Emergency disaster detection method and device

Abstract

The present invention provides an emergency disaster detection method and device, comprising: allocating drone-assisted terminals for disaster detection based on the channel conditions between each terminal and each drone in a target area, as well as the existing computing power on the drone; allocating channels to each terminal using a KM matching algorithm, ensuring that the sum of the link uplink transmission rates of all terminals and the SINR value of each communication link are greater than a first preset threshold; offloading the sequence of disaster images to be detected on each terminal to the drone assigned to each terminal via the channel allocated to each terminal; selecting a super-resolution network model on the drone to reconstruct the sequence of disaster images to be detected, and using the reconstructed sequence of disaster images to be detected for disaster detection, ensuring that the reconstructed image quality is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold. The present invention achieves a balance between disaster detection completion delay and detection accuracy.

Description

Technical Field

[0001] The present invention relates to the field of intelligent detection technology, and in particular to an emergency disaster detection method and device. Background Art

[0002] The emergency command center needs to dispatch rescue equipment in real time according to the disaster situation. Effectively sensing disaster information and efficiently transmitting the results back to the command end is the basis and key to ensuring accurate decision-making at the command end.

[0003] With the development of technologies such as fire detection and visual positioning, the effective perception of disaster information is becoming intelligent. Furthermore, with the introduction of mobile edge computing architectures, edge nodes can collaborate on communication and computation to meet the latency and accuracy requirements of perception tasks in diverse disaster environments. However, these intelligent disaster awareness tasks primarily rely on image and video data. However, spectrum resources are limited due to the damage to ground-based communication infrastructure. Furthermore, limited battery life and low computing power make it difficult for ground-based disaster information collection equipment to perform intelligent disaster detection solely on its own computing resources, making it extremely difficult for command centers to make accurate decisions.

[0004] Therefore, there is an urgent need to propose a method with low communication loss and flexibility to assist disaster awareness, assist ground devices in intelligent detection of collected data, and improve the accuracy and timeliness of emergency rescue. Summary of the Invention

[0005] The present invention provides an emergency disaster detection method and device to address the defect in the existing technology that intelligent disaster detection cannot be performed under the dual limitations of spectrum resources and the computing power of ground equipment. It realizes the use of drones to assist ground terminals in emergency disaster detection, achieving a balance between detection delay and detection accuracy.

[0006] The present invention provides an emergency disaster detection method, comprising:

[0007] Each terminal in the target area acquires a sequence of images of the disaster condition to be detected and offloads the sequence to a drone in the target area. The drone provides computing power for detecting the sequence of images of the disaster condition to be detected and constructs an optimization model to balance the accuracy and latency of detecting the sequence of images of the disaster condition to be detected.

[0008] Based on the optimization model, according to the channel conditions between each terminal and each drone, and the existing computing capacity of each drone, a drone is assigned to each terminal, so as to achieve balanced clustering of the terminals and drones based on the channel conditions and the existing computing capacity;

[0009] Based on the optimization model, a bipartite graph of the terminals and the channels in the target area is established, and a KM matching algorithm is used to allocate a channel to each terminal so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold. The sequence of disaster images to be detected on each terminal is offloaded to the drone allocated to each terminal via the channel allocated to each terminal, thereby optimizing the channel allocated to each terminal.

[0010] Based on the optimization model, multiple super-resolution network models on the drone are selected, the sequence of disaster images to be detected is reconstructed, and disaster detection is performed using the reconstructed sequence of disaster images to be detected. The reconstructed image quality is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold, thereby achieving a balance between disaster detection accuracy and overall delay of disaster detection.

[0011] According to an emergency disaster detection method provided by the present invention, the steps of allocating a drone to each terminal based on the optimization model according to the channel conditions between each terminal and each drone and the existing computing capacity of each drone include:

[0012] Obtaining a first distance between each terminal of each unassigned drone and each drone, and a second distance between each terminal and each drone assigned to the terminal, and using the first distance and the second distance as the channel condition;

[0013] The first distance, the second distance, and the existing calculation amount corresponding to each drone are weighted and added together, and the drone with the smallest addition result is allocated to the terminal of the unassigned drone.

