Distributed computing power model aggregation method, device and computer equipment

Abstract

The present application relates to a method, device and computer equipment for aggregating computing power models based on distribution. The method includes obtaining auxiliary medical imaging data and its image recognition tasks, and identifying each sub-image recognition task through an identification agent unit; based on the execution order of each sub-image recognition task, the auxiliary medical imaging data is processed through the sub-agent unit corresponding to each sub-image recognition task to obtain the sub-initial processing result fed back by each sub-agent unit; and then, through the detection agent unit, the agent feedback information of each sub-agent unit is generated, so as to return and adjust each sub-initial processing result, and obtain a new sub-initial processing result fed back by each sub-agent unit; each new sub-initial processing result replaces each sub-initial processing result, and returns to the above steps until there is no agent feedback information that does not meet the data output condition, and the target recognition result of the auxiliary medical imaging data is determined. The use of this method can improve the accuracy of medical image recognition under low-cost conditions.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence and distributed computing technology, and in particular to a distributed computing power model aggregation method, device and computer equipment. Background Art

[0002] With the advancement of medical imaging technology, medical equipment (such as CT (Computed Tomography), MRI (Magnetic Resonance Imaging), X-ray, etc.) generates a large amount of high-resolution image data every day. These data need to be analyzed quickly and accurately to help doctors diagnose diseases, detect lesions, and classify tumors. However, relying on traditional manual analysis is time-consuming and labor-intensive and easily affected by human factors, so more and more medical institutions have introduced AI (Artificial Intelligence) to assist in medical image recognition technology. Therefore, how to improve the computing power optimization of low-cost AI models is the research focus of improving auxiliary medical image recognition.

[0003] The traditional way to optimize the computing power of AI models is to comprehensively optimize the computing power of AI models through distributed computing frameworks, edge computing platforms, small model technologies, and model aggregation and fusion technologies. However, the above computing power optimization methods have high hardware requirements, slow processing capabilities, poor real-time performance, and low system fault tolerance, resulting in poor accuracy in medical image recognition under low-cost conditions. Summary of the invention

[0004] Based on this, it is necessary to provide a distributed computing power model aggregation method, device, computer equipment, computer-readable storage medium and computer program product to address the above technical problems.

[0005] In a first aspect, the present application provides a distributed computing power model aggregation method, comprising:

[0006] Acquire auxiliary medical image data and an image recognition task of the auxiliary medical image data, and identify each sub-image recognition task through an identification agent unit based on the image recognition task;

[0007] Based on each of the sub-image recognition tasks, query the sub-agent unit corresponding to each sub-image recognition task in the agent database, and based on the execution order of each of the sub-image recognition tasks, perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each of the sub-image recognition tasks to obtain the sub-initial processing results fed back by each of the sub-agent units;

[0008] Based on the sub-initial processing results fed back by each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, agent feedback information of each of the sub-agent units is generated by detecting the agent unit, and when there is agent feedback information that does not meet the data output condition, each of the sub-initial processing results is adjusted by each of the sub-agent units based on the agent feedback information of each of the sub-agent units to obtain a new sub-initial processing result fed back by each of the sub-agent units;

[0009] Each of the new sub-initial processing results replaces each of the sub-initial processing results, and returns to execute the sub-initial processing results based on the feedback of each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and generates the agent feedback information step of each of the sub-agent units through the detection agent unit, until there is no agent feedback information that does not meet the data output condition, and based on the new sub-initial processing results fed back by each of the sub-agent units obtained in the last iteration, generates the target recognition result of the auxiliary medical image data through the detection agent unit.

[0010] Optionally, the identifying each sub-image recognition task based on the image recognition task by using an identification agent unit includes:

[0011] Identify the task flow of the image recognition task by identifying the agent unit, and split the task flow into subtask flows of each flow node;

[0012] Identify the task type corresponding to each subtask process, and divide the subtask processes of the same task type into the same task group to obtain an identification task group corresponding to each task type;

[0013] All sub-task processes in each recognition task group and process nodes corresponding to each sub-task process in each recognition task group are used as sub-image recognition tasks.

[0014] Optionally, based on each of the sub-image recognition tasks, querying a proxy database for a sub-proxy unit corresponding to each sub-image recognition task includes:

[0015] Based on the task type corresponding to each sub-image recognition task, in the agent database, the initial sub-agent unit corresponding to each task type is queried, and for each task type, the current running state of each initial sub-agent unit of the task type and the running cost information of each initial sub-agent unit are collected;

[0016] Based on the current running status of each initial sub-agent unit and the running cost information of each initial sub-agent unit, the target sub-agent unit corresponding to the task type is screened through a resource optimization strategy, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

[0017] Optionally, based on the execution order of each of the sub-image recognition tasks, the auxiliary medical image data is subjected to data recognition by a sub-agent unit corresponding to each of the sub-image recognition tasks, and a sub-initial processing result fed back by each of the sub-agent units is obtained, including:

[0018] Based on the process nodes corresponding to the subtask processes in each sub-image recognition task, the data recognition order of each subtask process in each sub-image recognition task is identified, and based on the data recognition order of each subtask process in each sub-image recognition task, the interaction order of each sub-agent unit is generated;

[0019] Based on the subtask processes corresponding to each subagent unit and the interaction order of each subagent unit, data recognition is performed on the auxiliary medical imaging data through each subagent unit to obtain the data recognition result of each subtask process, and the data recognition processing results of all subtask processes processed by each subagent unit are used as the sub-initial processing results fed back by each subagent unit.

[0020] Optionally, the generating of agent feedback information of each of the sub-agent units by detecting the agent unit based on the sub-initial processing results fed back by each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units includes:

[0021] For each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit and the task content corresponding to each sub-task process, the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization correction information corresponding to each sub-task process are identified through the detection agent unit;

[0022] Sub-result evaluation information corresponding to all sub-task processes and sub-result optimization correction information corresponding to all sub-task processes are used as agent feedback information of the sub-agent unit.

