Method for preparing gallium oxide single crystal through edge-defined film-fed crystal method based on machine vision
By applying a machine vision discrimination model in the process of preparing gallium oxide single crystals in the guide mode method, automatically identifying and adjusting process parameters, the problem of difficult to determine crystal state and parameter adjustment hysteresis in the existing technology is solved, and high-quality single crystals are achieved Preparation and labor costs reduction.
Patent Information
- Application Number
- CN202411645444.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-18
- Publication Date
- 2025-02-18
AI Technical Summary
During the preparation of gallium oxide single crystals by the mode guide method, there are problems such as difficulty in distinguishing the crystal state, certain lag in parameter adjustment, and long personnel training cycle.
Using a machine vision-based method, the machine vision discriminant model is trained through the YOLOv7 algorithm, and the growth interface picture in the process of preparing gallium oxide single crystals is identified and judged, and the process parameters are automatically adjusted.
It realizes rapid response and timely adjustment of process parameters, improves single crystal rate, reduces dependence on operator experience, shortens training cycles, and saves labor costs.
Smart Images
Figure CN119465382A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of preparing gallium oxide single crystals by a guided mold method, and in particular to a method for preparing gallium oxide single crystals by a guided mold method based on machine vision. Background Art
[0002] The guided mold method is one of the methods for artificially producing single crystal materials from the melt, that is, the edge-limited thin film feeding and pulling growth technology. It is to place the melt in a mold with a capillary slit, and the molten liquid rises to the top of the mold by capillary action, forming a thin film and spreading around. At the same time, it is induced to crystallize by the seed crystal. The edge of the top of the mold can control the crystal to be produced in a sheet, tube or a certain geometric shape required. This method is used for shaping and growing single crystal materials, and has the advantages of fast growth rate and stable solid-liquid interface.
[0003] At present, in the process of preparing gallium oxide single crystals by the guided mold method, the growth situation is mainly judged by naked eyes based on the photos taken, and the power, pulling speed and other process parameters are manually adjusted according to the shape, angle, speed, etc. of the shoulder front to obtain high-quality single crystals. However, this growth method has the disadvantages of difficult to judge the crystal state, a certain lag in parameter adjustment, and a long personnel training cycle.
[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention
[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the present invention is to provide a method for preparing gallium oxide single crystals by a guided mold method based on machine vision, aiming to solve the problems in the process of preparing gallium oxide single crystals by the existing guided mold method, such as difficulty in distinguishing the crystal state, certain lag in parameter adjustment, and long personnel training cycle.
[0006] The technical solution of the present invention is as follows:
[0007] A first aspect of the present invention provides a method for preparing a gallium oxide single crystal by a guided mode method based on machine vision, which comprises the steps of:
[0008] Providing a sample library, wherein the sample library includes known image samples, wherein the image is an image of a growth interface in a process of preparing a gallium oxide single crystal by a guided mold method;
[0009] Classify and label the known image samples;
[0010] The YOLOv7 algorithm is used to train all the image samples after classification and annotation to obtain a machine vision discrimination model;
[0011] Obtain the growth interface image during the preparation of gallium oxide single crystals by the current guided mode method;
[0012] Identifying and judging the acquired growth interface image through the machine vision discrimination model to obtain a judgment result;
[0013] According to the judgment result, the process parameters of preparing the gallium oxide single crystal by the guided mode method are adjusted, and the preparation of the gallium oxide single crystal is completed under the adjusted process parameters.
[0014] Optionally, the step of classifying and labeling the known image samples specifically includes:
[0015] Dividing the known image samples into two categories, the two categories including single crystals and miscellaneous crystals;
[0016] Select the valid area of the single crystal image sample and the valid area of the mixed crystal image sample respectively, add a GOOD label in the single crystal valid area, and add a BAD label in the mixed crystal valid area, where the valid area refers to the area containing the crystal shoulder front.
[0017] Optionally, the step of using the YOLOv7 algorithm to train all the classified and labeled image samples specifically includes:
[0018] The YOLOv7 algorithm is used to cut all the image samples after classification and annotation into feature maps, and then the feature maps are characterized into detection box positions and category probabilities through a multi-layer convolutional network learned by a neural network to obtain prediction results.
