A data offloading method and device based on mobile edge computing

Abstract

The present invention provides a data offloading method and apparatus based on mobile edge computing. The method includes: obtaining at least one video sequence to be detected, and dividing the at least one video sequence to be detected into at least one scheduling period; for each scheduling period, obtaining preset channel gains corresponding to a preset number of unmanned aerial vehicles (UAVs), determining corresponding offloading situations and video sequence compression ratios through a preset optimization model according to the preset channel gains, and offloading data to the UAVs according to the corresponding offloading situations and video sequence compression ratios. In an emergency scenario with dual constraints on spectrum and computing power, the present invention considers using UAVs to assist in detecting and compressing and offloading data. Since the size of the compression ratio affects the task completion delay and detection accuracy, how to perform real-time scheduling of transmission and computing resources to achieve a balance between task detection accuracy and completion delay is the key problem to be solved by the present invention.

Description

Technical Field

[0001] The present invention relates to the technical field of computer networks, and in particular, to a data offloading method and device based on mobile edge computing. Background Art

[0002] Disaster sensing is the basis and key for ensuring emergency rescue. Especially due to the damage of ground communication facilities, the bandwidth is extremely limited, and the limited battery energy and low computing power make it difficult for ground collection devices to rely solely on their own computing resources for intelligent disaster detection, bringing great difficulties to the accurate decision-making of the command center. Therefore, there is an urgent need to propose a disaster sensing method with low latency and high precision to assist ground collection devices in efficiently detecting the collected data, and transmitting the results to the command end in real time to improve the accuracy and timeliness of the emergency rescue system.

[0003] In recent years, most studies have used computing offloading to alleviate the disadvantages of insufficient terminal computing power. In the research on task collaborative computing in the context of mobile edge computing to effectively sense information, it is generally divided into three optimization strategies for objectives. First, aiming at minimizing the latency of obtaining effective information, considering resource scheduling algorithms based on the Lyapunov function to reduce communication latency and computing latency; in addition, there are also those aiming at minimizing system energy consumption, using artificial fish swarm algorithms or genetic algorithms to allocate system spectrum and computing resources to reduce communication energy consumption and computing energy consumption; finally, collaborative computing is carried out aiming at reducing system energy consumption and meeting latency requirements, jointly optimizing variables such as the allocation of transmission and computing resources and data scheduling decisions, and considering reinforcement learning methods to obtain effective information. However, with the increasing number of computationally intensive tasks such as object detection and visual positioning, the above methods can no longer meet the latency requirements for the migration of multimedia data such as images and videos between nodes.

[0004] In addition, data compression, as a technology that can both retain useful information and effectively reduce the amount of data, has been widely used. In an edge computing system, the amount of offloaded data can be compressed, jointly optimizing offloading decisions and transmission and computing resources to minimize the energy consumption and service latency of users. Although a large number of solutions have been proposed for mobile edge computing problems, there is less research on the collaborative computing of unmanned aerial vehicles and terminal nodes, and the requirements for service accuracy in intelligent scenarios have been ignored. Summary of the Invention

[0005] The present invention provides a data offloading method and device based on mobile edge computing. The present invention mainly solves the problem of balancing the completion delay and deduction accuracy of intelligent detection tasks in the edge architecture in emergency scenarios. The ground data acquisition device (or ground terminal) collects disaster situation information and conducts intelligent detection. However, considering the limited computing power, battery and other device conditions of the rescue cart, it is unable to perform efficient operations. As an effective disaster situation perception carrier in the emergency command scene, the unmanned aerial vehicle (UAV) can receive the ground-collected data and execute computationally intensive model deduction tasks such as disaster situation detection. However, the on-site spectrum is insufficient and the link state is time-varying, which brings great challenges to the migration and offloading of multimedia data such as images and videos between the ground terminal and the UAV. The present invention introduces data compression technology, assigns the UAV to assist in the intelligent detection of disaster situation information, alleviates the pressure of insufficient on-site computing power, breaks through the bottleneck of limited communication resources in emergency scenarios, and at the same time considers the problems of data accuracy and transmission delay during the process of the terminal offloading the data to be detected to the UAV. The data compression technology is introduced to alleviate the communication link burden, improve the accuracy of the transmitted data, and reduce the transmission delay by controlling the compression ratio of the data. To achieve the balance between the task completion delay and the detection accuracy in the process of transmission and computing resource scheduling.

[0006] It mainly solves the problem of balancing the completion delay and deduction accuracy of intelligent detection tasks.

[0007] In a first aspect, the present invention provides a data offloading method based on mobile edge computing, which is applied to a ground terminal and includes: obtaining at least one video sequence to be detected, dividing the at least one video sequence to be detected into at least one scheduling period, each scheduling period including a preset number of video sequences to be detected, the preset number of video sequences to be detected corresponding to a preset number of UAVs one by one, the UAVs being used to provide computing power support for detecting the video sequences to be detected; for each scheduling period, obtaining a preset number of channel gains corresponding to the preset number of UAVs, and according to the preset number of channel gains, determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model, and offloading the data to the UAVs according to the corresponding offloading situation and video sequence compression ratio, the preset optimization model being a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

[0008] Further, the step of, for each scheduling period, obtaining a preset number of channel gains corresponding to the preset number of UAVs includes: for each scheduling period, generating a preset number of channel gains corresponding to the preset number of UAVs according to the Rayleigh distribution method.

