Unmanned platform multimode remote sensing and multi-processing task autonomous optimization scheduling method

Abstract

The invention discloses a multi-mode remote sensing and multi-processing task autonomous optimization scheduling method for an unmanned platform, and the method comprises the steps: collecting multi-mode remote sensing data in a disaster scene through the unmanned platform, and storing the multi-mode remote sensing data in edge equipment; performing priority ranking on the multi-modal remote sensing data, generating a task and defining the emergency degree of the task; performing priority definition and resource allocation on the tasks based on micro-service and containerization technologies; a container arrangement framework is adopted to manage an edge cluster of the unmanned platform, and dynamic scheduling and load balancing of the tasks are achieved; parallel processing of high-priority tasks is realized through a copy mechanism; and data sharing and aggregation of processing results are carried out by using a network file system, and are transmitted to a ground station to support post-disaster rescue decisions. According to the scheduling method, the total response time is reduced by 35%, and the task processing speed is increased by 146%; the resource utilization rate is improved by about 120%, and flexible expansion of the model is supported so as to adapt to diversified emergency scenes.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of remote sensing task scheduling, in particular to an unmanned platform multi-mode remote sensing and multi-processing task autonomous optimization scheduling method based on microservices and containerization. BACKGROUND

[0002] Natural disasters have far-reaching and complex characteristics in terms of damage to human society and the natural environment, with impacts far beyond direct casualties and property losses. Disasters such as earthquakes, floods, hurricanes, and typhoons not only destroy buildings, infrastructure, and farmland, but also can trigger serious secondary disasters, such as tsunamis triggered by earthquakes, climate anomalies caused by volcanic eruptions, or water supply system failures caused by floods. The chain effect of these disasters often plunges disaster-stricken areas into a situation of food shortages, disease outbreaks, and resource scarcity, significantly prolonging the time and difficulty of post-disaster recovery. In addition, the long-term impact of natural disasters on the social economy cannot be ignored, such as traffic disruptions, production stagnation, and ecological system destruction, further exacerbating the vulnerability of disaster-stricken areas. Therefore, rapid and efficient disaster emergency response has become a key link in reducing life and property losses and promoting post-disaster reconstruction.

[0003] However, traditional disaster response methods have many limitations in practical application. Ground rescue teams can provide direct assistance, but their mobility and coverage are limited by terrain, traffic conditions, and disaster size, making it difficult to fully grasp the disaster situation in a short time. On the other hand, large-scale remote sensing technology (such as satellites and manned aircraft) can cover a wide area, but due to low resolution, long data collection period, and delayed response time, it cannot meet the needs of real-time disaster assessment and emergency decision-making. In addition, the application of single-mode remote sensing data cannot provide multi-dimensional disaster information, for example, visible light imaging alone may not work effectively at night or in bad weather conditions. These defects make it difficult for traditional methods to achieve efficient and accurate emergency support in complex and variable disaster scenarios, and there is an urgent need for a more flexible and real-time technology to make up for the shortcomings. SUMMARY

[0004] The purpose of the present application is to solve the problems existing in the prior art, and to provide an unmanned platform multi-mode remote sensing and multi-processing task autonomous optimization scheduling method.

[0005] To achieve the above-mentioned purpose, the technical scheme adopted by the present application is: An unmanned platform multi-mode remote sensing and multi-processing task autonomous optimization scheduling method, the scheduling method comprising the following steps: Using the collection equipment carried by the unmanned platform, multi-modal remote sensing data is collected in the disaster scene, and the multi-modal remote sensing data is stored in the edge device; prioritizing the multi-modal remote sensing data, generating tasks and defining the urgency of the tasks; prioritizing the tasks and allocating resources based on microservices and containerization technology; managing the edge cluster of the unmanned platform using a container orchestration framework to achieve dynamic scheduling and load balancing of the tasks; achieving parallel processing of high-priority tasks through a replica mechanism; sharing data and aggregating processing results using a network file system and transmitting them to the ground station to support post-disaster rescue decision-making.

[0006] This scheduling method integrates multi-modal data fusion technologies such as hyperspectral imaging and low-light infrared imaging to construct an all-weather and comprehensive disaster view, significantly improving the comprehensiveness and accuracy of disaster assessment. To address the challenges of limited edge computing resources and concurrent multi-task processing, this strategy uses microservices architecture and containerization technology to optimize resource allocation through dynamic priority sorting and task replica mechanism to ensure rapid response of high-priority tasks. This method not only significantly reduces task response time, but also improves single-task processing efficiency and overall resource utilization, providing a practical solution for efficient application of unmanned aerial vehicles in disaster emergency response and promoting the intelligent and real-time development of disaster management technology.

