Task processing methods and related equipment

Abstract

Embodiments of the present application provide a task processing method, a first data set returned by an edge device set is acquired, the first data set is processed by a preset first visual model set to obtain a second data set; the first data set is processed to obtain a second data set; based on the second data set, the preset first visual model set is trained by a preset training method to obtain a second visual model set, the accuracy of the second visual model set is higher than that of the first visual model set; the second visual model set is sent to the edge device set, so that the edge device set replaces the first visual model set by the second visual model set for task processing. The present application can continuously update and iterate the model in the edge device, without the need for pre-customized algorithm model, so as to decouple the edge device and the algorithm model, and solve the problem of slow application speed of end-edge-cloud computing task.

Description

Technical Field

[0001] This invention relates to the field of image task processing, and more particularly to a task processing method and related equipment. Background Technology

[0002] With the development of digitalization and the continuous breakthroughs in technologies such as 5G, IoT, and AI, massive information flows and data have emerged. This has created a demand for near-data source, low latency, and high bandwidth computing, giving rise to edge computing architectures. Traditional edge-cloud AI computing tasks require pre-customization. For example, the front end transmits video image data, the edge device processes the video image data, and the cloud data center performs further data mining and analysis on the processed data transmitted from the edge computing, providing the results to upper-layer business applications based on business application needs. Edge computing tasks need to be pre-customized according to business application requirements; that is, customized AI computing processing of the content transmitted from the front end is required based on business requirements. This customization work needs to be completed in advance, such as developing algorithm models adaptable to edge devices. This process generates a significant workload. Limited by the differences in edge-cloud collaboration requirements across different scenarios and the development cycle of algorithm models, it is difficult to achieve rapid application for business scenarios. Therefore, existing edge-cloud computing tasks suffer from slow application speed. Summary of the Invention

[0003] This invention provides a task processing method that obtains a first dataset from a set of edge devices, performs data addition on the first dataset to obtain a second dataset, trains a preset first visual model set using the second dataset to obtain a second visual model set with higher accuracy, and then distributes the second visual model set to the set of edge devices for task processing. This method can continuously update and iterate the models in the edge devices without the need for pre-customized algorithm models, thereby decoupling the edge devices from the algorithm models and solving the problem of slow application speed of edge-cloud computing tasks.

[0004] In a first aspect, embodiments of the present invention provide a task processing method, the task processing method comprising:

[0005] The first dataset returned by the edge device set is obtained, and the first dataset is processed by a preset first visual model set.

[0006] The first dataset is processed to obtain the second dataset;

[0007] Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain a second visual model set, wherein the accuracy of the second visual model set is higher than that of the first visual model set.

[0008] The second visual model set is distributed to the edge device set so that the edge device set can replace the first visual model set for task processing using the second visual model set.

[0009] Optionally, the first dataset includes main data and labeled data, wherein the labeled data is a label for the main data, and the data processing of the first dataset to obtain the second dataset includes:

[0010] The labeled data is processed into data tags;

[0011] The main data is associated with the corresponding data tags to obtain the second dataset.

[0012] Optionally, before obtaining the first dataset returned by the edge device, and before processing the first data using a preset first visual model set, the method further includes:

[0013] Obtain the initial visual model set;

[0014] According to the predefined parsing requirements, the initial visual model set is pre-trained to obtain the first visual model set;

[0015] The first visual model set is distributed to the edge device set through the first task predefined record set, so that the edge device set can perform task processing through the first visual model set.

[0016] Optionally, before distributing the second visual model set to the edge device set to replace the first visual model set for task processing, the method further includes:

[0017] The first predefined record set of the task is updated according to the preset update strategy set to obtain the second predefined record set of the task;

[0018] The step of distributing the second visual model set to the edge device set, so that the second visual model set replaces the first visual model set for task processing, includes:

[0019] The second visual model set is distributed to the edge device set through the second task predefined record set, so that the edge device set can replace the first visual model set for task processing with the second visual model set.

[0020] Optionally, before obtaining the first dataset returned by the edge device set, the method further includes:

[0021] The task processing status of the current edge device set is evaluated based on a preset quality evaluation set.

[0022] If the evaluation result does not meet the preset evaluation conditions, a data backhaul request is initiated to the edge device set based on the preset backhaul strategy set, so that the edge device backhauls the first dataset based on the data backhaul request.