[0023] Optionally, adjusting each of the sub-initial processing results based on the proxy feedback information of each of the sub-agent units through each of the sub-agent units to obtain a new sub-initial processing result fed back by each of the sub-agent units includes:

[0024] For each subagent unit, based on the sub-result optimization correction information corresponding to each subtask process of the subagent unit, a processing correction strategy for each subtask process of the subagent unit is generated;

[0025] Based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction sequence of each sub-agent unit, the auxiliary medical image data is re-recognized through each sub-agent unit to obtain a new data recognition result of each sub-task process;

[0026] The new data recognition results of all sub-task processes of each sub-agent unit are used as the new sub-initial processing results fed back by each sub-agent unit.

[0027] In a second aspect, the present application also provides a distributed computing power model aggregation device, including:

[0028] An acquisition module, used to acquire auxiliary medical image data and image recognition tasks of the auxiliary medical image data, and identify each sub-image recognition task through an identification agent unit based on the image recognition task;

[0029] A processing module, configured to query the sub-agent unit corresponding to each sub-image recognition task in the agent database based on each of the sub-image recognition tasks, and based on the execution order of each of the sub-image recognition tasks, perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each of the sub-image recognition tasks to obtain a sub-initial processing result fed back by each of the sub-agent units;

[0030] an adjustment module, configured to generate agent feedback information of each of the sub-agent units by detecting the agent units based on the sub-initial processing results fed back by each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and to adjust each of the sub-initial processing results by each of the sub-agent units based on the agent feedback information of each of the sub-agent units when there is agent feedback information that does not meet the data output condition, so as to obtain a new sub-initial processing result fed back by each of the sub-agent units;

[0031] An iterative module is used to replace each of the sub-initial processing results with each of the new sub-initial processing results, and return to execute the sub-initial processing results based on the feedback of each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and generate the agent feedback information step of each of the sub-agent units through the detection agent unit until there is no agent feedback information that does not meet the data output condition, and then based on the new sub-initial processing results fed back by each of the sub-agent units obtained in the last iteration, the target recognition result of the auxiliary medical image data is generated through the detection agent unit.

[0032] Optionally, the acquisition module is specifically used to:

[0033] Identify the task flow of the image recognition task by identifying the agent unit, and split the task flow into subtask flows of each flow node;

[0034] Identify the task type corresponding to each subtask process, and divide the subtask processes of the same task type into the same task group to obtain an identification task group corresponding to each task type;

[0035] All sub-task processes in each recognition task group and process nodes corresponding to each sub-task process in each recognition task group are used as sub-image recognition tasks.

[0036] Optionally, the processing module is specifically used to:

[0037] Based on the task type corresponding to each sub-image recognition task, in the agent database, the initial sub-agent unit corresponding to each task type is queried, and for each task type, the current running state of each initial sub-agent unit of the task type and the running cost information of each initial sub-agent unit are collected;

[0038] Based on the current running status of each initial sub-agent unit and the running cost information of each initial sub-agent unit, the target sub-agent unit corresponding to the task type is screened through a resource optimization strategy, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

[0039] Optionally, the processing module is specifically used to:

[0040] Based on the process nodes corresponding to the subtask processes in each sub-image recognition task, the data recognition order of each subtask process in each sub-image recognition task is identified, and based on the data recognition order of each subtask process in each sub-image recognition task, the interaction order of each sub-agent unit is generated;

[0041] Based on the subtask processes corresponding to each subagent unit and the interaction order of each subagent unit, data recognition is performed on the auxiliary medical imaging data through each subagent unit to obtain the data recognition result of each subtask process, and the data recognition processing results of all subtask processes processed by each subagent unit are used as the sub-initial processing results fed back by each subagent unit.

[0042] Optionally, the adjustment module is specifically used to:

[0043] For each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit and the task content corresponding to each sub-task process, the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization correction information corresponding to each sub-task process are identified through the detection agent unit;

[0044] Sub-result evaluation information corresponding to all sub-task processes and sub-result optimization correction information corresponding to all sub-task processes are used as agent feedback information of the sub-agent unit.

[0045] Optionally, the adjustment module is specifically used to:

[0046] For each subagent unit, based on the sub-result optimization correction information corresponding to each subtask process of the subagent unit, a processing correction strategy for each subtask process of the subagent unit is generated;

[0047] Based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction sequence of each sub-agent unit, the auxiliary medical image data is re-recognized through each sub-agent unit to obtain a new data recognition result of each sub-task process;

[0048] The new data recognition results of all sub-task processes of each sub-agent unit are used as the new sub-initial processing results fed back by each sub-agent unit.

[0049] In a third aspect, the present application provides a computer device, wherein the computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of any one of the methods in the first aspect are implemented.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0051] In a fifth aspect, the present application provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0052] The above-mentioned distributed computing model aggregation method, device and computer equipment obtain auxiliary medical image data and image recognition tasks of the auxiliary medical image data, and based on the image recognition tasks, identify each sub-image recognition task through an identification agent unit; based on each sub-image recognition task, query the sub-agent unit corresponding to each sub-image recognition task in the agent database, and based on the execution order of each sub-image recognition task, perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each sub-image recognition task to obtain the sub-initial processing results fed back by each sub-agent unit; based on the sub-initial processing results fed back by each sub-agent unit and the sub-image recognition tasks corresponding to each sub-agent unit, generate agent feedback of each sub-agent unit through a detection agent unit. Information, and in the case that the proxy feedback information does not meet the data output conditions, based on the proxy feedback information of each sub-proxy unit, through each sub-proxy unit, adjust each sub-initial processing result to obtain a new sub-initial processing result fed back by each sub-proxy unit; replace each sub-initial processing result with each new sub-initial processing result, and return to execute the sub-initial processing result fed back by each sub-proxy unit and the sub-image recognition task corresponding to each sub-proxy unit, and generate the proxy feedback information step of each sub-proxy unit through the detection proxy unit, until there is no proxy feedback information that does not meet the data output conditions, based on the new sub-initial processing result fed back by each sub-proxy unit obtained in the last iteration, through the detection proxy unit, generate the target recognition result of the auxiliary medical image data. This solution significantly reduces the dependence on high-performance central computing resources by distributing the image recognition tasks to the proxy units corresponding to multiple small models with low computing power set on the edge device, thereby reducing the overall hardware resource demand and avoiding the dependence on high-performance hardware in traditional AI models. Secondly, through the independent operation and interactive processing of each sub-agent unit, even if the edge devices corresponding to some sub-agent units fail, other devices can continue to work, which effectively reduces the risk of single-point failure interfering with the overall data recognition processing and enhances the fault tolerance in the face of hardware failures or network problems. After that, because the image recognition task is performed in the sub-processing unit set on the edge device, local collaborative data recognition is performed, which significantly reduces the delay in data transmission and improves the response speed in real-time applications. This method solves the problem of data transmission delay in the traditional centralized model while improving the data recognition response time for application scenarios that require low latency, such as smart transportation and real-time monitoring.Finally, this solution uses distributed processing of sub-processing units corresponding to multiple small models and the comprehensive output of the aggregate model, and performs iterative data recognition optimization in the distributed processing process, thereby ensuring the data recognition accuracy of auxiliary medical imaging data while ensuring relatively low computing resources, thereby effectively improving the accuracy of medical image recognition under low-cost conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 This is a schematic diagram of an example operation flow structure of a distributed computing power model aggregation system in one embodiment;