[0007] Further, the acquisition device includes at least a hyperspectral camera, a low-light infrared camera, and a visible light camera, and the multi-modal remote sensing data includes at least acquired hyperspectral video data, low-light infrared data, and visible light data.

[0008] Further, the tasks include at least hyperspectral video tracking, low-light infrared video tracking, hyperspectral feature classification, and single-time-phase damage assessment. After the tasks are generated, the multi-modal remote sensing data and the information of the tasks are structured and packaged by the preprocessing module of the edge device, and transmitted to the subsequent task priority definition and scheduling stage.

[0009] Further, the implementation of priority definition and resource allocation is as follows: The tasks are prioritized according to the urgency of multi-modal remote sensing data acquisition and processing in the disaster scene. The tasks are then decomposed into independent schedulable microservice units using microservices architecture and packaged into containers through containerization technology to support rapid deployment and migration of tasks. Based on task priority and real-time computing requirements, the computing resources of the edge cluster are dynamically allocated to ensure rapid response of high-priority tasks.

[0010] Further, the container orchestration framework manages the edge cluster as follows: The automation function of the container orchestration framework is used to realize dynamic scheduling, load balancing and fault recovery of tasks through the scheduler thereof; based on a replica mechanism, the number of replicas is dynamically adjusted according to the urgency and load demand of the tasks, and parallel processing of high-priority tasks is supported; a node allocation filtering chain of the container orchestration framework is used to screen edge nodes meeting the conditions, and the filtering conditions include the memory usage of the node, manually set stains and node affinity; a minimum request priority and a balanced resource allocation strategy are used to select the optimal node through weighted scoring.

[0011] Further, the minimum request priority and the balanced resource allocation are realized in the following manner: The minimum request priority is realized by calculating the available resources of the node, and the node with the most sufficient resources is selected, and the scoring formula of the minimum request priority is as follows: , Wherein, score LRP represents the score of the minimum request priority, C cpu and C men are the total capacity of CPU and memory of the node, and ∑R cpu and ∑R mem represent the total amount of CPU and memory requested by the allocated tasks, and the score range is normalized to 0-10. The balanced resource allocation is realized by calculating the request proportion of each resource type, evaluating the resource utilization balance of the node, defining the resource score of the node as the request resource amount divided by the available resource amount, and calculating the distance of the two resource proportions through the variance algorithm, and the node with the smallest resource proportion difference is selected, and the scoring formula is as follows: , Wherein, score BPR represents the score of the balanced resource allocation, F cpu , F mem and F vol represent the resource request proportion of CPU, memory and storage respectively, and variance is the variance of the three, and the score range is normalized to 0-10.

[0012] Further, the replica mechanism realizes the parallel processing of high-priority tasks in the following manner: The number of replicas is determined according to the task priority and the calculation demand; the input multi-modal remote sensing data is divided according to the number of replicas by using the load balancing function of the container orchestration framework, so as to ensure that each replica processes an independent data subset; the replica instances are scheduled to different edge nodes, and the data stream is stored and accessed through the shared directory of the network file system, so as to maintain data consistency; then, the replicas process the tasks in parallel.

[0013] Further, the data sharing and aggregation method of the processing result is as follows: The results of parallel processing are uploaded to the shared directory of the network file system by each copy mechanism; the master node reads and aggregates the results of parallel processing of all copies from the network file system to obtain aggregated data; the aggregated data is transmitted to the ground station through a high-speed heterogeneous communication link, and the ground station performs real-time comprehensive evaluation based on the transmitted aggregated data to support post-disaster rescue decision-making and real-time guidance.

[0014] Further, the scheduling method further comprises dynamically adjusting the type of the task according to multiple disaster scenarios in emergency operation, and optimizing resource allocation efficiency through a dynamic resource allocation factor, and the calculation formula of the dynamic resource allocation factor is: , wherein, R alloc represents the dynamic resource allocation factor of the task, W task represents the calculation workload of the task, P task represents the priority of the task, C total represents the total calculation capacity of the cluster, and N avail represents the current available node number.

