing the adaptability of the algorithm and the reliability of the final result. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort. In the drawings: Figure 1 This is a flowchart illustrating a method for calculating detector parallelism provided in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or device. Without further limitation, an element defined by the phrase "comprising one" does not exclude the presence of other identical elements in the product or device that includes that element.

[0022] The embodiments of this application are described in detail below with reference to the accompanying drawings.

[0023] Example 1: like Figure 1 As shown in the figure, this application provides a method for calculating detector parallelism, including the following steps: S1. Obtain the coordinates of at least four positioning points on the lower surface of the detector, the coordinates of at least three reference points on the upper surface, and the coordinate data of all pixel point clouds on the upper surface.

[0024] In this embodiment, the coordinates of four positioning points on the lower surface of the detector, the coordinates of three reference points on the upper surface, and the coordinate data of all 9564 pixel point clouds on the upper surface are collected using a two-dimensional planar scanner, a comparator stand, and three measuring probes in the prior art.

[0025] In the technical solution of this application embodiment, the three reference points on the upper surface are the minimum and sufficient condition for defining the coarse plane of the upper surface. The fourth point among the four positioning points on the lower surface is used to verify whether the assumption of the plane of the lower surface is valid; if the four points are perfectly coplanar, then the plane fitted by the three points will naturally pass precisely through the fourth point. If the fourth point deviates significantly from the plane determined by the first three points, it indicates that there is a non-negligible deformation on the lower surface, or that there is a significant error at a certain measurement point. The number of positioning points and reference points can be increased according to the actual situation. Increasing the number of positioning points and reference points will lead to increased measurement time cost, increased measurement complexity, or the introduction of new errors; however, increasing the number of positioning points and reference points can make the acquired data smoother and improve resolution. Therefore, this application does not limit this, and the number of positioning points and reference points can be set according to specific circumstances.

[0026] Coordinates of the four positioning points on the lower surface: , .in, Indicates the lower surface of the detector The three-dimensional coordinates of each positioning point, i.e., the three-dimensional coordinates of the vertex; Indicates the lower surface number Each positioning point Axis coordinates; Indicates the lower surface number Each positioning point Axis coordinates; Indicates the lower surface number Each positioning point Axis coordinates. Coordinates of the three reference points on the upper surface: , ( ).in, Indicates the upper surface of the detector The three-dimensional coordinates of each reference point; Indicates the first of the upper surface benchmarks Axis coordinates; Indicates the first of the upper surface benchmarks Axis coordinates; Indicates the first of the upper surface benchmarks Axis coordinates. Data for all pixel point clouds on the upper surface: ). Indicates the first layer on the upper surface of the detector The three-dimensional coordinates of a pixel point cloud; Indicates the upper surface of the detector A pixel point cloud Axis coordinates; Indicates the first layer on the upper surface of the detector A pixel point cloud Axis coordinates; Indicates the first layer on the upper surface of the detector A pixel point cloud Axis coordinates.

[0027] S2. For the coordinates of all pixel point clouds on the upper surface, perform duplicate data removal and coordinate calibration sequentially. This application provides a preferred technical solution, employing a neighborhood deduplication method based on dual thresholds for x and y coordinates to remove duplicate data. The collected 9564 pixel point clouds on the upper surface are sorted in ascending order of x-axis and y-axis coordinates, ensuring that adjacent pixel points are continuously distributed on the x-y plane. Both x and y coordinate thresholds are set to 5 μm. Using the first point as a reference, the subsequent sorted pixel points are traversed, and the x-axis and y-axis coordinate differences between each subsequent point and the reference point are calculated sequentially. If both differences are less than the corresponding thresholds, the pixel is determined to be a duplicate and discarded. For example, if a point's Δx = 3 μm and Δy = 4 μm are both less than 5 μm, it is determined to be a duplicate and discarded. Here, Δx and Δy represent the x-axis and y-axis coordinate differences, respectively.