[0023] Secondly, embodiments of the present invention provide a task processing method, the task processing method comprising the following steps:

[0024] The first dataset is obtained by processing the task using a pre-set first visual model set;

[0025] The first dataset is uploaded to the data center so that the data center processes the first dataset to obtain the second dataset. Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain the second visual model set, so that the edge device can send back the first dataset based on the data back transmission request.

[0026] Receive the second visual model set sent by the data center, and use the second visual model set to replace the first visual model set for task processing.

[0027] Thirdly, embodiments of the present invention provide a data center device, the device comprising:

[0028] The first acquisition module is used to acquire the first dataset returned by the edge device set. The first dataset is obtained by processing the first visual model set through a preset first visual model set.

[0029] The processing module is used to process the first dataset to obtain the second dataset;

[0030] The first training module is used to train the preset first visual model set based on the second dataset using a preset training method to obtain the second visual model set.

[0031] The first distribution module is used to distribute the second visual model set to the edge device set so that the second visual model set replaces the first visual model set for task processing.

[0032] Thirdly, embodiments of the present invention provide an edge device, the edge device comprising:

[0033] The processing module is used to process tasks using a preset first visual model set to obtain the first dataset.

[0034] The upload module is used to upload the first dataset to the data center so that the data center can process the first dataset to obtain the second dataset, and train the preset first visual model set based on the second dataset using a preset training method to obtain the second visual model set, so that the edge device can send back the first dataset based on the data back transmission request.

[0035] The receiving module is used to receive the second visual model set sent by the data center and use the second visual model set to replace the first visual model set for task processing.

[0036] Fourthly, embodiments of the present invention provide a task processing system. The system includes an image device, a data center device as described in the embodiments of the present invention, and an edge device as described in the embodiments of the present invention, wherein the image device is signal-connected to the edge device, and the edge device is information-connected to the data center device.

[0037] Thirdly, embodiments of the present invention provide an electronic device, including: 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 steps in the task processing method provided in embodiments of the present invention.

[0038] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the task processing method provided in the embodiments of the invention.

[0039] In this embodiment of the invention, a first dataset returned by an edge device set is obtained. This first dataset is processed using a preset first visual model set. The first dataset is then further processed to obtain a second dataset. Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain a second visual model set, where the accuracy of the second visual model set is higher than that of the first visual model set. The second visual model set is then distributed to the edge device set, allowing the edge device set to replace the first visual model set for task processing. By obtaining the first dataset from the edge device set, processing it to obtain the second dataset, training the preset first visual model set with the second dataset to obtain a more accurate second visual model set, and then distributing the second visual model set to the edge device set for task processing, continuous updates and iterations of the models in the edge devices can be achieved without the need for pre-defined algorithm models. This decouples the edge devices from the algorithm models, solving the problem of slow application speed in edge-cloud computing tasks. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 This is an architecture diagram of a task processing system provided in an embodiment of the present invention;

[0042] Figure 2 This is a flowchart of a task processing method provided in an embodiment of the present invention;

[0043] Figure 3 This is a flowchart illustrating another task processing method provided in an embodiment of the present invention;

[0044] Figure 4 This is a schematic diagram of the structure of a data center device provided in an embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of the structure of an edge device provided in an embodiment of the present invention;

[0046] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0047] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Please see Figure 1 , Figure 1 This is an architecture diagram of a task processing system provided in an embodiment of the present invention, such as... Figure 1 As shown, the task processing system includes the following steps: an image device 101, a data center device 102, and an edge device 103, wherein the image device 101 is signal-connected to the edge device 103, and the edge device 103 is information-connected to the data center device 102.

[0049] The aforementioned image device 101 is used to acquire video or images. One edge device 103 can connect to multiple image devices 101. The image device 101 transmits the acquired video or image data to the edge device 103 via an edge network for task processing. The edge network can be a wired or wireless connection, used for communication between the image device 101 and the edge device 103. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB (ultra-wideband) connections, and other currently known or future known wireless connection methods.

[0050] Data center device 102 can be a server or a cloud server; furthermore, data center device 102 can be a server cluster, enabling large-scale management of the visual model set. Data center device 102 stores a preset first visual model set, which includes multiple visual models. Different visual models correspond to different algorithms, and different visual models can handle different computational tasks. One data center device 102 can connect to multiple edge devices 103, and the data center device 102 can distribute the first visual model set to the edge devices 103. The aforementioned data center device 102 can also be referred to as a data center.