[0015] Compared with the prior art, the beneficial effects of the present application are: 1. Significantly improve the task response efficiency: through the micro-service replication mechanism and dynamic priority sorting, high-priority tasks can be quickly processed in parallel, compared with the baseline method, the overall task response time is reduced by 60%-70%, and the task processing speed is increased by about 146% under the condition that the number of copies is limited to three, effectively meeting the low delay demand in the disaster emergency scenario; 2. Optimize resource utilization: based on the resource scheduling strategy of the Kubernetes framework, the calculation resources of the edge cluster are reasonably allocated, the resource utilization is improved by at least 120%, the waste of resources is avoided, and the overall efficiency of the system is improved; 3. Enhance system adaptability: support flexible extension of the model, which can dynamically adjust the scheduling strategy according to different disaster scenarios and task loads, adapt to various emergency scenarios, and improve the stability and reliability of the unmanned aerial vehicle edge cluster in complex environments; 4. Support real-time decision-making: the processing results and collected data are transmitted to the ground station in real time through a high-speed heterogeneous communication link to provide timely and accurate decision support for post-disaster rescue, significantly reducing the loss of life and property caused by disasters. BRIEF DESCRIPTION OF DRAWINGS

[0016] Figure 1 is a flowchart of the present application, a multi-mode remote sensing and multi-processing task autonomous optimization scheduling method for unmanned platform; Figure 2 is a schematic diagram of the application scenario of the scheduling method of the present application; Figure 3 is a schematic diagram of the overall flow of the scheduling strategy of the present application. Figure 4 The schematic diagram of memory usage comparison between the scheduling method of the present application and the baseline method when the number of copies is 1; Figure 5 The schematic diagram of memory usage comparison between the scheduling method of the present application and the baseline method when the number of copies is 2; Figure 6 The schematic diagram of memory usage comparison between the scheduling method of the present application and the baseline method when the number of copies is 3; DETAILED DESCRIPTION

[0017] The technical solutions of the present application will be described clearly and completely below in combination with the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0018] An unmanned platform multi-mode remote sensing and multi-processing task autonomous optimization scheduling method, the scheduling method comprising the following steps: Using the collection equipment carried by the unmanned platform, multi-modal remote sensing data are collected in a disaster scene, and the multi-modal remote sensing data are stored in an edge device; The multi-modal remote sensing data are prioritized, tasks are generated and the urgency of the tasks is defined; Based on microservices and containerization technology, the tasks are prioritized and resources are allocated; A container orchestration framework is used to manage the edge cluster of the unmanned platform, realizing dynamic scheduling and load balancing of the tasks; Through a replica mechanism, parallel processing of high-priority tasks is realized; A network file system is used for data sharing and aggregation of processing results, and the results are transmitted to the ground station to support post-disaster rescue decision-making.

[0019] The present scheduling method sorts tasks according to their urgency, parallel processes high-priority tasks through a microservice replication mechanism, ensures fast response and improves system performance; uses microservices and containerization technology to optimize task priority and resource allocation, reduces the overall response time by 35%, and improves task processing speed by 146%; optimizes resource allocation to improve resource utilization by about 120%, and supports flexible expansion of models to adapt to diversified emergency scenarios.

[0020] In addition, the multi-modal remote sensing task scheduling strategy is divided into three core stages: task generation and priority definition, replica allocation strategy, and task scheduling and processing results. Its application scenarios include deploying multi-modal cameras and edge processing devices under the constraints of unmanned aerial vehicle payloads, performing data acquisition tasks through aerial observation capabilities, storing real-time data on edge devices to generate tasks, transmitting processing results and acquisition data back to ground stations through high-speed communication links, and supporting real-time evaluation and rescue guidance.

[0021] An application scenario of the scheduling method is shown in Figure 2 The scheduling method is mainly used for real-time processing and transmission of remote sensing data in a space-ground collaborative platform. Specifically, under the constraints of unmanned aerial vehicle payloads, multi-modal cameras and edge processing devices are deployed, and multi-modal data acquisition tasks are performed using the aerial observation capabilities of unmanned aerial vehicles. Real-time data is stored on edge devices for subsequent processing tasks. To maximize the processing capacity of the platform, the data processing board integrates multiple edge processing modules to provide hardware support for multi-task parallel execution. Ideally, tasks are generated based on received data, and real-time processing results and acquisition data are transmitted back to ground stations through high-speed heterogeneous communication links, thereby supporting real-time comprehensive evaluation and post-disaster rescue guidance.