[0014] According to an emergency disaster detection method provided by the present invention, the step of allocating a channel to each terminal using the KM matching algorithm includes:

[0015] When the number of the terminals is less than the number of channels in the network, obtaining a SINR for each terminal on each channel for each drone, and obtaining a link uplink transmission rate for each terminal on each channel for each drone based on the SINR;

[0016] Add up the link uplink transmission rates of each terminal for each UAV on each channel to obtain the sum of the link uplink transmission rates of all terminals;

[0017] The terminal and the channel are taken as the matching parties, the link uplink transmission rate of each terminal for each drone on each channel is taken as the matching edge weight, and the KM matching algorithm is used to allocate the channel to each terminal, so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal for each drone on the allocated channel is greater than the first preset threshold.

[0018] According to an emergency disaster detection method provided by the present invention, the step of obtaining the SINR of each terminal for each drone on each channel includes:

[0019] When any terminal has been assigned to any drone, the SINR of the terminal for the drone on each channel is obtained based on the interference of the terminal assigned to the drone on each channel and the interference of the terminals assigned to other drones in the target area on each channel.

[0020] According to an emergency disaster detection method provided by the present invention, the step of obtaining the SINR of each terminal for each drone on each channel includes:

[0021] When any terminal has been assigned to any drone, obtain the number of terminals assigned to the drone;

[0022] When the number of terminals is greater than the number of channels, the SINR of the terminal for the drone on each channel is obtained according to the interference of the terminal assigned to the drone on the terminal on each channel.

[0023] According to an emergency disaster detection method provided by the present invention, the steps of selecting multiple super-resolution network models on the drone based on the optimization model and reconstructing the sequence of disaster images to be detected include:

[0024] Acquire a transmission delay for unloading the sequence of disaster images to be detected and a calculation delay for performing emergency disaster detection using the sequence of disaster images to be detected;

[0025] Obtaining the reconstruction delay and image quality of reconstructing the sequence of disaster images to be detected using each super-resolution network model;

[0026] The sum of the transmission delay, calculation delay and reconstruction delay is used as the overall delay of disaster detection corresponding to the terminal;

[0027] A super-resolution network model whose image quality is greater than a second preset threshold and whose overall delay in disaster detection is less than a third preset threshold is selected to reconstruct the sequence of disaster images to be detected.

[0028] The present invention also provides an emergency disaster detection device, comprising:

[0029] A construction module is configured to obtain a sequence of disaster image sequences to be detected by each terminal in the target area, and offload the sequence of disaster image sequences to be detected to a drone in the target area. The drone provides computing power for detecting the sequence of disaster image sequences to be detected, and construct an optimization model to balance the accuracy and delay of detecting the sequence of disaster image sequences to be detected;

[0030] A first allocation module is configured to allocate a drone to each terminal based on the optimization model according to the channel conditions between each terminal and each drone, and the existing computing capacity of each drone, so as to achieve balanced clustering of the terminals and drones based on the channel conditions and the existing computing capacity;

[0031] A second allocation module is configured to establish a bipartite graph of the terminals and the channels in the target area based on the optimization model, allocate channels to each terminal using a KM matching algorithm so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold, and offload the sequence of disaster images to be detected on each terminal to the drone allocated to each terminal via the channel allocated to each terminal, thereby optimizing the channel allocated to each terminal;

[0032] A selection module is used to select multiple super-resolution network models on the drone based on the optimization model, reconstruct the sequence of disaster images to be detected, and use the reconstructed sequence of disaster images to be detected to perform disaster detection. The reconstructed image quality is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold, so as to achieve a balance between disaster detection accuracy and overall delay of disaster detection.

[0033] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the emergency disaster detection method as described above is implemented.

[0034] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described emergency disaster detection methods.

[0035] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned emergency disaster detection methods.

