On-satellite load balancing scheduling system based on busy degree and algorithm model state

Abstract

The invention discloses an on-satellite load balancing scheduling system based on busy degree and an algorithm model state, and aims at carrying out load balancing scheduling on two or more than two AI modules. The busy degree-based AI module load balancing method specifically comprises a processing device for realizing load balancing, a processing device for executing a calculation task, a memory database redis for performing remote control instruction and telemetering data interaction, a relational database management system MySQL and a computer program. Scheduling load balancing processing is carried out on two or more than two AI modules, in the whole on-satellite calculation analysis processing process, almost no interference exists on the ground, calculation tasks are autonomously scheduled and distributed in the whole process, on-satellite calculation power resources are reasonably utilized, the autonomously scheduling function has a start-stop control position, and the system can be widely applied to the field of on-satellite calculation analysis processing. And the calculation task can also be allocated to the specific AI module from the ground, so that the satellite has the capability of automatically processing and scheduling the AI module to participate in the calculation task in real time.

Description

Technical Field

[0001] The present invention relates to the technical field of load balancing scheduling, and specifically to an on-satellite load balancing scheduling system based on busyness and algorithm model status. Background Art

[0002] Traditional remote sensing satellites do not have the conditions for on-satellite image processing and analysis. They only have the ability to control the camera payload to take pictures and store them, and then transmit the images to the ground through the data transmission channel for analysis and processing. With the extensive use of AI technology, traditional remote sensing satellites have also begun to carry AI processing modules and have the ability to process and analyze images in real time on the satellite, becoming intelligent remote sensing satellites different from traditional ones. The computing power of a single AI processing module is limited. Once the number of images to be processed increases significantly, its processing efficiency will decline. To improve the computing power of the entire satellite, multiple AI processing modules need to be carried on the intelligent satellite for parallel analysis and calculation. This brings the load balancing problem of how multiple modules cooperate with each other, which can not only meet the computing power requirements but also control the power consumption without wasting resources.

[0003] Currently, for load balancing between AI modules, ground commands can be sent for control. Through ground commands, single or multiple AI modules can be controlled to participate in image processing and analysis. However, controlling by sending ground commands requires real-time ground intervention in the satellite. Once the satellite is in overseas airspace, it is difficult for the ground to intervene and control the AI modules in real time. To enable the satellite to have the ability to process and schedule AI modules to participate in computing tasks autonomously in real time, this scheduling method is proposed. Summary of the Invention

[0004] The present invention provides an on-satellite load balancing scheduling system based on busyness and algorithm model status, which can effectively solve the problem in the above background art that for load balancing between AI modules currently, ground commands can be sent for control, and single or multiple AI modules can be controlled by ground commands to participate in image processing and analysis, but controlling by sending ground commands requires real-time ground intervention in the satellite. Once the satellite is in overseas airspace, it is difficult for the ground to intervene and control the AI modules.

[0005] To achieve the above object, the present invention provides the following technical solution: An on-satellite load balancing scheduling system based on busyness and algorithm model status, which performs load balancing scheduling for two or more AI modules carried. Specifically, it includes a processing device for realizing load balancing, a processing device for executing computing tasks, an in-memory database redis for remote control command and telemetry data interaction, a relational database management system MySQL, and a computer program.

[0006] According to the above technical solution, the processing device for implementing load balancing includes a processor PS and a memory M2. PS serves as the main control unit, responsible for monitoring the running status of each AI module, collecting the telemetry information of the AI module. PS also receives the remote control instructions uplinked from the ground and distributes them to each AI module;

[0007] PS is connected to the camera, adjusts the camera parameters, controls the camera to execute the photographing instruction, receives the image data returned by the camera, and stores the image data in the storage disk for backup. PS is also responsible for distributing the image data in the storage disk, distributing the image data to each AI module, and receiving the processing results of the AI module, storing the results returned by the AI module in the storage disk for downlink to the ground.

[0008] According to the above technical solution, the processing device for executing computing tasks includes 4 AI modules. The algorithms and hardware configurations of the 4 AI modules are completely the same. The AI module receives the image data, then performs recognition, segmentation, and compression processing on the image data, and immediately returns the calculation result to PS after the calculation is completed, and PS receives and stores it in the storage disk.