[0019] Optionally, after the step of using the YOLOv7 algorithm to train all the classified and labeled image samples, the step further includes:
[0020] First, the loss function is used to evaluate the loss between the predicted result and the true label;
[0021] Then, the loss is passed to each layer of the convolutional network through the back-propagation algorithm to update the convolutional network parameters;
[0022] Finally, through iterative training, the convolutional network gradually converges to the optimal state.
[0023] Optionally, the growth interface image obtained during each process of preparing gallium oxide single crystal by the guided mode method is added to the sample library.
[0024] Optionally, the process parameters include power and pulling speed.
[0025] Optionally, a camera is used to obtain a picture of the growth interface during the process of preparing gallium oxide single crystal by the current guided mode method.
[0026] A second aspect of the present invention provides a method for online monitoring of the growth interface during the preparation of gallium oxide single crystals by a guided mode method, comprising the steps of:
[0027] Providing a sample library, wherein the sample library includes known image samples, wherein the image is an image of a growth interface in a process of preparing a gallium oxide single crystal by a guided mold method;
[0028] Classify and label the known image samples;
[0029] The YOLOv7 algorithm is used to train all the image samples after classification and annotation to obtain a machine vision discrimination model;
[0030] Obtain the growth interface image during the preparation of gallium oxide single crystals by the current guided mode method;
[0031] The acquired growth interface image is identified and judged by the machine vision discrimination model to obtain a judgment result.
[0032] Optionally, the judgment result includes the probability of single crystal and the probability of impure crystal.
[0033] Beneficial effects: The present invention applies machine vision to the process of preparing gallium oxide single crystals by the guided mold method. According to the growth interface picture taken by the camera, the machine vision discrimination model is used to judge whether the interface of crystal growth is normal, so as to adjust the process parameters such as power and pulling speed, and then prepare high-quality single crystals. Compared with the existing method of judging crystal growth by naked eyes, the present invention uses the machine vision discrimination model to judge crystal growth, which has the advantages of rapid response, timely adjustment of process parameters, and saving labor costs. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 Schematic diagram for adding labels to the effective area of a single crystal.
[0035] Figure 2 A schematic diagram for adding labels to the effective area of the miscellaneous crystal.
[0036] Figure 3 This is a diagram of the judgment results obtained by applying the machine vision discrimination model in the process of preparing gallium oxide single crystals by the guided mold method.
[0037] Figure 4 Schematic diagram of the process of building and applying machine vision discrimination models. DETAILED DESCRIPTION
[0038] The present invention provides a method for preparing gallium oxide single crystals by a guided mode method based on machine vision. In order to make the purpose, technical solution and effect of the present invention clearer and more specific, the present invention is further described in detail below. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0039] At present, in the process of preparing gallium oxide single crystals by the guided mold method, the growth situation is mainly judged by naked eyes based on the photos taken, and the power, pulling speed and other process parameters are manually adjusted according to the shape, angle, speed, etc. of the shoulder front to obtain high-quality single crystals. However, this growth method has the disadvantages of difficult to judge the crystal state, a certain lag in parameter adjustment, and a long personnel training cycle.
[0040] Machine Vision (MV) is an inspection machine equipped with sensing visual instruments (such as autofocus cameras or sensors), of which optical inspection instruments account for a very high proportion. It can be used to detect defects in various products, or to judge and select objects, etc. It is used in automated production lines to calibrate and position materials. It is the most industrialized part of computer vision, and is mainly used in factory automation inspection and robotics industries.
[0041] The present invention applies machine vision to the process of preparing gallium oxide single crystals by the guided mold method. Based on the growth interface photos taken by the camera, the machine vision discrimination model is used to judge whether the interface of crystal growth is normal, thereby adjusting the process parameters such as power and pulling speed, and then obtaining high-quality gallium oxide single crystals. The present invention has the advantages of rapid response, timely adjustment, and saving labor costs.