[0009] Further, determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model includes: determining the corresponding offloading situation according to the preset number of channel gains and the preset number of preset video sequence compression ratios through the optimization model in combination with a first preset condition and a second preset condition; and determining the corresponding video sequence compression ratio according to the preset number of channel gains and the corresponding offloading situation through the optimization model in combination with the second preset condition and a third preset condition.

[0010] Further, the first preset condition indicates that the parameter of the offloading situation is a discrete variable, the second preset condition indicates that the deduction delay of each scheduling period is less than or equal to a first threshold, and the third preset condition indicates that the value of the fitting function of the video sequence compression ratio and the data accuracy is greater than or equal to a second threshold.

[0011] In a second aspect, the present invention further provides a data offloading device based on mobile edge computing, which is applied to a ground terminal and includes: a first processing module, configured to obtain at least one video sequence to be detected, divide the at least one video sequence to be detected into at least one scheduling period, each scheduling period includes a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponds to a preset number of unmanned aerial vehicles one by one, and the unmanned aerial vehicles are used to provide computing power support for detecting the video sequences to be detected; a second processing module, configured to, for each scheduling period, obtain a preset number of channel gains corresponding to the preset number of unmanned aerial vehicles, determine the corresponding offloading situation and video sequence compression ratio according to the preset number of channel gains through a preset optimization model, and offload data to the unmanned aerial vehicles according to the corresponding offloading situation and video sequence compression ratio, and the preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

[0012] In a third aspect, the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the program, the steps of the data offloading method based on mobile edge computing as described in any one of the above are implemented.

[0013] In a fourth aspect, the present invention further provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the data offloading method based on mobile edge computing as described in any one of the above are implemented.

[0014] In a fifth aspect, the present invention further provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the data offloading method based on mobile edge computing as described in any one of the above are implemented.

[0015] A data offloading method and device based on mobile edge computing provided by the present invention, when data needs to be offloaded to an unmanned aerial vehicle (UAV), considering the balance problem between the completion delay and the deduction accuracy of intelligent detection tasks, determines the corresponding offloading situation and video sequence compression ratio through a preset optimization model. By minimizing the data deduction delay of the overall task according to the obtained offloading situation and video sequence compression ratio, while ensuring the disaster detection accuracy, it also improves the completion efficiency of the collaborative detection task between the UAV and the ground terminal and reduces the system overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0017] Figure 1 is a schematic flowchart of some embodiments of the data offloading method based on mobile edge computing provided by the present invention;

[0018] Figure 2 is a schematic flowchart of other embodiments of the data offloading method based on mobile edge computing provided by the present invention;

[0019] Figure 3 is a schematic diagram of the scenario of the intelligent disaster detection and UAV-assisted computing offloading decision mechanism based on data compression;

[0020] Figure 4 Schematic diagram of the structure of the DQN decision framework based on data compression;

[0021] Figure 5 is a schematic diagram of the model convergence after 10,000 task cycles of generating channel gain information according to the Rayleigh distribution;

[0022] Figure 6 is a schematic diagram of the fitting function of the linear compression ratio and the accuracy;

[0023] Figure 7 is a schematic diagram of the fitting function of the non-linear compression ratio and the accuracy;

[0024] Figure 8 is a schematic diagram of the simulation analysis of the system overhead of different data offloading methods under different powers;

[0025] Figure 9 is a schematic diagram of the simulation analysis of the system overhead of different data offloading methods for different amounts of video;

[0026] Figure 10 It is the intention of the system overhead simulation analysis diagram of different data offloading methods for multiple scheduling cycles;

[0027] Figure 11 It is a schematic structural diagram of some embodiments of a data offloading device based on mobile edge computing provided by the present invention;

[0028] Figure 12 It is a schematic structural diagram of an electronic device provided by the present invention. Detailed implementation manners

[0029] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0030] In addition, it should be noted that for the sake of convenience of description, only parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present invention and the features in the embodiments may be combined with each other.

[0031] It should be noted that the concepts such as "first" and "second" mentioned in the present invention are only used to distinguish different devices, modules or units, and are not used to limit the order of functions executed by these devices, modules or units or their interdependent relationships.

[0032] It should be noted that the modifications of "one" and "multiple" mentioned in the present invention are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0033] The names of the messages or information exchanged between multiple devices in the embodiments of the present invention are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0034] The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0035] Please refer to Figure 1 , Figure 1 It is a schematic flowchart of some embodiments of a data offloading method based on mobile edge computing provided by the present invention. As Figure 1 shown, this method is applied to a ground terminal and includes the following steps:

[0036] Step 101: Obtain at least one video sequence to be detected, and divide the at least one video sequence to be detected into at least one scheduling period. Each scheduling period includes a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponds one-to-one to a preset number of drones. The drones are used to provide computing power support for detecting the video sequences to be detected.

