Round hole measurement method based on image feature recognition, medium and system

Abstract

The invention discloses a circular hole measurement method based on image feature recognition, a medium and a system. The method comprises the steps of S1, performing Gaussian filtering processing on an original image; s2, performing gray scale superposition projection through a sub-pixel circular boundary projection algorithm to generate a projection curve; s3, performing moving average filtering on the projection curve to generate a smooth curve; s4, performing differential operation based on the smooth curve to generate a differential curve; s5, screening an effective boundary feature point set on the differential curve; s6, for each extreme point in the effective boundary feature point set, calculating an adaptive threshold value in a neighborhood gray scale curve of the extreme point, and realizing sub-pixel-level positioning through gray scale straight line and smooth curve interpolation fitting; s7, generating an initial ellipse model from the sub-pixel feature points through a robustness ellipse fitting algorithm; and S8, iteratively optimizing a fitting result, removing abnormal points which deviate from the model and exceed a threshold value, and outputting an optimized ellipse parameter. The method has the advantages of high measurement precision and the like.

Description

TECHNICAL FIELD

[0001] The present application mainly relates to the field of electron beam technology, and in particular to a round hole measurement method based on image feature recognition, a medium and a system. BACKGROUND

[0002] The size of the key dimension of the semiconductor chip has developed from 125um to 50nm and below. The reduction of the key dimension makes the number of devices on each chip reach millions or even more. With the reduction of the size, it is more difficult to find the possible defects. The existing CDSEM acquires scanning electron microscope images and uses image algorithms to measure the key dimension in order to find possible defect points (including various types of line width, various types of round hole, various types of round corner, various types of ellipse, various types of gap, etc.) in the pattern.

[0003] Electronic beam measurement equipment, such as scanning electron microscope (CDSEM), is widely used in the field of semiconductor manufacturing due to its high resolution and high precision. When detecting the microstructure on the semiconductor wafer, it is necessary to accurately identify and measure various graphical features to ensure the performance and yield of the chip. One of the important indicators is repeatability, which is reflected in the size deviation of measuring the same graphical feature multiple times. If the multiple measurement values fluctuate within a very small error range, it means that the device has good repeatability in terms of size measurement accuracy.

[0004] Image round hole measurement technology combines computer vision and image processing algorithms to identify circular features in images. This technology is widely used in manufacturing, especially in the quality inspection process of parts, to ensure the shape quality and dimensional accuracy of products. As one of the main shapes of semiconductor measurement, it is very important to study robust round hole measurement algorithms.

[0005] Currently, there are many repeated round holes in 8 / 12-inch wafer patterns. During the electron microscope imaging process, there are very complex image interference factors, including: (1) Electron optical imaging resolution: 1. The collection efficiency and the conversion sensitivity of the detector directly affect the signal-to-noise ratio and the definition; 2. The beam spot size, spherical aberration, coma, astigmatism, distortion, and chromatic aberration also affect the resolution.

[0006] (2) Wafer itself difference and positioning influence: 1. Similar pattern interference of adjacent measurement pattern, similar image interference of adjacent measurement chip of the same wafer, leading to increased measurement difficulty; 2. Pattern difference of different measurement positions of the same wafer: difference of images to be measured of different photographing chips and random deviation of positioning, leading to increased measurement difficulty; 3. Difference of different wafers at the same measurement position: difference of measurement images of the same measurement chip and random deviation of positioning, leading to increased measurement difficulty; 4. Occlusion and shadow caused by imaging; 5. Many types of patterns to be measured. SUMMARY

[0007] In view of the technical problems existing in the prior art, the present application provides a circular hole measurement method based on image feature recognition for improving measurement accuracy, a medium and a system.

