Gallium oxide single crystal growth method based on active learning image recognition, automatic gallium oxide single crystal growth method, device, equipment, medium and product

Abstract

The invention discloses a gallium oxide single crystal growth method based on active learning image recognition, an automatic gallium oxide single crystal growth method, a device, equipment, a medium and a product, and relates to the technical field of single crystal growth. The method comprises the following steps: acquiring a crystal image in real time in the current growth process; inputting the crystal image into the crystal image recognition model after the tth training to perform primary recognition of the growth state of the crystal; when the growth state of the crystal is abnormal, outputting an alarm signal, and carrying out secondary identification on the growth state of the crystal; when the growth state of the crystal is abnormal, crystal growth parameters are adjusted; and otherwise, not adjusting the crystal growth parameters until the crystal growth is completed, marking the crystal image in the current growth process, constructing a data set of the current growth process, and carrying out secondary training on the crystal image recognition model after the tth training. The automatic gallium oxide single crystal growth is realized, the labor consumption is reduced, and the production efficiency is improved.

Description

Technical Field

[0001] The present application relates to the technical field of single crystal growth, and in particular to a gallium oxide single crystal growth method based on active learning image recognition, an automatic gallium oxide single crystal growth method, device, equipment, medium and product. Background Art

[0002] Gallium oxide is an oxide semiconductor with an ultra-wide bandgap of 4.7-4.9 eV. It has significant advantages such as high breakdown field strength (about 8 MeV / cm) and stable physical and chemical properties. It has broad application prospects in power devices and solar-blind ultraviolet detectors. It is the material most likely to be the first to achieve commercial application among ultra-wide bandgap semiconductors. In order to achieve the preparation of high-performance gallium oxide devices, high-quality single crystal growth and substrate development are key.

[0003] At present, the main growth methods of β-Ga2O3 bulk single crystals include edge-defined film-fed growth (EFG), Czochralski (CZ), vertical Bridgman (VB) and floating zone (FZ). Due to the problems of easy dissociation, strong anisotropy and low thermal conductivity of gallium oxide, it is very difficult to grow high-quality gallium oxide single crystals. In the process of growing gallium oxide single crystals by casting, the main exposed surface of the seed produced by spontaneous nucleation is not easy to be manually identified, and the monitoring of impurity crystals during the growth process requires a lot of manual time, which reduces production efficiency. In the process of Czochralski growth, the control of shoulder quality and the monitoring of defects such as twins and cracks also require a lot of manual labor, which greatly reduces production efficiency. Summary of the invention

[0004] The purpose of this application is to provide a gallium oxide single crystal growth method based on active learning image recognition, an automatic gallium oxide single crystal growth method, device, equipment, medium and product to achieve automatic gallium oxide single crystal growth, reduce labor consumption and improve production efficiency.

[0005] To achieve the above objectives, this application provides the following solutions.

[0006] In a first aspect, the present application provides a gallium oxide single crystal growth method based on active learning image recognition, comprising:

[0007] Real-time acquisition of crystal images during the current growth process;

[0008] Inputting the crystal image into the crystal image recognition model after the t-th training to perform a primary recognition of the growth state of the crystal, and obtaining a primary recognition result; the growth state of the crystal is normal or abnormal; t represents the total number of secondary training of the crystal image recognition model, t≥0;

[0009] When the primary identification result indicates that the growth state of the crystal is abnormal, an alarm signal is output, and a secondary identification of the growth state of the crystal is performed to obtain a secondary identification result;

[0010] When the secondary identification result indicates that the growth state of the crystal is abnormal, the crystal growth parameters are adjusted;

[0011] When the first recognition result indicates that the growth state of the crystal is normal or the second recognition result indicates that the growth state of the crystal is normal, no adjustment is made to the crystal growth parameters;

[0012] Return to the step of "acquiring crystal images in real time during crystal growth" until the crystal growth is completed, mark the crystal images in the current growth process, and construct a data set for the current growth process;

[0013] The crystal image recognition model after the t-th training is trained again using the data set of the current growth process to obtain the crystal image recognition model after the t+1-th training.