[0055] Figure 2 It is a flowchart of a distributed computing power model aggregation method in one embodiment;

[0056] Figure 3 A schematic diagram of a process of an example of aggregation based on a distributed computing power model in an embodiment;

[0057] Figure 4 It is a structural block diagram of a distributed computing power model aggregation device in one embodiment;

[0058] Figure 5 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0059] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0060] The distributed computing model aggregation method provided in the embodiment of the present application can be applied to the medical image recognition application environment of medical equipment. Figure 1The diagram shows an example operation flow structure diagram of a distributed computing model aggregation system. In this example, after the system obtains the auxiliary medical image data, it analyzes the data task of the auxiliary medical image data through AI agent 1, then performs data recognition through AI agent 2, AI agent 3, and AI agent 4, performs detection optimization through AI agent 5, and iterates again through AI agent 2, AI agent 3, and AI agent 4 based on the detection optimization results to perform optimization adjustment again, until AI agent 5 passes the detection, and outputs the target recognition result of the auxiliary medical image data through AI agent 5. Among them, the data system processing system includes: edge computing device: including multiple edge devices, each of which is deployed with a small model (i.e., AI agent, or agent unit). The edge computing device is responsible for receiving input data and performing local reasoning. Aggregation model unit: used to aggregate the small model reasoning results from each edge device, and finally generate a comprehensive output. Data transmission unit: used to connect the edge device with the aggregation model server to ensure efficient data transmission and communication. Among them, edge computing devices: distributed near the source of data generation (such as smart cameras, sensor nodes, etc.). Each device has a built-in small model responsible for preliminary reasoning and processing of data. Aggregation model unit: deployed in the central node or cloud, aggregates and integrates the output results of small models from multiple edge computing devices to generate the final inference results. Data transmission unit: connects all edge devices and the aggregation model server to support efficient data transmission and communication. Among them, the edge computing devices include: Controller: optional MCU or embedded processor, such as Raspberry Pi, NVIDIA Jetson, etc., responsible for running small models and processing data. Sensor: used to collect data, the specific selection depends on the application scenario, such as camera, temperature and humidity sensor, etc. Internal device composition: including processing unit, storage unit, communication interface, etc., to ensure that the device can perform inference tasks and connect to the data transmission network. Aggregation model unit includes: Controller: high-performance server or cloud computing platform, equipped with high-performance CPU / GPU, capable of processing and integrating data from multiple edge computing devices. Internal device composition: including high-performance processor, storage system, network interface, etc., to ensure that the aggregation model can be efficiently executed and large amounts of data can be processed.

[0061] Among them, the system can be set in a terminal, a server, or a device that combines a terminal and a server. Among them, the terminal is a personal computer, a laptop, etc. The server can be implemented by an independent server or a server cluster composed of multiple servers. Among them, the terminal distributes the image recognition task to the proxy units corresponding to multiple small models with low computing power set on the edge device, and performs interactive data recognition between the proxy units, which significantly reduces the dependence on high-performance central computing resources, thereby reducing the demand for overall hardware resources and avoiding the problem of dependence on high-performance hardware in traditional AI models. Secondly, through the independent operation and interactive processing of each sub-agent unit, even if the edge devices corresponding to some sub-agent units fail, other devices can continue to work, thereby effectively reducing the risk of single-point failure interfering with the overall data recognition processing, and enhancing the fault tolerance in the face of hardware failures or network problems. Afterwards, since the image recognition task is performed in the sub-processing unit set on the edge device, local collaborative data recognition is performed, thereby significantly reducing the delay of data transmission and improving the response speed in real-time applications. This method solves the problem of data transmission delay in the traditional centralized model while improving the data recognition response time of application scenarios that require low latency, such as intelligent transportation and real-time monitoring. Finally, this solution uses distributed processing of sub-processing units corresponding to multiple small models and the comprehensive output of the aggregate model, and performs iterative data recognition optimization in the distributed processing process, thereby ensuring the data recognition accuracy of auxiliary medical imaging data while ensuring relatively low computing resources, thereby effectively improving the accuracy of medical image recognition under low-cost conditions.

[0062] In an exemplary embodiment, Figure 2 As shown, a distributed computing power model aggregation method is provided, and the method is applied to a terminal as an example for explanation, including the following steps S201 to S204. Among them:

[0063] Step S201 , obtaining auxiliary medical image data and image recognition tasks of the auxiliary medical image data, and identifying each sub-image recognition task through an identification agent unit based on the image recognition task.

[0064] In this embodiment, the terminal responds to the staff's operation of collecting data on the area where medical image recognition is required through medical imaging equipment, the acquired data input operation, the acquisition of image data, and the recognition task corresponding to the data, and obtains auxiliary medical image data and the image recognition task of the auxiliary medical image data. Among them, the auxiliary medical image data can be the influence data of each image type, and the image type is, for example, a color image, a black and white image, a grayscale image, and a binary image. The image recognition task is a recognition task for performing data recognition on the acquired image data, and the recognition task is, for example, an image classification recognition, an image detection recognition, a target image range recognition, and the like. Then, the terminal identifies each sub-image recognition task in the image recognition task by setting a small model in the edge computing device based on the distributed computing power model aggregation system. Among them, each sub-image recognition task is a recognition task group corresponding to a different task type, wherein each recognition task group may include one or more sub-task processes, and all sub-task processes in all recognition task groups are sorted according to the process nodes corresponding to each sub-task process. The task process obtained is the task process corresponding to the image recognition task. The specific recognition process will be described in detail later. The order in which each process node is sorted is the order obtained by sorting the task positions of each process node in the task flow.