[0037] As an example, a single ground shooting device (or ground terminal) at the disaster site can collect on-site environment videos in real time. At the same time, multiple drones carrying computing resources and deploying detection networks can receive the data collected by the ground terminal and provide auxiliary deduction calculation services. As Figure 3 shown, assume that there are M drones in the system, and each drone computing node has different computing resources. The collected video data needs to be compressed and encoded and then migrated and transmitted from the ground device to the drones.

[0038] As an example, first, the collected video data is segmented into N video sequences to be detected (i.e., obtain at least one video sequence to be detected). Then, every M video sequences to be detected among the N video sequences to be detected are determined as a group, and each group is a scheduling period (i.e., divide the at least one video sequence to be detected into at least one scheduling period, and each scheduling period includes a preset number of video sequences to be detected), with a total of scheduling periods. Assume that the ground terminal knows the computing power situation of each drone. For each scheduling period, the required computing power of the M video sequences to be detected in the scheduling period can correspond one-to-one to the computing power of the M drones (i.e., the preset number of video sequences to be detected corresponds one-to-one to the preset number of drones). The M drones process the M video sequences to be detected in one scheduling period each time until they finish processing scheduling periods.

[0039] Step 102: For each scheduling period, obtain a preset number of channel gains corresponding to the preset number of drones. According to the preset number of channel gains, determine the corresponding offloading situation and video sequence compression ratio through a preset optimization model, and offload the data to the drones according to the corresponding offloading situation and video sequence compression ratio. The preset optimization model is a model for a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

[0040] The offloading situation refers to whether the M video sequences to be detected need to be downloaded to the corresponding M drones for data detection.

[0041] The video sequence compression ratio refers to the compression degree of the M video sequences to be detected.

[0042] As an example, for each scheduling period, it is necessary to separately obtain the channel conditions of M drones first, that is, obtain the preset number of channel gains corresponding to the preset number of drones. Through the preset optimization model and the obtained channel conditions of M drones, first determine whether the M video sequences to be detected in the current scheduling period can be offloaded to the corresponding M drones for detection (that is, determine the corresponding offloading situation). Then, according to the offloading situation and the obtained channel conditions of M drones, through the preset optimization model, determine the video sequence compression ratio of the video sequences to be detected that need to be downloaded to the drones (that is, according to the preset number of channel gains, determine the corresponding offloading situation and video sequence compression ratio through the preset optimization model, and offload the data to the drones according to the corresponding offloading situation and video sequence compression ratio).

[0043] As an example, use the parameter pair <α i,j ,β i,j > to represent the data volume size and the computational amount required for deduction for the j-th video sequence to be detected in the i-th scheduling period (as an example, it can be the i-th scheduling period in the -th scheduling period), and the units are bits and cycles respectively. The video sequences to be detected within a single scheduling period correspond one-to-one with the drones, and x i,j ∈{0,1} represents the optimal offloading decision, that is, whether the j-th sequence in the i-th scheduling period is transmitted to the edge (i.e., the drone) for detection (0 means not transmitted to the drone, 1 means transmitted to the drone). The preset optimization model can be defined that the system overhead is the product of the average completion delay (i.e., L in the following formula, or the system overhead) of all video sequences to be detected (i.e., at least one video sequence to be detected) and the detection accuracy. The optimization model can refer to:

[0044]

[0045] Among them, the deduction accuracy and video sequence compression ratio of the j-th sequence in the i-th period are: ρ i (R i,j ).

[0046] Among them, the delay of the ground terminal executing the intelligent detection task is expressed as T 0 (seconds). When the local (i.e., the ground terminal) executes the detection task, the terminal serially performs intelligent detection on at least one video sequence to be detected. r 0 (cysles / s) represents the terminal deduction computing power, and its overall deduction delay T 0 (in seconds) is:

[0047]

[0048] When unloading the video sequence to be detected to the UAV for intelligent detection, considering that the channels between UAVs are independent of each other, the uplink transmission rate U of the j-th video sequence to be detected in the i-th scheduling period i,j (bits / s) is:

[0049]

[0050] where w (MHz) is the communication system bandwidth, p (mW) represents the transmit power of the terminal node, and H i,j (dB) is the channel gain on the j-th channel in the i-th scheduling period, and σ 2 (dBm) is the additive white Gaussian noise.

[0051] where the video sequence compression ratio is defined as the ratio of the data volume of the original video sequence to be detected to the data volume of the compressed video sequence to be detected, denoted as R i,j , select R i,j to perform compressed transmission on the j-th detection sequence in the i-th period, and the transmission delay is:

[0052]

[0053] where the rotation speed of M UAVs for deduction is r j (cycles / s), and the intelligent detection delay of the j-th sequence in the i-th period is:

[0054]

[0055] The optimization model is a model for a two-layer optimization problem based on minimizing the delay of detecting the video sequence to be detected and maximizing the detection accuracy of the video sequence to be detected. That is, for a preset number of video sequences to be detected in each scheduling period, it can be determined whether to unload to the corresponding UAV and the video sequence compression ratio through the model of the two-layer optimization problem based on minimizing the delay of detecting the video sequence to be detected and maximizing the detection accuracy of the video sequence to be detected. To achieve a reasonable balance between the wireless link load and the final detection accuracy, and at the same time ensure the high timeliness of the overall detection task, it is necessary to effectively optimize and control the data compression parameters (i.e., the video sequence compression ratio) and the task offloading strategy (i.e., the offloading situation).