[0014] In a second aspect, the present application provides a gallium oxide single crystal growth device based on active learning image recognition, the gallium oxide single crystal growth device based on active learning image recognition applies the above-mentioned gallium oxide single crystal growth method based on active learning image recognition, and the gallium oxide single crystal growth device based on active learning image recognition includes:

[0015] A crystal image acquisition module is used to acquire the crystal image in real time during the current growth process;

[0016] A primary recognition module is used to input the crystal image into the crystal image recognition model after the t-th training to perform a primary recognition of the growth state of the crystal to obtain a primary recognition result; the growth state of the crystal is normal or abnormal; t represents the total number of secondary training of the crystal image recognition model, t≥0;

[0017] A secondary identification module, used for outputting an alarm signal and performing secondary identification of the growth state of the crystal to obtain a secondary identification result when the primary identification result indicates that the growth state of the crystal is abnormal;

[0018] A growth parameter adjustment module, used for adjusting the crystal growth parameters when the secondary recognition result indicates that the growth state of the crystal is abnormal; and not adjusting the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is normal or the secondary recognition result indicates that the growth state of the crystal is normal;

[0019] A marking module is used to return to the step of "acquiring crystal images in real time during the crystal growth process" until the crystal growth is completed, mark the crystal images in the current growth process, and construct a data set for the current growth process;

[0020] The secondary training module is used to perform secondary training on the crystal image recognition model after the t-th training using the data set of the current growth process to obtain the crystal image recognition model after the t+1-th training.

[0021] In a third aspect, the present application provides an automated gallium oxide single crystal growth method, comprising:

[0022] Real-time acquisition of crystal images during the current growth process;

[0023] The crystal image is input into a target crystal image recognition model to perform a primary recognition of the growth state of the crystal to obtain a primary recognition result; the growth state of the crystal is normal or abnormal; the recognition accuracy of the target crystal image recognition model is greater than an accuracy threshold; the target crystal image recognition model is obtained by executing the above-mentioned gallium oxide single crystal growth method based on active learning image recognition;

[0024] When the primary identification result indicates that the growth state of the crystal is abnormal, adjusting the crystal growth parameters;

[0025] When the primary identification result indicates that the growth state of the crystal is normal, no adjustment is performed on the crystal growth parameters;

[0026] Return to the step of "Acquiring real-time crystal images during crystal growth" until the crystal growth is complete.

[0027] In a fourth aspect, the present application provides an automated gallium oxide single crystal growth device, the automated gallium oxide single crystal growth device applies the automated gallium oxide single crystal growth method described above, and the automated gallium oxide single crystal growth device comprises:

[0028] A crystal image acquisition module is used to acquire the crystal image in real time during the current growth process;

[0029] A state recognition module is used to input the crystal image into a target crystal image recognition model to perform a primary recognition of the growth state of the crystal and obtain a primary recognition result; the growth state of the crystal is normal or abnormal; the recognition accuracy of the target crystal image recognition model is greater than an accuracy threshold;

[0030] A parameter adjustment module, used for adjusting the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is abnormal; and not adjusting the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is normal;

[0031] The return module is used to return to call the crystal image acquisition module until the crystal growth is completed.

[0032] In a fifth aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any one of the above-described methods for growing gallium oxide single crystals based on active learning image recognition, or the above-described automated method for growing gallium oxide single crystals.

[0033] In a sixth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned gallium oxide single crystal growth method based on active learning image recognition, or the above-mentioned automated gallium oxide single crystal growth method.

[0034] In a seventh aspect, the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above-mentioned gallium oxide single crystal growth method based on active learning image recognition, or the above-mentioned automated gallium oxide single crystal growth method.

[0035] According to the specific embodiments provided in this application, this application has the following technical effects.

[0036] The present application provides a gallium oxide single crystal growth method based on active learning image recognition, an automatic gallium oxide single crystal growth method, device, equipment, medium and product. The present application is based on the method of active learning image recognition, performs active learning in the monitoring process of single crystal growth, and constructs a crystal image recognition model that can realize automatic gallium oxide single crystal growth. The present application realizes automatic gallium oxide single crystal growth, reduces labor consumption and improves production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A schematic flow chart of a gallium oxide single crystal growth method based on active learning image recognition provided in one embodiment of the present application.