[0051] Edge device 103 processes video or image data transmitted from image device 101 and transmits the processed data to data center device 102 via an edge-cloud network. The edge-cloud network can be a wired or wireless connection, and is used for communication between data center device 102 and edge device 103. It should be noted that the aforementioned wireless connection methods may include, but are not limited to, 3G / 4G / 5G connections, WiFi connections, Bluetooth connections, WiMAX connections, Zigbee connections, UWB connections, and other currently known or future known wireless connection methods.

[0052] Please see Figure 2 , Figure 2 This is a flowchart of a task processing method provided in an embodiment of the present invention, such as... Figure 2 As shown, this task processing method is applied to a data center and specifically includes the following steps:

[0053] 201. Obtain the first dataset returned by the edge device set.

[0054] In this embodiment of the invention, the data center obtains a first dataset transmitted back from the edge device set. The first dataset can be obtained by processing a preset first visual model set. The first visual model set includes multiple first visual models, which can be trained and composed of multiple first visual models.

[0055] The aforementioned tasks can be visual tasks, or more specifically, necessary visual tasks in smart city scenarios, such as image or video-based target detection tasks, like personnel detection tasks and vehicle detection tasks.

[0056] The edge device cluster includes at least one edge device that communicates with the data center, and each edge device may be equipped with a pre-defined first visual model for task processing.

[0057] Edge devices acquire video or image data from imaging devices, perform task processing on the video or image data according to a first visual model, and obtain corresponding task processing data. The first dataset includes task processing data corresponding to at least one edge device.

[0058] Optionally, before step 201, the data center may obtain an initial visual model set; pre-train the initial visual model set according to predefined parsing requirements to obtain the first visual model set; and distribute the first visual model set to the edge device set through a first task predefined record set, so that the edge device set can perform task processing through the first visual model set. The initial visual model set includes multiple initial visual models.

[0059] Data centers can obtain different initial visual models from network channels, or they can obtain the corresponding initial visual models based on user uploads.

[0060] The aforementioned parsing requirements can be predefined based on the upper-layer visual business applications. Specifically, they can be predefined based on the target of the upper-layer visual business applications. For example, if the upper-layer business application is analyzing people, then the target of the upper-layer visual business application is people, and the parsing requirement in this case can be the detection and recognition of people. If the upper-layer business application is analyzing vehicles, then the target of the upper-layer visual business application is vehicles, and the parsing requirement in this case can be the detection and recognition of vehicles. Different initial visual models can be pre-trained according to different parsing requirements.

[0061] The aforementioned pre-training can involve training an initial visual model with a small amount of data to obtain a coarse-precision visual model as the first visual model. For example, for the resolution requirement of personnel detection and recognition, the initial visual model can be trained using a dataset containing a small number of personnel images to obtain a coarse-precision personnel detection visual model as the first visual model. Similarly, for the resolution requirement of vehicle detection and recognition, the initial visual model can be trained using a dataset containing a small number of vehicle images to obtain a coarse-precision vehicle detection visual model as the first visual model. Through pre-training, first visual models corresponding to different resolution requirements are obtained, thus forming a first visual model set. At this point, the first visual model set includes first visual models corresponding to different resolution requirements.

[0062] The aforementioned first task predefined record set can be a set of task records required to run by edge devices. The aforementioned first task predefined record set includes task records required to run by each edge device. Each task can include task conditions and task content. The task conditions can be the time of task execution, and the task content can be the visual model that needs to be run. For example, an edge device starts to perform a personnel detection task at 2 pm and needs to run a personnel detection visual model; it starts to perform a vehicle detection task at 6 pm and needs to run a vehicle detection visual model; it starts to perform a personnel detection task at 8 pm and needs to run a personnel detection visual model, etc.

[0063] The data center distributes the first visual model set to the edge device set through the first task predefined record set. Specifically, the data center can generate the first visual model set required by each edge device according to the first task predefined record set, and distribute the corresponding first visual model set to the corresponding edge device.

[0064] After receiving the first visual model set from the data center, each edge device in the edge device cluster deploys the first visual model in the first visual model set and deploys the video data or image data transmitted back by the image device through the deployed first visual model.