[0042] Specifically, an embodiment of the present invention provides a method for preparing a gallium oxide single crystal by a guided mode method based on machine vision, which comprises the steps of:
[0043] Providing a sample library, wherein the sample library includes known image samples, wherein the image is an image of a growth interface in a process of preparing a gallium oxide single crystal by a guided mold method;
[0044] Classify and label the known image samples;
[0045] The YOLOv7 algorithm is used to train all the image samples after classification and annotation to obtain a machine vision discrimination model;
[0046] Obtain the growth interface image during the preparation of gallium oxide single crystals by the current guided mode method;
[0047] Identifying and judging the acquired growth interface image through the machine vision discrimination model to obtain a judgment result;
[0048] According to the judgment result, the process parameters of preparing the gallium oxide single crystal by the guided mode method are adjusted, and the preparation of the gallium oxide single crystal is completed under the adjusted process parameters.
[0049] The embodiment of the present invention classifies and labels known image samples, and then uses the YOLOv7 algorithm to train all the image samples after classification and labeling. The machine vision discrimination model obtained after the training is completed can judge the crystal state of other unknown image samples. The camera continuously captures the current growth interface, uses the machine vision discrimination model to identify and judge the captured images, and gives a judgment result (including the probability of single crystal and the probability of impure crystal). The operator can respond quickly according to the judgment result given by the machine vision discrimination model, adjust the process parameters such as power and pulling speed, so as to reduce the probability of impure crystal and improve the single crystal rate. The method of the embodiment of the present invention has the advantages of rapid response and timely adjustment of process parameters.
[0050] The camera of the embodiment of the present invention continuously captures the growth interface, uses a machine vision discrimination model to identify and judge the captured pictures, and gives a judgment result, thereby achieving the purpose of real-time online monitoring of the growth interface during the growth of gallium oxide single crystals by the guided mode method.
[0051] The method of the embodiment of the present invention can accurately judge the growth of the crystal, shorten the growth time, improve the single crystal rate, and at the same time reduce the dependence of crystal growth on the operator's experience, shorten the operator's training cycle, and greatly reduce labor costs.
[0052] It should be noted that the known image samples in the sample library are images of growth interfaces in the process of preparing gallium oxide single crystals by the guide mold method. Known means that the crystal state shown in the growth interface image is known, that is, it is known whether the crystal shown in the growth interface image is a single crystal or a mixed crystal.
[0053] In one implementation, the step of classifying and labeling the known image samples specifically includes:
[0054] Dividing the known image samples into two categories, the two categories including single crystals and miscellaneous crystals;
[0055] Select the valid area of the single crystal image sample and the valid area of the mixed crystal image sample respectively, and add the GOOD label in the valid area of the single crystal (such as Figure 1 ), add a BAD tag in the effective area of the impurity crystal (as shown Figure 2 As shown), the effective area is the object in the image, and in the embodiment of the present invention, it refers to the area including the front edge of the crystal shoulder.
[0056] In one implementation, the step of using the YOLOv7 algorithm to train all the classified and labeled image samples specifically includes:
[0057] The YOLOv7 algorithm is used to cut all the image samples after classification and annotation into feature maps, and then the feature maps are characterized into detection box positions and category probabilities through a multi-layer convolutional network learned by a neural network to obtain prediction results.
[0058] The embodiment of the present invention is based on a fast target detection algorithm, that is, the YOLOv7 algorithm is used to locate and identify objects in an image. YOLOv7 cuts the input image into some feature maps, and then uses a multi-layer convolutional network learned by a neural network to characterize the feature maps into detection box positions and category probabilities, and finally forms a prediction result.
[0059] In one embodiment, after the step of using the YOLOv7 algorithm to train all the image samples after classification and annotation, the following steps are also included:
[0060] First, the loss function is used to evaluate the loss between the predicted result and the true label;
[0061] Then, the loss is passed to each layer of the convolutional network through the back-propagation algorithm to update the convolutional network parameters;
[0062] Finally, through iterative training, the model gradually converges to the optimal state.
[0063] That is to say, the training process of YOLOv7 mainly includes the steps of forward propagation, loss calculation, back propagation and optimization update. In the forward propagation stage, the input image is passed through the convolutional network to generate the prediction result; then, the loss function is calculated to evaluate the difference between the prediction result and the true label; then, the loss is passed to each layer of the network through the back propagation algorithm, and the convolutional network parameters are updated to optimize the model performance; finally, through iterative training, the model gradually converges to the optimal state, thereby obtaining the machine vision discrimination model.