[0028] Using the coordinate system established by the four positioning points on the lower surface as the global reference, the rigid transformation matrix between the three reference point cloud data points on the upper surface and the coordinate system of the lower surface is calculated. The rigid transformation matrix is ​​then applied to unify the coordinate systems of the reference points on the upper surface and the pixel point cloud through rigid transformation, thereby eliminating the systematic errors of translation and rotation caused by the installation and adjustment of the measurement and acquisition equipment, and completing the coordinate calibration.

[0029] S3. Obtain the planar parameters of the lower surface based on the positioning points. Construct equations based on the four positioning points of the lower surface, and transform the equations into matrix form. Solve for the matrix of the lower surface plane using the ordinary least squares method in existing technology to obtain the planar parameters of the lower surface.

[0030] The expression for the normal vector obtained by fitting the four positioning points on the lower surface is: in, This represents the normal vector of the lower surface; Represents the normal vector of the lower surface. Axial direction component; Represents the normal vector of the lower surface. Axial direction component; Represents the normal vector of the lower surface. Axial direction components, in the embodiments of this application The value of is -1, that is, when the plane parametric expression is written in standard form, The axial component is -1.

[0031] The expression for the planar parameters of the lower surface is: in, The plane equation representing the plane parameters of the lower surface; This represents the constant term in the equation of the plane of the lower surface.

[0032] S4. Based on the reference points, obtain the parameters of the rough plane of the upper surface. This is done using three upper surface reference points. , ( The equation for the coarse plane of the upper surface is constructed and transformed into matrix form. The matrix of the coarse plane of the upper surface is solved using ordinary least squares method to obtain the parameters of the coarse plane of the upper surface. In this embodiment, the coarse plane is the fitted upper surface, and since this is the first fitting, it is therefore a coarse plane.

[0033] The matrix of parameters for the coarse plane in the table above is expressed as follows: in, This represents the coefficient matrix for fitting the coarse plane on the upper surface. This represents the dependent variable vector fitted to the coarse plane on the upper surface.

[0034] The expression for the parameters of the rough plane on the upper surface is: in, The parameter vector representing the rough plane of the upper surface; In the equation of the rough plane of the upper surface The coefficient of the term; In the equation of the rough plane of the upper surface The coefficient of the term; This represents the constant term in the equation of the rough plane of the upper surface.

[0035] The equation for the rough plane of the upper surface is expressed as follows: in, This represents the equation of the rough plane on the upper surface.

[0036] S5. Calculate the distance from each pixel to the coarse plane and obtain the prior weights.

[0037] This application provides a preferred technical solution: calculating the vertical distance from each pixel to the coarse plane; and generating normalized prior weights based on the inverse relationship of the distances.

[0038] Calculate the vertical distance from each pixel to the coarse plane. The expression for the vertical distance is: in, Indicates the first of the upper surface The perpendicular distance from each pixel to the coarse plane; In the equation of the rough plane of the upper surface The coefficient of the term.

[0039] This application provides a preferred technical solution for vertical distance. Perform distance correction and add minute values. ;in, To represent minute values, avoid division by zero. The corrected vertical distance is obtained using the following expression: in, Indicates the corrected number The vertical distance from each pixel to the coarse plane.

[0040] Weight allocation is performed by generating prior weights based on the corrected inverse vertical distance ratio, and then normalizing them. The expression is as follows: in, Indicates the first of the upper surface The prior weight of each pixel; m represents the traversal index variable of the pixel point cloud on the upper surface; N represents the total number of valid pixel point clouds on the upper surface after deduplication and coordinate calibration; This represents the vertical distance from the corrected m-th pixel to the coarse plane.

[0041] While existing dynamic weights can suppress outliers based on iterative residuals, they do not consider the differences in physical reliability of pixels on the detector surface. The technical solution provided in this application uses a priori weights to transform the physical reliability of detector pixels into quantifiable fixed weights, providing an objective physical basis for weight allocation and overcoming the deficiency of single dynamic weights lacking prior information.