[0009] According to the above technical solution, the in-memory database redis for remote control instruction and telemetry data interaction is a key-value database. Redis stores data in memory, so the read and write operations are very fast. Redis supports the publish and subscribe mode, allowing messages to be transmitted in real time between different clients. The desktop can also connect to the redis server through TinyRDM to view the telemetry information and remote control instructions in real time;

[0010] PS will run a redis server, and the AI module accesses as a client. After PS receives the ground remote control instruction, it immediately publishes it through redis. The AI module subscribes to the redis service. Once PS publishes the instruction, the AI module can quickly receive and execute it. The telemetry information of the AI module is also written into redis regularly. The PS side queries the keywords regularly to obtain the telemetry information and then sends it to the ground operation control system.

[0011] According to the above technical solution, the relational database management system MySQL provides intuitive command-line and graphical interface tools, making the management and operation of the database simple and easy to use. It is also possible to connect to mysql using Navicat Premium for easy debugging and querying. PS runs the mysql service to maintain image and calcTask;

[0012] When the PS receives the image data, it is necessary to write the image data into the memory for storage, and at the same time update the file information to the image table. According to the actual task requirements, the PS extracts the file information from the image table and updates it to the calcTask table to execute the distributed image task;

[0013] The computer program runs on the PS and monitors the redis server, mysql system, and each process.

[0014] According to the above technical solution, the load balancing scheduling system specifically includes a camera module, a memory, an AI module, a PS module, and a ground operation management system;

[0015] The camera module provides a data source, the memory is used to store data, the AI module is used to execute calculation and analysis tasks, the PS module runs a software scheduling system to execute the scheduling scheme, and the ground operation management system sends remote control instructions and displays telemetry data on the ground;

[0016] In the load balancing scheduling system, there are also a storage process, an image distribution process, a scheduling process, and an AI module busy state detection process;

[0017] The storage process is responsible for collecting image resources, the image distribution process is responsible for transmitting the image file to the AI module, and the scheduling process, redis, and MySQL are at the core. Redis collects telemetry data and publishes remote control instructions;

[0018] The scheduling process is connected to redis to indirectly know the busy state of all AI modules. The scheduling process is connected to MySQL to indirectly know all image file resources, so that the scheduling process can adjust the allocation of image files in a timely manner according to the busy state of the AI module;

[0019] Moreover, in the load balancing scheduling system, the camera generates image data and transmits it to the PS. The PS stores the image in the memory, reads the image data from the memory, distributes it to the AI module, and the AI module returns the result to the PS after calculation. The PS stores the calculation result in the memory.

[0020] According to the above technical solution, the storage process specifically includes the following process:

[0021] S101. After the storage process receives the image data, it parses the required file information from the image data, especially the spectral band channel and the unique code reserve1;

[0022] S102. The image data is stored in the memory;

[0023] S103. The file information is updated to MySQL.

[0024] According to the above technical solution, the scheduling process specifically includes the following steps:

[0025] S201. The ground operation and management system sends an uplink remote control instruction to enable autonomous scheduling and execute Algorithm X;

[0026] S202. The scheduling process obtains instructions from the redis server through the subscription function, and starts polling the busy status of the AI module from redis. If the first AI module is idle, it adds it to the computing task. If the first module is busy, it pauses its task, then queries the busy status of the second AI module. If it is idle, it adds it to the computing task; if it is busy, it pauses, and polls in this way;

[0027] S203. After the scheduling process finds an idle AI module, it publishes an instruction through redis to notify the idle AI module to execute Algorithm X;

[0028] S204. After the idle AI module obtains its own algorithm instruction from redis through the subscription function, it starts to execute the algorithm;

[0029] S205. The AI module reports its own busy status to the redis service in real time;

[0030] S206. The scheduling process obtains the image file information to be calculated from the image file information table in MySQL;

[0031] S207. The scheduling process adds the image file information obtained in S206, the idle AI module identifier and the algorithm identifier obtained in S202, and then writes them into the computing task table in MySQL, indicating that the image file is assigned to the AI module for calculation;

[0032] The scheduling process loops through the steps from S202 to S207 to complete the allocation of the entire computing task.

