A method for real-time size measurement of irregular crystals

Abstract

The present application relates to the field of crystal measurement of the direct pulling method, and particularly to a real-time size measurement method for irregular crystals, comprising the following sequentially executed steps: S1: acquiring a crystal growth video in real time, converting the crystal growth video into images frame by frame, performing grayscale processing on each image, and obtaining a grayscale data set; S2: selecting a long-axis image and a short-axis image from the grayscale data set as a training set; S3: performing edge detection on the training set based on a Scharr operator and a multiple-neighborhood iterative completion strategy, and obtaining an edge feature map of each image; S4: performing ellipse fitting on the edge feature map extracted in step S3 by using an improved RANSAC algorithm; and S5: generating a crystal cross-sectional image according to model parameters of the ellipse, and outputting long-axis and short-axis data of the crystal cross-sectional image; the accuracy of crystal quality monitoring and the production efficiency are improved.

Description

Technical Field

[0001] This invention relates to the field of Czochralski crystal measurement, and more specifically to a method for real-time measurement of the dimensions of irregular crystals. Background Technology

[0002] The Czochralski method is a commonly used method for producing neodymium-doped yttrium vanadate (Nd:YVO4). During crystal growth, temperature, pulling speed, and rotation speed all have a certain impact on diameter uniformity. Therefore, real-time detection of crystal size during the growth process is very important for improving single crystal quality.

[0003] Traditional methods for measuring crystal diameter mainly include the diameter gauge method, the weight calculation method, and visual observation. The diameter gauge method uses a laser or optical rangefinder to directly measure the crystal diameter, offering advantages such as high precision and non-contact operation. However, it is susceptible to interference from high-temperature and high-brightness environments, and the equipment cost is relatively high. The weight calculation method calculates the diameter by measuring the crystal's weight and combining it with the material density. It is simple to operate and unaffected by light, but as an indirect measurement, it is prone to introducing errors. Visual observation of the crystal diameter by workers is inaccurate, easily influenced by subjective factors, and difficult to perform in high-temperature and high-brightness environments, potentially causing eye damage. Therefore, all three methods have certain limitations and cannot meet the high requirements of modern crystal growth.

[0004] The cross-sectional shape of single-crystal silicon is approximately elliptical or circular, thus visual edge detection and ellipse fitting methods are commonly used. However, the cross-section of neodymium-doped yttrium vanadate (Nd:YVO4) crystals exhibits an irregular polygonal shape with varying side lengths, making it difficult for traditional ellipse fitting methods to accurately describe its cross-sectional contour and achieve high-precision dimensional inspection. This complex geometry increases the difficulty of edge recognition and shape fitting, resulting in larger measurement errors based on conventional ellipse fitting algorithms, which cannot meet the stringent requirements for dimensional uniformity and accuracy during crystal growth. Therefore, existing technologies have significant shortcomings in dimensional inspection of neodymium-doped yttrium vanadate crystals, necessitating the development of detection algorithms and systems more suitable for complex polygonal cross-sections to achieve more accurate, real-time dimensional monitoring, improve crystal quality, and increase production efficiency. Summary of the Invention

[0005] The purpose of this invention is to provide a method for real-time size measurement of irregular crystals with improved measurement accuracy.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for real-time size measurement of irregular crystals includes the following steps performed sequentially:

[0008] S1: Acquire crystal growth video in real time, convert the crystal growth video frame by frame into images, perform grayscale processing on each image, and obtain grayscale dataset;

[0009] S2: Select a fixed row region in each grayscale image in the grayscale dataset as the region of interest, select a preset number of grayscale images, and calculate the average pixel grayscale value of the region of interest in each grayscale image. Select the grayscale image with the lowest average pixel grayscale value as the reference frame, calculate the position and corresponding angle of the current frame relative to the reference frame, and select the long axis image and short axis image as the training set from the grayscale dataset based on the corresponding angle of the reference frame and the preset long axis vertical viewing interval and short axis vertical viewing interval.

[0010] S3: Edge detection is performed on the training set based on the Scharr operator and a multi-neighbor iteration completion strategy to obtain edge feature maps of each image;

[0011] S4: The improved RANSAC algorithm is used to perform ellipse fitting on the edge feature map extracted in step S3. The specific fitting steps are as follows:

[0012] S4-1: Use the coordinates of each edge pixel in the edge feature map as a sample point set, and use the mean of the sample point set as the geometric center.