[0064] The trained machine vision discrimination model can identify the shoulder front of the crystal in the growth interface image, that is, the effective area mentioned above, match and calculate with the samples in the sample library, and give the category and similarity of the predicted target. In the present invention, the similarity is probability.
[0065] In one embodiment, the growth interface image obtained during each preparation of gallium oxide single crystal by the guided mode method is added as a sample to the sample library. Each time the preparation of gallium oxide single crystal by the guided mode method is completed, a sample is extracted and entered into the sample library, and the sample library is continuously expanded to improve the accuracy of the machine vision discrimination model.
[0066] The embodiment of the present invention provides a method for online monitoring of the growth interface in the process of preparing gallium oxide single crystal by the guided mode method, which comprises the steps of:
[0067] Providing a sample library, wherein the sample library includes known image samples, wherein the image is an image of a growth interface in a process of preparing a gallium oxide single crystal by a guided mold method;
[0068] Classify and label the known image samples;
[0069] The YOLOv7 algorithm is used to train all the image samples after classification and annotation to obtain a machine vision discrimination model;
[0070] Obtain the growth interface image during the preparation of gallium oxide single crystals by the current guided mode method;
[0071] The acquired growth interface image is identified and judged by the machine vision discrimination model to obtain a judgment result.
[0072] In the traditional growth process, when there is a slight change in the growth interface, it is often difficult to detect with the naked eye. As time goes by, when there is a significant change in the growth interface, it can be noticed by people. At this time, the crystal may no longer be a single crystal. The embodiment of the present invention uses machine vision to replace naked eye observation. The machine vision discrimination model can monitor the growth interface in real time online, compare it with the samples in the sample library, and quickly determine the probability of single crystal and impurity crystal on the growth interface. Figure 3 As shown in the figure (GOOD means that the model judges that the crystal state at this time is a single crystal, BAD means mixed crystal, that is, the label defined above, and the number indicates the similarity given by the model), the operator can respond quickly according to the prompt and adjust the process parameters when there is a slight change in the growth interface, which can avoid operations such as melting back to a certain extent; in addition, the model can be continuously updated and upgraded, and the picture samples of each growth process are added to the sample library, and the model is repeatedly trained and strengthened, so that the model can judge the growth interface and the characteristics of the single crystal under various conditions, and continuously improve the accuracy of the model judgment.
[0073] like Figure 4 The figure shows a flow chart of model training and application. After the model is trained with known samples, the interface during the crystal growth process is monitored online. The key innovation of the present invention is to apply machine vision to the field of gallium oxide single crystal growth by the guided mode method to assist in regulating the crystal growth process. The method of the present invention can accurately judge the growth of the crystal, shorten the growth time, and increase the single crystal rate. At the same time, it reduces the dependence of crystal growth on the operator's experience, shortens the operator's training cycle, and greatly reduces the labor cost.
[0074] It should be noted that the present invention only exemplifies the gallium oxide (001) model using a flat mold, and different models can be trained according to the shape of the mold, the material of the crystal, and the surface shape of the crystal to cope with different growth conditions.
[0075] The embodiment of the present invention can directly record the judgment results obtained by applying the machine vision discrimination model, that is, the recognized photos and the corresponding judgment results. These data can then be reviewed by specialized personnel to evaluate the performance of the model in practical applications. For cases where the judgment is wrong, the staff can make manual corrections to ensure the accuracy of the data. The corrected samples and their correct labels can be re-incorporated into the sample library as new training data.
[0076] This approach can achieve the following goals: first, it reduces the workload of traditional manual annotation and optimizes the data annotation process; second, this automated feedback and correction mechanism allows multiple model iterations in a shorter time, thereby quickly adapting to changing actual application scenarios. Ultimately, this will greatly improve the accuracy of the model, making it more reliable and effective in practical applications.