[0033] According to the above technical solution, the image distribution process specifically includes the following steps:

[0034] S301. The image distribution process obtains the tasks that have not participated in the calculation from the computing task table in MySQL;

[0035] S302. The image distribution process parses the task, obtains the image file information and the AI module identifier, and reads the image file data from the memory;

[0036] S303. According to the AI module identifier obtained in S302, the image data in S302 is sent to the AI module;

[0037] S304. The AI module analyzes and processes the image and returns the calculation result to the image distribution process;

[0038] S305. The image distribution process stores the calculation result in the memory;

[0039] S306. Set the calculation task as calculated and update it to the calculation task table in MySQL.

[0040] According to the above technical solution, the specific process of the busy state detection process of the AI module is as follows:

[0041] S401. The AI module subscribes to the redis service to obtain that it has been added to the calculation task;

[0042] S402. The AI module prepares to receive images and starts to receive image data;

[0043] S403. After the AI module performs calculation and analysis processing on the image, it returns the calculation result to the image distribution process;

[0044] S404. The AI module detects its remaining memory capacity. When the remaining memory capacity is less than the set threshold, it is determined to be busy;

[0045] If it is greater than the set threshold, it is determined to be idle, and then it reports the busy degree detection result and other telemetry information to redis.

[0046] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0047] 1. Based on the load balancing method of the AI module based on the busy degree, the present invention realizes the scheduling load balancing processing of two or more AI modules. During the entire on-orbit calculation and analysis process, there is almost no ground intervention, and the scheduling is autonomous throughout the process, allocating calculation tasks, reasonably utilizing the on-orbit computing resources, and the autonomous scheduling function has a start / stop control bit, and the ground can also allocate calculation tasks to specific AI modules, so that the satellite has the ability to process in real time and autonomously and schedule the AI module to participate in the calculation task.

[0048] 2. The redis and mysql are used to respectively control the AI module resources and image file resources, and the scheduling process is responsible for combining and allocating the two. The computer program running on the PS monitors the redis service, mysql service and each process. When the redis or mysql service or other abnormal exits, the daemon function will wake it up again, and the processes connected to redis and mysql all have the function of reconnecting after disconnection. Once the redis service and mysql service are woken up again, the scheduling process can also immediately connect, obtain the latest resources, and continue to allocate to the AI module. Description of the Drawings

[0049] The accompanying drawings are used to provide a further understanding of the present invention and form a part of the specification. They are used in conjunction with the embodiments of the present invention to explain the present invention, but do not constitute a limitation to the present invention.

[0050] In the accompanying drawings:

[0051] Figure 1 is the overall block diagram of the load scheduling system of the present invention;

[0052] Figure 2 is the flow diagram of the flow of image data and result data of the present invention;

[0053] Figure 3 is the flow diagram of the data flow in the process of storing images of the present invention;

[0054] Figure 4 is the program flow diagram of the process of storing images of the present invention;

[0055] Figure 5 is the flow diagram of the data flow of the scheduling process instruction of the present invention;

[0056] Figure 6 is the program flow diagram of the scheduling process of the present invention;

[0057] Figure 7 is the flow diagram of the data flow of the image distribution process of the present invention;

[0058] Figure 8 is the program flow diagram of the image distribution process of the present invention;

[0059] Figure 9 is the flow diagram of the data flow of AI of the present invention;

[0060] Figure 10 is the detection flow diagram of the busy state of the AI module of the present invention. Specific Embodiments

[0061] The following describes the preferred embodiments of the present invention with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention and are not used to limit the present invention.

[0062] Embodiment: The present invention provides a technical solution, a load balancing scheduling system based on the busy degree and the algorithm model state in space, which performs load balancing scheduling for two or more AI modules carried, and specifically includes a processing device for realizing load balancing, a processing device for executing computing tasks, a memory database redis for remote control instruction and telemetry data interaction, a relational database management system MySQL, and a computer program.

[0063] Based on the above technical solution, the processing device for implementing load balancing includes a processor PS and a memory M2. PS serves as the main control unit, responsible for monitoring the operating status of each AI module, collecting the telemetry information of the AI module. At the same time, PS receives the remote control instructions uplinked from the ground and distributes them to each AI module;

[0064] PS is connected to the camera, adjusts the camera parameters, controls the camera to execute the photographing instruction, receives the image data returned by the camera, and stores the image data in the storage disk for backup. PS is also responsible for distributing the image data in the storage disk, distributing the image data to each AI module, and receiving the processing results of the AI module, storing the results returned by the AI module in the storage disk for downlinking to the ground.