[0065] Step S202, based on each sub-image recognition task, query the sub-agent unit corresponding to each sub-image recognition task in the agent database, and based on the execution order of each sub-image recognition task, perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each sub-image recognition task to obtain the sub-initial processing result fed back by each sub-agent unit.

[0066] In this embodiment, the terminal queries the sub-agent unit corresponding to each sub-image recognition task in the agent database based on each sub-image recognition task. The task type described in each sub-image recognition task can correspond to one or more sub-agent units, that is, each sub-agent unit can process an image recognition task of a task type, and each task type can correspond to one or more sub-agent units. The number of sub-agent units corresponding to each image recognition task is one. The query process will be described in detail later.

[0067] Based on the execution order of each sub-image recognition task, the terminal performs data recognition on the auxiliary medical image data through the sub-agent units corresponding to each sub-image recognition task, and obtains the sub-initial processing results fed back by each sub-agent unit. Among them, the sub-initial processing structure fed back by each sub-agent unit includes the task processing results of each sub-task process in the recognition task group corresponding to each sub-agent unit.

[0068] Step S203, based on the sub-initial processing results fed back by each sub-agent unit and the sub-image recognition task corresponding to each sub-agent unit, by detecting the agent unit, agent feedback information of each sub-agent unit is generated, and in the case that the agent feedback information does not meet the data output condition, based on the agent feedback information of each sub-agent unit, each sub-initial processing result is adjusted through each sub-agent unit to obtain a new sub-initial processing result fed back by each sub-agent unit.

[0069] In this embodiment, the terminal generates proxy feedback information of each sub-agent unit through the detection agent unit based on the sub-initial processing results fed back by each sub-agent unit and the sub-image recognition task corresponding to each sub-agent unit, and adjusts each sub-initial processing result through each sub-agent unit based on the proxy feedback information of each sub-agent unit to obtain a new sub-initial processing result fed back by each sub-agent unit. The proxy unit feedback of each sub-agent unit includes sub-result evaluation information corresponding to each sub-task process and sub-result optimization correction information corresponding to each sub-task process. The sub-result evaluation information is the evaluation value obtained by the detection agent unit for evaluating the data recognition result of each sub-task process, and the data output condition is the evaluation value threshold. When there is a sub-task process below the evaluation value threshold, it is determined that there is a situation where the proxy feedback information does not meet the data output condition. When there is no sub-task process below the evaluation value threshold, it is determined that the proxy feedback information meets the data output condition. The process of adjusting each sub-initial processing result will be described in detail later.

[0070] Step S204, replace each sub-initial processing result with each new sub-initial processing result, and return to execute the sub-initial processing results based on the feedback of each sub-agent unit and the sub-image recognition task corresponding to each sub-agent unit, and generate the agent feedback information step of each sub-agent unit by detecting the agent unit, until there is no agent feedback information that does not meet the data output condition, based on the new sub-initial processing results fed back by each sub-agent unit obtained in the last iteration, generate the target recognition result of the auxiliary medical image data by detecting the agent unit.

[0071] In this embodiment, the terminal replaces each sub-initial processing result with each new sub-initial processing result, and returns to execute step S203, until there is no proxy feedback information that does not meet the data output condition, based on the new sub-initial processing result feedbacked by each sub-proxy unit obtained in the last iteration, the target recognition result of the auxiliary medical image data is generated by detecting the proxy unit. Among them, when the data recognition method of each sub-task process is a progressive processing method, the target recognition result of the auxiliary medical image data is the data recognition result of the last sub-task process, and when the data recognition method of each sub-task process is a parallel processing method, the target recognition result of the auxiliary medical image data is the data recognition result of each sub-task process, which is sorted and summarized according to the order of the process nodes corresponding to each sub-task process.

[0072] Based on the above scheme, by distributing the image recognition task to the proxy units corresponding to multiple small models with low computing power set on the edge device, interactive data recognition is performed between the proxy units, which significantly reduces the dependence on high-performance central computing resources, thereby reducing the overall demand for hardware resources and avoiding the dependence on high-performance hardware in traditional AI models. Secondly, through the independent operation and interactive processing of each sub-agent unit, even if the edge devices corresponding to some sub-agent units fail, other devices can continue to work, thereby effectively reducing the risk of single-point failure interfering with the overall data recognition processing and enhancing the fault tolerance in the face of hardware failures or network problems. After that, since the image recognition task is performed in the sub-processing unit set on the edge device, local collaborative data recognition is performed, which significantly reduces the delay of data transmission and improves the response speed in real-time applications. This method solves the problem of data transmission delay in the traditional centralized model while improving the data recognition response time of application scenarios that require low latency, such as smart transportation and real-time monitoring. Finally, this solution uses distributed processing of sub-processing units corresponding to multiple small models and the comprehensive output of the aggregate model, and performs iterative data recognition optimization in the distributed processing process, thereby ensuring the data recognition accuracy of auxiliary medical imaging data while ensuring relatively low computing resources, thereby effectively improving the accuracy of medical image recognition under low-cost conditions.

[0073] Optionally, based on the image recognition task, each sub-image recognition task is identified by identifying the agent unit, including: identifying the task flow of the image recognition task by identifying the agent unit, and splitting the task flow into sub-task flows of each flow node; identifying the task type corresponding to each sub-task flow, and dividing the sub-task flows of the same task type into the same task group to obtain an identification task group corresponding to each task type; taking all sub-task flows in each identification task group and the flow nodes corresponding to each sub-task flow in each identification task group as each sub-image recognition task.