[0036] The emergency disaster detection method and device provided by the present invention first cluster terminal nodes according to channel conditions and the amount of computing tasks carried by drones to reduce node selection delay; secondly, the KM matching algorithm is used to optimize the allocation of spectrum resources in the network; finally, based on relevant constraints such as delay and accuracy, an appropriate super-resolution model is loaded for the sequence to be detected of each terminal to restore high-quality disaster visual information. While ensuring the accuracy of disaster detection, the completion delay of the detection task is minimized, thereby improving the completion efficiency of collaborative detection tasks and reducing system overhead. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0038] Figure 1 This is a schematic diagram of a UAV-assisted multi-node disaster awareness scenario based on a super-resolution network in the emergency disaster detection method provided by the present invention;

[0039] Figure 2 This is one of the flow charts of the emergency disaster detection method provided by the present invention;

[0040] Figure 3 It is a bipartite graph of spectrum resource allocation of terminal nodes in the emergency disaster detection method provided by the present invention;

[0041] Figure 4 This is the second flow chart of the emergency disaster detection method provided by the present invention;

[0042] Figure 5 This is a schematic diagram of the effect of clustering by considering only the distance factor in the emergency disaster detection method provided by the present invention;

[0043] Figure 6 This is a schematic diagram of the clustering effect based on distance and task calculation amount in the emergency disaster detection method provided by the present invention;

[0044] Figure 7 This is a simulation analysis diagram of system overhead under different terminal transmission powers and clustering schemes in the emergency disaster detection method provided by the present invention;

[0045] Figure 8 This is a simulation analysis diagram of system overhead under different terminal transmission powers and carrier allocation modes in the emergency disaster detection method provided by the present invention;

[0046] Figure 9This is a simulation analysis diagram of system overhead under different terminal transmission powers and super-resolution model selections in the emergency disaster detection method provided by the present invention;

[0047] Figure 10 It is a structural diagram of the emergency disaster detection device provided by the present invention;

[0048] Figure 11 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0049] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0050] Before introducing the embodiments of the present invention, the improvement ideas and system model of the present invention are first introduced.

[0051] In response to emergency scenarios where both spectrum and computing power are limited, the present invention considers analysis from two perspectives: alleviating computing power pressure and reducing communication burden.

[0052] On the one hand, in disaster scenarios, the environment is time-varying, making it difficult to rely on fixed nodes to provide computing power for mobile terminals performing disaster awareness tasks. Considering the low cost of building and scheduling drones, and their ability to form a flexible mobile edge computing network with ground-based terminal collection nodes, drones can effectively improve the timeliness of disaster data and information perception. Previous research has addressed the problem of minimizing energy consumption in drone-carried small-scale MEC (Multi-access Edge Computing) systems. This proposed an efficient iterative optimization algorithm that minimizes the total energy consumption of the drone by jointly optimizing variables such as the drone's flight time and task scheduling strategy. However, these studies all considered computational offloading scenarios with small data volumes, primarily non-computationally intensive tasks. When intelligent detection tasks, such as images and videos, are involved, the surge in network traffic leads to excessive transmission latency. This is particularly true in emergency scenarios where spectrum resources are scarce, where the conflict between latency-sensitive disaster information detection tasks and insufficient communication bandwidth is particularly acute.

[0053] Super-resolution reconstruction technology, on the other hand, can map low-resolution images to high resolution, reducing the amount of data transmitted over the network while also enhancing the visual quality of the images. This technology holds great potential for applications in both surveillance equipment and remote sensing imagery. Previous research has explored super-resolution reconstruction of the entire remote sensing image to detect small targets in remote sensing images, while also enhancing the edges of small targets to achieve high-precision detection. Another study, to overcome the low-resolution image and video quality caused by hardware limitations of terminal devices or bandwidth constraints during transmission, employed a super-resolution method based on compressed sensing theory to restore high-quality images, achieving promising results. However, this research lacked coupled optimization of super-resolution restoration with disaster detection latency and accuracy, failing to ensure that task latency requirements were met while maintaining accuracy. Therefore, further research is needed on collaborative computing across multiple nodes using image and video data as computational tasks.

[0054] In summary, in the scenario of UAV-assisted mobile terminal disaster perception, it is considered that the mobile terminal collects low-resolution disaster images and transmits them to the UAV, and deploys a super-resolution network model on the UAV to restore high-quality images, and then performs disaster perception to alleviate computing power and spectrum pressure. However, while collecting low-quality images to reduce transmission delay, it also reduces detection accuracy, and the greater the computational delay of the super-resolution model, the better the quality of the recovered data, which leads to the problem of balancing the delay in completing the disaster perception task and the detection accuracy. Therefore, the present invention addresses the problem of balancing detection delay and accuracy in the UAV-assisted super-resolution reconstruction disaster perception system, and proposes a computational offloading matching mechanism based on a super-resolution network. First, the collaborative computing nodes are clustered in a way that is based on the balance between channel conditions and the computational load of the UAV; secondly, a matching algorithm is used to allocate spectrum resources in the network, and finally, a suitable super-resolution model is loaded on the UAV end to restore high-quality images, thereby achieving the purpose of reducing system overhead and improving disaster perception efficiency.