[0008] To solve the above technical problems, the technical solution provided by the present application is: A circular hole measurement method based on image feature recognition, comprising the steps of: S1, image preprocessing: Gaussian filtering processing is performed on the original image collected by the scanning electron microscope to obtain a denoising image; S2, projection denoising: taking the denoising image as input, performing gray superposition projection through a sub-pixel circular boundary projection algorithm to generate a projection curve representing the hole boundary; S3, smoothing denoising: performing moving average filtering on the projection curve to generate a smoothed curve eliminating noise interference; S4, differential processing: performing differential operation based on the smoothed curve to generate a differential curve containing edge feature information; S5, extreme value feature point search: scanning extreme value points on the differential curve with a radius r, and screening an effective boundary feature point set through a dynamic threshold; S6, sub-pixel feature positioning: for each extreme value point in the effective boundary feature point set, an adaptive threshold is calculated in its neighborhood gray curve, and sub-pixel level positioning is realized through gray straight line and smoothed curve interpolation fitting; S7, ellipse fitting: the sub-pixel feature points positioned in S6 are used as input to generate an initial ellipse model through a robust ellipse fitting algorithm; S8, abnormal point filtering: taking the initial ellipse model as input, using RANSAC algorithm to iteratively optimize the fitting result, removing abnormal points deviating from the model by more than a threshold, and outputting optimized ellipse parameters.

[0009] Preferably, the specific process of step S2 is: S201, dividing N equal small ROI regions based on a preset center C and a radius range [nR1, nR2], and calculating sub-pixel coordinates in each small ROI; S202, calculating the gray value of the projection point by sub-pixel bilinear interpolation; S203, accumulating and averaging the gray values of each small ROI along the arc direction to generate a projection curve.

[0010] Preferably, in step S201, the angle difference between the two adjacent small ROIs is fRadsOffset=2pi / N, the small ROI sector angle is nInspectArea, the counterclockwise rotation angle of the i-th small ROI is (i-1)*fRadsOffset (i=1, 2, 3...n), and the coordinates of point C are pointCenter(x, y), wherein pi is the circular ratio π; The specific process of step S201 is as follows: S2011, traversing j=nR1...nR2; S2012, calculating the number of elements on the arc nNumArc=ceil(nInspectArea*j) when the radius is j, wherein j=nR1...nR2; S2013, calculating the angle difference fTheterD corresponding to adjacent feature points on each arc in each small ROI: fTheterD=nInspectArea / (nNumArc*1.0); S2014, calculating the angle (i-1)*fRadsOffset+fTheterD*k corresponding to the k-th feature point on the j-th arc in the i-th small ROI: S2015, calculating the sub-pixel coordinates pt(x, y) corresponding to the k-th feature point on the j-th arc in the i-th small ROI: pt.x=pointCenter.x+cos(fRadsOffset+fTheterD*k)*j; pt.y=pointCenter.y+sin(fRadsOffset+fTheterD*k)*j; S2016, obtaining all the sub-pixel coordinates pt on the n equal division small ROIs by loop executing S2011-S2015.

[0011] Preferably, the specific process of step S202 is as follows: Set the distances between the projection point and the left and right field points X coordinates as dis_x1 and dis_x0, respectively, and the distances between the projection point and the upper and lower field points Y coordinates as dis_y1 and dis_y0, respectively. Set the gray values of the left upper, right upper, left lower, and right lower four field points as f(0, 0), f(1, 0), f(0, 1), and f(1, 1), respectively. Use the sub-pixel difference algorithm to calculate the gray value of the projection point, which is specifically: P0=f(1,0)*dis_x1+f(0,0)*dis_x0; P1=f(1,1)*dis_x1+f(0,1)*dis_x0; Gray=P0*dis_y0+P1*dis_y1; Wherein P0: sub-pixel upper field x direction gray interpolation; P1: sub-pixel lower field x direction gray interpolation; Gray: 4 field sub-pixel interpolation gray value.

[0012] Preferably, the specific process of step S5 is: In the extreme point search process, first set an extreme point search area, assume r; S501, from the r+1th element of the differential curve Start index; wherein the ith small ROI differential curve is: ; S502, if the sth element satisfies , it is considered that a candidate maximum value point is found; If the sth element satisfies , it is considered that a candidate minimum value point is found; S503, after obtaining all the maximum and minimum value points, calculate the maximum value , the minimum value , respectively, in the maximum and minimum value points, and search and filter out the maximum and minimum value points that meet the dynamic threshold condition with as the dynamic threshold; wherein is the threshold ratio.

[0013] Preferably, in step S6, within the range of nNumNearEdge elements left and right of the gray extreme point, the maximum / minimum gray value is searched respectively, and the maximum / minimum value position and gray value are recorded; set the maximum / minimum gray value as / , and the gray threshold percentage as fThresh, then the feature point position gray value is: ; The gray straight line and the gray smooth curve are interpolated and fitted to obtain the sub-pixel intersection point, which is the sub-pixel feature point.