[0074] In the present embodiment, the terminal identifies the task flow of the image recognition task by identifying the proxy unit, and splits the task flow into subtask flows of each process node. Then, the terminal identifies the task type corresponding to each subtask flow, and divides each subtask flow of the same task type into the same task group, and obtains the recognition task group corresponding to each task type. Among them, the task type can be but is not limited to image data recognition, image data classification, image data positioning, and image data rendering. In another embodiment, when the image data is divided into regions, frames, or frequency bands for data recognition, the subtask flows obtained by splitting the task flow are subtask flows of different process nodes corresponding to different regions, frames, or frequency bands.

[0075] Afterwards, the terminal uses all sub-task processes in each recognition task group and process nodes corresponding to each sub-task process in each recognition task group as each sub-image recognition task.

[0076] Based on the above solution, by splitting the task process according to the task type, or splitting the task process according to different areas, number of frames, and frequency bands, the comprehensiveness and accuracy of the splitting of image recognition tasks with different processing methods are ensured.

[0077] Optionally, based on each sub-image recognition task, the sub-agent unit corresponding to each sub-image recognition task is queried in the agent database, including: based on the task type corresponding to each sub-image recognition task, the initial sub-agent unit corresponding to each task type is queried in the agent database, and for each task type, the current operating status of each initial sub-agent unit of the task type and the operating cost information of each initial sub-agent unit are collected; based on the current operating status of each initial sub-agent unit and the operating cost information of each initial sub-agent unit, the target sub-agent unit corresponding to the task type is screened through a resource optimization strategy, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

[0078] In this embodiment, based on the task type corresponding to each sub-image recognition task, the terminal queries the initial sub-agent unit corresponding to each task type in the agent database, and for each task type, collects the current operation status of each initial sub-agent unit of the task type and the operation cost information of each initial sub-agent unit. In another embodiment, when the image data is divided into regions, frames, or frequency bands for data recognition, the initial sub-agent units obtained are the initial sub-agent units corresponding to the same task type.

[0079] Then, based on the current running state of each initial subagent unit and the running cost information of each initial subagent unit, the terminal selects the target subagent unit corresponding to the task type through the resource optimization strategy, and uses the target subagent unit as the subagent unit of the sub-image recognition task corresponding to the task type. Specifically, the terminal selects the initial subagent unit whose current running state is an idle state as the initial target subagent unit, and then selects the initial target subagent unit with the lowest running cost in the order of running cost information from low to high as the target subagent unit corresponding to the task type. In another embodiment, when image data is divided into regions, frames, or frequency bands for data recognition, the terminal selects the initial subagent units corresponding to the number of all process nodes contained in the task flow in the initial subagent units in each idle running state in the order of running cost information from low to high as the target subagent units corresponding to each subtask flow.

[0080] Based on the above scheme, after filtering the initial sub-agent units corresponding to each sub-task process of different task processes, resource optimization screening is performed to determine the target sub-agent units corresponding to each sub-task process, thereby improving the screening resource optimization degree and screening accuracy of the target sub-agent units.

[0081] Optionally, based on the execution order of each sub-image recognition task, data recognition is performed on the auxiliary medical imaging data through the sub-agent unit corresponding to each sub-image recognition task to obtain a sub-initial processing result fed back by each sub-agent unit, including: based on the process nodes corresponding to each sub-task process in each sub-image recognition task, the data recognition order of each sub-task process in each sub-image recognition task is identified, and based on the data recognition order of each sub-task process in each sub-image recognition task, an interaction order of each sub-agent unit is generated; based on each sub-task process corresponding to each sub-agent unit and the interaction order of each sub-agent unit, data recognition is performed on the auxiliary medical imaging data through each sub-agent unit to obtain a data recognition result of each sub-task process, and the data recognition processing results of all sub-task processes processed by each sub-agent unit are used as the sub-initial processing results fed back by each sub-agent unit.

[0082] In this embodiment, the terminal identifies the data recognition order of each subtask process in each sub-image recognition task based on the process nodes corresponding to each subtask process in each sub-image recognition task, and generates the interaction order of each sub-agent unit based on the data recognition order of each subtask process in each sub-image recognition task. The interaction order of each sub-unit is the data recognition order of each subtask process, and the order of sorting the sub-agent units to which each subtask process belongs, that is, there is a process of repeated calling of sub-agent units. For example, the subtask process includes a, b, c, d, e, f, and g. A, b, and c belong to the AI1 agent unit, d, e belong to the AI2 agent unit, and f and g belong to the AI3 agent unit. The data recognition order of the subtask process is a, e, g, d, b, f, and c, and the interaction order of each sub-agent unit is AI1 agent unit, AI2 agent unit, AI3 agent unit, AI2 agent unit, AI1 agent unit, AI3 agent unit, and AI1 agent unit. In another embodiment, when the image data is divided into regions, frames, or frequency bands for data recognition. There is no interaction order among the sub-agent units, but each sub-agent unit performs data recognition at the same time, and the sub-task processes contained in each sub-agent unit are sorted and processed in a random order.

[0083] Then, based on the subtask processes corresponding to each subagent unit and the interaction order of each subagent unit, the terminal performs data recognition on the auxiliary medical imaging data through each subagent unit to obtain the data recognition result of each subtask process, and uses the data recognition processing results of all subtask processes processed by each subagent unit as the sub-initial processing results fed back by each subagent unit.

[0084] Based on the above scheme, each sub-agent unit is interactively sorted through the data recognition order of the sub-task process, thereby ensuring that each sub-agent unit can independently process each sub-task process while improving the processing accuracy and efficiency of the task process.

[0085] Optionally, based on the sub-initial processing results fed back by each sub-agent unit and the sub-image recognition tasks corresponding to each sub-agent unit, the agent feedback information of each sub-agent unit is generated by detecting the agent unit, including: for each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit and the task content corresponding to each sub-task process, the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization and correction information corresponding to each sub-task process are identified by detecting the agent unit; the sub-result evaluation information corresponding to all sub-task processes and the sub-result optimization and correction information corresponding to all sub-task processes are used as the agent feedback information of the sub-agent unit.