[0042] S6. Based on the prior weights and Huber dynamic weights calculated from the residuals, iteratively fit the data of the upper surface pixel point cloud after removing duplicate data and calibrating the coordinates to obtain the parameters and normal vector of the fine plane of the upper surface.

[0043] Using the parameters of the coarse plane as initial values, calculate the residuals for each pixel in the current iteration. Initialization: Using the surface roughness parameters as initial values, the residuals are calculated iteratively. In the first... In this iteration, the expression for the residual of a pixel is: Where t represents the iteration number index; This represents the residual of the k-th pixel on the upper surface during the t-th iteration; In the plane equation at the t-th iteration The coefficient of the term; In the plane equation at the t-th iteration The coefficient of the term; Let represent the constant term in the plane equation at the t-th iteration.

[0044] According to residuals Calculate Huber's dynamic weights. The expression for Huber's dynamic weights is: in, This represents the Huber dynamic weight of the k-th pixel on the upper surface in the t-th iteration. This represents the threshold for Huber's dynamic weights.

[0045] The technical solutions of the embodiments of this application, The value is determined based on the real-time error distribution characteristics of the pixel point cloud on the upper surface of the detector. An iterative adaptive statistical calculation method is adopted to ensure that the threshold is highly compatible with the residual law of the current fitting stage. It is dynamically updated during the iterative process, and is carried out in the t-th iteration of the fine fitting of the upper surface.

[0046] First, based on the current planar parameters, calculate the residuals for all upper surface pixels. Then, perform statistical processing on the residuals, taking the absolute values ​​of each residual, sorting them, and extracting the median (| The median absolute median deviation (MAD) of the residuals is calculated based on this median. The calculation of MAD requires the introduction of a normality correction factor of 1.4826, i.e., MAD = 1.4826 × median (| The normality correction coefficient serves to make MAD equivalent to the standard deviation σ under a normal distribution, ensuring that the statistical scale conforms to the error law of precise measurement. Finally, multiplying MAD by the Huber function's standard coefficient of 1.345 yields the Huber threshold for the t-th iteration. ,Right now =1.345×MAD, the coefficient 1.345 is chosen because, under the normal distribution error scenario, this coefficient can make the threshold... It covers approximately 95% of the normal residuals, and only adjusts the linear loss weights for outliers outside this range, achieving the dual goals of preserving high accuracy for normal points and suppressing interference from outliers. After each iteration, the above process needs to be repeated to generate the threshold corresponding to the new iteration. The method for determining Huber dynamic weights adapts to the error fluctuations of the detector pixel point cloud at different fitting stages, ensuring that Huber dynamic weights always have a reliable basis for outlier judgment, and ultimately supporting the anti-interference ability and fitting accuracy of the weighted robust least squares algorithm.

[0047] The fusion weight is obtained by multiplying the prior weights by the Huber dynamic weights. The expression for the fusion weight is: in, This represents the final fusion weight of the k-th pixel on the upper surface in t iterations.

[0048] Based on the fusion weights, the weighted least squares method is used to iteratively update the plane parameters of the upper surface until the convergence condition is met, thereby obtaining the parameters and normal vector of the fine plane of the upper surface.

[0049] The fusion weights are normalized to ensure that the sum of the fusion weights is 1. A system of weighted least squares equations is constructed and solved iteratively until the convergence condition is met, thus obtaining the parameters of the fine plane of the upper surface.

[0050] The expression for the weighted least squares equation system is: The expression for the parameter vector of the fine plane of the upper surface is: in, This represents the upper surface plane parameter vector at the (t+1)th iteration; In the plane equation at the (t+1)th iteration The coefficient of the term; In the plane equation at the (t+1)th iteration The coefficient of the term; The table shows the constant terms in the plane equation at iteration t+1.

[0051] Normal vector of the fine plane of the upper surface The expression is: = Parameters of the fine plane of the upper surface The expression is: in, The normal vector of the fine plane of the upper surface; Parameters indicating the precision plane of the upper surface; The normal vector of the fine plane of the upper surface. Axial direction component; The normal vector of the fine plane of the upper surface. Axial direction component; The constant term in the equation representing the fine plane of the upper surface.