[0039] Figure 2 A schematic diagram of a gallium oxide single crystal growth method based on active learning image recognition provided in one embodiment of the present application.

[0040] Figure 3A flow chart of an automated gallium oxide single crystal growth method provided in one embodiment of the present application.

[0041] Figure 4 A schematic diagram of the structure of a computer device provided in one embodiment of the present application. DETAILED DESCRIPTION

[0042] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0043] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0044] In terms of automatic crystal growth, the existing patent CN 113026103 A provides a monitoring system for growing gallium oxide crystals by the guided mold method and a method for preparing gallium oxide crystals. The solution electrically connects the pressure sensor, the infrared camera and the external computer, which can timely identify the phenomenon of overheating or overcooling in the process of gallium oxide crystal growth, and adjust the heating power of the heating element in the first time, effectively reducing the problem of crystal growth failure caused by overheating or overcooling, which is conducive to the large-scale production of high-quality and low-cost gallium oxide single crystals. However, the solution only obtains information such as crystal weight and width through pressure sensors and infrared cameras, and feeds back crystal growth. The information obtained is relatively small, and it is impossible to judge the spontaneous nucleation process of the crystal and abnormal phenomena such as impurities and cracking during the crystal growth process. In addition, no machine learning training is performed, and the obtained data cannot be effectively summarized. The scope of use is limited, and the auxiliary effect on workers' operations is low, making it difficult to achieve fully automated production.

[0045] The existing patent CN 111754516 B provides an intelligent control method for the growth of rutile single crystals based on computer vision feedback. The scheme uses a high-definition video acquisition system to complete the acquisition of image information and pre-process the image through the high-definition video acquisition system; the computer uses OPENCV to analyze the image features to achieve image segmentation and feature point positioning; the processed image uses a convolutional neural network to analyze the current real-time morphology of the crystal growth state characteristics; the crystal morphology is judged whether it is normal after comparing the analysis results with the edge coordinate curve of the healthy morphology of the crystal growth obtained by machine learning; if so, the original given function curve is continued to be used for growth control; if not, the position of the combustion table and the feeding parameters are adjusted through self-service adjustment, and the feedback image is implemented. It can accurately control various parameters of crystal growth and ensure the temperature distribution, composition distribution and mechanical distribution in the growth chamber. This scheme has no active learning conditions and cannot develop and improve itself in the long-term growth time. It has problems such as low efficiency, limited scope of use, and inability to keep up with technological development in a timely manner, and has low auxiliary effect on workers.

[0046] Therefore, it is necessary to use an automated gallium oxide single crystal growth method based on active learning image recognition to replace manual identification of nucleation and defects during the growth of gallium oxide single crystals and improve the growth efficiency.

[0047] In an exemplary embodiment, Figure 1 and Figure 2 As shown, a gallium oxide single crystal growth method based on active learning image recognition is provided, including the following steps 101 to 107.

[0048] Step 101, acquiring a crystal image in real time during the current growth process.

[0049] Step 102, input the crystal image into the crystal image recognition model after the t-th training to perform a primary recognition of the growth state of the crystal to obtain a primary recognition result; the growth state of the crystal is normal or abnormal; t represents the total number of secondary training of the crystal image recognition model, t≥0.

[0050] Step 103: When the primary identification result indicates that the growth state of the crystal is abnormal, an alarm signal is output, and a secondary identification of the growth state of the crystal is performed to obtain a secondary identification result.

[0051] Step 104: When the secondary identification result indicates that the growth state of the crystal is abnormal, the crystal growth parameters are adjusted.

[0052] Step 105: When the first recognition result indicates that the growth state of the crystal is normal or the second recognition result indicates that the growth state of the crystal is normal, the crystal growth parameters are not adjusted.

[0053] Step 106, return to the step of "acquiring crystal images in real time during the crystal growth process" until the crystal growth is completed, mark the crystal images in the current growth process, and construct a data set for the current growth process.

[0054] Step 107 , performing a second training on the crystal image recognition model after the t-th training using the data set of the current growth process, to obtain the crystal image recognition model after the t+1-th training.