[0056] In some alternative implementation manners, the first preset condition indicates that the parameter of the offloading situation is a discrete variable, the second preset condition indicates that the deduction delay of each scheduling period is less than or equal to the first threshold, and the third preset condition indicates that the value of the fitting function of the video sequence compression ratio and the data accuracy is greater than or equal to the second threshold.

[0057] As an example, aiming to minimize the system overhead of the edge-end collaborative detection system (i.e., minimizing the delay of detecting the video sequence to be detected), the compression ratio R of the j-th sequence in the i-th cycle i,j and the task offloading decision variable X i are modeled as the following optimization problem. For the model of the two-layer optimization problem based on minimizing the delay of detecting the video sequence to be detected and maximizing the detection accuracy of detecting the video sequence to be detected, reference can be made to:

[0058]

[0059] where C1 indicates that the above problem is a two-dimensional offloading decision problem, C2 indicates that the overall completion delay of the task should be less than the delay threshold, and C3 indicates the deduction accuracy ρ i (R i,j ) should be greater than the set threshold ε, and the result does not affect the accuracy of the decision. C1, C2, and C3 are the first preset condition, the second preset condition, and the third preset condition respectively. Formula (6) indicates that under the conditions of minimizing the delay of detecting the video sequence to be detected and the detection accuracy of detecting the video sequence to be detected, the optimal video compression ratio R i,j and the offloading strategy X i are selected.

[0060] where T i * is the task completion delay of each scheduling cycle and is the maximum completion delay of the data processing of each corresponding drone in the current cycle. That is, for each scheduling cycle, multiple drones perform the deduction calculation of the optimization model on the received video sequences to be detected respectively:

[0061] T i * = max j∈M T i，j Formula (7)

[0062] For the data offloading method based on mobile edge computing disclosed in some embodiments of the present invention, when the data needs to be offloaded to the drone, considering the balance problem between the completion delay of the intelligent detection task and the deduction accuracy, the corresponding offloading situation and video sequence compression ratio are determined through a preset optimization model. According to the obtained offloading situation and video sequence compression ratio, the completion delay of the overall video sequence to be detected is minimized. While ensuring the disaster detection accuracy, the completion efficiency of the collaborative detection task between the drone and the ground terminal is also improved, and the system overhead is reduced.

[0063] Please refer to Figure 2 , Figure 2 which is a flowchart of some other embodiments of the data offloading method based on mobile edge computing according to the present invention. As shown in Figure 2As shown, this method is applied to a ground terminal and includes the following steps:

[0064] Step 201, obtain at least one video sequence to be detected, divide the at least one video sequence to be detected into at least one scheduling period, each scheduling period includes a preset number of video sequences to be detected, the preset number of video sequences to be detected corresponds to a preset number of unmanned aerial vehicles (UAVs) one by one, and the UAVs are used to provide computing power support for detecting the video sequences to be detected.

[0065] In some embodiments, for the specific implementation of step 201 and the technical effects brought by it, reference can be made to Figure 1 step 101 in the corresponding embodiment, which will not be elaborated here.

[0066] Step 202, for each scheduling period, generate a preset number of channel gains corresponding to the preset number of UAVs according to the Rayleigh distribution method, determine the corresponding offloading situation and video sequence compression ratio through a preset optimization model according to the preset number of channel gains, and offload the data to the UAVs according to the corresponding offloading situation and video sequence compression ratio. The preset optimization model is a model for a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

[0067] Rayleigh Distribution: When the two components of a random two-dimensional vector are independent, have a mean of 0, and have the same variance, the magnitude of this vector follows a Rayleigh distribution.

[0068] As an example, other methods can also be used to randomly determine the channel gains to simulate the channel gain situation of UAVs in a real scenario.

[0069] In some optional implementation manners, determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model includes: determining the corresponding offloading situation according to the preset number of channel gains and the preset number of preset video sequence compression ratios through the optimization model in combination with the first preset condition and the second preset condition; determining the corresponding video sequence compression ratio according to the preset number of channel gains and the corresponding offloading situation through the optimization model in combination with the second preset condition and the third preset condition.

