Low Latency Computing Offloading And Resource Scheduling Mechanism For Uav-Aided Disaster Perception

In emergency scenarios,accurate and timely disaster perception is crucial for rescue.However,the destruction of communication infrastructure may lead to bad transmission of image data.Insufficient computing power of on-site equipment may result in bad performance of intelligent tasks such as disaster detection and tracking of the wounded,which greatly reduces the rescue efficiency.To solve the above problem,we use data compression and super-resolution network to relieve communication burden.UAVs acted as auxiliaries of acquisition terminals in ground are also dispatched for disaster detection.On one hand,with a high degree of compression,the transmission latency will be reduced,but the detection sequence will be seriously distorted and the detection accuracy will be reduced.On the other hand,with a more complex super-resolution network model,the image data is better restored,which leads to a higher detection accuracy.Therefore,to achieve a tradeoff between the detection accuracy and the transmission latency,we investigate an UAV-assisted disaster detection in emergency scenarios.The allocation of communication and computation resources during disaster detection are optimized to improve the rescue efficiency.The main contents and contributions are summarized as follows:Firstly,we investigate the scenario of multi-UAVs-assisted a single mobile terminal in disaster detection.To resolve the contradiction between insufficient resources and high requirements of latency and precision in the emergency network,we propose a data compression based computation offloading scheme in an UAV-assisted single node scenario.The optimization of task offloading and data compression is modeled as a mixed integer nonlinear programming problem.A Deep Reinforcement Learning(DRL)based approach is proposed to obtain the decision of task offloading,whether the terminals offload task to UAVs or not.Then,the compression ratio is obtained by convex optimization.Our goal is to minimize the overall task completion latency while ensuring the accuracy of the disaster detection.The numerical results show that the proposed algorithm reduced system costs about 27%,which greatly improve the efficiency of disaster perception under the accuracy constraint of the detection.Secondly,we extend the scenario to multi-UAVs-assisted multimobile terminals in disaster detection,and propose a super-resolution network based computation offloading scheme.The optimization of node selection,super resolution model selection,and allocation of spectrum resources is modeled as a mixed integer linear programming problem.A clustering scheme based on channel conditions and load balancing is proposed to select the UAV assisted nodes.Then,the Kuhn-Munkras(KM)matching algorithm is used to allocate spectrum resources.Then,different super-resolution network are loaded on UAVs so that the quality of the lowresolution visual data can be improved.By this way,the disaster detection accuracy can be enhanced and the task detection latency can be minimized.Simulation results show that the proposed algorithm can greatly reduce the system costs about 35.7%and improve the detection accuracy.Lastly,to verify the performance of our proposed algorithms,we build disaster sensing system under the cloud side interaction based on the intelligent communication computing service platform.According to the request made by the user,the cloud can schedule edge collection devices and UAVs to serve the user,and the detection results can display in real time.Specifically,after the cloud obtains the requirements of the user,it dispatches the edge device.First,the acquisition device obtains the disaster information and compresses the disaster video.Then,it transmits the compressed data to the edge UAVs to disaster detection.Finally,the detection results are displayed on the UI interface of the user.The practice results show that the system can effectively reduce the completion delay of disaster detection task and improve the efficiency of emergency rescue under the condition of ensuring the accuracy of task detection.