[0014] Preferably, in step S8, the process of iterative optimization fitting result by using RANSAC algorithm is: Randomly select several points to fit an ellipse, then calculate the distance from all points to the ellipse, and divide the inliers and outliers according to the distance threshold; Repeat the above random process, and the ellipse corresponding to the maximum number of inliers is the final result; The process of iterative optimization is to obtain inner points according to the fitting ellipse and the distance threshold to form a new fitting ellipse, and the process is iterated continuously, and the result obtained when the number of inner points no longer increases is the optimal ellipse.

[0015] The application further discloses a computer program product comprising a computer program which executes the steps of the method when run by a processor.

[0016] The application further discloses a computer readable storage medium having a computer program stored thereon, which executes the steps of the method when run by a processor.

[0017] The application further discloses a circle hole measurement system based on image feature recognition, comprising a memory and a processor connected to each other, wherein the memory stores a computer program which executes the steps of the method when run by the processor.

[0018] Compared with the prior art, the application has the following advantages: The circle hole measurement method based on image feature recognition of the application significantly improves the precision stability and scene adaptability of the semiconductor wafer circle hole size measurement through multi-step collaborative innovation: first, the sub-pixel circle boundary projection algorithm (sub-pixel coordinate formula calculation) is combined with the bilinear interpolation gray scale calculation to carry out projection denoising, effectively suppressing the adjacent pattern interference and positioning deviation in electron beam imaging, and reducing the wafer false alarm rate; second, through the differential extreme point dynamic screening mechanism and the gray / differential hyperbolic curve space mapping, the feature point robustness in the weak contrast scene is enhanced, the low signal-to-noise ratio circle hole detection rate is improved, and the electron optical resolution limit is overcome; in addition, the RANSAC iterative optimization random sampling fitting ellipse model is used to filter out the abnormal points caused by the wafer difference, and the circle hole measurement repeatability error is reduced; at the same time, the algorithm framework is compatible with multiple types of patterns such as line width / ellipse, realizes the reduction of wafer production misjudgment rate under the premise of avoiding photoresist shrinkage, promotes the optimization and upgrading of electron beam measurement from single point precision to complex scene system stability new paradigm, forms a closed-loop technology chain of projection denoising-fractionation-enhanced positioning-dynamic threshold anti-interference-iterative optimization refinement, and systematically solves the measurement drift problem in semiconductor manufacturing. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 The circle hole measurement method based on image feature recognition of the application is in the flowchart of the embodiment.

[0020] Figure 2 The projection point coordinate diagram in the application.

[0021] Figure 3 The four fields (green points) of the red projection points in the application.

[0022] Figure 4 A noise (red) scene diagram in the present application.

[0023] Figure 5 An abnormal point filtering diagram in the present application. DETAILED DESCRIPTION

[0024] The present application is further described below in conjunction with the accompanying drawings and specific embodiments.

[0025] As shown in the drawings, Figure 1 the image feature recognition-based round hole measurement method provided by the embodiments of the present application comprises the following steps: S1, image preprocessing: performing Gaussian filtering processing on the original image collected by the scanning electron microscope to obtain a denoising image; Through simple Gaussian filtering, Gamma correction, histogram equalization and the like for image preprocessing, multi-frame superposition is used to increase the electron dose and improve the stability of SEM image imaging to a certain extent.

[0026] S2, projection denoising algorithm.

[0027] Taking the denoising image as input, the sub-pixel round boundary projection algorithm is used to realize gray superposition projection denoising to generate a projection curve. Specifically, the following steps are included: S201, calculating the projection point coordinates.

[0028] As shown in the drawings, Figure 2 x / y is the image coordinate system, green is each pixel point (integer) of the image, and the area between the two yellow common centers (the center is C, and the radii are nR1 / nR2, nR1<nR2) is the ROI area (abbreviated as ROI).

[0029] It is assumed that the ROI area is evenly divided into N equal small ROI areas (abbreviated as small ROIs), and the small ROI is Figure 2 the area between the two red solid lines, the angle difference fRadsOffset between two adjacent small ROIs is 2pi / N, and the small ROI sector angle is nInspectArea; wherein pi is the circular constant π. The counterclockwise rotation angle of the i-th small ROI is (i-1)*fRadsOffset (i=1, 2, 3...n). The C point coordinates are pointCenter(x, y).