[0022] Further, the unmanned platform is an unmanned aerial vehicle, which uses multi-modal cameras and edge processing devices carried by the unmanned aerial vehicle to achieve multi-source data acquisition and task generation in disaster scenarios, supporting subsequent real-time disaster assessment and emergency decision-making. The acquisition device includes at least a hyperspectral camera, a low-light infrared camera, and a visible light camera carried on the unmanned aerial vehicle, and the multi-modal remote sensing data includes at least acquired hyperspectral video data, low-light infrared data, and visible light data.

[0023] The hyperspectral camera, the low-light infrared camera, and the visible light camera are used to acquire the hyperspectral video data, the low-light infrared data, and the visible light data, respectively. The hyperspectral video data captures the spectral characteristics of ground objects in the disaster area through high-resolution spectral imaging technology, providing rich spatial and spectral information; the low-light infrared data records thermal radiation characteristics under low-illumination or night conditions, which is used to identify hidden targets or heat sources; and the visible light data provides intuitive images of the disaster area, which is used to assist scene understanding and damage assessment. These multi-modal data are acquired in real time during the flight of the unmanned aerial vehicle and temporarily stored in the edge device through the on-board high-speed storage module.

[0024] Further, based on the collected multi-source data, a plurality of processing tasks are further generated in this stage to meet the diversified needs of disaster response. The tasks at least include hyperspectral video tracking, low-light infrared video tracking, hyperspectral feature classification, and single-time-phase damage assessment. After the tasks are generated, the multi-modal remote sensing data and the information of the tasks are structured and packaged by the preprocessing module of the edge device, and are transmitted to the subsequent task priority definition and scheduling stage.

[0025] Specifically, hyperspectral video tracking realizes dynamic positioning and tracking of moving targets (such as rescue vehicles or trapped personnel) in the disaster area by analyzing the spectral characteristics of the targets in the hyperspectral video data; low-light infrared video tracking uses the thermal imaging characteristics of low-light infrared data to continuously monitor and track the target path under low-visibility conditions; hyperspectral feature classification accurately classifies the feature types (such as vegetation, water, buildings, etc.) in the disaster area based on the multi-band information of hyperspectral video data, providing a basis for resource allocation and rescue planning; single-time-phase damage assessment quickly evaluates the damage degree of buildings, infrastructure, etc. at a certain time phase after the disaster by fusing visible light data and hyperspectral video data.

[0026] Further, the implementation method of priority definition and resource allocation is as follows: According to the urgency of each multi-modal remote sensing data collection and processing in the disaster scene, the tasks are prioritized; the tasks are decomposed into independent schedulable micro-service units using the micro-service architecture, and are packaged into containers through containerization technology to support rapid deployment and migration of the tasks; based on the task priority and real-time computing requirements, the computing resources of the edge cluster are dynamically allocated to ensure rapid response of high-priority tasks.

[0027] Further, the method for the container orchestration framework to manage the edge cluster is as follows: Using the automation function of the container orchestration framework, the dynamic scheduling, load balancing, and fault recovery of the tasks are realized through the scheduler thereof; based on the replica mechanism, the number of replicas is dynamically adjusted according to the urgency and load demand of the tasks to support parallel processing of high-priority tasks; the node allocation filtering chain of the container orchestration framework is used to screen the edge nodes that meet the conditions, and the filtering conditions include the memory usage of the node, manually set stains, and node affinity; the minimum request priority and balanced resource allocation strategy are used to select the optimal node through weighted scoring.

[0028] In this embodiment, the container orchestration framework refers to Kubernetes framework (Kubernetes is abbreviated as k8s), and the method is based on task scheduling and dynamic replica adjustment of Kubernetes. Kubernetes is used to manage and orchestrate containers in edge clusters, and its automatic functions are used to realize task scheduling, load balancing and fault recovery. The core functions of Kubernetes include dynamic creation and destruction of Pods, automatic scaling of services and health check of nodes, which provide strong support for disaster emergency tasks. Specifically, the method supports replica mechanism for high-priority tasks in parallel processing, and the number of replicas is dynamically adjusted according to the task urgency and current load. This dynamic adjustment mechanism can be expressed by the following formula: , where C i represents the load of node (i), represents the average load of all nodes, and (N) represents the total number of nodes in the edge cluster.

