Diamond surface quality online sorting method and system for MPCVD

Abstract

The invention provides a diamond surface quality online sorting method and system for MPCVD, relates to the technical field of diamond growth monitoring, is applied to MPCVD equipment, and the MPCVD equipment is used for generating diamond crystals. Preprocessing the original image to obtain a preprocessed image; s2, extracting and processing the preprocessed image through multi-dimensional surface features to respectively obtain texture features, light intensity distribution features and geometric morphology features of the diamond; s3, a quality evaluation model is constructed and used for inputting texture features, light intensity distribution features and geometric morphology features of the diamond and outputting a diamond surface quality detection result, online real-time monitoring of the surface quality in the MPCVD diamond growth process is achieved, a traditional off-line detection mode is broken through, and the quality of the diamond is evaluated. The problem that the surface defect of the crystal cannot be dynamically tracked within a hundreds of hours of growth cycle is solved.

Description

Technical Field

[0001] This invention relates to the field of diamond growth monitoring technology, and in particular to an online sorting method and system for the surface quality of MPCVD diamonds. Background Technology

[0002] Microwave plasma chemical vapor deposition (MPCVD) is a core technology for preparing high-quality, large-size single-crystal diamonds. This process is typically carried out in a high-temperature (approximately 1000°C), low-pressure, and strong plasma environment, with a single growth cycle lasting hundreds of hours, resulting in extremely high energy, gas, and time costs.

[0003] Currently, diamond quality testing and analysis technologies mainly focus on offline processes after growth, and none of them can solve the problem of real-time monitoring and decision-making during the growth process. Specifically, they have the following limitations: The problem of "lack of real-time monitoring" of surface quality during growth: Existing technologies are all offline and post-event detection, which cannot dynamically track whether defects such as polycrystalline structures, inclusions, abnormal steps, or cracks appear on the crystal surface during a growth cycle of hundreds of hours. The monitoring methods and growth equipment suffer from insufficient integration and automation; currently, the methods that rely on manual visual inspection or removal for inspection after growth cannot form a closed-loop control.

[0004] The problem of "lack of automated sorting decision-making" based on real-time quality status; even if monitoring data is obtained, the existing technology system lacks a mechanism for automatically judging and executing decisions at the process level. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide an online, real-time, and automated quality monitoring and sorting method and system for the MPCVD diamond growth process, and a technical means to automatically execute "continue growth" or "interrupt sorting" decisions based on the judgment results. This reduces the blindness of the production process, high scrap rates, and huge resource waste.

[0006] In a first aspect, embodiments of the present invention provide an online sorting method for the surface quality of diamond obtained by MPCVD, applied to an MPCVD device used to generate diamond crystals, comprising: An image acquisition device, configured to acquire raw images of diamond growth in a target wavelength band in response to a control command, and a deposition chamber for holding diamond crystals, characterized in that it comprises: S1. Obtain the original image of diamond growth through an image acquisition device, and preprocess the original image to obtain a preprocessed image; S2. The preprocessed image is processed by multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features and geometric morphology features of diamond. S3. Construct a quality assessment model, which is used to input the texture features, light intensity distribution features and geometric morphology features of the diamond, and output the diamond surface quality test results.

[0007] Furthermore, the steps for preprocessing the original image to obtain the preprocessed image include: The first reference image is obtained by performing non-uniformity correction on the acquired original image using background subtraction. ; For the first reference image The second reference image is obtained by performing two-dimensional median filtering on each color channel. ; The second reference image Multiple block histograms were obtained by segmentation. The histogram of each block is cropped, restricted, and equalized, and the inter-block boundaries are eliminated by interpolation to obtain the preprocessed image.

[0008] Furthermore, in step S2, where the preprocessed image is processed through multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features, and geometric morphology features of the diamond, the texture features are determined in the following manner: The local traditional gray-level co-occurrence matrix is ​​calculated using the sliding window method on the preprocessed image. Based on the local traditional gray-level co-occurrence matrix, the texture disorder index (TDI) is obtained. The texture disorder index (TDI) is used to quantify texture abrupt changes caused by polycrystalline nucleation or surface contamination. Where TDI is the standard deviation of the LTU values ​​of all pixels in the region; ; in, For the current diamond region R diamond Local contrast in the neighborhood For local homogeneity, where To prevent small constants from being divided by zero, For local texture uniformity. Further, in step S2, where the preprocessed image is processed through multi-dimensional surface feature extraction to obtain the diamond's texture features, light intensity distribution features, and geometric morphology features, the light intensity distribution features are determined using the following methods: Calculate the statistical characteristics of the surface intensity gradient of the preprocessed image, and obtain the coefficient of variation of the gradient magnitude. and the proportion of low gradient uniform regions ; Among them, the coefficient of variation of the gradient magnitude is obtained. and the proportion of low gradient uniform regions The steps include: Calculate the diamond region Inner grayscale image gradient magnitude plot ; Based on gradient magnitude map Calculate the coefficient of variation of gradient magnitude The coefficient of variation Used to measure light intensity uniformity: ; in, and Gradient magnitude diagrams Mean and standard deviation within the region; Based on gradient magnitude map Calculate the proportion of low gradient uniform regions The proportion of the low gradient uniform region Used to reflect the specific gravity of flat, smooth parts of a surface: ; in, θ —Empirical threshold —Counting function.