[0055] By implementing the above steps 101 to 107, based on the method of active learning image recognition, active learning is performed during the monitoring process of single crystal growth to construct a crystal image recognition model that can realize automatic growth of gallium oxide single crystals. The present application realizes automatic growth of gallium oxide single crystals, reduces labor consumption, and improves production efficiency.

[0056] In another exemplary embodiment of the present application, before performing secondary training, the embodiment of the present application first uses an image recognition program developed based on the pytorch neural network framework to pre-train the crystal image recognition model for a certain number of artificially calibrated crystal growth results and process data to obtain the crystal image recognition model after the 0th training.

[0057] In the pre-training process, the initial crystal growth image data is manually classified to create an initial data set of crystal growth images, and the created initial data set is trained through the crystal image recognition model developed based on the pytorch neural network framework to establish a preliminary recognition model (i.e., the crystal image recognition model after the 0th training) to identify whether the crystal growth process is abnormal in real time and assist crystal growth. Among them, the pytorch neural network framework is the python version of torch. It is an open source mathematical library that is mainly used to operate multidimensional matrix data and has a wide range of applications. One of the important applications is image recognition. For any phenomenon in the crystal growth process that needs to be paid attention to, it is necessary to accumulate certain experimental data and manually accurately calibrate to create a data set. The research object can be the spontaneous nucleation phenomenon in the casting single crystal growth process, or the cracking phenomenon, the impurity phenomenon, or the monitoring of the solid-liquid interface quality during the Czochralski growth process. For any object of interest, it is necessary to manually create an initial data set and correspond the camera images to the results one by one.

[0058] In another exemplary embodiment of the present application, during the crystal growth process, the growth status of the crystal is recorded in real time by a high-definition camera, that is, the crystal image is acquired by the high-definition camera.

[0059] During the crystal growth process, a computer will run the established crystal image recognition model in real time, and at the same time, a high-definition camera will sample the crystal growth situation in real time. The sampled image data will be transmitted to the crystal image recognition model, and the model will determine which category the collected image is more in line with in the model. If the category is an abnormal growth category, an alarm will be issued; if the category is a normal growth category, the next step of identification will be carried out. In another exemplary embodiment of the present application, Figure 2 As shown, the specific implementation process of the above steps 102 to 105 includes:

[0060] Step 1-1, use the pytorch neural network framework to preprocess the crystal image and enhance the image data, including tensor quantization, grayscale conversion, etc., and input it into the existing crystal image recognition model after the tth training to identify the growth state.

[0061] Step 1-2: If an abnormality is identified, the crystal growth parameters are adjusted through the growth control program built into the single crystal growth control system to adjust to appropriate growth conditions, and an alarm is issued to notify the workers.

[0062] Steps 1-3, if the recognition is normal, continue to grow the crystal in the existing growth mode and grow high-quality crystals.

[0063] In another exemplary embodiment of the present application, Figure 2 As shown in the figure, after the crystal growth is completed, the abnormal and normal crystal growth image records and growth results in the growth process are calibrated by artificial means to construct a data set of the current growth process; the uncertainty Poisson model is used to screen the pictures of the current growth process data set to obtain data that has a positive training effect on the existing crystal image recognition model; on the basis of the existing crystal image recognition model (i.e., the crystal image recognition model after the tth training), the newly added screening pictures are actively learned and trained to improve the prediction accuracy of the model.

[0064] The specific operation steps of the uncertainty Poisson model are:

[0065] Step 2-1, based on the existing pytorch crystal image recognition model, calculate the uncertainty of the image in the newly added crystal growth image data. This value can be a function of probability distribution or other indicators used to measure model uncertainty. The uncertainty means that the model's recognition result for the image is not accurate enough to accurately determine which category the image belongs to. The less accurate the judgment, the higher the uncertainty.

[0066] Step 2-2, based on the uncertainty value, select samples for learning, and give priority to learning images with high uncertainty to improve the adaptability of the model to complex situations in the crystal growth process. The adaptability means that the model has not yet accommodated the data of the crystal growth image data, so the uncertainty of the judgment of this data is high. By actively learning this data, inaccurate recognition can be reduced and the recognition accuracy of different crystal growth conditions can be improved.