[0052] The technical solutions of this application embodiment divide the convergence conditions for iterative fitting of the upper surface pixel point cloud data into two categories: core convergence conditions and auxiliary stopping conditions.

[0053] The core convergence condition is that the change in the plane normal vector between two consecutive iterations is less than a preset tolerance. Specifically, it involves calculating the change in the magnitude of the normal vector in the (t+1)th iteration compared to the normal vector in the tth iteration. If this change satisfies the condition of being less than the preset tolerance, which in this embodiment is... If the iteration converges, the parameter update is stopped and the parameters of the fine plane on the upper surface are output. When the change in the plane normal vector is sufficiently small, it indicates that the fitted plane has approximated the true physical shape of the detector's upper surface, and the improvement in plane accuracy from further iteration is negligible, which meets the requirements of precision optical measurement for fitting stability.

[0054] The auxiliary stopping condition is reaching a preset maximum number of iterations. To avoid the iteration from falling into an infinite loop, a maximum of 30 iterations is preset as an auxiliary stopping condition during the weighted robust fitting process. If the core convergence condition that the change in the normal vector is less than the tolerance is not met when the number of iterations t reaches this maximum number of iterations, the iteration will be forcibly stopped to ensure computational efficiency. Other maximum iteration numbers can also be preset according to actual conditions; this application does not impose any limitations on this.

[0055] S7. Calculate the angle between the upper and lower surfaces of the detector based on the normal vector of the lower surface and the normal vector of the precision plane. The expression for the angle between the upper and lower surfaces of the detector is: in, This indicates the angle between the upper and lower surfaces of the detector; The radian measure of the angle θ between the normal vectors of the upper and lower surfaces.

[0056] The expression for converting units to arcseconds is: in, An angular representation of the angle θ between the normal vectors of the upper and lower surfaces; An arcsecond representation of the angle θ between the normal vectors of the upper and lower surfaces.

[0057] The technical solution of this application embodiment completes deduplication and coordinate calibration by acquiring the coordinate data of four positioning points on the lower surface, three reference points on the upper surface, and the point cloud of all pixels on the upper surface. For the lower surface plane fitting, ordinary least squares fitting is applied to the four positioning points to obtain the lower surface plane parameters and normal vector. Ordinary least squares fitting is applied to the three reference points to obtain the coarse plane parameters, and the vertical distance from each pixel to the coarse plane is calculated. Prior weights are generated and normalized based on the inverse distance ratio. The upper surface fine fitting is an iterative process, first initializing the parameters and the number of iterations, and then calculating the current residual. Then calculate Huber's dynamic weights. And merge to obtain the final weight. Solve the weighted least squares system of equations and continuously update the parameters. After each iteration, a convergence check is performed. If the conditions are met, the fine plane parameters are output. and normal vector ,otherwise The process is iterated, and finally, the parallelism is calculated and the result is output.

[0058] The technical solution of this application's embodiment employs a weighted robust least squares algorithm that combines prior weights and Huber dynamic weights through multiplication. Prior weights are generated based on the distance from the pixel to the coarse plane, reflecting the physical reliability of the pixel. This solves the problem that traditional methods cannot distinguish the physical reliability of pixels, leading to outliers interfering with fitting accuracy. Huber dynamic weights dynamically adjust weights based on iterative residuals, suppressing the influence of outliers. This addresses the issue that Huber fitting relies solely on residuals, lacking prior judgment of pixel physical location, and is prone to misjudgment. Furthermore, the fusion of prior weights and Huber weights achieves dual suppression of outliers, solving the technical problem that a single weighting method cannot simultaneously handle static anomalies and dynamic noise.