[0070] Still taking formula (6) as an example, the determination according to the binary offloading mechanism and the selection of the video sequence compression ratio during the offloading process can divide formula (6) into two sub-problems for solution. First, considering the confirmation of the offloading mechanism, sub-problem one can be obtained, that is, determining the corresponding offloading situation according to the preset number of channel gains and the preset number of preset video sequence compression ratios through the optimization model in combination with the first preset condition and the second preset condition, which can be referred to:

[0071]

[0072] In sub - problem P1, considering the given channel gain and the video sequence compression ratio (the video sequence compression ratio can be defaulted to 1) during the offloading process, the optimal offloading mechanism for each scheduling period is confirmed. That is, in each scheduling period, first obtain the system overheads of the corresponding preset number (assume the preset number is 2) of video sequences to be detected, such as L1 and L2, select the maximum value of the system overheads as the system overhead of the current scheduling period. For example, select L1 with a larger system overhead value, and then minimize the overall task completion delay of the above - mentioned system overhead L1 of the current scheduling period (i.e., minimize L1) to determine the final offloading strategy X i 。

[0073] After obtaining the final offloading strategy X i In this case, then consider the optimization problem of the video sequence compression ratio during the offloading process, and sub - problem two can be obtained. That is, according to the preset number of channel gains and the corresponding offloading situations, through the optimization model, combined with the second preset condition and the third preset condition, determine the corresponding video sequence compression ratio. The reference can be as follows:

[0074]

[0075] When confirming whether each video sequence to be detected in each scheduling period is offloaded to the corresponding UAV (i.e., the final offloading strategy X obtained through formula (8) i ) and the channel gain, the optimal video sequence compression ratio for each scheduling period is confirmed. Sub - problem P2 is a convex optimization problem regarding the compression ratio. That is, in each scheduling period, first obtain the system overheads of the corresponding preset number (assume the preset number is 2) of video sequences to be detected, such as L3 and L4, select the maximum value of the system overheads as the system overhead of the current scheduling period. For example, select L3 with a larger system overhead value, and then minimize the overall task completion delay of the above - mentioned system overhead L3 of the current scheduling period (i.e., minimize L3) to determine the final optimal video sequence compression ratio R i,j 。

[0076] From Figure 2 it can be seen that compared with the descriptions of some corresponding embodiments Figure 1 Figure 2 ​In some corresponding embodiments, the data offloading method based on mobile edge computing reflects that according to the Rayleigh distribution method, a preset number of channel gains corresponding to a preset number of unmanned aerial vehicles (UAVs) are generated. It can be seen that by simulating the channel gains of real UAVs through the Rayleigh distribution method, the efficiency of the collaborative detection task between UAVs and ground terminals is improved while ensuring the accuracy of disaster detection in a real application scenario, and the system overhead is reduced.

[0077] As an example, referring to Figure 4 , determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model involves the optimization of a continuous variable R i,j and the determination of a discrete variable X i , which belongs to a mixed-integer non-linear programming problem. The following steps can be referred to:

[0078] Step 1: The input data includes: the channel gain H i,j between the ground terminal and the UAV; the transmission power p; the amount of video sequence data α i,j to be detected and the amount of computation β i,j required for its detection and deduction; the computing power r j of the UAV;

[0079] Step 2: Obtain the optimal offloading decision: Input the channel gain information H i,j between the terminal (ground terminal) and the UAV into the deep neural network, and use the K-nearest neighbor algorithm to map the multi-dimensional offloading decision for the output result. Define the negative value of the benefit function as the reward function of the deep Q network, and select the optimal offloading decision Take the current channel gain as the action, and the optimal offloading decision as the reward and store them in the memory network. Randomly extract samples from the memory network to train the deep neural network until the network converges;

[0080] Step 3: Obtain the adaptive selection of the compression ratio: Use MobileNet-v2 to identify images at different compression ratios, fit the corresponding relationship between the accuracy ρ(R i,j ) and the video sequence compression ratio as a linear function, and then solve the optimal video sequence compression ratio through convex optimization according to the optimal offloading decision;

[0081] Step 4: Obtain the minimum completion delay under the condition of tolerating the disaster detection accuracy according to the optimal offloading decision and the optimal video sequence compression ratio, so that the system overhead L i is minimized, and finally obtain the minimum value of the sum of the system overheads of all detection tasks.

[0082] Among them, the offloading strategy X i is obtained by using the deep Q learning method, and then for a fixed offloading strategy, the continuous variable compression ratio R i，jOptimal selection.

[0083] Among them, the channel gain H between the ground terminal and the UAV i，j The channel gain can be randomly generated based on the Rayleigh distribution. The randomly generated channel gain is input into the deep neural network. The output result is mapped by the K-nearest neighbor algorithm for multi-dimensional offloading decision-making. The negative value of the benefit function is defined as the reward function of the deep Q network. The optimal offloading decision is selected. The current channel gain is used as the action, and the optimal offloading decision is stored as the reward corresponding to the current action in the memory network. Samples in the memory network are randomly extracted to train the deep neural network until convergence. To implement the above algorithm, first obtain the channel gain results randomly generated by the Rayleigh distribution as the input, referring to the following table (assuming M = 5 UAVs, the first row represents the channel gains of 5 UAVs in the first scheduling period):

[0084] Table 1 Input channel gain information generated by Rayleigh distribution

[0085] 8.5038e-07 3.05791e-06 2.2405e-06 5.4385e-07 6.1290e-07 6.0602e-06 1.1033e-05 1.0021e-07 1.2161e-06 1.9613e-06 ... ... ... ... ... 1.8916e-06 4.8858e-06 3.8989e-06 6.8934e-07 7.0364e-06 2.3827e-05 5.6775e-06 2.9798e-06 3.1422e-06 1.1908e-06

[0086] After iterative training of the deep neural network, the following convergence performance results are obtained as Figure 5 shown.