[0029] The optimization objective aims to minimize the load variance, thereby achieving uniform distribution of resources. Experiments show that this method significantly improves resource utilization, shortens response time and enhances the adaptability of the system in disaster emergency scenarios. In addition, the system further enhances stability through a fault recovery mechanism. When a node fails, Kubernetes automatically migrates tasks to healthy nodes and reallocates resources according to priority. This adaptive ability ensures real-time and reliability of multi-modal data processing.

[0030] In addition to the above functions, the method also supports dynamic priority adjustment. For example, when a new high-urgency task enters the system, the task queue can be reordered by the following priority calculation formula: , where P n represents the priority of task (n), T n represents the urgency of the task, D n represents the data volume of the task, and w1 and w2 are weight parameters; in this way, the system can flexibly adjust the task order according to the disaster changes, for example, a task with small data volume but high urgency may be processed first, thereby improving the real-time decision support capability.

[0031] Copy allocation strategy: In different disaster scenarios, the urgency and demand of multi-modal remote sensing data collection and processing are different. Therefore, this project introduces microservices and containerization technology to fully leverage the advantages of edge clusters and accelerate the processing efficiency of emergency tasks. Specifically, the microservices architecture supports independent and flexible deployment and execution of tasks, while the containerization technology ensures efficient migration and rapid startup of tasks between different nodes, thereby achieving dynamic adjustment of task processing priority and rapid allocation of resources, maximizing the computing power of the cluster. Through this method, the real-time processing needs of multi-modal remote sensing data in disaster emergency scenarios can be effectively met, and the response speed of the system to emergency tasks is improved.

[0032] Combining the least request priority (LRP) and balanced resource allocation (BPR) mechanisms to improve the resource utilization efficiency and system stability of the UAV edge cluster in multi-modal remote sensing task processing. The LRP strategy quantifies the remaining resource capacity of the node based on the CPU core number (capacity) and the number of received task requests (request), selects the node with the most available resources for task allocation, thereby reducing task waiting time and improving scheduling efficiency; the BPR strategy calculates the difference in resource utilization ratio (i.e., the ratio of Pod resource request to node available resource) of CPU, memory and storage using variance algorithm, and selects the node with the smallest resource utilization ratio difference for task scheduling to ensure balanced resource allocation and avoid performance degradation or failure due to resource overload in single node. The two strategies calculate the node score respectively through the following formulas to guide the task scheduling process.

[0033] The least request priority selects the node with the most resources by calculating the available resources of the node, and the score formula of the least request priority is: , where score LRP represents the score of the least request priority, C cpu and C men are the total capacity of CPU and memory of the node, ∑R cpu and ∑R mem represent the total request of CPU and memory of the allocated tasks, and the score range is normalized to 0 to 10; The balanced resource allocation evaluates the resource utilization balance of the node by calculating the request ratio of each resource type, and its resource score is defined as the ratio of the request resource amount of the node to the available resource amount. The distance between the two resource utilization ratios is calculated by the variance algorithm, and the node with the smallest resource utilization ratio difference is selected. Its score formula is: , , 、 , wherein score BPR represents the score of balanced resource allocation, F cpu , F mem and F vol represent the resource request proportion of CPU, memory and storage respectively, variance is the variance of the three, and the score range is normalized to 0-10.

[0034] In implementation, the present application integrates LRP and BPR strategies in the Kubernetes scheduler, and realizes the automation optimization of task allocation through a custom algorithm. The LRP strategy is suitable for high-load scenarios to quickly allocate tasks, and the BPR strategy is suitable for resource-intensive scenarios to maintain balanced allocation, and the two strategies are dynamically switched according to task priority and cluster load. The scheduler calculates the score according to the real-time state of the node, and combines the node affinity rules and the stain tolerance mechanism to optimize the task deployment location. The present application supports adjusting the strategy parameters through a configuration file, such as assigning different weights to CPU and memory score, or extending to network bandwidth and other resource dimensions, to adapt to different hardware configurations and task requirements. Experimental verification shows that the present application significantly reduces task scheduling delay, reduces node load variance, improves resource utilization and system stability, and provides efficient support for multi-modal data processing in disaster emergency scenarios.

[0035] Further, the replica mechanism realizes the parallel processing of high-priority tasks in the following manner: According to the task priority and computing requirements, the number of replicas is determined; the input multi-modal remote sensing data is divided according to the number of replicas by using the load balancing function of the container orchestration framework, ensuring that each replica processes an independent data subset; the replica instances are scheduled to different edge nodes, and the data stream is stored and accessed through the shared directory of the network file system, maintaining data consistency; then the replicas process the tasks in parallel.