[0065] Furthermore, the first visual model set also includes a first task predefined record set corresponding to the edge device. The edge device deploys the first visual models in the first visual model set according to the first task predefined record set. The first task predefined record set also includes the deployment time of each first visual model. For example, if the edge device needs to perform a personnel detection task at 2 pm, it can deploy the personnel detection visual model at 1:50 pm and start running the personnel detection visual model at 2 pm. If the edge device needs to perform a vehicle detection task at 6 pm, it can deploy the vehicle detection visual model at 5:50 pm and start running the vehicle detection visual model at 6 pm, and so on.

[0066] 202. Process the first dataset to obtain the second dataset.

[0067] In this embodiment of the invention, the first dataset can be processed using a preset data processing method. This preset data processing method can be various methods, and it can be related to the training method required by each first visual model. For example, for a first visual model, if the corresponding training method is supervised learning, the data processing method is to label all task data in the first dataset to obtain the second dataset; if the corresponding training method is semi-supervised learning, the data processing method can be to label a portion of the task data in the first dataset to obtain the second dataset; if the corresponding training method is unsupervised learning, the data processing method can be to perform simple data processing on the task data in the first dataset to obtain the second dataset.

[0068] The aforementioned pre-defined data processing methods can also be related to the amount of training data required by each first vision model. For example, when the amount of task data in the first dataset does not meet the training data requirements of the first vision model, data augmentation can be performed on the task data in the first dataset to increase the data volume. For example, data augmentation can be performed through generative adversarial networks or image space transformation to obtain a second dataset. When the amount of task data in the first dataset exceeds the training data required by the first vision model, data cleaning can be performed on the task data in the first dataset. For example, hard sample mining can be performed on the task data in the first dataset, retaining hard samples and removing easy samples to obtain a second dataset.

[0069] Optionally, in this embodiment of the invention, the first dataset includes subject data and labeled data. The labeled data is an annotation of the subject data. Both the subject data and the labeled data are obtained by the edge device through task processing using the first visual model in the first visual model set. The subject data can be visual data, and the labeled data can be attribute data. For example, if the subject data is an image of a car, the labeled data can be attribute data such as the car's color, brand, license plate number, time of appearance, and location of appearance. For example, if the subject data is an image of a target person, the labeled data can be attribute data such as the target person's height, clothing color, hair color, time of appearance, and location of appearance.

[0070] Specifically, the data center can process the aforementioned labeled data into data labels according to a preset data processing method; then, it can associate the aforementioned subject data with the corresponding data labels to obtain a second dataset. Different labeled data can be processed into corresponding data labels for different first visual models. For example, for a face detection visual model, only the face-related labeled data from the personnel labeled data will be extracted as the face data label, without processing the body labeled data, thus obtaining the second dataset for the face detection visual model. Simultaneously, negative sample labels can be applied to non-person subject data, or to non-face subject data, to obtain the second dataset for the face detection visual model.

[0071] In one possible embodiment, the first dataset may further include raw data, which is video data or image data collected by an image device. The data center can process the raw data, such as by labeling or data augmentation, so that the processed raw data can be added to the second dataset.

[0072] In another possible embodiment, the above-mentioned raw data can be processed through a first visual model set deployed in the data center to obtain the main data and labeled data, and then processed according to a preset data processing method. In this way, the hardware error between the edge device and the data center can be added to the second dataset, so that when the visual model is trained through the second dataset, it can learn this hardware error, obtain a more adaptive visual model, and increase the robustness of the visual model.

[0073] 203. Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain the second visual model set.

[0074] In this embodiment of the invention, after obtaining a second dataset according to a preset data processing method, the data center trains each first visual model in a preset first visual model set using the second dataset. The training method for each first visual model can be determined when acquiring or constructing the model. The preset training method can be supervised learning, semi-supervised learning, unsupervised learning, reinforcement learning, ensemble learning, or other training methods.

[0075] After training the corresponding first visual model using the second dataset, the corresponding second visual model is obtained, and thus the corresponding second visual model set is obtained. It should be noted that in this embodiment of the invention, the first visual model and the second visual model are distinguished by visual models with different training times. The second visual model is a visual model obtained by retraining on the basis of the corresponding first visual model, and the accuracy of the second visual model is higher than that of the first visual model.

[0076] 204. Distribute the second vision model set to the edge device set so that the edge device set can replace the first vision model set for task processing using the second vision model set.