[0086] In this embodiment, the terminal identifies the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization correction information corresponding to each sub-task process through the detection agent unit for each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit and the task content corresponding to each sub-task process. The detection agent unit is a small model used to evaluate the data recognition results of each sub-task process and to optimize and correct the data recognition results. The evaluation method for the result evaluation of different task processes and the optimization correction method for the result optimization correction are the evaluation procedures preset by the staff in the small model and the methods executed by the optimization correction program. This scheme is not limited.

[0087] Finally, the terminal uses the sub-result evaluation information corresponding to all sub-task processes and the sub-result optimization correction information corresponding to all sub-task processes as the agent feedback information of the sub-agent unit.

[0088] Based on the above scheme, data detection and optimization correction are performed by individually selecting proxy units to ensure that the terminal can fully utilize the computing power of each proxy unit and gradually improve and optimize the results of data recognition.

[0089] Optionally, based on the agent feedback information of each sub-agent unit, each sub-initial processing result is adjusted through each sub-agent unit to obtain a new sub-initial processing result fed back by each sub-agent unit, including: for each sub-agent unit, based on the sub-result optimization correction information corresponding to each subtask process of the sub-agent unit, generating a processing correction strategy for each sub-task process of the sub-agent unit; based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction sequence of each sub-agent unit, re-data recognition is performed on the auxiliary medical imaging data through each sub-agent unit to obtain a new data recognition result for each subtask process; the new data recognition results of all subtask processes of each sub-agent unit are used as the new sub-initial processing results fed back by each sub-agent unit.

[0090] In this embodiment, the terminal generates a processing correction strategy for each sub-task process of the sub-agent unit based on the sub-result optimization correction information corresponding to each sub-task process of the sub-agent unit for each sub-agent unit. The sub-result optimization correction information includes the process optimization strategy of the sub-task process corresponding to the data recognition result, and then the terminal identifies the target optimization parameters corresponding to the process optimization strategy and the optimization values ​​corresponding to the target optimization parameters. Then, the terminal uses the optimization values ​​of the target optimization parameters of the sub-task process as the processing correction strategy of the sub-task process. In the subsequent iterative data recognition, the parameter values ​​of each target optimization parameter are first adjusted to the optimization values ​​corresponding to each target optimization parameter through the processing correction strategy before executing the sub-task process.

[0091] Afterwards, based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction order of each sub-agent unit, the terminal re-recognizes the auxiliary medical imaging data through each sub-agent unit to obtain the new data recognition result of each sub-task process. Finally, the terminal uses the new data recognition results of all sub-task processes of each sub-agent unit as the new sub-initial processing results fed back by each sub-agent unit.

[0092] Based on the above scheme, each subtask process is optimized by optimizing the correction information of the sub-results corresponding to each subtask process, thereby ensuring that the new data recognition results of each subtask process obtained by iteration can converge and be more accurate, effectively improving the optimization effect of the data recognition results, thereby ensuring the accuracy of the data recognition results obtained by collaborative distributed processing of multiple small models.

[0093] This application also provides an example of aggregation based on distributed computing power model, such as Figure 3 As shown, the specific processing process includes the following steps:

[0094] Step S301, obtaining auxiliary medical image data and an image recognition task of the auxiliary medical image data.

[0095] Step S302 , identifying the task flow of the image recognition task by identifying the agent unit, and splitting the task flow into sub-task flows of each flow node.

[0096] Step S303 , identifying the task type corresponding to each subtask process, and grouping the subtask processes of the same task type into the same task group, to obtain an identified task group corresponding to each task type.

[0097] Step S304: All sub-task processes in each recognition task group and process nodes corresponding to each sub-task process in each recognition task group are used as sub-image recognition tasks.

[0098] Step S305, based on the task type corresponding to each sub-image recognition task, query the initial sub-agent unit corresponding to each task type in the agent database, and for each task type, collect the current operating status of each initial sub-agent unit of the task type and the operating cost information of each initial sub-agent unit.

[0099] Step S306, based on the current running state of each initial sub-agent unit and the running cost information of each initial sub-agent unit, through the resource optimization strategy, the target sub-agent unit corresponding to the task type is screened, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

[0100] Step S307, based on the process nodes corresponding to each sub-task process in each sub-image recognition task, identify the data recognition order of each sub-task process in each sub-image recognition task, and based on the data recognition order of each sub-task process in each sub-image recognition task, generate the interaction order of each sub-agent unit.

[0101] Step S308, based on the sub-task processes corresponding to each sub-agent unit and the interaction order of each sub-agent unit, data recognition is performed on the auxiliary medical imaging data through each sub-agent unit to obtain the data recognition result of each sub-task process, and the data recognition processing results of all sub-task processes processed by each sub-agent unit are used as the sub-initial processing results fed back by each sub-agent unit.

[0102] Step S309, for each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit, and the task content corresponding to each sub-task process, the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization and correction information corresponding to each sub-task process are identified by detecting the agent unit.

[0103] Step S310: Sub-result evaluation information corresponding to all sub-task processes and sub-result optimization correction information corresponding to all sub-task processes are used as agent feedback information of the sub-agent unit.

[0104] Step S311, when there is agent feedback information that does not meet the data output condition, for each subagent unit, based on the sub-result optimization correction information corresponding to each subtask process of the subagent unit, a processing correction strategy for each subtask process of the subagent unit is generated.

[0105] Step S312, based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction sequence of each sub-agent unit, the auxiliary medical image data is re-recognized through each sub-agent unit to obtain new data recognition results of each sub-task process.

[0106] Step S313: The new data recognition results of all subtask processes of each subagent unit are used as new sub-initial processing results fed back by each subagent unit.

[0107] Step S314, replace each sub-initial processing result with each new sub-initial processing result, and return to execute the sub-initial processing results based on the feedback of each sub-agent unit and the sub-image recognition task corresponding to each sub-agent unit, and generate the agent feedback information step of each sub-agent unit by detecting the agent unit, until there is no agent feedback information that does not meet the data output condition, based on the new sub-initial processing results fed back by each sub-agent unit obtained in the last iteration, generate the target recognition result of the auxiliary medical image data by detecting the agent unit.