[0087] Among them, the MobileNet-v2 network and the kaggle dataset are used for object recognition. The mapping relationship between the compression ratio and the recognition accuracy can be referred to the following table:

[0088] Table 2 Accuracy relationship corresponding to different compression ratios under the MobileNet-v2 model

[0089]

[0090] Furthermore, different fitting functions are used to obtain the mapping relationship between the compression ratio and the accuracy of the video sequence to be detected, as Figure 6 and Figure 7 shown.

[0091] Among them, the output result after network convergence is the offloading decision output after randomly adopting the channel gain. Based on this fixed offloading strategy, and further solving the optimal compression ratio through convex optimization to achieve low-latency and high-precision disaster detection.

[0092] Taking the linear fitting function of accuracy and video sequence compression ratio as an example:

[0093] ρ i (R i，j ) = -0.0085R i，j + 0.95 Formula (10)

[0094] The system overhead of offloading the video sequence to the ground terminal for intelligent deduction can be rewritten as:

[0095]

[0096] By replacing the variables in the above function, the system overhead can be rewritten as:

[0097]

[0098] Solve the above equation to find the optimal compression ratio of the jth sequence in the i-th period, and the closed-form solution to the above problem is:

[0099]

[0100] As an example, the experimental environment and parameter settings can be: built based on the Tensorflow environment, using the Intel_i5-10210U_CPU_@_1.60GHz processor, the input data is a random gain generated by Rayleigh distribution, and there are 5 drones assisting the ground terminal nodes for intelligent detection. The ground terminal node transmission power is 300mW, the CPU speed is 100cycles / s, the noise power in the network is -96dBm, the transmission bandwidth is 20MHz, the CPU speed of the 5 drone computing nodes is set to [2000,2000,5000,2000,5000]cycles / s, and the video sequence compression ratio range is set to [1,100].

[0101] Based on the above experimental environment and parameter settings, the data offloading method based on mobile edge computing provided by the present invention (i.e., the optimally compressed DQN architecture) is compared with the following schemes:

[0102] (1) Deep Q Network (DQN) architecture, or uncompressed DQN architecture: does not consider data compression technology, and only uses existing deep reinforcement learning methods to make offload decisions for deduction tasks;

[0103] (2) DQN architecture based on maximum compression ratio, or maximum compression DQN architecture: The compression ratio is set to the maximum value, and the DQN offloading mechanism is combined to make offloading decisions for deduction tasks;

[0104] (3) DQN architecture based on random compression ratio, or randomly compressed DQN architecture: randomly generate compression ratio between [0, 20], and combine DQN offloading mechanism to make offloading decisions for deduction tasks.

[0105] The comparison results can be found in Figure 8 , Figure 9 and Figure 10 , Figure 8The horizontal axis represents the terminal transmission power, with values ranging from [0, 250] mW. The vertical axis is the system overhead. Taking the uncompressed DQN architecture as the benchmark, the different system overheads under the maximum compression DQN architecture, random compression DQN architecture, and optimal compression DQN architecture are compared. The results show that when the power is 10 mW, the method proposed in the present invention (corresponding to the optimal compression DQN architecture) can effectively reduce the system overhead caused by intelligent task detection, approximately 41.2%. When the power value is 250 mW, the system overhead is reduced by approximately 7.5%. Therefore, under different powers, the method proposed in the present invention can reduce the system overhead to the greatest extent. And when the power is greater, the offloading transmission delay is shorter, and the influence brought by the compression ratio is also smaller.

[0106] Figure 9 The horizontal axis represents different amounts of offloaded data of the terminal, with values ranging from [0, 18.5] MB. The vertical axis is the system overhead. Taking the uncompressed DQN architecture as the benchmark, the different system overheads under the optimal compression DQN architecture, random compression DQN architecture, and maximum compression DQN architecture are compared. The results show that when the offloaded data amount is 18 MB, the method proposed in the present invention (corresponding to the optimal compression DQN architecture) can effectively reduce the system overhead caused by intelligent task detection, approximately 21.1%. When the offloaded data amount is 1 MB, the system overhead is reduced by approximately 1.9%. Therefore, under different amounts of offloaded data, the method proposed in the present invention can reduce the system overhead to the greatest extent. And when the offloaded data amount is large, the required delay is smaller, and the influence brought by the compression ratio is also greater.

[0107] Figure 10 It shows the comparison of the system overhead of the optimal compression DQN architecture, uncompressed DQN architecture, random compression DQN architecture, and maximum compression DQN architecture. The results show that in the case of randomly obtaining 200 channel gains, the performance of the method proposed in the present invention is better than that of the method using only deep reinforcement learning (i.e., better than the method based on the uncompressed DQN architecture), and the overall system overhead is reduced by approximately 27%.