[0013] Calculate the polar angle of each sample point relative to the geometric center, and then project the sample point set into polar coordinate space, dividing the sample point set into multiple non-overlapping angular subsets.

[0014] Calculate the radial distance from each sample point to the geometric center, and select representative points from each angle subset whose difference between the radial distance and the median radial distance is within a preset range threshold to construct the first sampling point set;

[0015] S4-2: In each RANSAC iteration, several points are randomly selected uniformly from each first sampling point set as fitting sampling points;

[0016] An elliptical model is fitted based on the fitted sampling points;

[0017] Calculate the distance from all the fitted sampling points to the elliptical model, and determine the set of interior points;

[0018] S4-3: The dynamic distance threshold is calculated using the following formula:

[0019] ;

[0020] in, For dynamic thresholds, As the initial threshold, Let be the standard deviation of the point set. Let the density of the point set be... As an index for estimating crystal growth rate, , , These are the weighting coefficients;

[0021] The crystal growth rate is calculated using the following formula. :

[0022] ;

[0023] in, To increase lifting speed, The crystal rotation speed, These are the weighting coefficients;

[0024] Based on the preset dynamic distance threshold Re-evaluate the set of interior points, and update the number of interior points and model quality;

[0025] S4-4: Repeat steps S4-2 and S4-3 multiple times to record the ellipse model with the most interior points and stable model parameters as the final fitting result, that is, to obtain the model parameters of the ellipse.

[0026] S5: Generate a crystal cross-sectional image based on the model parameters of the ellipse, and output the major and minor axis data of the crystal cross-sectional image.

[0027] Preferably, the specific detection steps for edge detection in step S3 are as follows:

[0028] S3-1: Select the complete lower edge and part of the left and right edges of the crystal from the images in the training set as detection images;

[0029] S3-2: The Scharr operator is used to calculate the gradients of each detection image in the horizontal and vertical directions respectively, and the gradients in the two directions are combined into a gradient magnitude map.

[0030] S3-4: Automatic thresholding segmentation effectively separates edge regions from the background and acquires edge region images;

[0031] S3-5: By using a multi-iteration method of taking the maximum pixel value in the neighborhood of each pixel in the edge region image, the edge break of the edge region image is completed, and the feature extraction image is obtained.

[0032] S3-6: Scan vertically from top to bottom in the upper part of the feature extraction image to find the first edge pixel as the starting detection point. Establish an access matrix with the same size as the input feature extraction image. This access matrix is ​​used to record the pixel access status. Initially, all pixels are marked as unaccessed. Use the starting detection point as a node and check the 8 neighboring pixels of the starting detection point to determine whether the neighboring pixels are edges and unaccessed. Add newly discovered edge pixels to the processing queue and mark the edge pixels as accessed. Iterate through the nodes in the access matrix until no new connected edge pixels are found, thus completing edge extraction and obtaining the edge feature map.

[0033] Preferably, step S3 further includes the following step of refining the edge feature map:

[0034] S3-7: The crystal rotation angle increment corresponding to each frame's edge feature map is... N represents the number of frames in one rotation cycle; assuming the edge point coordinates of the current frame are... Calculate the polar angle of the edge point in the edge feature map. :

[0035] ;

[0036] in, Represents the coordinates of the geometric center;

[0037] S3-8: Cache the edge point set of three consecutive cycles: current frame The edge point set of the previous period Edge point set of the previous two cycles For the current frame Points in ,exist and The search term finds corresponding points that satisfy both angle matching and positional proximity, with angle matching being the target point. polar angle Must meet The proximity of locations satisfies the spatial distance requirement. , The distance is Euclidean.

[0038] S3-9: Calculate edge points based on the matching results within 3 cycles. Trust level :

[0039] ;

[0040] in: This indicates that the current frame was successfully matched. This indicates that the previous period's match was successful. This indicates that the first two cycles were successfully matched; a value of 0 indicates that no match was found. , , These are the weighting coefficients;

[0041] S3-10: Perform edge point filtering: retain Edge points, remove point;

[0042] S3-11: To A high-confidence point, if it is not detected in a certain period, is based on the crystal rotation direction and pulling speed. It predicts the position of the edge within the cycle and performs a small-scale search and completion near the predicted position to avoid the real edge breaking due to brief detection failures.