[0030] The specific steps of calculating the projection point coordinates are as follows: S2011, traversing j=nR1...nR2; S2012, calculate the number of elements on the arc when the radius is j (j = nR1... nR2), nNumArc = ceil (nInspectArea * j); S2013, calculate the angle difference corresponding to each adjacent feature point on each arc in each small ROI: fTheterD = nInspectArea / (nNumArc * 1.0); S2014, calculate the angle corresponding to the kth feature point on the jth arc in the ith small ROI: (i-1) * fRadsOffset + fTheterD * k; S2015, calculate the sub-pixel coordinates pt (x, y) corresponding to the kth feature point on the jth arc in the ith small ROI: pt.x = pointCenter.x + cos (fRadsOffset + fTheterD * k) * j; pt.y = pointCenter.y + sin (fRadsOffset + fTheterD * k) * j; By executing S2011-S2015 in a loop, all sub-pixel coordinates pt on the n equal division small ROIs can be obtained.

[0031] S202, calculate the gray scale of the projection point.

[0032] The sub-pixel coordinates of each projection point have been calculated in the previous step. Select one red point for analysis (hereinafter referred to as the projection point). The four green points are the four field points of the red projection point. X and Y are the image coordinate system, as shown in Figure 3 .

[0033] Assume that the projection point is at a distance of dis_x1 and dis_x0 from the left and right field points X in the X coordinate, and at a distance of dis_y1 and dis_y0 from the top and bottom field points Y in the Y coordinate. Assume that the gray scale values of the top-left, top-right, bottom-left, and bottom-right four field points of the projection point are f(0,0), f(1,0), f(0,1), and f(1,1), respectively. Use the sub-pixel interpolation algorithm to calculate the gray scale value of the projection point, which is: P0 = f(1,0) * dis_x1 + f(0,0) * dis_x0; P1 = f(1,1) * dis_x1 + f(0,1) * dis_x0; Gray = P0 * dis_y0 + P1 * dis_y1.

[0034] Where P0: sub-pixel upper field x direction gray scale interpolation; P1: sub-pixel lower field x direction gray scale interpolation; Gray: 4 field sub-pixel interpolation gray scale value.

[0035] S203, calculate the projection gray curve.

[0036] By The first small ROI projection gray curve is calculated by the following formula:

[0037] Where a1, a2...an is the number of sub-pixel points on the nth arc.

[0038] The basic principle is to track and accumulate the gray value in the arc direction. Each arc can calculate an average dose gray, and n arcs can generate a gray curve of n gray.

[0039] Similarly, the i-th small ROI projection gray curve (i=1, 2, 3...n) can be expressed as: .

[0040] S3, smooth denoising.

[0041] Smooth denoising, mainly through mean / median smoothing denoising of gray projection curve; or through clustering algorithm to filter out abnormal points, to improve data stability.

[0042] The i-th small ROI projection gray curve (i=1, 2, 3...N) obtained by the above algorithm is expressed as: ; Where n is the number of curve elements.

[0043] The gray smooth curve is calculated by the following formula (p is the smoothing coefficient):

[0044] The i-th small ROI gray smooth curve is: , the number of elements is n-2*p.

[0045] S4, differential processing.

[0046] Differential processing, mainly to the above smooth projection curve for differential processing, generate differential curve.

[0047] The i-th small ROI gray smooth curve obtained in the last section is: , the number of elements is n-2*p; The differential processing curve (q is the differential coefficient) is calculated by the following formula, and the value of each element is calculated:

[0048] S5, extreme value feature point search.

[0049] The above process has yielded the differential curve (shown as a dashed line). The differential curve of the i-th small ROI is: The next step is to search for all candidate extreme points (maximum / minimum).

[0050] In the process of extreme point search, we first set an extreme point search region, let's say it's r.

[0051] S501, from the (r+1)th element of the differential curve Start indexing; S502, if the s-th element satisfies If so, then a candidate maximum point has been found; (s = r+1, r+2, ..., n-2*p-2*qr-1); If the s-th element satisfies If so, then a candidate minimum point has been found.

[0052] Using the search algorithm described above, filters can be removed. Figure 4 The impact of scene noise.