[0009] Furthermore, in step S2, where the preprocessed image is processed through multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features, and geometric morphology features of the diamond, the geometric morphology features are determined using the following methods: Using the Canny algorithm Edge detection is performed on the region to obtain a binary edge map E; Based on the binary edge map E, all edge contours are extracted, and polygon approximation is performed on each contour to obtain the contour line segments of the polygon. Calculate the rate of change of direction for each contour line segment, and calculate the standard deviation of the rate of change of direction to obtain the contour curvature disorder. CCD ; By obtaining the contour curvature disorder CCD A high curvature disorder profile is obtained by comparing it with the preset profile curvature disorder. The density of anomalous edges is calculated based on statistical analysis of edge pixels belonging to high curvature disorder contours. AED : ; Where N is the diamond analysis region. The total number of pixels within.

[0010] Furthermore, the steps in constructing the quality assessment model using the acquired diamond growth history data include: Calculate the quality score using a weighted scoring model. ; ; Norm() is a normalization function that maps eigenvalues ​​to the interval [0,1]. w1—First weighting coefficient; w2—Second weighting coefficient; w3—Third weighting coefficient; w4—Fourth weighting coefficient; w1-4 was determined through optimization using historical data; The weighted scoring model uses a sorting threshold. Determine the current crystal quality.

[0011] Furthermore, the steps in constructing the quality assessment model using the acquired diamond growth history data include: The historical data of diamond growth was labeled to obtain image samples of normal growth stage and image samples of abnormal growth stage. A binary classification SVM model is trained by combining the image samples from the growth stage and their feature vectors. The binary classification SVM model is used to output normal or abnormal category labels and corresponding confidence scores; The binary classification SVM model triggers sorting when M consecutive frames are classified as abnormal and the average confidence level is higher than a preset threshold.

[0012] Furthermore, the image acquisition device also includes an LED array light source, used to provide supplemental lighting for the image acquisition device when acquiring the original image of diamond growth.

[0013] Furthermore, the image acquisition device also includes a filter for enabling the CMOS camera to acquire reflected light images of the diamond crystal surface under near-infrared filtering conditions of a preset wavelength.

[0014] An online sorting system for the surface quality of MPCVD diamond, employing any one of the above-described methods for online sorting of the surface quality of MPCVD diamond, includes: An online image acquisition module is used to acquire raw images of diamond growth and preprocess the raw images to obtain preprocessed images; The image processing and sorting decision module is used to process the preprocessed image to obtain the texture features, light intensity distribution features and geometric morphology features of the diamond, and generate the diamond surface quality detection results through the quality assessment model. The sorting execution module is used to make sorting decisions based on the diamond surface quality test results.

[0015] The embodiments of the present invention bring the following beneficial effects: 1. This invention enables real-time, online quality monitoring of the growth process, filling a gap in process monitoring. Compared to existing technologies that can only perform offline inspections after growth, this invention, by integrating a high-temperature resistant optical imaging system into the MPCVD equipment, achieves for the first time dynamic, non-destructive, in-situ monitoring of the surface morphology and optical characteristics of diamond crystals under a continuous high-temperature plasma growth environment lasting hundreds of hours. This breaks the "black box" state of the growth process, enabling process engineers to perceive the growth status in real time.

[0016] 2. Transforming subjective experience-based judgment into objective, quantitative, and automated sorting decisions. Unlike experience-based judgments relying on human visual inspection, this invention extracts multi-dimensional quantitative features such as texture contrast, light intensity standard deviation, and edge density through image processing algorithms, and performs real-time analysis based on a pre-set quality assessment model (such as weighted scoring or a machine learning classifier). In simulated tests, using a pre-trained SVM model to identify abnormal growth images containing polycrystalline structures and inclusions, the accuracy rate reaches over 95%, significantly outperforming the stability and consistency of human judgment.

[0017] 3. An automated closed loop of "monitoring-decision-execution" has been formed, enabling proactive intervention at the process level. This is a fundamental advancement that distinguishes it from all offline inspection technologies. When the system determines a quality anomaly, it can automatically trigger the sorting mechanism. In contrast, traditional methods may only detect problems after hundreds of hours of growth. This closed-loop control directly avoids all wasted energy, gas, and time from defect generation to the end of growth, shifting quality control from "post-inspection" to "in-process intervention."