[0065] Based on the above technical solution, the processing device for executing computing tasks includes 4 AI modules. The algorithms and hardware configurations of the 4 AI modules are exactly the same. The AI module receives the image data, then performs recognition, segmentation, and compression processing on the image data, and immediately returns the calculation result to PS after the calculation is completed, and PS receives and stores it in the storage disk.

[0066] Based on the above technical solution, the in-memory database redis for remote control instruction and telemetry data interaction is a key-value database. Redis stores data in memory, so the read and write operations are very fast. Redis supports the publish and subscribe mode, allowing messages to be transmitted in real time between different clients. The desktop can also connect to the redis server through TinyRDM to view the telemetry information and remote control instructions in real time;

[0067] PS will run a redis server, and the AI module accesses as a client. After PS receives the ground remote control instruction, it immediately publishes it through redis. The AI module subscribes to the redis service. Once PS publishes the instruction, the AI module can quickly receive and execute it. The telemetry information of the AI module is also written into redis regularly. The PS side queries the keywords regularly to obtain the telemetry information and then distributes it to the ground operation and control system.

[0068] Based on the above technical solution, the relational database management system MySQL provides intuitive command-line and graphical interface tools, making the management and operation of the database simple and easy to use. It is also possible to connect to mysql using Navicat Premium for easy debugging and query. At the same time, MySQL is open source and can be used for free, reducing the software development and maintenance costs. PS runs the mysql service to maintain image and calcTask. mage is the image file information table, and calcTask is the calculation task table. The image file information table is shown in Table 1 below, and the calculation task table is shown in Table

[0069] Table 2 as follows:

[0070]

[0071]

[0072] Table 1 - Image File Information Table

[0073]

[0074] Table 2 - Calculation Task Table

[0075] When the PS receives the image data, it needs to write the image data into the memory for storage, and at the same time update the file information to the image table. According to the actual task requirements, the PS extracts the file information from the image table, updates it to the calcTask table, and executes the distribution of the image task;

[0076] The computer program runs on the PS and monitors the redis server, the mysql system, and each process.

[0077] As Figure 1-2 shown, based on the above technical solution, the load balancing and scheduling system specifically includes a camera module, a memory, an AI module, a PS module, and a ground operation and management system;

[0078] The camera module provides the data source, the memory is used to store the data, the AI module is used to execute the calculation and analysis tasks, the PS module runs the software scheduling system and executes the scheduling plan, while the ground operation and management system sends remote control instructions and displays the telemetry data on the ground;

[0079] In the load balancing and scheduling system, there are also a storage process, an image distribution process, a scheduling process, and an AI module busy state detection process;

[0080] The storage process is responsible for collecting image resources, the image distribution process is responsible for transmitting the image files to the AI module, and the scheduling process, redis, and MySQL are at the core. Redis collects telemetry data and publishes remote control instructions;

[0081] The scheduling process is connected to redis and indirectly grasps the busy state of all AI modules. The scheduling process is connected to MySQL and indirectly grasps all image file resources, so that the scheduling process can adjust the allocation of image files in a timely manner according to the busy state of the AI module;

[0082] Moreover, in the load balancing and scheduling system, the camera generates image data and transmits it to the PS. The PS stores the image in the memory, reads the image data from the memory, distributes it to the AI module, and the AI module returns the result to the PS after calculation. The PS stores the calculation result in the memory.

[0083] As Figure 3-4 shown, based on the above technical solution, the storage process specifically includes the following processes:

[0084] S101. After the storage process receives the image data, parse out the file information required in Table 1 from the image data, especially the spectral band channel and the unique code reserve1;

[0085] S102. Store the image data in the memory;

[0086] S103. Update the file information to MySQL.

