Data processing method, system and device for online training of visual model and medium

Abstract

A data processing method, system, device and medium for online training of a visual model, relating to the technical field of data processing. The method comprises: receiving a data processing request sent by a client, and analyzing the type of the data processing request; if the data processing request is a change instruction, detecting whether there is a conflict, and if there is a conflict, executing a conflict processing strategy, and synchronizing the effective data after the conflict processing to the client; if the data processing request is a query instruction, detecting whether the cache hits, and if not, executing a query optimization strategy, and storing the query result after merging into the cache and returning to the client. By adopting the scheme, the classification processing of the change instruction and the query instruction in the online training of the visual model is realized, which can not only effectively solve the data conflict problem of the change instruction, but also improve the response efficiency of the query instruction through the cache mechanism, while guaranteeing the data consistency of the client and the platform, and overall improving the data processing stability and running efficiency of the online training platform of the visual model.

Description

Technical Field

[0001] This application relates to the field of data processing technology, specifically to a data processing method, system, device, and medium for online training of visual models. Background Technology

[0002] In the field of industrial product quality inspection, using machine vision technology to quickly locate and identify the type of defects on the product surface is one of the mainstream solutions. Typically, the image of the product to be inspected is divided into several grid areas, and combined with a preset defect type library, the probability and specific type of defects appearing in each grid are determined to achieve accurate detection.

[0003] Traditional visual inspection solutions of this type require the defect detection model to be deployed to offline devices, and each device independently completes the model training before putting it into use. In order to improve efficiency, some factories choose to complete the defect detection model training on a single device and then synchronize the model to other devices through the control center. However, this method still requires retraining the model when the product defect type is adjusted or the inspection standard is updated, which is inefficient.

[0004] To address these issues, the industry is increasingly adopting online AI cloud platform solutions: pre-setting various defect information corresponding to products in the cloud platform, centrally completing the training and updating of defect detection models, distributing the models to various offline devices, and receiving detection image data uploaded by the devices, thereby improving the efficiency of model deployment and updates.

[0005] However, during the operation of this online AI cloud platform, it is necessary to process multi-source data simultaneously, such as detection image data uploaded from different devices, updated data from the platform's defect database, and adjusted model parameters. The concurrent transmission and update of related data can easily lead to data conflicts in the background, affecting the accuracy and real-time performance of defect detection. Summary of the Invention

[0006] This application provides a data processing method, system, device, and medium for online training of visual models, which addresses the problems of insufficient stability and efficiency in data processing during existing online training of visual models.

[0007] Firstly, this application provides a data processing method for online training of visual models, the method comprising: Receive data processing requests sent by clients and parse the type of the data processing requests; If the data processing request is a change instruction, check whether the data conflicts. If there are conflicts, execute the conflict handling strategy and synchronize the valid data after conflict handling to the client. If the data processing request is a query command, check if the cache is hit. If not, execute the query optimization strategy, merge the query results, store them in the cache, and return them to the client.

[0008] By adopting the above technical solutions, the classification and processing of change commands and query commands in online training of visual models can be realized. This can effectively solve the data conflict problem of change commands, improve the response efficiency of query commands through caching mechanisms, and ensure data consistency between the client and the platform, thereby improving the overall data processing stability and operating efficiency of the online training platform for visual models.

[0009] In a specific feasible implementation, conflict resolution strategies include: Perform automatic merging operations on conflicting data; If automatic merging fails, mark the conflict as unresolved and save all conflicted versions; Compare the timestamps of all conflicting operations, filter out the operation with the latest timestamp, and retain the dataset corresponding to the latest operation.

[0010] By adopting the above technical solutions, a layered conflict handling process is constructed. Automatic merging is prioritized to reduce the cost of manual intervention. When merging fails, conflict versions are saved to ensure data traceability. Timestamps are used to filter the latest operations to ensure data timeliness, thereby improving the automation, reliability, and timeliness of conflict handling.

[0011] In a specific feasible implementation, the automatic merging of conflicting data includes: Identify the target object type corresponding to the conflicting data, wherein the target object type includes at least model parameters, defect library data, and mesh configuration; If the target object is a model parameter, extract the gradient tensor corresponding to each of the change instructions; Each gradient tensor is input into a pre-trained gradient fusion network to calculate the compatibility weights of each gradient tensor in different dimensions. If the compatibility score output by the gradient fusion network is higher than the compatibility threshold, the gradient tensors are weighted and fused according to the compatibility weights to generate a merged update instruction.