[0053] S503. All maximum and minimum points were obtained using the above algorithm. The maximum value was calculated among the maximum and minimum points respectively. ) value, minimum value ( ), and then As a dynamic threshold, the search filters out the maximum and minimum points that satisfy the dynamic threshold condition; where This is the threshold ratio, ranging from 0 to 1, with a default of 0.7.

[0054] S6, Threshold Subpixel Feature Point Search Algorithm.

[0055] 1. Gray-scale extreme point search algorithm GrayAverageCurve1, GrayAverageCurve2, GrayAverageCurve3, and GrayAverageCurve4 represent four grayscale smooth projection curves (referred to as grayscale curves in this section); GradCurve1, GradCurve2, GradCurve3, and GradCurve4 represent four differential curves, which have been aligned with the grayscale projection curves. Grad1, Grad2, Grad3, and Grad4 represent the extreme feature points found in the search; Grad2GrayLine1, Grad2GrayLine2, Grad2GrayLine3, and Grad2GrayLine4 represent the mapping lines between the grayscale projection curve and the differential curve, respectively. Gray1, Gray2, Gray3, Gray4 represent the mapping pixel index position (referred to as gray index point) on the gray projection curve respectively.

[0056] Through the above algorithm, the first Number=1 extreme value feature point of the four small ROIs (red / green / blue / yellow) has been obtained: (i=1 / 2 / 3 / 4), u is the index of the extreme value feature point in the differential curve, which is obtained by synchronous mapping (i=1 / 2 / 3 / 4), v is the index of the extreme value feature point in the gray curve.

[0057] 2. Gray threshold sub-pixel feature point search algorithm The gray threshold sub-pixel feature point search algorithm process: in the range of nNumNearEdge elements left and right of the gray extreme point, respectively search the maximum / minimum gray value, and record the maximum / minimum value position and gray value. Assuming that the maximum / minimum gray value is / , the gray threshold percentage is fThresh, then the feature point position gray value is: .

[0058] The interpolation fitting of the gray straight line and the gray smooth curve can obtain the sub-pixel intersection point, that is, the sub-pixel feature point.

[0059] The above threshold sub-pixel feature point search algorithm uses the gray curve correction and differential curve correction algorithm to improve the anti-interference ability of the threshold and improve the robustness of the Threshold algorithm feature point detection; S7, feature point fitting.

[0060] Use to perform ellipse feature point fitting, and the calculation process is as follows: General equation of ellipse:

[0061] Ellipse parameter solving formula group:

[0062] In the original measured N (N≥5) groups of data (x i , y i )(i=1, 2, 3, …, N), according to the general equation of the ellipse and the least square method principle, the minimum value of the objective function is used to determine the parameters A, B, C, D and E. Let the partial derivative of F(A, B, C, D, E) with respect to each parameter be zero, and the following equation group is obtained:

[0063] Solving this linear equation set can solve A, B, C, D and E, that is, the parameters of the fitted ellipse.

[0064] S8, abnormal point filtering.

[0065] The Ransac algorithm is used for filtering of abnormal points, and the detection robustness and accuracy are improved. The process of Ransac is: 5 points are randomly selected for ellipse fitting, then the distance of all points to the ellipse is calculated, and the inliers and outliers are divided according to the distance threshold, the above random process is repeated, and the ellipse corresponding to the maximum number of inliers is obtained as the final result; the iterative optimization idea is to obtain new fitting ellipse according to the fitting ellipse and distance threshold, and the process is iterated constantly, and when the number of inliers no longer increases, the result obtained is the optimal ellipse. The abnormal points can be filtered out, and the circular hole measurement accuracy is improved, such as Figure 5 The red points are abnormal points.

[0066] The circle hole measurement method based on image feature recognition of the application improves the precision stability and scene adaptability of the semiconductor wafer circle hole size measurement through multi-step cooperative innovation: first, the sub-pixel circle boundary projection algorithm (sub-pixel coordinate formula) is used in combination with the bilinear interpolation gray calculation for projection denoising, which effectively suppresses the adjacent pattern interference and positioning deviation in electron beam imaging, and reduces the wafer false alarm rate; second, through the differential extreme point dynamic screening mechanism and the gray / differential hyperbolic space mapping, the feature point robustness in the weak contrast scene is enhanced, the low signal-to-noise ratio circle hole detection rate is improved, and the electron optical resolution limit is overcome; in addition, the RANSAC iterative optimization random sampling fitting ellipse model is used to filter out the abnormal points caused by wafer differences, and the circle hole measurement repeatability error is reduced; at the same time, the algorithm framework is compatible with multiple types of patterns such as line width / ellipse, realizes the reduction of wafer production misjudgment rate under the premise of avoiding photoresist shrinkage, promotes the electron beam measurement from single-point precision optimization to complex scene system stability new paradigm, forms a closed-loop technology chain of projection denoising foundation-differential enhancement positioning-dynamic threshold anti-interference-iterative optimization refinement, and systematically solves the measurement drift problem in semiconductor manufacturing.