[0108] In summary, through the real-time scheduling of communication and computing resources, the present invention provides support for the collaborative deduction of intelligent detection tasks in emergency scenarios, and can effectively improve the disaster perception efficiency. In the deduction task offloading architecture with edge-end linkage, a DQN-based offloading decision mechanism is proposed to solve the binary offloading problem of deduction tasks. At the same time, data compression technology is introduced during the offloading process to alleviate the communication link burden brought by media data migration. Experimental results show that when the power is 10 mW, the method proposed by the present invention can effectively reduce the system overhead caused by intelligent task detection by approximately 41.2%; when the offloaded data volume is 18 MB, the method proposed by the present invention can effectively reduce the system overhead caused by intelligent task detection by approximately 21.1%; when randomly obtaining 200 channel gains, the method of the present invention can effectively reduce the system overhead by approximately 27%.

[0109] Please refer to Figure 11 , Figure 11 FIG. is a schematic structural diagram of some embodiments of a data offloading device based on mobile edge computing provided by the present invention. As an implementation of the methods shown in the above figures, the present invention also provides some embodiments of a data offloading device based on mobile edge computing. These device embodiments correspond to some method embodiments shown in Figure 1 and the device can be applied to various electronic devices.

[0110] As Figure 11 shown, a data offloading device based on mobile edge computing in some embodiments, which is applied to a ground terminal, includes a first processing module 1101 and a second processing module 1102: The first processing module 1101 is configured to obtain at least one video sequence to be detected, divide the at least one video sequence to be detected into at least one scheduling period, each scheduling period includes a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponds to a preset number of unmanned aerial vehicles one by one. The unmanned aerial vehicle is used to provide computing power support for detecting the video sequence to be detected; The second processing module 1102 is configured to, for each scheduling period, obtain a preset number of channel gains corresponding to the preset number of unmanned aerial vehicles, determine the corresponding offloading situation and video sequence compression ratio through a preset optimization model according to the preset number of channel gains, and offload the data to the unmanned aerial vehicle according to the corresponding offloading situation and video sequence compression ratio. The preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequence to be detected and maximizing the detection accuracy of detecting the video sequence to be detected.

[0111] In an alternative implementation manner of some embodiments, the second processing module 1102 is further configured to: for each scheduling period, generate a preset number of channel gains corresponding to the preset number of unmanned aerial vehicles according to the Rayleigh distribution method.

[0112] In an optional implementation of some embodiments, the second processing module 1102 is further used to: determine the corresponding unloading situation based on a preset number of channel gains and a preset number of preset video sequence compression ratios through an optimization model in combination with the first preset condition and the second preset condition; determine the corresponding video sequence compression ratio based on the preset number of channel gains and the corresponding unloading situation through an optimization model in combination with the second preset condition and the third preset condition.

[0113] In optional implementations of some embodiments, the first preset condition indicates that the parameter of the unloading situation is a discrete variable, the second preset condition indicates that the deduced delay of each scheduling cycle is less than or equal to a first threshold, and the third preset condition indicates that the value of the fitting function of the video sequence compression ratio and data accuracy is greater than or equal to the second threshold.

[0114] It is understandable that the modules described in the device are similar to those described in the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the device and the modules and units contained therein, and will not be described in detail here.

[0115] Figure 12 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 12 As shown, the electronic device may include: a processor (processor) 1210, a communication interface (Communications Interface) 1220, a memory (memory) 1230 and a communication bus 1240, wherein the processor 1210, the communication interface 1220, and the memory 1230 communicate with each other through the communication bus 1240. The processor 1210 can call the logic instructions in the memory 1230 to execute a data unloading method based on mobile edge computing, which includes: obtaining at least one video sequence to be detected, dividing the at least one video sequence to be detected into at least one scheduling period, each scheduling period including a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponds one-to-one to a preset number of drones, and the drones are used to provide computing power support for detecting the video sequences to be detected; for each scheduling period, obtaining a preset number of channel gains corresponding to the preset number of drones, and according to the preset number of channel gains, determining the corresponding unloading situation and video sequence compression ratio through a preset optimization model, and unloading the data to the drone according to the corresponding unloading situation and video sequence compression ratio, and the preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequence to be detected and maximizing the detection accuracy of the video sequence to be detected.

[0116] In addition, when the logical instructions in the above-mentioned memory 1230 can be implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the above-mentioned methods in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical discs that can store program codes.