[0012] By adopting the above technical solutions, differentiated automatic merging logic is realized for different target objects. In particular, gradient fusion network is used to perform weighted merging for model parameter conflicts, avoiding the loss of effective gradient information, improving the rationality of model parameter conflict merging, and ensuring the accuracy and completeness of model parameter updates.

[0013] In a specific feasible implementation, query optimization strategies include: Based on the query type of the query instruction, the query is routed to the corresponding query engine; The query types include at least point queries, range queries, full-text searches, and aggregate queries, and are respectively routed to primary key indexes, composite indexes, Elasticsearch (ES) indexes, and materialized views.

[0014] By adopting the above technical solutions, different types of query commands are routed to the appropriate index engine, giving full play to the performance advantages of each index engine, reducing resource redundancy consumption during the query process, significantly improving the execution speed of different types of query commands, and optimizing the overall performance of query operations.

[0015] In one specific implementation scheme, routing to the corresponding query engine based on the type of the query instruction includes: Extract the query pattern features of the query instruction, input the query pattern features into the query classification model, and output the matching probability of each query type; Based on the query type with the highest matching probability and the current load status of each data source, the optimal routing path is selected from the preset strategy matrix.

[0016] By adopting the above technical solutions, accurate identification of query types can be achieved. At the same time, the routing path can be dynamically selected based on the load status of the data source, avoiding overload of a single query engine, improving the utilization efficiency of query resources, and ensuring the stability and efficiency of query command response.

[0017] In one specific feasible implementation, the method further includes: Collect historical conflict handling records, extract conflict object types, conflict mode characteristics, and the effectiveness of corresponding resolution strategies, and use these to update the compatibility weight threshold of the conflict handling strategies; Collect historical query execution data, extract the correlation between query type and execution time and resource consumption, and use it to optimize the routing priority of the query optimization strategy.

[0018] By adopting the above technical solutions, historical data is used to iteratively optimize conflict handling and query optimization strategies, enabling the strategies to adapt to changes in actual business scenarios, continuously improving the success rate of conflict handling and the performance of query operations, and enhancing the adaptability of the solutions.

[0019] In one specific feasible implementation, the method further includes: Based on historical versions of conflict data, operational contexts, and corresponding resolution strategies, the system generates visual distribution charts of conflict data, impact scope analysis reports, and conflict resolution recommendations based on historical cases.

[0020] By adopting the above technical solutions, relevant information of conflict data can be displayed intuitively, the scope of the conflict's impact can be clarified, and solutions based on historical cases can be provided to help managers quickly locate and handle conflicts, reducing the negative impact of unresolved conflicts on business operations.

[0021] A second aspect of this application provides a data processing system for online training of visual models, the system comprising: The change processing module is configured to process change commands and detect conflicts in the target data. The conflict resolution module is configured to execute a conflict handling strategy to obtain valid data after a conflict. The query processing module is configured to process query commands and perform cache hit detection; The query optimization module is configured to execute query optimization strategies and route query commands to the corresponding engines.

[0022] A third aspect of this application provides an electronic device, comprising: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute the above-described method steps.

[0023] A fourth aspect of this application provides a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the method steps described above. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a data processing method for online training of a visual model provided in an embodiment of this application. Detailed Implementation

[0025] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0026] In the description of the embodiments of this application, the words "for example" or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design that is described as "for example" or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design options. Rather, the use of the words "for example" or "for instance" is intended to present the relevant concepts in a specific manner.

[0027] In the description of the embodiments of this application, the term "multiple" means two or more. For example, multiple systems means two or more systems, and multiple screen terminals means two or more screen terminals. 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 indicated technical features. Thus, a feature defined with "first" or "second" may explicitly or implicitly include one or more of that feature. The terms "comprising," "including," "having," and variations thereof all mean "including but not limited to," unless otherwise specifically emphasized.

[0028] Please refer to Figure 1 A flowchart illustrating a data processing method for online training of visual models is presented. This method can be implemented using a computer program, a microcontroller, or run on a data processing system for online training of visual models. The computer program can be integrated into a computer device or run as a standalone application. Specifically, the method includes steps S100 to S300, as follows: S100: Receive data processing requests sent by the client and parse the type of the data processing request; A data processing request is an operation instruction sent by the client to the online training platform for visual models. This instruction is used to request the platform to perform data-related operations.