[0067] Among them, the uncertainty Poisson model is used to screen the crystal images in the data set of the current growth process, and the crystal images that have a positive training effect on the crystal image recognition model after the t-th training are obtained to form a training set, which specifically includes: dividing the data set of the current generation process into multiple subsets; calculating the uncertainty of each subset; determining the crystal images in the subset whose uncertainty is not greater than a preset threshold, and forming the training set.

[0068] The calculation formula for uncertainty is:

[0069]

[0070] Where M is the uncertainty of the subset, p(y i |x i ) represents the crystal image recognition model after the t-th training for the i-th crystal image x in the subset i The prediction belongs to x i Label y i The probability of N is the number of crystal images in the subset.

[0071] The numerator is a weighted sum of the uncertainty of all samples, based on the product of the model's prediction probability and information entropy. The idea of ​​this part is to measure the model's prediction confidence.

[0072] The denominator is the sum of the predicted probabilities of all samples and is used for normalization.

[0073] The iteration rule of active learning is:

[0074] θ t is the parameter vector of the model at the tth iteration, α is the learning rate, which determines the step size of each iteration update, L(θ t ,x i ,y i ) is the loss function, which means the error of the model under a given sample. is the loss function for the parameter vector θ t The gradient of , which represents the derivative of the error with respect to the change of parameters.

[0075] The above process will adjust the model parameters according to the gradient of the sample in each iteration to minimize the loss function. Selecting recognition samples with high uncertainty for active learning to update the model means that the model will focus on samples that have not been well learned or predicted inaccurately, thereby improving model performance more efficiently.

[0076] Based on the same inventive concept, the embodiment of the present application also provides a gallium oxide single crystal growth device based on active learning image recognition for implementing the gallium oxide single crystal growth method based on active learning image recognition involved above. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above method, so the specific limitations in one or more embodiments of the gallium oxide single crystal growth device based on active learning image recognition provided below can refer to the limitations of the gallium oxide single crystal growth method based on active learning image recognition above, and will not be repeated here.

[0077] In an exemplary embodiment, a distributed photovoltaic power prediction device is provided, comprising:

[0078] The crystal image acquisition module is used to acquire the crystal image in real time during the current growth process.

[0079] The primary recognition module is used to input the crystal image into the crystal image recognition model after the t-th training to perform a primary recognition of the growth state of the crystal and obtain a primary recognition result; the growth state of the crystal is normal or abnormal; t represents the total number of secondary training of the crystal image recognition model, t≥0.

[0080] The secondary recognition module is used to output an alarm signal and perform secondary recognition of the growth state of the crystal to obtain a secondary recognition result when the primary recognition result indicates that the growth state of the crystal is abnormal.

[0081] The growth parameter adjustment module is used to adjust the crystal growth parameters when the secondary recognition result indicates that the growth state of the crystal is abnormal; when the primary recognition result indicates that the growth state of the crystal is normal or the secondary recognition result indicates that the growth state of the crystal is normal, the crystal growth parameters are not adjusted.

[0082] The marking module is used to return to the step of "acquiring crystal images in real time during the crystal growth process" until the crystal growth is completed, mark the crystal images in the current growth process, and construct a data set for the current growth process.

[0083] The secondary training module is used to perform secondary training on the crystal image recognition model after the t-th training using the data set of the current growth process to obtain the crystal image recognition model after the t+1-th training.

[0084] By repeatedly applying the crystal image recognition model to the prediction and active learning training of crystal growth, high-precision prediction can be achieved, achieving the goal of fully automatic crystal growth instead of manual work.

[0085] In an exemplary embodiment, Figure 3 As shown, an automated gallium oxide single crystal growth method is provided, which can apply the high-precision target crystal image recognition model obtained by the above-mentioned active learning training to perform fully automatic crystal growth, including the following steps 201 to 204. Among them:

[0086] Step 201, input the crystal image into the target crystal image recognition model to perform a primary recognition of the growth state of the crystal to obtain a primary recognition result; the growth state of the crystal is normal or abnormal; the recognition accuracy of the target crystal image recognition model is greater than the accuracy threshold; the target crystal image recognition model is obtained by executing the above-mentioned gallium oxide single crystal growth method based on active learning image recognition.