[0036] Further, the data sharing and processing result aggregation method is as follows: The results of parallel processing of each replica of the replica mechanism are uploaded to the shared directory of the network file system; the master node reads and aggregates all the parallel processing results of the replicas from the network file system to obtain aggregated data; the aggregated data is transmitted to the ground station through a high-speed heterogeneous communication link, and the ground station performs real-time comprehensive evaluation based on the transmitted aggregated data to support post-disaster rescue decision-making and real-time guidance.

[0037] In the experimental verification process, the effectiveness of the scheduling strategy of the method is fully verified by comparison with the baseline method without using the scheduling strategy. The baseline method deploys the task processing model on a single node, only supports single-instance sequential execution, and the node sequentially completes data collection, preprocessing and analysis after the task is started. The present application supports task parallel processing and resource elastic allocation by introducing the dynamic scheduling strategy of the Kubernetes (k8s) framework. To quantify the scheduling effect, the present application sets three configurations with replica number limits of 1 to 3 for testing. The experimental results show that, compared with the baseline method, after applying the scheduling strategy of the present application, the overall task response time is reduced by about 60% to 75%, and the resource utilization is increased by at least 146%. As shown in Figures 4-6 , the resource utilization comparison under different replica limits intuitively reflects the optimization effect of the scheduling strategy. To further measure the efficiency of task parallel processing, the present application also defines the speedup ratio of task completion time, and the calculation formula is: , In the formula, S represents the speedup ratio, T base represents the task completion time of the baseline method, and T parallel represents the task completion time after parallel scheduling of the present application. The present application supports multi-instance parallel processing to shorten the response time by reasonably setting the number of replicas when the cluster resources are sufficient, while avoiding high replica number that causes instance creation delay and resource congestion, thereby ensuring efficient and stable completion of the task.

[0038] In addition, the present application fully utilizes the independence of the containerized model environment and the elastic management characteristics of the Kubernetes framework to support dynamic adjustment of multi-modal remote sensing task types according to various disaster scenarios in emergency operations, such as switching image analysis or sound wave detection tasks by priority adjustment, and adapting to the payload and power consumption requirements of different unmanned aerial vehicles through expansion and contraction of hardware resources. To optimize resource allocation efficiency, the present application designs a dynamic resource allocation factor, and the calculation formula is: , In the formula, R alloc represents the dynamic resource allocation factor of the task, W task represents the computing workload of the task, P task represents the priority of the task, C total represents the total computing capacity of the cluster, and N avail represents the current available node number.

[0039] Through the formula, the system can dynamically allocate resources according to task requirements and cluster states. This method can flexibly adjust resource allocation to cope with different task loads, ensure stable operation of the system in a resource-limited environment, and significantly improve the utilization efficiency of computing resources, providing reliable technical support for multi-modal data processing in disaster emergency scenarios.

[0040] In specific implementation, the present application decomposes tasks into independent modules through a micro-service architecture, each module running in a containerized environment, and using the Pod management and scheduling function of Kubernetes to realize automatic deployment and execution of tasks. For example, high-priority tasks can be quickly expanded through the replica mechanism, and low-load nodes can be preferentially scheduled through node affinity rules, thereby optimizing the overall processing flow. The present application also supports a fault recovery mechanism, when a node fails, Kubernetes automatically migrates tasks to healthy nodes, ensuring continuous operation of the system. Experimental results show that the resource utilization of this method is improved from 35% of the baseline to 86%, and the data processing delay is reduced by about 70%, significantly improving the adaptability and real-time response capability of the unmanned aerial vehicle edge cluster in complex environments.

[0041] Although embodiments of the present application have been shown and described, it will be understood by those having ordinary skill in the art that various changes, modifications, substitutions and alterations can be made therein without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An unmanned platform multi-mode remote sensing and multi-processing task autonomous optimization scheduling method, characterized in that, The scheduling method comprises the following steps: Using the collection equipment carried by the unmanned platform, multi-modal remote sensing data is collected in a disaster scene, and the multi-modal remote sensing data is stored in an edge device; The multi-modal remote sensing data is prioritized, a task is generated, and the urgency of the task is defined; Based on microservices and containerization technology, the task is prioritized and resources are allocated; A container orchestration framework is used to manage the edge cluster of the unmanned platform, to realize dynamic scheduling and load balancing of the task; Through the replica mechanism, parallel processing of high-priority tasks is realized; Data sharing and processing result aggregation are performed using a network file system, and the results are transmitted to the ground station to support post-disaster rescue decision-making.