[0029] In some embodiments, after the visual model online training platform starts, it continuously listens to the network port to receive data processing requests sent by clients. Upon receiving the data processing request, the visual model online training platform parses the data processing request message, extracts the request type identifier field from the message, and uses this identifier field to determine the type of data processing request.

[0030] S200. If the data processing request is a change instruction, check whether the data is conflicted. If there is a conflict, execute the conflict handling strategy and synchronize the valid data after conflict handling to the client. In this embodiment of the application, the change instruction refers to an instruction used to request modification of the data content in the online training platform for the visual model. The modification content corresponding to the instruction includes the parameters of the visual model, the data of the product defect library, the grid configuration of the detection image, etc.

[0031] In some embodiments, after the visual model online training platform determines that the data processing request is a change instruction, it obtains the current data status of the target data object corresponding to the change instruction in the platform, and at the same time queries the platform's operation log to obtain other change instructions for the target data object within a preset time window.

[0032] The online visual model training platform compares the modified content of the currently received change instruction with the modified content of other change instructions in the operation log. If the comparison results show that the modifications target the same attribute field of the same target data object and the content is different, a data conflict is determined. If a conflict is determined, the online visual model training platform executes a conflict resolution strategy. After the conflict is resolved, the online visual model training platform sends the conflict-resolved valid data to the client that initiated the change instruction through the network transmission channel. After receiving the valid data, the client updates the corresponding data content stored locally.

[0033] S201. Based on the above embodiments, as another optional embodiment, the conflict handling strategy includes: Perform an automatic merge operation on conflicting data; if the automatic merge fails, mark the conflict as unresolved and save all conflicting versions; compare the timestamps corresponding to all conflicting operations, filter out the operation with the latest timestamp and retain the dataset corresponding to the latest operation.

[0034] In this embodiment of the application, the automatic merging operation refers to the operation of merging the modification content corresponding to multiple conflicting change instructions into one executable modification content according to preset rules.

[0035] In some embodiments, when the online visual model training platform executes a conflict resolution strategy, it first performs an automatic merging operation on the conflicting data. After completing the automatic merging operation, the online visual model training platform determines whether the automatic merging operation was successful. The determination is based on whether the merged content can fully cover the legitimate modification intentions of each conflicting change instruction and whether there are any logical contradictions.

[0036] If the automatic merging fails, the online visual model training platform marks the conflict as unresolved and stores the conflict versions corresponding to all conflicting change instructions in the platform's historical version database. The platform then extracts the timestamp for each change instruction from the message information corresponding to those instructions, compares the values ​​of these timestamps, and selects the change instruction corresponding to the timestamp with the largest value. The dataset corresponding to this change instruction is then designated as the retained dataset.

[0037] S202. Based on the above embodiments, as another optional embodiment, the conflicting data is automatically merged, including: Identify the target object type corresponding to the conflicting data. The target object type includes at least model parameters, defect library data, and grid configuration. If the target object is model parameters, extract the gradient tensor corresponding to each change instruction. Input each gradient tensor into a pre-trained gradient fusion network and calculate the compatibility weights of each gradient tensor in different dimensions. If the compatibility score output by the gradient fusion network is higher than the compatibility threshold, then the gradient tensors are weighted and fused according to the compatibility weights to generate the merged update instruction.

[0038] In this embodiment, the target object type refers to the category of the modified data corresponding to the conflicting data. Here, model parameters refer to the numerical parameters used in the visual defect detection model to achieve the detection function; these parameters determine the detection accuracy and effect of the visual defect detection model. Defect database data refers to the dataset stored in the online training platform of the visual model, used to describe the surface defect features of various products. Grid configuration refers to the parameter settings used to divide the detection image into several regions; these settings determine the region division method of the detection image.

[0039] A gradient tensor is a multidimensional array of partial derivatives of model parameters during visual model training. This array is used to represent the direction and magnitude of model parameter updates. A compatibility score is a value output by a gradient fusion network that represents the probability that multiple gradient tensors can be merged.