[0059] Example 2: This application also provides a system for calculating detector parallelism, including a data acquisition module, a coarse plane fitting module, a prior weight generation module, a weighted robust fitting module, and a parallelism calculation module. The data acquisition module is used to acquire the three-dimensional coordinates of at least four positioning points on the lower surface of the detector, the three-dimensional coordinates of at least three reference points on the upper surface, and the three-dimensional coordinates of all pixel point clouds on the upper surface. The coarse plane fitting module is used to obtain the coarse plane parameters of the upper surface using the least squares method based on the at least three reference points on the upper surface. The prior weight generation module is used to generate prior weights reflecting the physical reliability of pixels based on the distance from the pixel to the coarse plane. The weighted robust fitting module is used to fuse the prior weights with Huber dynamic weights calculated based on residuals, and perform a weighted robust least squares iterative fitting on all pixel point clouds on the upper surface to obtain the fine plane parameters of the upper surface. The parallelism calculation module is used to calculate the parallelism of the upper and lower surfaces of the detector based on the fine plane parameters of the upper surface and the plane parameters of the lower surface.

[0060] The weighted robust fitting module includes: a Huber dynamic weight calculation submodule, a weight fusion submodule, and an iterative fitting submodule. The Huber dynamic weight calculation submodule is used to dynamically calculate the Huber dynamic weights based on the current iteration residual using the Huber piecewise loss function; the weight fusion submodule is used to multiply the prior weights with the Huber dynamic weights to obtain the final fused weights; the iterative fitting submodule is used to iteratively update the upper surface fine plane parameters based on the final fused weights using the weighted least squares method until the convergence condition is met.

[0061] The data acquisition module includes a deduplication submodule and a coordinate calibration submodule. The deduplication submodule is used to deduplicatize all pixel point clouds on the upper surface using a dual-threshold neighborhood deduplication method based on xy coordinates. The coordinate calibration submodule is used to calibrate the coordinates of the upper surface reference point and the entire pixel point cloud based on the lower surface positioning point.

[0062] In this embodiment, the effectiveness of the technical solution of this application is verified through comparative experiments. The specific data is shown in Table 1: Table 1 Comparison of Experimental Results Through 12 sets of comparative experiments, the performance of three methods—weighted least squares method, Huber method and the method of this application—was compared under different outlier ratios of 0%, 6%, 12%, and 18% and different noise intensities of 0.06 μm, 0.12 μm, and 0.24 μm. The core evaluation indicators were the average parallelism error and the average RMSE.

[0063] Without outlier interference, the basic fitting accuracy of the proposed method is comparable to existing methods, without introducing additional bias due to algorithmic complexity. In real-world scenarios with outliers, compared to weighted least squares, the parallelism error of the proposed method is significantly lower in all outlier scenarios. For example, with 12% outliers and 0.24 μm noise, the weighted least squares error is 0.282 arcseconds, while the proposed method's error is only 0.127 arcseconds, representing an accuracy improvement of over 50%. Compared to Huber's method, which is highly sensitive to outliers and whose parallelism error increases sharply with the proportion of outliers, typically ranging from 0.5 to 0.7 arcseconds, the proposed method, under the same conditions, maintains a consistently low error of 0.13 to 0.15 arcseconds, demonstrating far greater stability than Huber's method. Even when outliers severely contaminate the data, resulting in high RMSE, the proposed method can still calculate exceptionally accurate plane normals and parallelism. This demonstrates that the method proposed in this application can minimize the impact of outliers on plane parameters even if the coordinates of every outlier cannot be completely repaired through a weighting mechanism. The method proposed in this application is a high-precision plane fitting and parallelism calculation method particularly suitable for complex real-world measurement environments. It significantly improves the accuracy and reliability of calculations in data containing outliers without sacrificing the accuracy of clean data.

[0064] Finally, it should be noted that 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 or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0065] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method of calculating the parallelism of a probe, characterized by, The method comprises the following steps: obtaining the coordinates of at least four positioning points under the surface of the detector, the coordinates of at least three reference points on the upper surface, and the coordinates of all pixel point clouds on the upper surface; performing duplicate data removal and coordinate calibration on the reference points on the upper surface and the coordinates of all pixel point clouds on the upper surface; obtaining the plane parameters of the lower surface according to the positioning points; obtaining the parameters of the coarse plane of the upper surface according to the reference points; calculating the distance of each pixel to the coarse plane to obtain prior weights; performing iterative fitting on the data of the upper surface pixel point cloud after duplicate data removal and coordinate calibration according to the prior weights and Huber dynamic weights calculated based on residuals to obtain the parameters of the fine plane of the upper surface; calculating the angle of the upper surface and the lower surface of the detector according to the plane parameters of the lower surface and the parameters of the fine plane of the upper surface.