[0108] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0109] Based on the same inventive concept, the embodiment of the present application also provides a distributed computing model aggregation device for implementing the above-mentioned distributed computing model aggregation method. The implementation scheme for solving the problem provided by the device is similar to the implementation scheme recorded in the above-mentioned method, so the specific limitations in one or more embodiments of the distributed computing model aggregation device provided below can refer to the limitations of the distributed computing model aggregation method above, and will not be repeated here.

[0110] In an exemplary embodiment, Figure 4 As shown, a distributed computing power model aggregation device is provided, including: an acquisition module 410, a processing module 420, an adjustment module 430 and an iteration module 440, wherein:

[0111] An acquisition module 410 is used to acquire auxiliary medical image data and an image recognition task of the auxiliary medical image data, and identify each sub-image recognition task through an identification agent unit based on the image recognition task;

[0112] The processing module 420 is used to query the sub-agent unit corresponding to each sub-image recognition task in the agent database based on each sub-image recognition task, and perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each sub-image recognition task based on the execution order of each sub-image recognition task, so as to obtain the sub-initial processing result fed back by each sub-agent unit;

[0113] The adjustment module 430 is used to generate agent feedback information of each sub-agent unit by detecting the agent unit based on the sub-initial processing results fed back by each sub-agent unit and the sub-image recognition tasks corresponding to each sub-agent unit, and to adjust each sub-initial processing result by each sub-agent unit based on the agent feedback information of each sub-agent unit when there is agent feedback information that does not meet the data output condition, so as to obtain a new sub-initial processing result fed back by each sub-agent unit;

[0114] The iteration module 440 is used to replace each of the sub-initial processing results with each of the new sub-initial processing results, and return to execute the sub-initial processing results based on the feedback of each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and generate the agent feedback information step of each of the sub-agent units through the detection agent unit until there is no agent feedback information that does not meet the data output condition, and then based on the new sub-initial processing results fed back by each of the sub-agent units obtained in the last iteration, the target recognition result of the auxiliary medical image data is generated through the detection agent unit.

[0115] Optionally, the acquisition module 410 is specifically configured to:

[0116] Identify the task flow of the image recognition task by identifying the agent unit, and split the task flow into subtask flows of each flow node;

[0117] Identify the task type corresponding to each subtask process, and divide the subtask processes of the same task type into the same task group to obtain an identification task group corresponding to each task type;

[0118] All sub-task processes in each recognition task group and process nodes corresponding to each sub-task process in each recognition task group are used as sub-image recognition tasks.

[0119] Optionally, the processing module 420 is specifically configured to:

[0120] Based on the task type corresponding to each sub-image recognition task, in the agent database, the initial sub-agent unit corresponding to each task type is queried, and for each task type, the current running state of each initial sub-agent unit of the task type and the running cost information of each initial sub-agent unit are collected;

[0121] Based on the current running status of each initial sub-agent unit and the running cost information of each initial sub-agent unit, the target sub-agent unit corresponding to the task type is screened through a resource optimization strategy, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

[0122] Optionally, the processing module 420 is specifically configured to:

[0123] Based on the process nodes corresponding to the subtask processes in each sub-image recognition task, the data recognition order of each subtask process in each sub-image recognition task is identified, and based on the data recognition order of each subtask process in each sub-image recognition task, the interaction order of each sub-agent unit is generated;

[0124] Based on the subtask processes corresponding to each subagent unit and the interaction order of each subagent unit, data recognition is performed on the auxiliary medical imaging data through each subagent unit to obtain the data recognition result of each subtask process, and the data recognition processing results of all subtask processes processed by each subagent unit are used as the sub-initial processing results fed back by each subagent unit.

[0125] Optionally, the adjustment module 430 is specifically configured to:

[0126] For each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit and the task content corresponding to each sub-task process, the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization correction information corresponding to each sub-task process are identified through the detection agent unit;

[0127] Sub-result evaluation information corresponding to all sub-task processes and sub-result optimization correction information corresponding to all sub-task processes are used as agent feedback information of the sub-agent unit.

[0128] Optionally, the adjustment module 430 is specifically configured to:

[0129] For each subagent unit, based on the sub-result optimization correction information corresponding to each subtask process of the subagent unit, a processing correction strategy for each subtask process of the subagent unit is generated;

[0130] Based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction sequence of each sub-agent unit, the auxiliary medical image data is re-recognized through each sub-agent unit to obtain a new data recognition result of each sub-task process;

[0131] The new data recognition results of all sub-task processes of each sub-agent unit are used as the new sub-initial processing results fed back by each sub-agent unit.

[0132] Each module in the above-mentioned distributed computing power model aggregation device can be implemented in whole or in part by software, hardware, or a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0133] In an exemplary embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as shown in FIG. Figure 5 As shown. The computer device includes a processor, a memory, an input / output interface, a communication interface, a display unit and an input device. Among them, the processor, the memory and the input / output interface are connected through a system bus, and the communication interface, the display unit and the input device are connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and the external device. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be realized through WIFI, a mobile cellular network, NFC (near field communication) or other technologies. When the computer program is executed by the processor, a distributed computing power model aggregation method is implemented. The display unit of the computer device is used to form a visually visible picture, which can be a display screen, a projection device or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device shell, or an external keyboard, touchpad or mouse.

[0134] Those skilled in the art will understand that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0135] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of any one of the methods in the first aspect are implemented.

[0136] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of any one of the methods in the first aspect are implemented.

[0137] In one embodiment, a computer program product is provided, comprising a computer program, which, when executed by a processor, implements the steps of any one of the methods in the first aspect.

[0138] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0139] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment method can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data recognition logic device based on quantum computing, etc., but are not limited to this.