[0043] ;

[0044] in These are the coordinates of the point corresponding to the previous cycle. To increase lifting speed, These are the weighting coefficients.

[0045] By adopting the aforementioned design scheme, the beneficial effects of the present invention are as follows: The present invention acquires and processes video images, adopts a measurement perspective judgment method based on angle and pixel mean, uses the Scharr operator and multiple neighborhood iteration completion strategy to realize crystal edge detection, performs connected component analysis based on neighborhood and access matrix, accurately extracts target edge features, and effectively improves the accuracy and robustness of ellipse fitting by adopting the improved polar angle layered sampling and dynamic threshold adjustment RANSAC algorithm.

[0046] This invention solves the problems of low accuracy and large error in traditional methods, and overcomes the shortcomings of traditional fitting methods in terms of insufficient accuracy for crystals with irregular polygonal cross sections. It realizes real-time, high-precision, non-contact detection of dimensions during the growth process of this type of crystal, thereby improving the accuracy of crystal quality monitoring and production efficiency. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the structure of the Claymore furnace and the camera setup of the present invention;

[0048] Figure 2 This is an example of a one-cycle crystal image according to the present invention;

[0049] Figure 3 This is a schematic diagram of the pixel mean curve of the present invention;

[0050] Figure 4 This is a schematic diagram of the edge feature map of the present invention;

[0051] Figure 5This is a schematic diagram of the purification edge feature map of the present invention;

[0052] Figure 6 This is a diagram showing the crystal ellipse fitting effect of the present invention;

[0053] Figure 7 This is a flowchart of the method of the present invention;

[0054] Figure 8 The results of the major and minor axis measurements are for this invention. Detailed Implementation

[0055] 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. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0056] The terms "first," "second," "third," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or apparatuses.

[0057] A method for real-time size measurement of irregular crystals, such as Figure 7 As shown, the steps are executed sequentially as follows:

[0058] S1: Real-time acquisition of crystal growth video, such as Figure 2 As shown, the crystal growth video is converted into images frame by frame, and each image is processed to reduce computational complexity, thus obtaining a grayscale dataset.

[0059] In this embodiment, the grayscale dataset is also subjected to Gaussian filtering for noise reduction to suppress noise interference.

[0060] like Figure 1 As shown, the Czochralski furnace is a key piece of equipment in the crystal growth process, typically consisting of a furnace body and a crucible. The furnace body serves as the outer shell of the entire system, responsible for providing the high-temperature environment required for crystal growth. The crucible, located inside the furnace body, holds the molten material, and the crystal grows by rotating within the crucible.

[0061] A camera is installed on the side of the furnace to observe and record the crystal growth process and collect crystal growth videos.

[0062] In the Czochralski (CZ) method for growing Nd:YVO4 crystals, the CZ furnace is equipped with only one observation window, and the crystal cross-section is an irregular polygon. If an ellipse is fitted to the crystal cross-section from any viewing angle, the resulting ellipse shape will have significant errors when the major and minor axes of the crystal are not nearly perpendicular to the line of sight, failing to accurately reflect the actual size of the crystal. Therefore, it is necessary to determine the precise measurement angle and only perform fitting near angles where the major and minor axes of the crystal are nearly perpendicular to the line of sight. This ensures fitting accuracy and lays the foundation for subsequent dimensional measurements.

[0063] S2: Select a fixed row region in each grayscale image of the grayscale dataset as the region of interest, select a preset number of grayscale images, and calculate the average pixel grayscale value of the region of interest in each grayscale image to obtain the time series. The grayscale image with the lowest average pixel grayscale value is taken as the reference frame. The position and corresponding angle of the current frame relative to the reference frame are calculated. Based on the corresponding angle of the reference frame and the preset vertical viewing intervals of the major and minor axes, the major and minor axis images are selected from the grayscale dataset as the training set. The operation of step S2 is represented by the following formula:

[0064] ;

[0065] ;

[0066] Among them, For frame number, Based on the historical pixel average, For areas of interest;

[0067] In this embodiment, the core reason is that during crystal growth, the upper region of the crystal has a relatively stable morphology, and the pixel features (such as grayscale values) of this region in the image differ significantly from the surrounding melt region, thus more clearly reflecting the crystal's contour changes. The selected fixed row region needs to cover the main part of the upper crystal while avoiding including too many interfering pixels from the melt region. This can usually be determined through preliminary experimental observation, such as selecting the middle row of the image as the region of interest. The preset frame rate can be set according to actual needs, such as setting the preset frame rate to 100 frames.