[0117] On the other hand, the present invention also provides a computer program product. The above-mentioned computer program product includes a computer program stored on a non-transitory computer-readable storage medium. The above-mentioned computer program includes program instructions. When the above-mentioned program instructions are executed by a computer, the computer can execute the data offloading method based on mobile edge computing provided by the above-mentioned various methods. The method includes: obtaining at least one video sequence to be detected, dividing the at least one video sequence to be detected into at least one scheduling period, each scheduling period including a preset number of video sequences to be detected, the preset number of video sequences to be detected corresponding to a preset number of unmanned aerial vehicles one by one, and the unmanned aerial vehicles being used to provide computing power support for detecting the video sequences to be detected; for each scheduling period, obtaining a preset number of channel gains corresponding to the preset number of unmanned aerial vehicles, and based on the preset number of channel gains, determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model, and offloading data to the unmanned aerial vehicles according to the corresponding offloading situation and video sequence compression ratio. The preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

[0118] In another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is used to execute the data offloading method based on mobile edge computing provided above. The method includes: obtaining at least one video sequence to be detected, dividing the at least one video sequence to be detected into at least one scheduling period, each scheduling period including a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponding to a preset number of unmanned aerial vehicles one by one. The unmanned aerial vehicles are used to provide computing power support for detecting the video sequences to be detected; for each scheduling period, obtaining a preset number of channel gains corresponding to the preset number of unmanned aerial vehicles, and based on the preset number of channel gains, determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model, and offloading data to the unmanned aerial vehicles according to the corresponding offloading situation and video sequence compression ratio. The preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

[0119] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative labor.

[0120] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, also by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disc, etc., including several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute each embodiment or some parts of the above method of the embodiment.

[0121] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A data offloading method based on mobile edge computing, characterized in that, applied to a ground terminal, including: Obtain at least one video sequence to be detected, divide the at least one video sequence to be detected into at least one scheduling period, each scheduling period includes a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponds to a preset number of unmanned aerial vehicles (UAVs) one by one, and the UAVs are used to provide computing power support for detecting the video sequences to be detected; For each scheduling period, obtain a preset number of channel gains corresponding to the preset number of UAVs, and according to the preset number of channel gains, determine the corresponding offloading situation and video sequence compression ratio through a preset optimization model, and offload data to the UAVs according to the corresponding offloading situation and video sequence compression ratio. The preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of detecting the video sequences to be detected and maximizing the detection accuracy of detecting the video sequences to be detected.

2. The data offloading method based on mobile edge computing according to claim 1, characterized in that, the step of, for each scheduling period, obtaining a preset number of channel gains corresponding to the preset number of UAVs includes: For each scheduling period, generate a preset number of channel gains corresponding to the preset number of UAVs according to the Rayleigh distribution method.

3. The data offloading method based on mobile edge computing according to any one of claims 1 to 2, characterized in that, the step of determining the corresponding offloading situation and video sequence compression ratio through a preset optimization model includes: According to the preset number of channel gains and a preset number of preset video sequence compression ratios, through the optimization model, combined with a first preset condition and a second preset condition, determine the corresponding offloading situation; According to the preset number of channel gains and the corresponding offloading situation, through the optimization model, combined with the second preset condition and a third preset condition, determine the corresponding video sequence compression ratio.

4. The data offloading method based on mobile edge computing according to claim 3, characterized in that, the first preset condition indicates that the parameter of the offloading situation is a discrete variable, the second preset condition indicates that the deduced delay of each scheduling period is less than or equal to a first threshold, and the third preset condition indicates that the value of the fitting function of the video sequence compression ratio and the data accuracy is greater than or equal to a second threshold.

5. A data offloading device based on mobile edge computing, characterized in that, applied to a ground terminal, including: A first processing module, configured to obtain at least one video sequence to be detected, divide the at least one video sequence to be detected into at least one scheduling period, each scheduling period includes a preset number of video sequences to be detected, and the preset number of video sequences to be detected corresponds to a preset number of unmanned aerial vehicles (UAVs) one by one, and the UAVs are used to provide computing power support for detecting the video sequences to be detected; A second processing module, configured to, for each scheduling period, obtain a preset number of channel gains corresponding to the preset number of drones, determine corresponding offloading situations and video sequence compression ratios through a preset optimization model according to the preset number of channel gains, and offload data to the drones according to the corresponding offloading situations and video sequence compression ratios. The preset optimization model is a model of a two-layer optimization problem based on minimizing the delay of the video sequence to be detected and maximizing the detection accuracy of the video sequence to be detected.

6. The mobile edge computing-based data offloading device according to claim 5, wherein, the second processing module is further configured to: for each scheduling period, generate a preset number of channel gains corresponding to the preset number of drones according to the Rayleigh distribution method.

7. The mobile edge computing-based data offloading device according to any one of claims 5 to 6, wherein, the second processing module is further configured to: determine the corresponding offloading situation according to the preset number of channel gains and a preset number of preset video sequence compression ratios through the optimization model in combination with a first preset condition and a second preset condition; determine the corresponding video sequence compression ratio according to the preset number of channel gains and the corresponding offloading situation through the optimization model in combination with the second preset condition and a third preset condition.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein, when the processor executes the program, the steps of the mobile edge computing-based data offloading method according to any one of claims 1 to 4 are implemented.

9. A non-transitory computer-readable storage medium, on which a computer program is stored, wherein, when the computer program is executed by a processor, the steps of the mobile edge computing-based data offloading method according to any one of claims 1 to 4 are implemented.

10. A computer program product, comprising a computer program, wherein, when the computer program is executed by a processor, the steps of the mobile edge computing-based data offloading method according to any one of claims 1 to 4 are implemented.

Images