[0040] In some embodiments, when the online visual model training platform performs automatic merging of conflicting data, it first extracts the attribute information of the target data object corresponding to the conflicting data. This attribute information includes a data identifier field, a storage path prefix, and a business association tag. The online visual model training platform identifies the target object type corresponding to the conflicting data by comparing this attribute information with a preset object type feature library. This target object type includes model parameters, defect library data, and mesh configuration. If the identified target object type is a model parameter, the online visual model training platform first parses the parameter update logs of each conflicting change instruction to determine the specific dimension and index range of the model parameter to be updated. Then, based on the training framework of the visual model, it calls the gradient calculation interface to solve for the partial derivatives of the model parameter update amount corresponding to each change instruction, obtaining the gradient tensor corresponding to each change instruction. For example, when the model parameter to be updated is the convolutional kernel weight of a convolutional layer, and its dimension is 3×3×64, the dimension of the gradient tensor is consistent with the dimension of the convolutional kernel weight, and each element corresponds to the partial derivative value of the convolutional kernel weight parameter.

[0041] The online training platform for the visual model inputs all extracted gradient tensors into a pre-trained gradient fusion network. The input layer of the gradient fusion network receives each gradient tensor and performs standardization, mapping the numerical range of the gradient tensors to the [-1,1] interval. The feature extraction layer extracts local and global features of each gradient tensor through alternating operations of three convolutional layers and two pooling layers. The convolutional kernel sizes of the convolutional layers are 3×3, 2×2, and 1×1, respectively, and the pooling layers use max pooling. The dimension matching layer maps the features of gradient tensors from different sources to the same dimensional space through a fully connected layer, eliminating the problem of inconsistent feature dimensions caused by differences in client training environments. The attention mechanism layer evaluates the importance of the matched features and calculates the attention weights of each gradient tensor feature. These attention weights are positively correlated with the influence of the features on the model's detection accuracy. The weight calculation layer combines the attention weights and feature similarity to output the compatibility weights of each gradient tensor in different dimensions.

[0042] Simultaneously, the scoring output layer of the gradient fusion network calculates and outputs a compatibility score based on the feature similarity of each gradient tensor, the consistency of update direction, and the uniformity of compatibility weight distribution. After obtaining this compatibility score, the online visual model training platform compares it with a preset compatibility threshold. The platform then performs a weighted summation operation on all gradient tensors according to their corresponding compatibility weights, generating a merged gradient tensor. The formula for the weighted summation operation includes: G = w1G1 + w2G2 + ... + w n G n ; Where G represents the merged gradient tensor, w1 to w n G1 to Gn represent the compatibility weights corresponding to the 1st to nth conflicting gradient tensors, respectively. n Let represent the first to nth conflict gradient tensors, where n represents the number of conflicting change instructions, and n≥2.

[0043] The online training platform for visual models generates a merged update instruction based on the merged gradient tensor, combined with the current values ​​of the model parameters and the learning rate. This update instruction includes the final updated values ​​of the model parameters and the effective time.

[0044] In some embodiments, the compatibility threshold can be set to 0.8. This compatibility threshold is set based on statistical analysis of historical conflict merging data. The online visual model training platform reviewed 10,000 past model parameter conflict merging cases and found that when the compatibility threshold was set to 0.8, the model detection accuracy loss after gradient tensor merging was less than 5%, and the merging success rate reached 92%. If the business scenario requires higher model detection accuracy, such as high-precision electronic component defect detection, the compatibility threshold can be adjusted to 0.85. At this point, the merging success rate drops to 88%, but the model detection accuracy loss rate can be controlled within 3%. If the business scenario prioritizes merging efficiency, such as rapid detection in large-scale batch production, the compatibility threshold can be adjusted to 0.75, increasing the merging success rate to 95%, and controlling the model detection accuracy loss rate within 8%.

[0045] S300: If the data processing request is a query command, check if the cache is hit. If not, execute the query optimization strategy, merge the query results, store them in the cache, and return them to the client.

[0046] In this embodiment of the application, the query instruction refers to an instruction used to request data content in the online training platform for visual models. The query content corresponding to the instruction includes parameters of the visual model, data in the product defect database, grid configuration of the detection image, etc.

[0047] Cache hit refers to the situation where the query result corresponding to the query command is already stored in the cache area; query optimization strategy refers to the preset process and method of the visual model online training platform to improve the execution efficiency of query commands.

[0048] In some embodiments, after determining that the data processing request is a query instruction, the online visual model training platform extracts the query keywords corresponding to the query instruction and generates a cache key-value pair based on the query keywords. The online visual model training platform queries the cache area using the cache key-value pair. If the stored data corresponding to the cache key-value pair is found in the cache area, a cache hit is determined.