2. The computational method of claim 1, wherein, The duplicate data removal on the reference points on the upper surface and the coordinates of all pixel point clouds on the upper surface comprises: performing duplicate data removal processing by using a neighborhood de-duplication method based on x-y coordinate double thresholds.

3. The computational method of claim 2, wherein, The coordinate calibration comprises: taking the positioning points of the lower surface as a global reference to perform coordinate calibration on the reference points on the upper surface and the data of all pixel point clouds on the upper surface.

4. The computational method of claim 1, wherein, The calculation of the distance of each pixel to the coarse plane to obtain prior weights comprises: calculating the vertical distance of each pixel to the coarse plane; generating normalized prior weights according to the inverse relationship of the distance.

5. The computational method of claim 4, wherein, The obtaining of the parameters of the fine plane of the upper surface comprises: calculating the residuals of each pixel point in the current iteration with the parameters of the coarse plane as the initial value; calculating Huber dynamic weights according to the residuals; multiplying the prior weights and the Huber dynamic weights to obtain fusion weights; iteratively updating the plane parameters of the upper surface by using a weighted least squares method according to the fusion weights until a convergence condition is met to obtain the parameters of the fine plane of the upper surface.

6. The computational method of claim 5, wherein, The convergence condition comprises: a core convergence condition and an auxiliary stop condition: the core convergence condition is that the change of the plane normal vector of adjacent two iterations is less than a preset tolerance; the auxiliary stop condition is that a preset maximum iteration number is reached.

7. The computational method of claim 1, wherein, The method of obtaining the parameters of the coarse plane of the upper surface and the method of obtaining the plane parameters of the lower surface are solved by using a least squares algorithm.

8. A computing system for implementing the calculation of the parallelism of a probe according to any one of claims 1-7, characterized in that, The method comprises: a data acquisition module configured to obtain the three-dimensional coordinates of at least four positioning points on the lower surface of the detector, the three-dimensional coordinates of at least three reference points on the upper surface, and the three-dimensional coordinates of all pixel point clouds on the upper surface; a coarse plane fitting module configured to fit the parameters of the coarse plane of the upper surface based on the at least three reference points on the upper surface by using a least squares method; a prior weight generation module configured to generate prior weights reflecting the physical credibility of pixels based on the distance of the pixels to the coarse plane; a weighted robust fitting module configured to fuse the prior weights and Huber dynamic weights calculated based on residuals to perform weighted robust least squares iterative fitting on all pixel point clouds on the upper surface to obtain the parameters of the fine plane of the upper surface. The parallelism calculation module is configured to calculate parallelism of the upper surface and the lower surface of the detector based on the parameters of the fine plane of the upper surface and the plane parameters of the lower surface.

9. The computing system of claim 8, wherein, The weighted robust fitting module comprises: A Huber dynamic weight calculation submodule is configured to dynamically calculate a Huber dynamic weight based on a Huber segmented loss function and a current iteration residual; A weight fusion submodule is configured to multiply the prior weight and the Huber dynamic weight to obtain a fusion weight; An iteration fitting submodule is configured to iteratively update the parameters of the fine plane of the upper surface based on the fusion weight and a weighted least square method until a convergence condition is met.

10. The computing system of claim 8, wherein, The data acquisition module comprises: A de-duplication submodule is configured to perform de-duplication processing on all pixel point clouds of the upper surface based on an x-y coordinate double-threshold neighborhood de-duplication method; A coordinate calibration submodule is configured to perform coordinate calibration on reference points of the upper surface and data of all pixel point clouds of the upper surface based on positioning points of the lower surface.

Images