[0068] In this embodiment, the frame number is used as the horizontal axis, and the average pixel grayscale value of the corresponding frame's region of interest is used as the vertical axis to plot the graph as shown below. Figure 3 The pixel mean curve shown is used as a reference frame by selecting the lowest point of the pixel mean curve. Record the reference frame number at the moment when the crystal's long axis is perpendicular to the line of sight.

[0069] Known crystal rotation period Video frame rate The number of frames corresponding to the calculation period Based on the number of frames corresponding to the crystal rotation period. The continuous image frames are divided into multiple time windows, each time window corresponding to a complete rotation cycle of the crystal.

[0070] With reference frame Using this as a baseline, calculate the current frame. With reference frame relative frame count ; and then according to Calculate the crystal rotation angle corresponding to the current frame.

[0071] The angle ranges are determined based on the geometric characteristics of crystal growth and the accuracy requirements of ellipse fitting. When the major axis of the crystal is perpendicular to the line of sight, the projection of the crystal cross-section in the image is closest to the direction of the major axis of the ellipse, and the fitted ellipse can best reflect the actual major axis size of the crystal. Similarly, when the minor axis of the crystal is perpendicular to the line of sight, the projection is closest to the direction of the minor axis of the ellipse, and the fitted ellipse can best reflect the actual minor axis size of the crystal. Through statistical analysis of extensive experimental data, the angle ranges corresponding to the major and minor axes being perpendicular to the line of sight are determined, and angle tolerances are introduced. The specific adjustment can be made according to the actual measurement accuracy requirements to ensure that effective ellipse fitting can be performed within a reasonable angle range.

[0072] Long axis vertical viewing range: based on the long axis vertical angle corresponding to the reference frame. and Centered on, combined with angular tolerance , set as and .

[0073] The minor axis perpendicular to the line of sight: Because the major and minor axes of the crystal are perpendicular to each other, when the crystal is rotated 90°, the minor axis is perpendicular to the line of sight. Therefore, it is considered as... and Centered on, combined with angular tolerance , set as and .

[0074] The crystal axis direction is determined by combining the angle range and the pixel mean. If the current angle is within the range perpendicular to the major axis, and the current pixel mean satisfies... If the current angle is within the vertical range of the minor axis, and the current pixel mean satisfies the condition, then it is determined to be a line of sight perpendicular to the major axis; If the angle is perpendicular to the minor axis, it is determined to be a line of sight; otherwise, it is determined to be "other angles". This refers to the set of valley frame numbers corresponding to the "long axis perpendicular line of sight". This refers to the set of peak frame numbers corresponding to the "short axis vertical line of sight". For set The standard deviation of gray level For set The standard deviation of gray level;

[0075] S3: Edge detection is performed on the training set based on the Scharr operator and a multi-neighbor iterative completion strategy to obtain, for example... Figure 4 Edge feature maps of each image shown;

[0076] In this embodiment, the specific detection steps for edge detection in step S3 are as follows:

[0077] S3-1: Select the complete lower edge and part of the left and right edges of the crystal from the images in the training set as detection images; focus on processing the region containing the "energy release ring" in the selected detection images. The gradient change of the ring is significant, and the inner circle is the edge of the crystal.

[0078] S3-2: The Scharr operator is used to calculate the gradients of each detected image in the horizontal (x-direction) and vertical (y-direction) directions respectively. The gradient in the horizontal direction is... and gradient in the vertical direction The calculation formula is the convolution operation:

[0079] ;

[0080] ;

[0081] in, The input is a grayscale image. It is a convolution operation;

[0082] The gradients from the two directions are combined into a gradient magnitude map, calculated using the following formula:

[0083] ;

[0084] S3-4: Automatic thresholding segmentation effectively separates edge regions from the background and acquires edge region images;

[0085] In this embodiment, the optimal binarization threshold is determined using the Otsu's method to achieve effective separation of the edge region from the background. The calculation formula is as follows:

[0086] ;

[0087] in, It is a threshold. It is the variance between classes. , It refers to the weights of the foreground and background regions. , It is the mean of the two classes.