[0049] If a cache miss is detected, the online visual model training platform executes a query optimization strategy. After the optimization strategy is completed, the platform merges the query results to generate query result data in a unified format. The platform then stores this query result data in the cache area and sends it to the client that initiated the query command via the network transmission channel.

[0050] S301. Based on the above embodiments, as another optional embodiment, the query optimization strategy includes: Based on the query type of the query command, the query is routed to the corresponding query engine; the query types include at least point query, range query, full-text search and aggregation query, and are routed to the primary key index, composite index, ES index and materialized view respectively.

[0051] In this embodiment of the application, the query type refers to the category categorized according to the query method and target of the query instruction. Among them, point query refers to the query method used to query a single target data object; range query refers to the query method used to query multiple data objects that meet a certain numerical range condition; full-text search refers to the query method used to query text data containing specific keywords; and aggregation query refers to the query method used to perform statistical calculations on the numerical values ​​of multiple data objects.

[0052] A query engine is a software module used to execute specific types of query operations; a primary key index is an index structure built based on the primary key field of a data object, which is used to accelerate the execution speed of point queries; a composite index is an index structure built based on multiple fields of a data object, which is used to accelerate the execution speed of range queries; an ES index is an index structure built based on Elasticsearch (ES), which is used to accelerate the execution speed of full-text search; a materialized view is a view that has pre-computed the data and stored the results, which is used to accelerate the execution speed of aggregation queries.

[0053] In some embodiments, when the online visual model training platform executes a query optimization strategy, it first parses the content of the query instruction to determine the corresponding query type. The online visual model training platform has a pre-defined mapping between query types and query engines. In this mapping, point queries correspond to primary key indexes, range queries correspond to composite indexes, full-text searches correspond to Elasticsearch indexes, and aggregation queries correspond to materialized views. Based on this mapping, the online visual model training platform routes the query instruction to the corresponding query engine. Upon receiving the query instruction, the corresponding query engine executes the query operation and returns the query results.

[0054] S302. Based on the above embodiments, as another optional embodiment, routing to the corresponding query engine according to the type of query instruction includes: Extract query pattern features from query commands, input these features into a query classification model, and output the matching probability of each query type. Based on the query type with the highest matching probability and the current load status of each data source, select the optimal routing path from a preset strategy matrix.

[0055] Query pattern features refer to the characteristic information that reflects the query method of a query instruction. This characteristic information includes the number of query conditions, the type of query fields, and the type of query operators. Matching probability refers to the probability value output by the query classification model that the query instruction belongs to a certain query type.

[0056] In some embodiments, after parsing the content of a query instruction, the online training platform for visual models extracts the query pattern features of the query instruction. These features include the number of query conditions, the type of query fields, and the type of query operators.

[0057] The online visual model training platform inputs the extracted query pattern features into a pre-trained query classification model. This model processes the query pattern features and outputs the matching probability of the query instruction belonging to each query type. From the output matching probabilities, the online visual model training platform determines the query type with the highest matching probability.

[0058] Simultaneously, the online visual model training platform obtains the current load status of each data source, including the current number of tasks and CPU resource utilization of each data source. The platform queries a preset strategy matrix, which stores routing paths corresponding to different query types and different data source load statuses. Based on the determined query type and the current data source load status, the platform selects the optimal routing path from the preset strategy matrix and routes the query command to the corresponding query engine according to that path.

[0059] In some embodiments, the number of training samples for the query classification model may be no less than 10,000 query instruction samples.

[0060] S303. Based on the above embodiments, as another optional embodiment, it further includes: Collect historical conflict handling records, extract conflict object types, conflict mode characteristics, and the effectiveness of corresponding resolution strategies, and use these to update the compatibility weight threshold of conflict handling strategies; collect historical query execution data, extract the correlation between query type and execution time and resource consumption, and use this to optimize the routing priority of query optimization strategies.

[0061] In this embodiment of the application, historical conflict handling records refer to the operation logs and result data of past execution of conflict handling strategies stored on the online training platform for visual models.

[0062] Conflict object type refers to the type of target data object corresponding to the conflict in the historical conflict handling record; conflict pattern characteristics refer to information that can reflect the conflict type and characteristics, including the number of change instructions in the conflict, the type of attribute field in the conflict, etc.