[0018] 4. It provides a traceable process quality data chain. The system continuously records time-series images and feature data, providing a dynamic dataset that is unavailable through traditional methods for analyzing defect generation mechanisms and optimizing growth process parameters, thus facilitating continuous process improvement. Attached Figure Description

[0019] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0020] Figure 1A flowchart of an online sorting method for the surface quality of MPCVD diamond is provided in an embodiment of the present invention; Figure 2 A structural diagram of an online sorting system for the surface quality of MPCVD diamond is provided in another embodiment of the present invention; Figure 3 This is a schematic diagram of the working principle of an image processing and sorting decision module provided in another embodiment of the present invention; Figure 4 A schematic diagram illustrating the working principle of online surface quality sorting provided in another embodiment of the present invention; Figure label: 100—Online image acquisition module; 200—Image processing and sorting decision module; 300—Sorting execution module; 400—MPCVD equipment; 101—High-temperature resistant optical window; 102—Camera; 103—Filter; 104—LED light source; 105—Cooling protective sleeve; 201—Image preprocessing unit; 202—Feature extraction unit; 203—Quality assessment model unit; 204—Sorting decision unit; 301—Movable robotic arm or pneumatic push rod; 302—End effector; 303—Waste temporary storage bin; 401—Deposition chamber; 402—Deposition stage; 403—Diamond crystal. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Currently, existing MPCVD single-crystal diamond growth technology suffers from a series of technical defects due to the lack of online, real-time, and automated surface quality monitoring and sorting capabilities. Based on this, the present invention provides an online sorting method and system for MPCVD diamond surface quality, which can identify crystal surface quality anomalies and perform sorting decisions in real time and automatically during the growth process.

[0023] To facilitate understanding of this embodiment, we will first describe an online sorting method for the surface quality of diamond in MPCVD, as disclosed in this embodiment of the invention, applied to an MPCVD equipment. The MPCVD equipment is used to generate diamond crystals and includes: An image acquisition device, configured to acquire raw images of diamond growth at a preset image acquisition frequency in response to control commands, and a deposition chamber for holding diamond crystals, characterized in that, as Figure 1 As shown, the method steps include: S1. Obtain the original image of diamond growth through an image acquisition device, and preprocess the original image to obtain a preprocessed image; S2. The preprocessed image is processed by multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features and geometric morphology features of diamond. S3. Construct a quality assessment model, which takes as input the texture features, light intensity distribution features, and geometric morphology features of the diamond, and outputs the diamond surface quality inspection results (e.g., Figure 4 (As shown).

[0024] In this embodiment, online real-time monitoring of surface quality during MPCVD diamond growth solves the "lack of real-time performance" problem, which prevents dynamic tracking of crystal surface defects (such as polycrystalline structures, inclusions, abnormal steps, and cracks) over growth cycles lasting hundreds of hours. By combining multi-dimensional surface feature extraction (texture, light intensity distribution, and geometric morphology) with a quality assessment model, the detection dimensions are comprehensive, significantly improving the accuracy of surface quality judgment and avoiding the limitations of single-feature detection. By constructing a closed-loop online sorting mechanism, image acquisition, feature extraction, quality assessment, and sorting decisions are integrated, overcoming the shortcomings of existing technologies in terms of "insufficient integration and automation" and "lack of automated sorting decisions," reducing manual intervention costs and improving production efficiency.

[0025] In conjunction with the first aspect, embodiments of the present invention provide a possible implementation method, wherein, as Figure 3 As shown, the steps in S1 for preprocessing the original image to obtain the preprocessed image include: The first reference image is obtained by performing non-uniformity correction on the acquired original image using background subtraction. ; Specifically, the system first acquires multiple background images and calculates their pixel grayscale average values ​​under a stable state where no substrate is present, only hydrogen gas is introduced, and the plasma is ignited, to generate a first reference image. .

[0026] In actual testing, the images acquired in real time will be... Correct according to the formula: ; Here, C is a constant offset value (usually 10% of the median gray level of the image) used to prevent negative values ​​or overall over-darkness after correction. This step can significantly improve the overall uniformity of the image.

[0027] Background caused by optical window contamination or uneven illumination is subtracted through non-uniformity correction.

[0028] For the first reference image The second reference image is obtained by performing two-dimensional median filtering on each color channel. ; Specifically, two-dimensional median filtering is performed independently on each color channel (R, G, B) of the corrected image. A 5×5 pixel square sliding window W is used, with the center pixel defined as... The set of neighboring pixels is: ; The output value of median filtering Calculated using the following formula: .

[0029] Median filtering is performed using a 5x5 pixel window to eliminate shot noise.