[0140] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0141] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A distributed computing power model aggregation method, characterized in that: The method comprises: Acquiring auxiliary medical image data and an image recognition task of the auxiliary medical image data; Identify the task flow of the image recognition task by identifying the agent unit, and split the task flow into subtask flows of each flow node; Identify the task type corresponding to each subtask process, and divide the subtask processes of the same task type into the same task group to obtain an identification task group corresponding to each task type; All subtask processes in each recognition task group and process nodes corresponding to each subtask process in each recognition task group are used as each sub-image recognition task; Based on each of the sub-image recognition tasks, query the sub-agent unit corresponding to each sub-image recognition task in the agent database, and based on the execution order of each of the sub-image recognition tasks, perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each of the sub-image recognition tasks to obtain the sub-initial processing results fed back by each of the sub-agent units; Based on the sub-initial processing results fed back by each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, agent feedback information of each of the sub-agent units is generated by detecting the agent unit, and when there is agent feedback information that does not meet the data output condition, each of the sub-initial processing results is adjusted by each of the sub-agent units based on the agent feedback information of each of the sub-agent units to obtain a new sub-initial processing result fed back by each of the sub-agent units; Each of the new sub-initial processing results replaces each of the sub-initial processing results, and returns to execute the sub-initial processing results based on the feedback of each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and generates the agent feedback information step of each of the sub-agent units through the detection agent unit, until there is no agent feedback information that does not meet the data output condition, and based on the new sub-initial processing results fed back by each of the sub-agent units obtained in the last iteration, generates the target recognition result of the auxiliary medical image data through the detection agent unit.

2. The method according to claim 1, characterized in that The querying of the sub-agent unit corresponding to each sub-image recognition task in the agent database based on each sub-image recognition task includes: Based on the task type corresponding to each sub-image recognition task, in the agent database, the initial sub-agent unit corresponding to each task type is queried, and for each task type, the current running state of each initial sub-agent unit of the task type and the running cost information of each initial sub-agent unit are collected; Based on the current running status of each initial sub-agent unit and the running cost information of each initial sub-agent unit, the target sub-agent unit corresponding to the task type is screened through a resource optimization strategy, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

3. The method according to claim 1, characterized in that Based on the execution order of each of the sub-image recognition tasks, the auxiliary medical image data is subjected to data recognition by the sub-agent units corresponding to each of the sub-image recognition tasks, and the sub-initial processing results fed back by each of the sub-agent units are obtained, including: Based on the process nodes corresponding to the subtask processes in each sub-image recognition task, the data recognition order of each subtask process in each sub-image recognition task is identified, and based on the data recognition order of each subtask process in each sub-image recognition task, the interaction order of each sub-agent unit is generated; Based on the subtask processes corresponding to each subagent unit and the interaction order of each subagent unit, data recognition is performed on the auxiliary medical imaging data through each subagent unit to obtain the data recognition result of each subtask process, and the data recognition processing results of all subtask processes processed by each subagent unit are used as the sub-initial processing results fed back by each subagent unit.

4. The method according to claim 3, characterized in that The generating of agent feedback information of each of the sub-agent units by detecting the agent unit based on the sub-initial processing results fed back by each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units includes: For each sub-agent unit, based on the data recognition results of each sub-task process fed back by the sub-agent unit and the task content corresponding to each sub-task process, the sub-result evaluation information corresponding to each sub-task process and the sub-result optimization correction information corresponding to each sub-task process are identified through the detection agent unit; Sub-result evaluation information corresponding to all sub-task processes and sub-result optimization correction information corresponding to all sub-task processes are used as agent feedback information of the sub-agent unit.

5. The method according to claim 4, characterized in that The adjusting each of the sub-initial processing results based on the proxy feedback information of each of the sub-agent units through each of the sub-agent units to obtain a new sub-initial processing result fed back by each of the sub-agent units includes: For each subagent unit, based on the sub-result optimization correction information corresponding to each subtask process of the subagent unit, a processing correction strategy for each subtask process of the subagent unit is generated; Based on the processing correction strategy of each sub-agent unit for each sub-task process and the interaction sequence of each sub-agent unit, the auxiliary medical image data is re-recognized through each sub-agent unit to obtain a new data recognition result of each sub-task process; The new data recognition results of all sub-task processes of each sub-agent unit are used as the new sub-initial processing results fed back by each sub-agent unit.

6. A distributed computing power model aggregation device, characterized in that: The device comprises: An acquisition module is used to acquire auxiliary medical image data and an image recognition task of the auxiliary medical image data; identify the task flow of the image recognition task through an identification agent unit, and split the task flow into subtask flows of each flow node; identify the task type corresponding to each subtask flow, and divide the subtask flows of the same task type into the same task group to obtain an identification task group corresponding to each task type; use all subtask flows in each identification task group and the flow nodes corresponding to each subtask flow in each identification task group as each sub-image recognition task; A processing module, configured to query the sub-agent unit corresponding to each sub-image recognition task in the agent database based on each of the sub-image recognition tasks, and based on the execution order of each of the sub-image recognition tasks, perform data recognition on the auxiliary medical image data through the sub-agent unit corresponding to each of the sub-image recognition tasks to obtain a sub-initial processing result fed back by each of the sub-agent units; an adjustment module, configured to generate agent feedback information of each of the sub-agent units by detecting the agent units based on the sub-initial processing results fed back by each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and to adjust each of the sub-initial processing results by each of the sub-agent units based on the agent feedback information of each of the sub-agent units when there is agent feedback information that does not meet the data output condition, so as to obtain a new sub-initial processing result fed back by each of the sub-agent units; An iterative module is used to replace each of the sub-initial processing results with each of the new sub-initial processing results, and return to execute the sub-initial processing results based on the feedback of each of the sub-agent units and the sub-image recognition tasks corresponding to each of the sub-agent units, and generate the agent feedback information step of each of the sub-agent units through the detection agent unit until there is no agent feedback information that does not meet the data output condition, and then based on the new sub-initial processing results fed back by each of the sub-agent units obtained in the last iteration, the target recognition result of the auxiliary medical image data is generated through the detection agent unit.

7. The device according to claim 6, characterized in that The processing module is specifically used for: Based on the task type corresponding to each sub-image recognition task, in the agent database, the initial sub-agent unit corresponding to each task type is queried, and for each task type, the current running state of each initial sub-agent unit of the task type and the running cost information of each initial sub-agent unit are collected; Based on the current running status of each initial sub-agent unit and the running cost information of each initial sub-agent unit, the target sub-agent unit corresponding to the task type is screened through a resource optimization strategy, and the target sub-agent unit is used as the sub-agent unit of the sub-image recognition task corresponding to the task type.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 5 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images