[0088] S3-5: Due to the obvious edge breaks in the neodymium-doped yttrium vanadate crystal image, an iterative method of taking the maximum pixel value in the neighborhood of each pixel in the edge region image is used to complete the edge breaks, improve the integrity and continuity of the edges, and obtain the feature extraction image; the calculation formula is as follows:

[0089] ;

[0090] in, Indicates the current pixel in the pixel's neighborhood. Offset in direction Indicates the current pixel in the pixel's neighborhood. Offset in direction Represents the neighborhood set of a pixel;

[0091] S3-6: To remove interfering components such as fine lines and the outer circle of the energy release ring in the molten metal, and reduce their impact on subsequent fitting analysis, the upper part of the feature extraction image is scanned vertically from top to bottom to find the first edge pixel as the starting detection point. An access matrix with the same size as the input feature extraction image is established. This access matrix is ​​used to record the pixel access status. Initially, all pixels are marked as unaccessed. The starting detection point is used as a node. The eight neighboring pixels of the starting detection point are checked to determine whether the neighboring pixels are edges and unaccessed. Newly discovered edge pixels are added to the processing queue and marked as accessed to avoid repeated access. The nodes in the access matrix are iterated continuously until no new connected edge pixels are found, completing the edge extraction and obtaining the edge feature map. The calculation formula for step S3-6 is as follows:

[0092] ;

[0093] in, Indicates the starting detection point is at Neighborhood offset in the direction, Indicates the starting detection point is at Neighborhood offset in the direction;

[0094] As a preferred embodiment, based on the periodic removal of instantaneous noise using crystal rotation, a precise rotation angle matching and dynamic confidence update mechanism are introduced to improve the recognition accuracy of stable edge points and enhance adaptability to slow edge changes during crystal growth. Step S3 further includes the following steps to purify the edge feature map:

[0095] S3-7: Known crystal rotation period Video frame rate The number of frames corresponding to the calculation period The increment of the crystal rotation angle corresponding to each frame's edge feature map is Assuming the coordinates of the edge points in the current frame are Calculate the polar angle of the edge point in the edge feature map. :

[0096] ;

[0097] in, Represents the coordinates of the geometric center;

[0098] S3-8: Cache the edge point set of three consecutive cycles: current frame The edge point set of the previous period Edge point set of the previous two cycles For the current frame Points in ,exist and The search term finds corresponding points that satisfy both angle matching and positional proximity, with angle matching being the target point. polar angle Must meet The proximity of locations satisfies the spatial distance requirement. , The distance is Euclidean.

[0099] S3-9: Dynamic Trust Scoring and Screening: Calculating Edge Points Based on Matching Results Over 3 Periods Trust level :

[0100] ;

[0101] in: This indicates that the current frame was successfully matched. This indicates that the previous period's match was successful. This indicates that the first two cycles were successfully matched; a value of 0 indicates that no match was found. , , This is a weighting coefficient that decays over time, enhancing the influence of recent matches;

[0102] S3-10: Perform edge point filtering: retain Edge points (meeting at least the current frame + 1 historical period matching) are removed. Points (instantaneous noise such as fine lines in the molten metal);

[0103] S3-11: Edge continuity compensation: for A high-confidence point, if it is not detected in a certain period, is based on the crystal rotation direction and pulling speed. Predict its position within the cycle; and perform a small-scale search and completion near the predicted position to avoid breaks in the real edge due to brief detection failures, obtaining results such as... Figure 5 The purification edge feature map shown;

[0104] ;

[0105] in These are the coordinates of the point corresponding to the previous cycle. To increase lifting speed, These are the weighting coefficients.

[0106] S4: An improved RANSAC algorithm is used to fit the purified edge feature map extracted in step S3 to an ellipse. In crystal section fitting, the traditional RANSAC algorithm has limited effectiveness due to the uneven distribution of edge points and potential noise. Therefore, we propose an improved RANSAC algorithm that combines an angle-layer-based sampling strategy and dynamic threshold adjustment to improve the accuracy and stability of ellipse fitting. The specific fitting steps are as follows:

[0107] S4-1: Using the coordinates of each edge pixel in the refined edge feature map as a sample point set, and the mean of this sample point set as the geometric center; the formula for estimating the geometric center is as follows:

[0108] , , ;

[0109] in, The center of the point set, It is a point set The average value of the coordinates, It is the average value of the y-coordinates of the point set.