[0030] The second reference image Multiple block histograms were obtained by segmentation. The histogram of each block is cropped, restricted, and equalized, and the inter-block boundaries are eliminated by interpolation to obtain the preprocessed image.

[0031] Specifically, the image is divided into blocks, the histogram of each block is cropped and restricted to suppress noise, then equalized to enhance local contrast, and finally interpolation is used to eliminate the boundaries between blocks.

[0032] ; in G is the block size, G is the number of gray levels, and ClipLimit is the clipping threshold.

[0033] Enhance the visibility of surface microstructures through contrast-limited adaptive histogram equalization.

[0034] In this embodiment, background subtraction effectively corrects the non-uniformity of the original image, eliminating background interference such as optical window contamination and uneven illumination, providing a clean image foundation for subsequent feature extraction and improving detection accuracy. Two-dimensional median filtering is used to specifically eliminate shot noise, avoiding noise interference with surface features and ensuring the integrity of subtle features such as texture and light intensity. Block histogram cropping and equalization processes suppress noise amplification and enhance local contrast, highlighting the fine structure of the diamond surface (such as minute defects and texture variations). Simultaneously, interpolation eliminates inter-block boundaries, ensuring image coherence and providing a high-quality pre-processed image for multi-dimensional feature extraction.

[0035] In conjunction with the first aspect, this embodiment of the invention provides a possible implementation method, wherein, in step S2, the texture features, light intensity distribution features, and geometric morphology features of diamond are obtained by multi-dimensional surface feature extraction processing of the preprocessed image, the texture features are determined in the following manner: The local traditional gray-level co-occurrence matrix is ​​calculated using the sliding window method on the preprocessed image. Based on the local traditional gray-level co-occurrence matrix, the texture disorder index (TDI) is obtained. The texture disorder index (TDI) is used to quantify texture abrupt changes caused by polycrystalline nucleation or surface contamination. Where TDI is the standard deviation of the LTU values ​​of all pixels in the region; ; in, For the current diamond region R diamond Local contrast in the neighborhood For local homogeneity, where To prevent small constants from being divided by zero, This refers to local texture uniformity. Traditional gray-level co-occurrence matrices (GLCMs) are affected by plasmon glow gradients when calculating texture features across the entire image. This application addresses this issue by optimizing the uniformity of textures in each extracted diamond region. Within the diamond region, the sliding window method is used to calculate the local GLCM, and the "texture disorder index" is defined as shown in the following formula to quantify texture abrupt changes caused by polycrystalline nucleation or surface contamination. The texture disorder index TDI for the entire diamond region is the standard deviation of the LTU values ​​of all pixels in the region. High-quality single-crystal surfaces are uniform and have low TDI values; when polycrystalline, defective, or contaminated surfaces appear, local texture abrupt changes occur, and the TDI value increases significantly.

[0036] In this embodiment, a sliding window method is used to calculate the local gray-level co-occurrence matrix of the preprocessed image, avoiding interference from the plasma glow gradient in the full-image calculation and accurately focusing on the local texture features of the diamond region. By defining the Texture Disorder Index (TDI), texture abrupt changes caused by polycrystalline nucleation and surface contamination can be quantified, enabling accurate identification of subtle texture defects, distinguishing between high-quality single crystals and defective crystals, and providing a core texture dimension basis for quality assessment. By calculating the ratio of local contrast to local homogeneity, combined with a small constant to prevent division by zero, computational stability is ensured, while enhancing sensitivity to texture abrupt changes and improving the reliability of texture feature detection.

[0037] In conjunction with the first aspect, this embodiment of the invention provides a possible implementation method, wherein, in step S2, the diamond texture features, light intensity distribution features, and geometric morphology features are obtained by extracting multi-dimensional surface features from the preprocessed image, and the light intensity distribution features are determined in the following manner: Calculate the statistical characteristics of the surface intensity gradient of the preprocessed image, and obtain the coefficient of variation of the gradient magnitude. and the proportion of low gradient uniform regions ; Among them, the coefficient of variation of the gradient magnitude is obtained. and the proportion of low gradient uniform regions The steps include: Calculate the diamond region Inner grayscale image gradient magnitude plot ; Based on gradient magnitude map Calculate the coefficient of variation of gradient magnitude The coefficient of variation Used to measure light intensity uniformity: ; in, and Gradient magnitude diagrams Mean and standard deviation within the region; Based on gradient magnitude map Calculate the proportion of low gradient uniform regions The proportion of the low gradient uniform region Used to reflect the specific gravity of flat, smooth parts of a surface: ; in, θ —Empirical threshold (e.g., 50% of the gradient mean); —Counting function.

[0038] A well-grown surface should have low GCV and high LGP.