[0087] As Figure 5-6 shown, based on the above technical solution, the scheduling process specifically includes the following processes:

[0088] S201. The ground operation management system sends an uplink remote control instruction to start autonomous scheduling and execute Algorithm X;

[0089] S202. The scheduling process obtains instructions from the redis server through the subscription function, and starts polling the busy status of the AI module from redis. If the first AI module is idle, add it to the computing task. If the first module is busy, suspend its task, then query the busy status of the second AI module. If it is idle, add it to the computing task. If it is busy, suspend it, and poll in this way;

[0090] S203. After the scheduling process finds an idle AI module, send an instruction through redis to notify the idle AI module to execute Algorithm X;

[0091] S204. After the idle AI module obtains its own algorithm instruction from redis through the subscription function, start executing the algorithm;

[0092] S205. The AI module reports its own busy status to the redis service in real time;

[0093] S206. The scheduling process obtains the image file information to be calculated from the image file information table in MySQL;

[0094] S207. The scheduling process adds the image file information obtained in S206, the idle AI module identifier and the algorithm identifier obtained in S202, and then writes them into the computing task table in MySQL, indicating that the image file is assigned to the AI module to participate in the calculation;

[0095] The scheduling process loops through the steps of S202 to S207 to complete the allocation of the entire computing task.

[0096] As Figure 7-8As shown in the figure, based on the above technical solution, the image distribution process specifically includes the following steps:

[0097] S301. The image distribution process obtains tasks that have not participated in the calculation from the calculation task table in MySQL;

[0098] S302. The image distribution process parses the tasks, obtains the image file information and the AI module identifier, and reads the image file data from the memory;

[0099] S303. According to the AI module identifier obtained in S302, the image data in S302 is sent to the AI module;

[0100] S304. The AI module analyzes and processes the image, and returns the calculation result to the image distribution process;

[0101] S305. The image distribution process stores the calculation result in the memory;

[0102] S306. Set the calculation task as calculated and update it to the calculation task table in MySQL.

[0103] As Figure 9-10 shown in the figure, based on the above technical solution, the AI module busy state detection process specifically includes the following steps:

[0104] S401. The AI module subscribes to the redis service to obtain that it has been added to the calculation task;

[0105] S402. The AI module prepares to receive images and starts receiving image data;

[0106] S403. After the AI module analyzes and processes the image calculation, it returns the calculation result to the image distribution process;

[0107] S404. The AI module detects its remaining memory capacity. When the remaining memory capacity is less than the set threshold, it is determined to be busy;

[0108] If it is greater than the set threshold, it is determined to be idle, and then it reports the busy degree detection result and other telemetry information to redis.

[0109] Finally, it should be noted that the above are only the preferred examples of the present invention and are not used to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. The load balancing scheduling system on the satellite based on busyness and algorithm model status is characterized by: Load balancing scheduling is performed for systems equipped with two or more AI modules, specifically including a processing device for implementing load balancing, a processing device for executing computing tasks, an in-memory database redis for interacting with remote control commands and telemetry data, a relational database management system MySQL, and a computer program.

2. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 1, characterized in that: The processing device for realizing load balancing includes a processor PS and a memory M2. The PS, as the main control unit, is responsible for monitoring the operation status of each AI module and collecting the telemetry information of the AI ​​module. The PS also receives remote control commands from the ground and distributes them to each AI module. PS is connected to the camera, adjusts the camera parameters, controls the camera to execute the photo command, receives the image data returned by the camera, and stores the image data in the storage disk for backup. PS is also responsible for distributing the image data in the storage disk, distributing the image data to each AI module, receiving the processing results of the AI ​​module, and storing the results returned by the AI ​​module in the storage disk for transmission to the ground.

3. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 1, characterized in that: The processing device used to perform computing tasks includes 4 AI modules. The algorithms and hardware configurations run by the 4 AI modules are completely consistent. The AI ​​modules receive image data, and then identify, segment and compress the image data. After the calculation is completed, the calculation results are immediately returned to the PS, which receives them and stores them in the storage disk.

4. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 1, characterized in that: Redis, an in-memory database used for interaction between remote control commands and telemetry data, is a key-value database. Redis stores data in memory, so read and write operations are very fast. Redis supports publish and subscribe modes, allowing messages to be transmitted in real time between different clients. The desktop can also connect to the redis server through TinyRDM to view telemetry information and remote control commands in real time. The PS will run a redis server, and the AI ​​module will access it as a client. After the PS receives the ground remote control command, it will immediately publish it through redis. The AI ​​module subscribes to the redis service. Once the PS publishes the command, the AI ​​module can quickly receive and execute it. The telemetry information of the AI ​​module is also written to redis regularly. The PS end will query the keywords regularly to obtain the telemetry information, and then send it to the ground operation and control system.

5. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 1, characterized in that: The relational database management system MySQL provides intuitive command line and graphical interface tools, making database management and operation simple and easy to use. You can also use Navicat Premium to connect to MySQL for debugging queries, PS to run MySQL services, and maintain image and calcTask; When the PS receives the image data, it needs to write the image data into the memory for storage and update the file information into the image table. The PS extracts the file information from the image table according to the actual task requirements, updates it into the calcTask table, and executes the image distribution task. The computer program runs on the PS and performs daemon monitoring on the redis server, the mysql system and each process.

6. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 1 is characterized in that: The load balancing scheduling system specifically includes a camera module, a memory, an AI module, a PS module and a ground transportation management system; The camera module provides the data source, the memory is used to save the data, the AI ​​module is used to perform the calculation and analysis tasks, the PS module runs the software scheduling system and executes the scheduling plan, and the ground operation management system sends the remote control instructions and displays the telemetry data on the ground; The load balancing scheduling system also includes a storage process, an image distribution process, a scheduling process, and an AI module busy state detection process; The storage process is responsible for collecting image resources, the image distribution process is responsible for transmitting image files to the AI ​​module, and the scheduling process, redis and MySQL are at the core. Redis collects telemetry data and issues remote control commands. The scheduling process is connected to redis to indirectly grasp the busy status of all AI modules. The scheduling process is connected to MySQL to indirectly grasp all image file resources, so that the scheduling process can adjust the allocation of image files in time according to the busy status of the AI ​​module; Moreover, in the load balancing scheduling system, the camera generates image data and transmits it to the PS, which stores the image in the memory. The PS reads the image data from the memory and distributes it to the AI ​​module. The AI ​​module returns the result to the PS after calculation, and the PS stores the calculation result in the memory.

7. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 6 is characterized in that: The storage process specifically includes the following process: S101, after receiving the image data, the storage process parses the image data to obtain the required file information, especially the spectrum channel and the unique code reserve1; S102, the image data is stored in a memory; S103. The file information is updated in MySQL.

8. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 6, characterized in that: The scheduling process specifically includes the following process: S201, the ground operation management system sends a remote control command to start autonomous dispatch and execute algorithm X; S202, the scheduling process obtains instructions from the redis server through the subscription function, and starts to poll the busy status of the AI ​​modules from redis. If the first AI module is idle, it is added to the calculation task. If the first module is busy, its task is suspended, and then the busy status of the second AI module is queried. If it is idle, it is added to the calculation task. If it is busy, it is suspended, and the polling is repeated in this way; S203, after the scheduling process finds an idle AI module, it issues a command through redis to notify the idle AI module to execute algorithm X; S204, the idle AI module obtains its own algorithm instructions from redis through the subscription function and starts to execute the algorithm; S205, the AI ​​module reports its busy status to the redis service in real time; S206, the scheduling process obtains the image file information to be calculated from the image file information table of MySQL; S207, the scheduling process adds the image file information obtained in S206, the idle AI module identifier and the algorithm identifier obtained in S202, and then writes them into the MySQL computing task table, indicating that the image file is assigned to the AI ​​module to participate in the computing; The scheduling process cyclically executes steps S202 to S207 to complete the allocation of the entire computing task.

9. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 6, characterized in that: The image distribution process specifically includes the following steps: S301, the image distribution process obtains the tasks that have not yet participated in the calculation from the calculation task table of MySQL; S302, image distribution process parsing task, obtaining image file information and AI module identification, and reading image file data from the memory; S303, sending the image data of S302 to the AI ​​module according to the AI ​​module identifier obtained in S302; S304, the AI ​​module analyzes and processes the image and returns the calculation result to the image distribution process; S305, the image distribution process stores the calculation result into the memory; S306: Set the calculation task as calculated, and update it to the calculation task table of MySQL.

10. The on-board load balancing scheduling system based on busyness and algorithm model status according to claim 6, characterized in that: The AI ​​module busy state detection process specifically includes the following process: S401, the AI ​​module obtains that it has been added to the computing task by subscribing to the redis service; S402, the AI ​​module is ready to receive images and starts receiving image data; S403, after the AI ​​module calculates and analyzes the image, it returns the calculation result to the image distribution process; S404, the AI ​​module detects its own remaining memory capacity, and when the remaining memory capacity is less than a set threshold, it is determined to be busy; If it is greater than the set threshold, it is judged as idle, and then the busyness detection result and other telemetry information are reported to redis.

Images