[0110] Calculate the polar angle of each sample point relative to the geometric center, thereby projecting the sample point set into polar coordinate space and dividing the sample point set into multiple non-overlapping angular subsets; the polar angle calculation formula is as follows:

[0111]

[0112] in, For point Relative to the center The polar angle.

[0113] Calculate the radial distance from each sample point to the geometric center. Representative points whose radial distance to the median radial distance falls within a preset threshold value are selected from each angular subset to construct the first sampling point set; the calculation formula is as follows:

[0114] ;

[0115] ;

[0116] in, For the sample set, The radial distance from the point to the center. The median distance from all points to the center. This is a preset range threshold.

[0117] S4-2: In each RANSAC iteration, several points are randomly selected uniformly from each set of first sampling points as fitting sampling points; this ensures that the sampling points are representative at all angle levels and avoids the concentration of sampling points in local areas.

[0118] An elliptical model is fitted based on the fitted sampling points;

[0119] Calculate the distance from all the fitted sampling points to the elliptical model. This distance is usually the nearest distance or geometric distance from the point to the ellipse, and determine the set of interior points.

[0120] S4-3: The dynamic distance threshold is calculated using the following formula:

[0121] ;

[0122] in, For dynamic thresholds, As the initial threshold, Let be the standard deviation of the point set. Let the density of the point set be... As an index for estimating crystal growth rate, , , These are the weighting coefficients;

[0123] The crystal growth rate is calculated using the following formula. :

[0124] ;

[0125] in, To increase lifting speed, The crystal rotation speed, These are the weighting coefficients;

[0126] Based on the preset dynamic distance threshold Re-evaluate the set of interior points, and update the number of interior points and model quality;

[0127] S4-4: Repeat steps S4-2 and S4-3 multiple times for iteration, and record the ellipse model with the most interior points and stable model parameters as the final fitting result. The fitting result is as follows: Figure 6 As shown, this is how to obtain the model parameters of the ellipse;

[0128] S5: Generate a crystal cross-sectional image based on the ellipse model parameters, and output the major and minor axis data of the crystal cross-sectional image, such as... Figure 8 As shown.

[0129] In summary, this invention acquires and processes video images, employs a measurement perspective judgment method based on angle and pixel mean, uses the Scharr operator and a multiple neighborhood iteration completion strategy to achieve crystal edge detection, performs connected component analysis based on neighborhood and access matrix to accurately extract target edge features, and effectively improves the accuracy and robustness of ellipse fitting by using an improved polar angle hierarchical sampling and dynamic threshold adjustment RANSAC algorithm.

[0130] This invention solves the problems of low accuracy and large error in traditional methods, and overcomes the shortcomings of traditional fitting methods in terms of insufficient accuracy for crystals with irregular polygonal cross sections. It realizes real-time, high-precision, non-contact detection of dimensions during the growth process of this type of crystal, thereby improving the accuracy of crystal quality monitoring and production efficiency.