[0063] Historical query execution data refers to the operation logs and performance data of past executed query commands stored on the online training platform for visual models; routing priority refers to the priority order of query engines corresponding to query types.

[0064] In some embodiments, the online visual model training platform initiates a historical data collection process, which continuously collects historical conflict resolution records from the platform. The online visual model training platform parses the collected historical conflict resolution records to extract conflict object types, conflict pattern characteristics, and the effectiveness data of corresponding resolution strategies.

[0065] The visual model online training platform inputs the extracted data into a preset optimization model. Based on this data, the optimization model calculates the adjustment value of the compatibility weight threshold in the conflict handling strategy. The visual model online training platform then updates the compatibility weight threshold of the conflict handling strategy according to this adjustment value.

[0066] Meanwhile, the historical data collection process also continuously collects historical query execution data from the platform. The online training platform for visual models analyzes this historical query execution data and extracts data on query type, execution time, and resource consumption.

[0067] The online training platform for visual models analyzes the extracted data to determine the correlation between the execution time and resource consumption of each query engine for different query types. Based on this correlation, the routing priority of the query engines corresponding to each query type in the query optimization strategy is adjusted.

[0068] S304. Based on the above embodiments, as another optional embodiment, it further includes: generating a visual distribution chart of conflict data, an impact range analysis report, and conflict resolution suggestions based on historical cases, based on historical versions of conflict data, operational contexts, and corresponding resolution strategies.

[0069] In some embodiments, the online visual model training platform acquires historical versions of conflict data, operational contexts, and corresponding resolution strategies for the conflict. The online visual model training platform then invokes a pre-defined visualization generation module, which generates a visualization distribution chart showing the data distribution of each conflict version based on the historical versions of the conflict data.

[0070] Meanwhile, the online visual model training platform analyzes the business modules associated with the target data object corresponding to the conflicting data based on the operational context, and generates an impact analysis report explaining the business scope that may be affected if the conflict is not handled correctly.

[0071] In addition, the online visual model training platform queries the historical conflict resolution case library, filters out historical cases that match the current conflict's conflict object type and conflict pattern characteristics, and generates conflict resolution suggestions based on the resolution strategies of these historical cases. The online visual model training platform stores the generated visualization distribution charts, impact scope analysis reports, and conflict resolution suggestions in the platform's report database, and also pushes them to the platform's management interface for administrators to view.

[0072] Based on the above embodiments, as another optional embodiment, this application also provides a data processing system for online training of visual models, the system comprising: The change processing module is configured to process change commands and detect conflicts in the target data; the conflict resolution module is configured to execute conflict handling strategies to obtain valid data after the conflict; the query processing module is configured to process query commands and perform cache hit detection; and the query optimization module is configured to execute query optimization strategies and route query commands to the corresponding engines.

[0073] In some embodiments, the system further includes: a request receiving module configured to receive data processing requests sent by a client; a type parsing module configured to parse the request type corresponding to the data processing request; a strategy optimization module configured to update parameters of conflict and query strategies based on historical data; and a report generation module configured to generate visual charts of conflict data and resolution suggestions.

[0074] It should be noted that the system provided in the above embodiments is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the system and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process can be found in the method embodiments, which will not be repeated here.

[0075] Based on the above embodiments, as another optional embodiment, the present application embodiment may further include a computer storage medium, which may store multiple instructions adapted for loading by a processor and executing a method of the above embodiments. For the specific execution process, please refer to the detailed description of the above embodiments, which will not be repeated here.

[0076] Based on the above embodiments, as another optional embodiment, this application embodiment may further include an electronic device. The electronic device may include: at least one processor, at least one communication bus, a user interface, at least one network interface, and a memory.

[0077] The communication bus is used to enable communication between these components.

[0078] The user interface may include a display screen and a camera. Optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0079] The network interface may include standard wired interfaces and wireless interfaces (such as Wi-Fi interfaces).

[0080] The processor may include one or more processing cores. It connects to various parts of the server via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various server functions and process data. Optionally, the processor may be implemented using at least one of the following hardware forms: Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), and Programmable Logic Array (PLA). The processor may integrate one or more of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content displayed on the screen; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0081] The memory may include random access memory (RAM) or read-only memory. Optionally, the memory may include a non-transitory computer-readable storage medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. As a computer storage medium, the memory may include an operating system, a network communication module, a user interface module, and an application program of one method.