[0077] In this embodiment of the invention, the second visual model set can be distributed to the edge device set through the first task predefined record set, so that the edge device set can replace the first visual model set for task processing with the second visual model set.

[0078] Optionally, the first task predefined record set can be updated according to a preset update strategy set to obtain a second task predefined record set; the second visual model set can be distributed to the edge device set through the second task predefined record set so that the edge device set can perform task processing through the first visual model set.

[0079] The aforementioned update strategy set can be determined by the user based on adjustments made to the upper-layer business application. These adjustments can be understood as applications to new business scenarios. For example, during a sudden outbreak, the upper-layer business application might add epidemic analysis. The update strategy set would then include corresponding epidemic prevention and control strategies to guide the generation of corresponding task records, resulting in a second predefined task record set. For instance, an edge device might start performing personnel detection tasks at 2 PM, requiring the running of a personnel detection visual model; start performing vehicle detection tasks at 6 PM, requiring the running of a vehicle detection visual model; and start performing personnel detection tasks at 8 PM, requiring the running of a personnel detection visual model. During an epidemic, it might perform mask-wearing detection tasks throughout the day, requiring the running of a personnel mask detection visual model, and so on. It should be noted that the first and second predefined task record sets distinguish between predefined task record sets that do not require updates. The second predefined task record set is a predefined task record set updated based on the corresponding first predefined task record set according to the preset update strategy set.

[0080] In one possible embodiment, after the data center updates the first predefined record set of the task with the predefined record set of the task through a preset update strategy set to obtain the second predefined record set of the task, it will send a data back transmission request to the edge device set to obtain the first dataset of the edge device set.

[0081] In one possible embodiment, prior to step 201, the data center can initiate a data backhaul request to the edge device set using a preset backhaul strategy set. This strategy set determines the task data to be backed up by each edge device. For example, if the data center adds a new visual model, it can instruct edge devices performing similar task types to backhaul the corresponding task data to train the visual model for that task type. Or, for instance, if an upper-layer application needs to analyze nighttime travel rates, it can instruct each edge device to backhaul task data from 7 PM to 6 AM the following day.

[0082] Of course, the return policy set can also be the same as the update policy set or the task predefined record set, so that the edge device returns the corresponding task data based on the task it is performing.

[0083] Optionally, the task processing status of the current edge device set can be evaluated based on a preset quality evaluation set; if the evaluation result does not meet the preset evaluation conditions, a data backhaul request is initiated to the edge device set based on a preset backhaul strategy set so that the edge device set can back the first dataset.

[0084] In this embodiment of the invention, the preset quality evaluation set includes different quality evaluation methods, each corresponding to a different task, or more specifically, different visual models. These quality evaluation methods can evaluate metrics such as accuracy, recall, robustness, and speed of the model. For example, for an object detection visual model, evaluation can be performed using recall and / or accuracy; for a face recognition visual model, evaluation can be performed using accuracy and / or speed.

[0085] In this embodiment of the invention, a first dataset returned by an edge device set is obtained. This first dataset is processed using a preset first visual model set. A second dataset is obtained by processing the first dataset using a preset data processing method. Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain a second visual model set, which has higher accuracy than the first visual model set. The second visual model set is then distributed to the edge device set so that the edge device set can replace the first visual model set for task processing. By obtaining the first dataset from the edge device set, performing data addition on the first dataset to obtain the second dataset, training the preset first visual model set with the second dataset to obtain a more accurate second visual model set, and then distributing the second visual model set to the edge device set for task processing, the models in the edge devices can be continuously updated and iterated without the need for pre-defined algorithm models. This decouples the edge devices from the algorithm models, solving the problem of slow application speed in edge-cloud computing tasks. Meanwhile, edge devices can autonomously form computing capabilities and complete computing tasks based on a small amount of predefined information through interactive communication with the data center. They have the characteristic of continuous iterative optimization of computing capabilities, thereby improving the adaptability of edge computing devices to the edge computing needs of business scenarios and realizing the rapid application of artificial intelligence technology to business scenarios.

[0086] It should be noted that the task processing method provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that are capable of task processing.

[0087] Please see Figure 3 , Figure 3 This is a flowchart of another task processing method provided in an embodiment of the present invention, such as... Figure 3 As shown, this task processing method is applied to edge devices and specifically includes the following steps:

[0088] 301. The task is processed using a pre-set first visual model set to obtain the first dataset.