[0039] Directly calculating the overall strength standard deviation is susceptible to the influence of uniform but overall brightness differences. This invention introduces gradient domain analysis, which captures micro-area reflective anomalies caused by uneven growth temperature, impurity doping, or stress by calculating the statistical characteristics of the surface strength gradient. It is more sensitive to weak defects.

[0040] In this embodiment, the light intensity distribution characteristics are analyzed by gradient amplitude map. Compared with directly calculating the overall intensity standard deviation, it is more resistant to the interference of uniform brightness differences and more sensitive to micro-area reflection anomalies caused by uneven growth temperature, impurity doping or stress, so as to accurately capture weak light intensity defects.

[0041] Among them, the gradient amplitude variation coefficient (GCV) is used to quantify the uniformity of light intensity, and the proportion of low gradient uniform regions (LGP) is used to reflect the flatness and smoothness of the surface. The two indicators characterize the light intensity distribution from two dimensions: "uniformity" and "smoothness proportion", comprehensively characterizing the light reflection characteristics of the diamond surface and providing key light intensity dimension support for quality assessment.

[0042] By combining empirical thresholds with counting functions, the calculation of the proportion of low-gradient uniform regions is made more consistent with actual detection scenarios, enhancing the practicality and accuracy of the indicators and helping to distinguish between normally grown and abnormally grown crystal surfaces. In conjunction with the first aspect, this embodiment of the invention provides a possible implementation method, wherein, in step S2, the preprocessed image is processed through multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features, and geometric morphology features of diamond, the geometric morphology features are determined in the following manner: Using the Canny algorithm Edge detection is performed on the region to obtain a binary edge map E; Based on the binary edge map E, all edge contours are extracted, and polygon approximation is performed on each contour to obtain the contour line segments of the polygon. Calculate the rate of change of direction for each contour line segment, and calculate the standard deviation of the rate of change of direction to obtain the contour curvature disorder. CCD ; By obtaining the contour curvature disorder CCD A high curvature disorder profile is obtained by comparing it with the preset profile curvature disorder. The density of anomalous edges is calculated based on statistical analysis of edge pixels belonging to high curvature disorder contours. AED : ; Where N is the diamond analysis region. The total number of pixels within.

[0043] Existing technologies cannot distinguish between regular edges resulting from normal crystal step growth and messy edges resulting from abnormal three-dimensional island growth, cracks, or attachments by simply using edge density; this invention combines edge detection and contour curvature analysis to distinguish edge types.

[0044] In this embodiment, by combining Canny edge detection and contour curvature analysis, the limitations of traditional simple edge density detection are overcome. It can effectively distinguish between the regular edges of normal crystal step growth and the messy edges caused by three-dimensional island-like abnormal growth, cracks or attachments, thereby improving the accuracy of geometric defect identification.

[0045] Contour curvature disorder (CCD) quantifies contour regularity by the standard deviation of the rate of change of direction, while abnormal edge density (AED) statistically measures the percentage of edge pixels in highly disordered contours. The two indicators, CCD and AED, describe the geometric shape from two dimensions: "contour regularity" and "abnormal edge percentage," comprehensively capturing surface morphology defects.

[0046] Polygon approximation processing simplifies the complexity of contour calculation, improves computational efficiency while ensuring detection accuracy, adapts to the timeliness requirements of online real-time detection, and ensures fast and reliable extraction of geometric features.

[0047] In conjunction with the first aspect, this embodiment of the invention provides a possible implementation method, wherein the step of constructing a quality assessment model using the diamond growth history data obtained in S3 includes: Calculate the quality score using a weighted scoring model. ; ; Norm() is a normalization function that maps eigenvalues ​​to the interval [0,1]. w1—First weighting coefficient; w2—Second weighting coefficient; w3—Third weighting coefficient; w4—Fourth weighting coefficient; w1-4 was determined through optimization using historical data; The weighted scoring model uses a sorting threshold. Determine the current crystal quality.

[0048] In this embodiment, a weighted scoring model is used to integrate multi-dimensional features such as texture (TDI), light intensity distribution (GCV, LGP), and geometric shape (AED). Combined with weighting coefficients optimized from historical data, the features are scientifically fused, and the quality score Q can comprehensively and objectively reflect the surface quality of diamond.

[0049] Sorting threshold Q th The system provides clear quality judgment standards, is easy to operate, can be calibrated according to actual production scenarios, adapts to production needs with different quality requirements, realizes standardized and automated quality sorting, provides immediate support for sorting decisions, and improves the continuity of the production process.

[0050] In conjunction with the first aspect, this embodiment of the invention provides another possible implementation, wherein the step of constructing a quality assessment model using the diamond growth history data obtained in S3 includes: The historical data of diamond growth was labeled to obtain image samples of normal growth stage and image samples of abnormal growth stage. A binary classification SVM model is trained by combining the image samples from the growth stage and their feature vectors. The binary classification SVM model is used to output normal or abnormal category labels and corresponding confidence scores; The binary classification SVM model triggers sorting when M consecutive frames are classified as abnormal and the average confidence level is higher than a preset threshold.