[0077] In summary, the present invention provides a method for preparing gallium oxide single crystals by a guided mold method based on machine vision. The present invention applies machine vision to the process of preparing gallium oxide single crystals by the guided mold method, and judges whether the interface of crystal growth is normal based on the growth interface picture taken by the camera through a machine vision discrimination model, thereby adjusting process parameters such as power and pulling speed, and then preparing high-quality single crystals. The present invention has the advantages of rapid response and timely adjustment of process parameters by judging crystal growth through a machine vision discrimination model. The method of the present invention can accurately judge the growth of the crystal, shorten the growth time, and improve the single crystal rate. At the same time, it reduces the dependence of crystal growth on the operator's experience, shortens the operator's training cycle, and greatly reduces the labor cost.
[0078] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.
Claims
1. A method for preparing gallium oxide single crystals by guided mode method based on machine vision, characterized in that: Includes steps: Providing a sample library, wherein the sample library includes known image samples, wherein the image is an image of a growth interface in a process of preparing a gallium oxide single crystal by a guided mold method; Classify and label the known image samples; The YOLOv7 algorithm is used to train all the image samples after classification and annotation to obtain a machine vision discrimination model; Obtain the growth interface image during the current process of preparing gallium oxide single crystals by the guided mode method; Identifying and judging the acquired growth interface image through the machine vision discrimination model to obtain a judgment result; According to the judgment result, the process parameters of preparing the gallium oxide single crystal by the guided mode method are adjusted, and the preparation of the gallium oxide single crystal is completed under the adjusted process parameters.
2. The method for preparing gallium oxide single crystal by guided mode method based on machine vision according to claim 1, characterized in that: The step of classifying and labeling the known image samples specifically includes: Dividing the known image samples into two categories, the two categories including single crystals and miscellaneous crystals; Select the valid area of the single crystal image sample and the valid area of the mixed crystal image sample respectively, add a GOOD label in the single crystal valid area, and add a BAD label in the mixed crystal valid area, where the valid area refers to the area containing the crystal shoulder front.
3. The method for preparing gallium oxide single crystal by guided mode method based on machine vision according to claim 1, characterized in that: The step of using the YOLOv7 algorithm to train all the image samples after classification and annotation specifically includes: The YOLOv7 algorithm is used to cut all the image samples after classification and annotation into feature maps, and then the feature maps are characterized into detection box positions and category probabilities through a multi-layer convolutional network learned by a neural network to obtain prediction results.
4. The method for preparing gallium oxide single crystal by guided mode method based on machine vision according to claim 3, characterized in that: After the step of using the YOLOv7 algorithm to train all the classified and labeled image samples, the following steps are also included: First, the loss function is used to evaluate the loss between the predicted result and the true label; Then, the loss is passed to each layer of the convolutional network through the back-propagation algorithm to update the convolutional network parameters; Finally, through iterative training, the convolutional network gradually converges to the optimal state.
5. The method for preparing gallium oxide single crystal by guided mode method based on machine vision according to claim 1, characterized in that: The growth interface image obtained during each process of preparing gallium oxide single crystal by the guided mode method is added to the sample library.
6. The method for preparing gallium oxide single crystal by guided mode method based on machine vision according to claim 1, characterized in that: The process parameters include power and drawing speed.
7. The method for preparing gallium oxide single crystal by guided mode method based on machine vision according to claim 1, characterized in that: The camera is used to obtain images of the growth interface during the current guided mode method for preparing gallium oxide single crystals.
8. A method for online monitoring of the growth interface during the preparation of gallium oxide single crystals by a guided mode method, characterized in that: Includes steps: Providing a sample library, wherein the sample library includes known image samples, wherein the image is an image of a growth interface in a process of preparing a gallium oxide single crystal by a guided mold method; Classify and label the known image samples; The YOLOv7 algorithm is used to train all the image samples after classification and annotation to obtain a machine vision discrimination model; Obtain the growth interface image during the preparation of gallium oxide single crystals by the current guided mode method; The acquired growth interface image is identified and judged by the machine vision discrimination model to obtain a judgment result.
9. The method for online monitoring of the growth interface in the process of preparing gallium oxide single crystal by the guided mode method according to claim 8, characterized in that: The judgment result includes the probability of single crystal and the probability of impure crystal.