[0089] In this embodiment of the invention, the preset first visual model set is distributed through a data center.

[0090] 302. Upload the first dataset to the data center so that the data center can process the first dataset to obtain the second dataset. Based on the second dataset, train the preset first visual model set using the preset training method to obtain the second visual model set.

[0091] 303. Receive the second visual model set sent by the data center, and use the second visual model set to replace the first visual model set for task processing.

[0092] In this embodiment of the invention, a first dataset obtained by processing a task using a preset first visual model set is uploaded to a data center. The data center can then use the first dataset to process a second dataset to train the preset first visual model into a more accurate second visual model. After receiving the second visual model set from the data center, the more accurate second visual model is used to replace the first visual model for task processing. This allows for continuous updates and iterations of the model in the edge device without the need for pre-customized algorithm models. This decouples the edge device from the algorithm model and solves the problem of slow application speed of edge-cloud computing tasks.

[0093] It should be noted that the task processing method provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that are capable of task processing.

[0094] Optional, please see Figure 4 , Figure 4 This is a structural schematic diagram of a data center device provided in an embodiment of the present invention, such as... Figure 4 As shown, the data center equipment includes:

[0095] The first acquisition module 401 is used to acquire the first dataset returned by the edge device set. The first dataset is obtained by task processing through a preset first visual model set.

[0096] Processing module 402 is used to process the first dataset to obtain the second dataset;

[0097] Training module 403 is used to train the preset first visual model set based on the second dataset using a preset training method to obtain a second visual model set, wherein the accuracy of the second visual model set is higher than that of the first visual model set.

[0098] The first distribution module 404 is used to distribute the second visual model set to the edge device set, so that the edge device set can replace the first visual model set with the second visual model set for task processing.

[0099] Optionally, the first dataset includes main data and labeled data, wherein the labeled data is a label for the main data, and the processing module 402 is further configured to process the labeled data into data labels according to a preset data processing method; and associate the main data with the corresponding data labels to obtain the second dataset.

[0100] Optionally, the device further includes:

[0101] The second acquisition module is used to acquire the initial visual model set;

[0102] The second training module is used to pre-train the initial visual model set according to predefined parsing requirements to obtain the first visual model set.

[0103] The second distribution module is used to distribute the first visual model set to the edge device set through the first task predefined record set, so that the edge device set can perform task processing through the first visual model set.

[0104] Optionally, the device further includes:

[0105] The update module is used to update the first predefined record set of the task according to the preset update strategy set to obtain the second predefined record set of the task.

[0106] The first distribution module 404 further includes: distributing the second visual model set to the edge device set through the second task predefined record set, so that the edge device set can perform task processing through the first visual model set.

[0107] Optionally, the device further includes:

[0108] The evaluation module is used to evaluate the task processing status of the current set of edge devices based on a preset quality evaluation set.

[0109] The request module is used to initiate a data backhaul request to the edge device set based on a preset backhaul strategy set if the evaluation result does not meet the preset evaluation conditions.

[0110] It should be noted that the data center equipment provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that can perform task processing.

[0111] The data center equipment provided in this embodiment of the invention can implement all the processes of the task processing method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, further details are omitted here.

[0112] Optional, please see Figure 5 , Figure 5 This is a schematic diagram of the structure of an edge device provided in an embodiment of the present invention, such as... Figure 5 As shown, the edge device includes:

[0113] Processing module 501 is used to perform task processing using a preset first visual model set to obtain a first dataset;

[0114] Upload module 502 is used to upload the first dataset to the data center so that the data center processes the first dataset to obtain a second dataset, and trains the preset first visual model set based on the second dataset using a preset training method to obtain a second visual model set, wherein the accuracy of the second visual model set is higher than that of the first visual model set.

[0115] The receiving module 503 is used to receive the second visual model set sent by the data center and use the second visual model set to replace the first visual model set for task processing.

[0116] It should be noted that the data center equipment provided in this embodiment of the invention can be applied to devices such as smartphones, computers, and servers that can perform task processing.

[0117] The data center equipment provided in this embodiment of the invention can implement all the processes of the task processing method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, further details are omitted here.