[0067] The application can greatly improve the virtual feature points caused by noise by searching for feature points in the area near the extreme feature point search result. The application fully considers the imaging of SEM circle hole images in various complex scenes, and realizes stable measurement of circle holes in various complex scenes by using a series of key algorithms.

[0068] The application further discloses a computer program product comprising a computer program which, when executed by a processor, performs the steps of the method described above. The application further discloses a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the method described above. The application further discloses an image feature recognition-based round hole measuring system comprising a memory and a processor connected to each other, wherein the memory stores a computer program which, when executed by the processor, performs the steps of the method described above. The product, the medium and the system of the application correspond to the method described above, and have the advantages of the method described above.

[0069] The application can realize all or part of the processes in the above-mentioned embodiment method, and can also be completed by hardware related to computer program instructions. The computer program can be stored in a computer readable storage medium, and the computer program can realize the steps of the above-mentioned method embodiment when executed by a processor. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable storage medium includes any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier wave signal, telecommunication signal and software distribution medium, etc. The memory is used to store computer programs and / or modules, and the processor realizes various functions by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can include high-speed random access memory, and can also include non-volatile memory such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash memory device or other volatile solid-state memory device, etc.

[0070] Noun explanation: Ransac (Random Sample Consensus): random sample consensus algorithm; CDSEM (Critical Dimension Scanning Electron Microscope): critical dimension scanning electron microscope; SEM (Scanning Electron Microscope): scanning electron microscope; ROI(region of interest): region of interest

[0071] The above are only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiments. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the art, some improvements and refinements without departing from the principles of the present application shall be considered as the protection scope of the present application.

Claims

1. A round hole measurement method based on image feature recognition, characterized in that, The method comprises the steps of: S1, image preprocessing: Gaussian filtering is performed on the original image collected by the scanning electron microscope to obtain a denoising image; S2, projection denoising: taking the denoising image as input, performing gray superposition projection through a sub-pixel circular boundary projection algorithm to generate a projection curve representing a hole boundary; S3, smoothing denoising: performing moving average filtering on the projection curve to generate a smoothed curve eliminating noise interference; S4, differential processing: performing differential operation based on the smoothed curve to generate a differential curve containing edge feature information; S5, extreme value feature point searching: scanning extreme value points on the differential curve with a radius r, and screening an effective boundary feature point set through a dynamic threshold; S6, sub-pixel feature positioning: for each extreme value point in the effective boundary feature point set, an adaptive threshold is calculated in the neighborhood gray curve thereof, and sub-pixel level positioning is realized through gray straight line and smoothed curve interpolation fitting; S7, ellipse fitting: generating an initial ellipse model through a robust ellipse fitting algorithm for the sub-pixel feature points positioned in S6; S8, abnormal point filtering: taking the initial ellipse model as input, iteratively optimizing the fitting result by using a RANSAC algorithm, removing abnormal points deviating from the model by more than a threshold, and outputting optimized ellipse parameters.

2. The image feature recognition-based round hole measurement method according to claim 1, characterized in that, The specific process of step S2 is as follows: S201, dividing N equal small ROI regions based on a preset center C and a radius range [nR1, nR2], and calculating sub-pixel coordinates in each small ROI; S202, calculating the gray value of the projection point by using sub-pixel bilinear interpolation; S203, accumulating and averaging the gray values in the direction of the circular arc of each small ROI to generate a projection curve.