[0082] In electronic devices, the user interface is primarily used to provide an input interface for users and to acquire user input data; while the processor can be used to call an application program stored in memory that represents a method. When executed by one or more processors, this causes the electronic device to perform one or more methods as described in the above embodiments. It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps can be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0083] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0084] In the various embodiments provided in this application, it should be understood that the disclosed apparatus 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 coupling or direct coupling or communication connection shown or discussed may be through some service interface; the indirect coupling or communication connection between apparatuses or units may be electrical or other forms.

[0085] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0086] 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 as a software functional unit.

[0087] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, portable hard drives, magnetic disks, or optical disks.

[0088] The above are merely exemplary embodiments of this disclosure and should not be construed as limiting the scope of this disclosure. Any equivalent changes and modifications made in accordance with the teachings of this disclosure shall still fall within the scope of this disclosure. Other embodiments of this disclosure will readily conceive of those skilled in the art upon consideration of the specification and the disclosure of practical truths.

[0089] This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not described in this disclosure. The specification and embodiments are to be considered exemplary only, and the scope and spirit of this disclosure are defined by the claims.

Claims

1. A data processing method for online training of a visual model, characterized in that, The method includes: Receive data processing requests sent by clients and parse the type of the data processing requests; If the data processing request is a change instruction, check whether the data conflicts. If there are conflicts, execute the conflict handling strategy and synchronize the valid data after conflict handling to the client. If the data processing request is a query command, check if the cache is hit. If not, execute the query optimization strategy, merge the query results, store them in the cache, and return them to the client.

2. The data processing method for online training of visual models according to claim 1, characterized in that, The conflict resolution strategy includes: Perform automatic merging operations on conflicting data; If automatic merging fails, mark the conflict as unresolved and save all conflicted versions; Compare the timestamps of all conflicting operations, filter out the operation with the latest timestamp, and retain the dataset corresponding to the latest operation.

3. The data processing method for online training of visual models according to claim 2, characterized in that, The automatic merging of conflicting data includes: Identify the target object type corresponding to the conflicting data, wherein the target object type includes at least model parameters, defect library data, and mesh configuration; If the target object is a model parameter, extract the gradient tensor corresponding to each of the change instructions; Each gradient tensor is input into a pre-trained gradient fusion network to calculate the compatibility weights of each gradient tensor in different dimensions. If the compatibility score output by the gradient fusion network is higher than the compatibility threshold, the gradient tensors are weighted and fused according to the compatibility weights to generate a merged update instruction.

4. The data processing method for online training of visual models according to claim 1, characterized in that, The query optimization strategies include: Based on the query type of the query instruction, the query is routed to the corresponding query engine; The query types include at least point queries, range queries, full-text searches, and aggregate queries, and are respectively routed to the primary key index, composite index, ES index, and materialized view.

5. The data processing method for online training of visual models according to claim 4, characterized in that, Based on the type of the query instruction, the query is routed to the corresponding query engine, including: Extract the query pattern features of the query instruction, input the query pattern features into the query classification model, and output the matching probability of each query type; Based on the query type with the highest matching probability and the current load status of each data source, the optimal routing path is selected from the preset strategy matrix.

6. The data processing method for online training of visual models according to claim 1, characterized in that, Also includes: Collect historical conflict handling records, extract conflict object types, conflict mode characteristics, and the effectiveness of corresponding resolution strategies, and use these to update the compatibility weight threshold of the conflict handling strategies; Collect historical query execution data, extract the correlation between query type and execution time and resource consumption, and use it to optimize the routing priority of the query optimization strategy.

7. The data processing method for online training of visual models according to claim 6, characterized in that, Also includes: Based on historical versions of conflict data, operational contexts, and corresponding resolution strategies, the system generates visual distribution charts of conflict data, impact scope analysis reports, and conflict resolution recommendations based on historical cases.

8. A data processing system for online training of a visual model, characterized in that, The system includes: The change processing module is configured to process change commands and detect conflicts in the target data. The conflict resolution module is configured to execute a conflict handling strategy to obtain valid data after a conflict. The query processing module is configured to process query commands and perform cache hit detection; The query optimization module is configured to execute query optimization strategies and route query commands to the corresponding engines.

9. An electronic device, characterized in that, It includes a processor, a memory, a user interface, and a network interface. The memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory to cause the electronic device to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted to be loaded by a processor and executed as described in any one of claims 1-7.

Images