[0051] In this embodiment, a binary classification SVM model is trained using historical labeled data. Utilizing image samples and feature vectors from both normal and abnormal growth stages, the model possesses accurate category recognition capabilities, outputting clear normal / abnormal labels and confidence levels, thus improving the intuitiveness and reliability of quality judgment. The trigger condition of N consecutive abnormal frames with an average confidence level exceeding a preset threshold effectively filters out single-shot misjudgment interference, enhancing the anti-interference capability of sorting decisions and avoiding missorting due to accidental frame anomalies, ensuring production stability. The binary classification characteristics of the SVM model adapt to the core "qualified / unqualified" requirement of online sorting. The trained model has fast inference speed, meeting the timeliness requirements of real-time detection, and enabling rapid identification and sorting triggering of abnormally grown crystals.

[0052] In conjunction with the first aspect, this embodiment of the invention provides a possible implementation method, wherein the image acquisition device further includes: an LED array light source, used to provide supplementary lighting for the image acquisition device when acquiring the original image of diamond growth.

[0053] In this embodiment, the LED array light source provides supplementary lighting support for image acquisition. Its adjustable brightness adapts to the lighting environment of different growth stages within the MPCVD equipment, ensuring clear diamond surface images can still be acquired even under high temperature and strong plasma conditions, thus improving the quality of the original image. The collaborative work between the light source and the image acquisition device adapts to the requirements of online real-time detection scenarios, ensuring the stability and consistency of image acquisition and providing hardware support for continuous and reliable quality monitoring.

[0054] In conjunction with the first aspect, this embodiment of the invention provides a possible implementation method, wherein the image acquisition device further includes: a filter for enabling the CMOS camera to acquire an image of reflected light from the surface of a diamond crystal under near-infrared filtering conditions of a preset wavelength.

[0055] Specifically, under normal MPCVD growth conditions, the image acquisition module is triggered synchronously. The camera acquires images with an exposure time of 1-5 seconds under near-infrared filtering conditions with a wavelength of 700-800nm ​​through a filter, effectively suppressing plasma emission interference.

[0056] In this embodiment, a near-infrared filter of a specific wavelength can effectively filter out strong interference light emitted by the plasma in the MPCVD equipment (such as the 656nm Hα line of hydrogen atoms), highlight the reflected light signal on the diamond crystal surface, improve the image signal-to-noise ratio, and ensure that the original image can truly reflect the surface features.

[0057] By combining filters with a CMOS camera, the image acquisition device can adapt to the complex growth environment of MPCVD, achieving interference-free, high-fidelity image acquisition and providing key hardware support for online real-time monitoring.

[0058] Secondly, embodiments of the present invention also provide an online sorting system for the surface quality of MPCVD diamond, which applies the online sorting method for the surface quality of MPCVD diamond described in any of the above claims, including: An online image acquisition module is used to acquire raw images of diamond growth and preprocess the raw images to obtain preprocessed images; The image acquisition settings are as follows: Filtering wavelength: preferably 700-800 nm, to avoid the strongest emission line of the plasma.

[0059] Acquisition frequency: 1-60 frames / minute, which can be adjusted according to the growth stage (more frequent in the early stage).

[0060] Illumination angle: The LED light source is grazing at an angle of 20-45 degrees to the surface normal.

[0061] Feature threshold (example): A contrast ratio (after normalization) > 0.6 may be abnormal.

[0062] Edge density > 0.05 may be abnormal.

[0063] Sorting trigger conditions: Mass fraction Q_th: Recommended range 0.65-0.75, can be calibrated through initial experiments.

[0064] The number of consecutive abnormal frames N is 2-5 frames, providing anti-interference capability.

[0065] The image processing and sorting decision module is used to process the preprocessed image to obtain the texture features, light intensity distribution features and geometric morphology features of the diamond, and generate the diamond surface quality detection results through the quality assessment model. The sorting execution module is used to make sorting decisions based on the diamond surface quality test results.

[0066] Specifically, such as Figure 2 As shown, this online surface quality sorting system mainly comprises three functional modules: an online image acquisition module 100, an image processing and sorting decision module 200, and a sorting execution module 300. This system is deeply integrated with the MPCVD equipment 400.

[0067] The online image acquisition module 100 is responsible for acquiring real-time images of the diamond crystal surface in the growth environment; it includes a high-temperature resistant optical window 101, a high-speed color or high dynamic range CMOS camera 102, a specific wavelength filter 103, an adjustable brightness LED array light source 104, and a cooling protective sleeve 105.