3. The image feature recognition based round hole measurement method according to claim 2, characterized in that, In step S201, the angle difference fRadsOffset between two adjacent small ROIs is 2pi / N, and the small ROI sector angle is nInspectArea; the counterclockwise rotation angle of the i-th small ROI is (i-1)*fRadsOffset, i=1, 2, 3...n; the point center C has coordinates pointCenter(x, y); wherein pi is the circular ratio pi; The specific process of step S201 is as follows: S2011, iterating j=nR1...nR2; S2012, calculating the number of elements nNumArc on the circular arc when the radius is j: nNumArc=ceil(nInspectArea*j); wherein j=nR1...nR2; S2013, calculating the angle difference fTheterD corresponding to adjacent feature points on each circular arc in each small ROI: fTheterD=nInspectArea / (nNumArc*1.0); S2014, calculating the angle (i-1)*fRadsOffset+fTheterD*k corresponding to the k-th feature point on the j-th circular arc in the i-th small ROI; S2015, calculating the sub-pixel coordinates pt(x, y) corresponding to the k-th feature point on the j-th circular arc in the i-th small ROI: pt.x=pointCenter.x+cos(fRadsOffset+fTheterD*k)*j; pt.y = pointCenter.y + sin(fRadsOffset + fTheterD * k) * j; S2016, through the loop execution S2011 ~ S2015, obtain all sub-pixel coordinates pt on n equal small ROI.

4. The image feature recognition-based round hole measurement method according to claim 3, characterized in that, The specific process of step S202 is: Set the projection point distance left and right field point X coordinate distance dis_x1, dis_x0, respectively, and the distance between the upper and lower field points Y coordinate distance dis_y1, dis_y0; Set the projection point upper left, upper right, lower left and lower right four field points, and the gray values are f(0,0), f(1,0), f(0,1), f(1,1) respectively, use sub-pixel difference value algorithm to realize the calculation of the gray value of the projection point, specifically: P0 = f(1,0) * dis_x1 + f(0,0) * dis_x0; P1 = f(1,1) * dis_x1 + f(0,1) * dis_x0; Gray = P0 * dis_y0 + P1 * dis_y1; Wherein P0: sub-pixel upper field x direction gray interpolation; P1: sub-pixel lower field x direction gray interpolation; Gray: 4 field sub-pixel interpolation gray value.

5. The image feature recognition based round hole measurement method according to any one of claims 1-4, characterized in that, The specific process of step S5 is: In the process of searching for extreme points, first set an extreme point search area, which is assumed to be r; S501、from the differential curve the r+1 element Start index; where the i-th small ROI differential curve is: ; S502, if the s-th element satisfies a candidate maximum point is considered to be found; If the s-th element satisfies a candidate minimum point is considered to be found; S503. After obtaining all maximum and minimum points, calculate the maximum value among the maximum and minimum points respectively. Minimum value , and then As a dynamic threshold, the search filters out the maximum and minimum points that satisfy the dynamic threshold condition; where This represents the threshold ratio.

6. The image feature recognition based round hole measurement method according to any one of claims 1-4, characterized in that, In step S6, in the range of nNumNearEdge elements around the left and right of the gray extremum point, the maximum / minimum gray value is searched respectively, and the maximum / minimum value position and the gray value are recorded; the maximum / minimum gray value is set as / , and the gray threshold percentage is fThresh, and the feature point position gray value is: ​ ； The gray straight line and the gray smooth curve are fitted to obtain the sub-pixel intersection point, which is the sub-pixel feature point.

7. The image feature recognition based round hole measurement method according to any one of claims 1-4, wherein, In step S8, the process of iterative optimization of the fitting result by using RANSAC algorithm is: Randomly select several points to fit an ellipse, then calculate the distance from all points to the ellipse, and divide the inner points and outer points according to the distance threshold; Repeat the above random process to obtain the ellipse corresponding to the maximum number of inner points as the final result; The process of iterative optimization is to obtain new fitting ellipse by fitting ellipse and distance threshold to obtain inner points, and the result obtained when the number of inner points no longer increases is the optimal ellipse.

8. A computer program product comprising a computer program, characterized in that, The computer program, when executed by a processor, performs the steps of the method of any one of claims 1-7.

9. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, performs the steps of the method of any one of claims 1-7.

10. A round hole measurement system based on image feature recognition, comprising a memory and a processor connected to each other, and a computer program is stored on the memory, characterized in that, The computer program, when executed by a processor, performs the steps of the method of any one of claims 1-7. The computer program, when executed by a processor, performs the steps of the method of any one of claims 1-7.

Images