2. The method of claim 1, wherein, The collection equipment at least includes a hyperspectral camera, a low-light infrared camera and a visible light camera, and the multi-modal remote sensing data at least includes collected hyperspectral video data, low-light infrared data and visible light data.

3. The method of claim 2, wherein, The task at least includes hyperspectral video tracking, low-light infrared video tracking, hyperspectral feature classification and single-time-phase damage assessment. After the task is generated, the multi-modal remote sensing data and the information of the task are structured and packaged by the preprocessing module of the edge device, and are transmitted to the subsequent task priority definition and scheduling stage.

4. The method of claim 1, wherein, The implementation of priority definition and resource allocation is as follows: According to the urgency of each multi-modal remote sensing data collection and processing in the disaster scene, the tasks are prioritized; the tasks are then decomposed into independent schedulable microservice units using a microservice architecture, and are packaged into containers using containerization technology to support rapid deployment and migration of the tasks; based on task priority and real-time computing requirements, the computing resources of the edge cluster are dynamically allocated to ensure rapid response of high-priority tasks.

5. The method of claim 1, wherein, The method for managing the edge cluster by the container orchestration framework is as follows: Using the automation function of the container orchestration framework, the scheduler thereof realizes dynamic scheduling, load balancing and fault recovery of the tasks; based on the replica mechanism, the number of replicas is dynamically adjusted according to the urgency and load demand of the tasks to support parallel processing of high-priority tasks; the node allocation filter chain of the container orchestration framework is used to screen edge nodes that meet the conditions, and the filter conditions include memory usage of the nodes, manually set stains and node affinity; the minimum request priority and the balanced resource allocation strategy are used to select the optimal node through weighted scoring.

6. The method of claim 5, wherein, The minimum request priority and the balanced resource allocation are realized in the following manner: The minimum request priority selects the node with the most sufficient resources by calculating the available resources of the node, and the scoring formula of the minimum request priority is as follows: ， where score LRP represents the minimum request priority score, C cpu and C men are the total CPU and memory capacities of the node,∑R cpu and∑R mem represent the total amount of CPU and memory requested by the allocated tasks, and the score range is normalized to 0 to 10; The balanced resource allocation evaluates the resource utilization balance of the node by calculating the request proportion of each resource type, and the resource score of the node is defined as the request resource amount of the node divided by the available resource amount, and the distance between the two resource proportions is calculated through the variance algorithm, and the node with the smallest difference in resource proportion is selected, and the scoring formula is as follows: ， where score BPR represents the score of balanced resource allocation, F cpu , F mem , and F vol represent the resource request proportion of CPU, memory, and storage respectively, variance is the variance of the three, and the score range is normalized to 0 to 10.

7. The method of claim 1, wherein, The replica mechanism realizes high-priority task parallel processing in the following manner: According to the task priority and the computing requirement, the number of replicas is determined; the input multi-modal remote sensing data is divided according to the number of replicas by using the load balancing function of the container orchestration framework, so as to ensure that each replica processes an independent data subset; the replica instances are scheduled to different edge nodes, the data stream is stored and accessed through the shared directory of the network file system, and data consistency is maintained; then, the replicas process the tasks in parallel.

8. The method of claim 1, wherein, The data sharing and the aggregation method of the processing result are as follows: The results of the parallel processing of each replica in the replica mechanism are uploaded to the shared directory of the network file system; the master node reads and aggregates the results of the parallel processing of all replicas from the network file system to obtain aggregated data; the aggregated data is transmitted to the ground station through a high-speed heterogeneous communication link, and the ground station performs real-time comprehensive evaluation based on the transmitted aggregated data to support post-disaster rescue decision-making and real-time guidance.

9. The method of claim 1, wherein, The type of the task is dynamically adjusted according to various disaster scenarios in emergency operation, and the resource allocation efficiency is optimized by a dynamic resource allocation factor, and the calculation formula of the dynamic resource allocation factor is as follows: ， wherein R alloc represents the dynamic resource allocation factor of the task, W task represents the computing workload of the task, P task represents the task priority, C total represents the total computing capacity of the cluster, N avail represents the current available node number.

Images