[0068] Connection and Position: The high-temperature resistant optical window 101 is embedded in the side wall or top of the deposition chamber 401 of the MPCVD equipment 400, communicating with the interior. The camera 102 and the light source 104 are placed outside the chamber, directly facing the optical window 101. A filter 103 is mounted in front of the camera lens to filter out specific strong interfering light emitted by the plasma (such as the 656nm Hα line of hydrogen atoms), enhancing the contrast of surface features. A cooling protective sleeve 105 ensures stable operation of the camera in an environment close to the high-temperature chamber.

[0069] Working principle: During the growth process, the LED array light source 104 illuminates the diamond crystal 403 on the deposition stage 402 using low-angle grazing illumination or coaxial illumination. The camera 102 captures the reflected light image of the crystal surface at a preset frequency (e.g., 1 frame per minute) through the filter 103 and optical window 101. Grazing illumination can highlight morphological features such as surface steps and polycrystalline protrusions.

[0070] The image processing and sorting decision module 200 is the "brain" of the system, responsible for analyzing images and making sorting decisions. It includes an image preprocessing unit 201, a feature extraction unit 202, a quality assessment model unit 203, and a sorting decision unit 204. These units run as software algorithms in an industrial computer or embedded processor.

[0071] Connection: This module receives raw image data from camera 102 via a data cable. Its decision result (electrical signal) is output to the sorting execution module 300.

[0072] Working principle: (Dynamic process) The original image first enters the image preprocessing unit 201 for noise reduction, contrast enhancement, and edge sharpening. The processed image is then sent to the feature extraction unit 202 to calculate key feature values. These feature values ​​are input into the quality assessment model unit 203 for real-time scoring. Finally, the sorting decision unit 204 generates control commands for "continue growing" or "execute sorting" according to preset logic based on the scoring results.

[0073] The sorting execution module 300 is responsible for physically executing sorting decisions; it includes a movable robotic arm or pneumatic push rod 301, an end effector 302, such as a vacuum suction cup or a high-temperature resistant gripper, and a waste storage bin 303.

[0074] Connection and Position Relationship: The robotic arm 301 is installed inside the deposition chamber 401 or connected to the main chamber via a secondary chamber. Its range of motion covers the deposition stage 402. The waste storage bin 303 is located in a non-interference area within the chamber.

[0075] Working principle: Upon receiving the "execute sorting" command, the control system first adjusts the MPCVD process parameters to a safe standby state (e.g., reducing power). Then, the robotic arm 301 moves above the target crystal, and the end effector 302 grasps or adsorbs the diamond wafer deemed abnormal, removing it from the deposition stage 402 and placing it in the waste storage bin 303. After completion, the system can reset and continue or begin the growth of other crystals.

[0076] The system provided in this application embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0077] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0078] The foregoing has described specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0079] In the description of the embodiments of this application, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the embodiments of this application. In the embodiments of this application, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in a suitable manner in any one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in the embodiments of this application, as well as the features of different embodiments or examples.

[0080] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of embodiments of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0081] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order according to the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0082] In the embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the displayed or discussed mutual couplings, direct couplings, or communication connections may be through some interfaces; indirect couplings or communication connections between devices or units may be electrical, mechanical, or other forms.

[0083] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0084] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present application should be included within the scope of protection of the present application.

Claims

1. A method for online sorting of diamond surface quality in MPCVD, characterized in that, Applied to an MPCVD (Multi-Level Video Deposition) device, the MPCVD device being used to generate diamond crystals, comprising: An image acquisition device, configured to acquire raw images of diamond growth at a preset image acquisition frequency in response to control commands, and a deposition chamber for holding diamond crystals, comprising: S1. Obtain the original image of diamond growth through an image acquisition device, and preprocess the original image to obtain a preprocessed image; S2. The preprocessed image is processed by multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features and geometric morphology features of diamond. S3. Construct a quality assessment model, which is used to input the texture features, light intensity distribution features and geometric morphology features of the diamond, and output the diamond surface quality test results.

2. The method for online sorting of surface quality of MPCVD diamond according to claim 1, characterized in that, The steps for preprocessing the original image to obtain the preprocessed image include: The first reference image is obtained by performing non-uniformity correction on the acquired original image using background subtraction. ; For the first reference image The second reference image is obtained by performing two-dimensional median filtering on each color channel. ; The second reference image Multiple block histograms were obtained by segmentation. The histogram of each block is cropped, restricted, and equalized, and the inter-block boundaries are eliminated by interpolation to obtain the preprocessed image.