[0118] See Figure 6 , Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention, such as... Figure 6 As shown, it includes: a memory 602, a processor 601, and a computer program for a task processing method stored in the memory 602 and executable on the processor 601, wherein:

[0119] The electronic device is used in a data center. The processor 601 is used to call the computer program stored in the memory 602 and execute the following steps:

[0120] The first dataset returned by the edge device set is obtained, and the first dataset is processed by a preset first visual model set.

[0121] The first dataset is processed to obtain the second dataset;

[0122] Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain a second visual model set, wherein the accuracy of the second visual model set is higher than that of the first visual model set.

[0123] The second visual model set is distributed to the edge device set so that the edge device set can replace the first visual model set for task processing using the second visual model set.

[0124] Optionally, the first dataset includes main data and labeled data, wherein the labeled data is a label for the main data, and the processor 601 performs data processing on the first dataset to obtain a second dataset, including:

[0125] The labeled data is processed into data tags according to a preset data processing method;

[0126] The main data is associated with the corresponding data tags to obtain the second dataset.

[0127] Optionally, before obtaining the first dataset returned by the edge device, and before the first data is processed by a preset first visual model set, the method executed by the processor 601 further includes:

[0128] Obtain the initial visual model set;

[0129] According to the predefined parsing requirements, the initial visual model set is pre-trained to obtain the first visual model set;

[0130] The first visual model set is distributed to the edge device set through the first task predefined record set, so that the edge device set can perform task processing through the first visual model set.

[0131] Optionally, before the second visual model set is distributed to the edge device set to replace the first visual model set for task processing, the method executed by the processor 601 further includes:

[0132] The first predefined record set of the task is updated according to the preset update strategy set to obtain the second predefined record set of the task;

[0133] The step of processor 601 sending the second visual model set to the edge device set so that the second visual model set replaces the first visual model set for task processing includes:

[0134] The second visual model set is distributed to the edge device set through the second task predefined record set, so that the edge device set can replace the first visual model set for task processing with the second visual model set.

[0135] Optionally, before acquiring the first dataset returned by the edge device set, the method executed by the processor 601 further includes:

[0136] The task processing status of the current edge device set is evaluated based on a preset quality evaluation set.

[0137] If the evaluation result does not meet the preset evaluation conditions, a data backhaul request is initiated to the edge device set based on the preset backhaul strategy set, so that the edge device backhauls the first dataset based on the data backhaul request.

[0138] Optionally, the electronic device is applied to an edge device, and the task processing method executed by the processor 601 further includes the following steps:

[0139] The first dataset is obtained by processing the task using a pre-set first visual model set;

[0140] The first dataset is uploaded to the data center so that the data center processes the first dataset to obtain the second dataset. Based on the second dataset, the preset first visual model set is trained using a preset training method to obtain the second visual model set. The accuracy of the second visual model set is higher than that of the first visual model set.

[0141] Receive the second visual model set sent by the data center, and use the second visual model set to replace the first visual model set for task processing.

[0142] It should be noted that the electronic device provided in the embodiments of the present invention can be applied to devices such as smartphones, computers, and servers that can perform task processing.

[0143] The electronic device provided in this embodiment of the invention can implement each process of the task processing method in the above-described method embodiments, and can achieve the same beneficial effects. To avoid repetition, further details are omitted here.

[0144] This invention also provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it implements the various processes of the task processing method or application-side task processing method provided in this invention and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0145] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0146] The above description discloses only preferred embodiments of the present invention and should not be construed as limiting the scope of the present invention. Therefore, equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A task processing method characterized by, The method comprises the following steps: obtaining a first data set returned by an edge device set, wherein the first data set is obtained by performing task processing on the first data set through a preset first visual model set; the first data set comprises original data, and the original data is video data or image data collected by an image device; performing data processing on the first data set to obtain a second data set; specifically, the original data is processed through the first visual model set deployed in a data center to obtain subject data and labeled data, and then the subject data and the labeled data are processed according to a preset data processing method, and hardware errors of the edge device and the data center are added to the second data set; based on the second data set, a preset training method is used to train the preset first visual model set, and when the first visual model set is trained through the second data set, the hardware errors are learned, and a second visual model set is obtained, wherein the accuracy of the second visual model set is higher than that of the first visual model set; the second visual model set is sent to the edge device set, so that the edge device set replaces the first visual model set with the second visual model set to perform task processing.