[0087] Step 202: When the primary identification result indicates that the growth state of the crystal is abnormal, adjust the crystal growth parameters.

[0088] Step 203: When the primary identification result indicates that the growth state of the crystal is normal, the crystal growth parameters are not adjusted.

[0089] Step 204, returning to the step of "acquiring a crystal image in real time during the crystal growth process" until the crystal growth is completed.

[0090] Based on the same inventive concept, the embodiment of the present application also provides an automated gallium oxide single crystal growth device for implementing the automated gallium oxide single crystal growth method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more automated gallium oxide single crystal growth device embodiments provided below can refer to the above limitations based on the automated gallium oxide single crystal growth method, and will not be repeated here.

[0091] In an exemplary embodiment, an automated gallium oxide single crystal growth apparatus is provided, comprising:

[0092] The crystal image acquisition module is used to acquire the crystal image in real time during the current growth process.

[0093] The state recognition module is used to input the crystal image into the target crystal image recognition model to perform a first recognition of the growth state of the crystal to obtain a recognition result; the growth state of the crystal is normal or abnormal; the recognition accuracy of the target crystal image recognition model is greater than the accuracy threshold.

[0094] The parameter adjustment module is used to adjust the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is abnormal; when the primary recognition result indicates that the growth state of the crystal is normal, the crystal growth parameters are not adjusted.

[0095] The return module is used to return to call the crystal image acquisition module until the crystal growth is completed.

[0096] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store crystal images and recognition results. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a gallium oxide single crystal growth method or an automatic gallium oxide single crystal growth method based on active learning image recognition is implemented.

[0097] Those skilled in the art will understand that Figure 4 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components. In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0098] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0099] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0100] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0101] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0102] The above-mentioned technical scheme of the present application can continuously improve the image recognition of the growth state of gallium oxide crystals during their growth process through active learning, including the type of casting method and defects such as cracking and impure crystals, and can improve the accuracy of the prediction model in continuous production, ultimately replacing manual labor and achieving the purpose of fully automated crystal growth.

[0103] The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. The non-relational database may include a distributed database based on blockchain, etc., but is not limited thereto. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but is not limited thereto.

[0104] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0105] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A gallium oxide single crystal growth method based on active learning image recognition, characterized in that: include: Real-time acquisition of crystal images during the current growth process; Inputting the crystal image into the crystal image recognition model after the t-th training to perform a primary recognition of the growth state of the crystal, and obtaining a primary recognition result; the growth state of the crystal is normal or abnormal; t represents the total number of secondary training of the crystal image recognition model, t≥0; When the primary identification result indicates that the growth state of the crystal is abnormal, an alarm signal is output, and a secondary identification of the growth state of the crystal is performed to obtain a secondary identification result; When the secondary identification result indicates that the growth state of the crystal is abnormal, the crystal growth parameters are adjusted; When the first recognition result indicates that the growth state of the crystal is normal or the second recognition result indicates that the growth state of the crystal is normal, no adjustment is made to the crystal growth parameters; Return to the step of "acquiring crystal images in real time during crystal growth" until the crystal growth is completed, mark the crystal images in the current growth process, and construct a data set for the current growth process; The crystal image recognition model after the t-th training is trained again using the data set of the current growth process to obtain the crystal image recognition model after the t+1-th training.

2. The gallium oxide single crystal growth method based on active learning image recognition according to claim 1, characterized in that: When t=0, the crystal image recognition model after the tth training is obtained in the following way; Build a crystal image recognition model based on the pytorch neural network framework; The crystal image recognition model is pre-trained based on the historical data set to obtain the crystal image recognition model after the 0th training.

3. The gallium oxide single crystal growth method based on active learning image recognition according to claim 1, characterized in that: The crystal image recognition model after the t-th training is trained again using the data set of the current growth process to obtain the crystal image recognition model after the t+1-th training, which specifically includes: The crystal images in the data set of the current growth process are screened using an uncertain Poisson model to obtain crystal images that have a positive training effect on the crystal image recognition model after the t-th training, forming a training set; The training set is used to perform secondary training on the crystal image recognition model after the t-th training to obtain the crystal image recognition model after the t+1-th training.