3. The method for online sorting of surface quality of MPCVD diamond according to claim 1, characterized in that, In step S2, where the preprocessed image is extracted using multi-dimensional surface features to obtain the texture features, light intensity distribution features, and geometric morphology features of diamond, the texture features are determined in the following manner: The local traditional gray-level co-occurrence matrix is ​​calculated using the sliding window method on the preprocessed image. Based on the local traditional gray-level co-occurrence matrix, the texture disorder index (TDI) is obtained. The texture disorder index (TDI) is used to quantify texture abrupt changes caused by polycrystalline nucleation or surface contamination. Where TDI is the standard deviation of the LTU values ​​of all pixels in the region; ； in, For the current diamond region R diamond Local contrast in the neighborhood For local homogeneity, where To prevent small constants from being divided by zero, This refers to the local texture uniformity.

4. The method for online sorting of diamond surface quality in MPCVD according to claim 3, characterized in that, In step S2, where the preprocessed image is extracted using multi-dimensional surface features to obtain the texture features, light intensity distribution features, and geometric morphology features of the diamond, the light intensity distribution features are determined using the following methods: Calculate the statistical characteristics of the surface intensity gradient of the preprocessed image, and obtain the coefficient of variation of the gradient magnitude. and the proportion of low gradient uniform regions ; Among them, the coefficient of variation of the gradient magnitude is obtained. and the proportion of low gradient uniform regions The steps include: Calculate the diamond region Inner grayscale image gradient magnitude plot ; Based on gradient magnitude map Calculate the coefficient of variation of gradient magnitude The coefficient of variation Used to measure light intensity uniformity: ； in, and Gradient magnitude diagrams Mean and standard deviation within the region; Based on gradient magnitude map Calculate the proportion of low gradient uniform regions The proportion of the low gradient uniform region Used to reflect the specific gravity of flat, smooth parts of a surface: ； in, θ —Empirical threshold; —Counting function.

5. The method for online sorting of surface quality of MPCVD diamond according to claim 3, characterized in that, In step S2, where the preprocessed image is processed through multi-dimensional surface feature extraction to obtain the texture features, light intensity distribution features, and geometric morphology features of diamond, the geometric morphology features are determined using the following methods: Using the Canny algorithm Edge detection is performed on the region to obtain a binary edge map E; Based on the binary edge map E, all edge contours are extracted, and polygon approximation is performed on each contour to obtain the contour line segments of the polygon. Calculate the rate of change of direction for each contour line segment, and calculate the standard deviation of the rate of change of direction to obtain the contour curvature disorder. CCD ; By obtaining the contour curvature disorder CCD A high curvature disorder profile is obtained by comparing it with the preset profile curvature disorder. The density of anomalous edges is calculated based on statistical analysis of edge pixels belonging to high curvature disorder contours. AED : ； Where N is the diamond analysis region. The total number of pixels within.

6. The method for online sorting of diamond surface quality in MPCVD according to claim 5, characterized in that, The steps in constructing the quality assessment model using the acquired diamond growth history data include: Calculate the quality score using a weighted scoring model. ; ； Norm() is a normalization function that maps eigenvalues ​​to the interval [0,1]. w1—First weighting coefficient; w2—Second weighting coefficient; w3—Third weighting coefficient; w4—Fourth weighting coefficient; w1-4 was determined through optimization using historical data; The weighted scoring model uses a sorting threshold. Determine the current crystal quality.

7. The method for online sorting of diamond surface quality in MPCVD according to claim 1, characterized in that, The steps in constructing the quality assessment model using the acquired diamond growth history data include: The historical data of diamond growth was labeled to obtain image samples of normal growth stage and image samples of abnormal growth stage. A binary classification SVM model is trained by combining the image samples from the growth stage and their feature vectors. The binary classification SVM model is used to output normal or abnormal category labels and corresponding confidence scores; The binary classification SVM model triggers sorting when M consecutive frames are classified as abnormal and the average confidence level is higher than a preset threshold.

8. The method for online sorting of surface quality of diamond in MPCVD according to claim 1, characterized in that, The image acquisition device also includes an LED array light source, used to provide supplemental lighting for the image acquisition device when acquiring the original image of diamond growth.

9. The method for online sorting of surface quality of MPCVD diamond according to claim 1, characterized in that, The image acquisition device further includes a filter for enabling the CMOS camera to acquire reflected light images of the diamond crystal surface under near-infrared filtering conditions of a preset wavelength.

10. An online sorting system for the surface quality of diamonds in MPCVD, characterized in that, An online sorting method for the surface quality of MPCVD diamond using any one of claims 1-9 includes: An online image acquisition module is used to acquire raw images of diamond growth and preprocess the raw images to obtain preprocessed images; The image processing and sorting decision module is used to process the preprocessed image to obtain the texture features, light intensity distribution features and geometric morphology features of the diamond, and generate the diamond surface quality detection results through the quality assessment model. The sorting execution module is used to make sorting decisions based on the diamond surface quality test results.

Images