2. The task processing method of claim 1, wherein, The first data set comprises subject data and labeled data, the labeled data is a label of the subject data, and the data processing on the first data set to obtain the second data set comprises: processing the labeled data into data labels; associating the subject data with the corresponding data labels to obtain the second data set.

3. The task processing method of claim 1, wherein, Before the first data set returned by the edge device is obtained, the first data set is obtained by performing task processing on the first data set through a preset first visual model set, the method further comprises: obtaining an initial visual model set; pre-training the initial visual model set according to a predefined analysis requirement to obtain the first visual model set; sending the first visual model set to the edge device set through a first task predefined record set, so that the edge device set performs task processing through the first visual model set.

4. The task processing method of claim 3, wherein, Before the second visual model set is sent to the edge device set to replace the first visual model set to perform task processing, the method further comprises: updating the first task predefined record set according to a preset update strategy set to obtain a second task predefined record set; the second visual model set is sent to the edge device set through the second task predefined record set, so that the edge device set replaces the first visual model set with the second visual model set to perform task processing. Before the first data set returned by the edge device set is obtained, the method further comprises:

5. The task processing method of claim 1, wherein, based on a preset quality evaluation set, evaluating the task processing of the current edge device set; if the evaluation result does not meet the preset evaluation condition, initiating a data return request to the edge device set based on a preset return strategy set, so that the edge device returns the first data set based on the data return request. ​ 6. A task processing method characterized by, The method comprises the following steps: Task processing is performed through a preset first visual model set to obtain a first data set; the first data set comprises original data, and the original data is video data or image data collected by an image device; The first data set is uploaded to a data center, so that the data center processes the first data set to obtain a second data set; specifically, the original data is processed through the first visual model set deployed in the data center to obtain subject data and labeled data, and then the subject data and the labeled data are processed according to a preset data processing method, and hardware errors of edge devices and the data center are added in the second data set; and based on the second data set, a preset training method is used to train the preset first visual model set, and when the first visual model set is trained through the second data set, the hardware errors are learned, to obtain a second visual model set, and the accuracy of the second visual model set is higher than that of the first visual model set; The second visual model set issued by the data center is received, and the second visual model set is used to replace the first visual model set for task processing.

7. A data center device, comprising: The data center device comprises: A first acquisition module is configured to acquire a first data set returned by an edge device set, wherein the first data set is obtained through task processing by a preset first visual model set; the first data set comprises original data, and the original data is video data or image data collected by an image device; A processing module is configured to process the first data set to obtain a second data set; specifically, the original data is processed through the first visual model set deployed in the data center to obtain subject data and labeled data, and then the subject data and the labeled data are processed according to a preset data processing method, and hardware errors of edge devices and the data center are added in the second data set; A first training module is configured to train the preset first visual model set based on the second data set by using a preset training method, and learn the hardware errors when the first visual model set is trained through the second data set, to obtain a second visual model set, and the accuracy of the second visual model set is higher than that of the first visual model set; A first issuing module is configured to issue the second visual model set to the edge device set, so that the second visual model set replaces the first visual model set for task processing.

8. An edge device, characterized by The edge device comprises: A processing module is configured to perform task processing through a preset first visual model set to obtain a first data set; the first data set comprises original data, and the original data is video data or image data collected by an image device; The uploading module is configured to upload the first data set to the data center, so that the data center processes the first data set to obtain a second data set; specifically, the original data is processed by a first visual model set deployed in the data center to obtain subject data and labeled data, and then the subject data and the labeled data are processed according to a preset data processing method, and a hardware error of the edge device and the data center is added in the second data set; and the data center trains the preset first visual model set based on the second data set by using a preset training method to obtain a second visual model set, and when the first visual model set is trained by using the second data set, the hardware error is learned, and the accuracy of the second visual model set is higher than that of the first visual model set; The receiving module is configured to receive the second visual model set issued by the data center, and replace the first visual model set with the second visual model set for task processing.

9. A task processing system characterized by comprising: The system comprises an image device, the data center device of claim 7, and the edge device of claim 8, wherein the image device is signal connected with the edge device, and the edge device is information connected with the data center device.

10. An electronic device, comprising: Comprise: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the task processing method according to any one of claims 1 to 5 when executing the computer program, or the processor implements the steps of the task processing method according to claim 6 when executing the computer program.

11. A computer readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the task processing method according to any one of claims 1 to 5, or the processor executes the computer program to implement the steps of the task processing method according to claim 6.

Images