[0131] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for real-time size measurement of irregular crystals, characterized in that: The steps are as follows, performed sequentially: S1: Acquire crystal growth video in real time, convert the crystal growth video frame by frame into images, perform grayscale processing on each image, and obtain grayscale dataset; S2: Select a fixed row region in each grayscale image in the grayscale dataset as the region of interest, select a preset number of grayscale images, and calculate the average pixel grayscale value of the region of interest in each grayscale image. Select the grayscale image with the lowest average pixel grayscale value as the reference frame, calculate the position and corresponding angle of the current frame relative to the reference frame, and select the long axis image and short axis image as the training set from the grayscale dataset based on the corresponding angle of the reference frame and the preset long axis vertical viewing interval and short axis vertical viewing interval. S3: Edge detection is performed on the training set based on the Scharr operator and a multi-neighbor iteration completion strategy to obtain edge feature maps of each image; S4: The improved RANSAC algorithm is used to perform ellipse fitting on the edge feature map extracted in step S3. The specific fitting steps are as follows: S4-1: Use the coordinates of each edge pixel in the edge feature map as a sample point set, and use the mean of the sample point set as the geometric center. Calculate the polar angle of each sample point relative to the geometric center, and then project the sample point set into polar coordinate space, dividing the sample point set into multiple non-overlapping angular subsets. Calculate the radial distance from each sample point to the geometric center, and select representative points from each angle subset whose difference between the radial distance and the median radial distance is within a preset range threshold to construct the first sampling point set; S4-2: In each RANSAC iteration, several points are randomly selected uniformly from each first sampling point set as fitting sampling points; An elliptical model is fitted based on the fitted sampling points; Calculate the distance from all the fitted sampling points to the elliptical model, and determine the set of interior points; S4-3: The dynamic distance threshold is calculated using the following formula: ； in, For dynamic thresholds, As the initial threshold, Let be the standard deviation of the point set. Let the density of the point set be... As an index for estimating crystal growth rate, , , These are the weighting coefficients; The crystal growth rate is calculated using the following formula. : ； in, To increase lifting speed, The crystal rotation speed, These are the weighting coefficients; Based on the preset dynamic distance threshold Re-evaluate the set of interior points, and update the number of interior points and model quality; S4-4: Repeat steps S4-2 and S4-3 multiple times to record the ellipse model with the most interior points and stable model parameters as the final fitting result, that is, to obtain the model parameters of the ellipse. S5: Generate a crystal cross-sectional image based on the model parameters of the ellipse, and output the major and minor axis data of the crystal cross-sectional image.

2. The method for real-time size measurement of irregular crystals as described in claim 1, characterized in that: The specific detection steps for edge detection in step S3 are as follows: S3-1: Select the complete lower edge and part of the left and right edges of the crystal from the images in the training set as detection images; S3-2: The Scharr operator is used to calculate the gradients of each detection image in the horizontal and vertical directions respectively, and the gradients in the two directions are combined into a gradient magnitude map. S3-4: Automatic thresholding segmentation effectively separates edge regions from the background and acquires edge region images; S3-5: By using a multi-iteration method of taking the maximum pixel value in the neighborhood of each pixel in the edge region image, the edge break of the edge region image is completed, and the feature extraction image is obtained. S3-6: Scan vertically from top to bottom in the upper part of the feature extraction image to find the first edge pixel as the starting detection point. Establish an access matrix with the same size as the input feature extraction image. This access matrix is ​​used to record the pixel access status. Initially, all pixels are marked as unaccessed. Use the starting detection point as a node and check the 8 neighboring pixels of the starting detection point to determine whether the neighboring pixels are edges and unaccessed. Add newly discovered edge pixels to the processing queue and mark the edge pixels as accessed. Iterate through the nodes in the access matrix until no new connected edge pixels are found, thus completing edge extraction and obtaining the edge feature map.

3. The method for real-time size measurement of irregular crystals as described in claim 2, characterized in that: Step S3 also includes the following steps to refine the edge feature map: S3-7: The crystal rotation angle increment corresponding to each frame's edge feature map is... N represents the number of frames in one rotation cycle. Assume the edge point coordinates of the current frame are... Calculate the polar angle of the edge point in the edge feature map. : ； in, Represents the coordinates of the geometric center; S3-8: Cache the edge point set of three consecutive cycles: current frame The edge point set of the previous period Edge point set of the previous two cycles For the current frame Points in ,exist and The search term finds corresponding points that satisfy both angle matching and positional proximity, with angle matching being the target point. polar angle Must meet The proximity of locations satisfies the spatial distance requirement. , The distance is Euclidean. S3-9: Calculate edge points based on the matching results within 3 cycles. Trust level : ； in: This indicates that the current frame was successfully matched. This indicates that the previous period's match was successful. This indicates that the first two cycles were successfully matched; a value of 0 indicates that no match was found. , , These are the weighting coefficients; S3-10: Perform edge point filtering: retain Edge points, remove point; S3-11: To A high-confidence point, if it is not detected in a certain period, is based on the crystal rotation direction and pulling speed. It predicts the position of the edge within the cycle and performs a small-scale search and completion near the predicted position to avoid the real edge breaking due to brief detection failures. ； in These are the coordinates of the point corresponding to the previous cycle. To increase lifting speed, These are the weighting coefficients.

Images