4. The gallium oxide single crystal growth method based on active learning image recognition according to claim 3, characterized in that: The uncertain Poisson model is used to screen the crystal images in the data set of the current growth process, and the crystal images that have a positive training effect on the crystal image recognition model after the t-th training are obtained to form a training set, which specifically includes: Divide the data set of the current generation process into multiple subsets; The uncertainty of each subset is calculated using the following formula; Where M is the uncertainty of the subset, p(y i |x i ) represents the crystal image recognition model after the t-th training for the i-th crystal image x in the subset i The prediction belongs to x i Label y i The probability of , N is the number of crystal images in the subset; It is determined that the crystal images in the subset whose uncertainty is not greater than a preset threshold constitute the training set.

5. A gallium oxide single crystal growth device based on active learning image recognition, characterized in that: A gallium oxide single crystal growth device based on active learning image recognition applies the gallium oxide single crystal growth method based on active learning image recognition as described in any one of claims 1 to 4, and a gallium oxide single crystal growth device based on active learning image recognition comprises: A crystal image acquisition module is used to acquire the crystal image in real time during the current growth process; A primary recognition module is used to input the crystal image into the crystal image recognition model after the t-th training to perform a primary recognition of the growth state of the crystal to obtain a primary recognition result; the growth state of the crystal is normal or abnormal; t represents the total number of secondary training of the crystal image recognition model, t≥0; A secondary identification module, used for outputting an alarm signal and performing secondary identification of the growth state of the crystal to obtain a secondary identification result when the primary identification result indicates that the growth state of the crystal is abnormal; A growth parameter adjustment module, used for adjusting the crystal growth parameters when the secondary recognition result indicates that the growth state of the crystal is abnormal; and not adjusting the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is normal or the secondary recognition result indicates that the growth state of the crystal is normal; A marking module is used to return to the step of "real-time acquisition of crystal images during crystal growth" until the crystal growth is completed, mark the crystal images in the current growth process, and construct a data set for the current growth process; The secondary training module is used to perform secondary training on the crystal image recognition model after the t-th training using the data set of the current growth process to obtain the crystal image recognition model after the t+1-th training.

6. An automated gallium oxide single crystal growth method, characterized in that: include: Real-time acquisition of crystal images during the current growth process; The crystal image is input into a target crystal image recognition model to perform a first recognition of the growth state of the crystal to obtain a first recognition result; the growth state of the crystal is normal or abnormal; the recognition accuracy of the target crystal image recognition model is greater than the accuracy threshold; the target crystal image recognition model is obtained by executing the gallium oxide single crystal growth method based on active learning image recognition according to any one of claims 1 to 4; When the primary identification result indicates that the growth state of the crystal is abnormal, adjusting the crystal growth parameters; When the primary identification result indicates that the growth state of the crystal is normal, no adjustment is performed on the crystal growth parameters; Return to the step "Acquiring real-time crystal images during crystal growth" until crystal growth is complete.

7. An automated gallium oxide single crystal growth device, characterized in that: The automated gallium oxide single crystal growth device applies the automated gallium oxide single crystal growth method according to claim 6, and the automated gallium oxide single crystal growth device comprises: A crystal image acquisition module is used to acquire the crystal image in real time during the current growth process; A state recognition module is used to input the crystal image into a target crystal image recognition model to perform a primary recognition of the growth state of the crystal and obtain a primary recognition result; the growth state of the crystal is normal or abnormal; the recognition accuracy of the target crystal image recognition model is greater than an accuracy threshold; A parameter adjustment module, used for adjusting the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is abnormal; and not adjusting the crystal growth parameters when the primary recognition result indicates that the growth state of the crystal is normal; The return module is used to return to call the crystal image acquisition module until the crystal growth is completed.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the gallium oxide single crystal growth method based on active learning image recognition as described in any one of claims 1 to 4, or the automated gallium oxide single crystal growth method as described in claim 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for growing a gallium oxide single crystal based on active learning image recognition described in any one of claims 1 to 4 or the automated method for growing a gallium oxide single crystal described in claim 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for growing a gallium oxide single crystal based on active learning image recognition described in any one of claims 1 to 4 or the automated method for growing a gallium oxide single crystal described in claim 6 is implemented.

Images