[0055] like Figure 1 The scene shown in Figure 1 is a scene where multiple drones assist multiple mobile terminals in disaster detection. There are I collection terminals in the scene, and the collection terminal set is recorded as Take the center of the scene graph as the origin of the two-dimensional plane coordinates, and use d i =(x i ,y i ) represents the location of the i-th collection terminal. M drones are distributed above the disaster area, and the set Assuming that all drones are hovering at height z for auxiliary calculation, the two-dimensional coordinates of the mth drone are represented as u m =(x m ,y m ). Use α i(bits) represents the amount of video data collected by the i-th mobile terminal, and β i (cycles) represents the amount of computation required to deduce the video data collected by the i-th mobile terminal, and t i (s) represents the time delay limit for completing disaster information detection for the video data collected by the i-th mobile terminal. Assuming that each acquisition terminal node carries a disaster image sequence to be detected, considering that the mobile terminal collects a low-resolution disaster video sequence, the video sequence to be detected is unloaded to the drones distributed along the way to perform the disaster detection task. Before detection, K super-resolution network models can be loaded on the drone. The image restoration delay and accuracy corresponding to each model are different. i,m ∈{0,1} indicates whether the video sequence of the i-th acquisition terminal is unloaded to the m-th UAV. The mobile acquisition terminal uses Orthogonal Frequency Division Multiplexing (OFDM) to access the UAV, and the channel bandwidth is BMHz.

[0056] The communication model in the system model of this application is introduced below.

[0057] Assume that there are F available orthogonal spectrum resources in the system, and different terminals and drones have channel competition and multiplexing relationships in the communication links. The distance between the i-th terminal and the m-th drone is: The distance between the i-th acquisition terminal and the m-th UAV is defined as the channel gain on the f-th (f=1,2,...,F) channel is expressed as:

[0058]

[0059] in, represents the large-scale channel power enhancement, l0 represents the channel gain when the reference distance is 1 meter, θ represents the channel fading coefficient, represents the small-scale fading coefficient, which obeys the Rice channel fading model.

[0060] The signal to interference plus noise ratio (SINR) between the i-th acquisition terminal and the m-th UAV on the f-th channel is expressed as:

[0061]

[0062] Among them, q i,f ∈{0,1} indicates whether the i-th acquisition terminal unloads the data to be detected on the f-th channel, and p indicates the transmission power of the acquisition terminal. is the gain between the i-th terminal and the m-th UAV on the f-th channel, σ2 is the white noise power, The channel gain of the nth acquisition terminal and the mth UAV transmitting the sequence to be detected on the fth channel.

[0063] Therefore, the uplink data transmission rate of the disaster video unloaded from the i-th disaster collection terminal to the m-th UAV on the f-th channel is as follows:

[0064]

[0065] The calculation model in the system model of this application is introduced below.

[0066] In this scenario, the acquisition terminal is responsible for data collection and needs to offload all sequences to be detected to the edge drone to perform the disaster detection task. In order to improve spectrum utilization and reduce the transmission delay of the data to be detected, the terminal collects low-resolution disaster sequences. The transmission delay of the sequence to be detected from the i-th acquisition terminal to the m-th drone is expressed as:

[0067]

[0068] In order to improve the quality of the sequence to be detected and enhance the detection accuracy, we consider adaptively loading the kth (k=1,...,K) super-resolution model on the edge drone, and using the Peak Signal to Noise Ratio (PSNR) of the image as the image quality indicator. Assuming that the optimal PSNR value of the restored image is ξ, the variable for selecting the kth super-resolution network model for quality enhancement of the sequence to be detected of the i-th terminal is expressed as ω i,k ∈{0,1},ω i,k =1 means that the sequence to be detected of the i-th terminal selects the k-th super-resolution network model for high-quality recovery, ω i,k = 0 means that the kth super-resolution network is not selected, and the quality of the recovery is expressed as The required recovery time is denoted as T i k .

[0069] The computing power of the drone's server is r sever , then the computational delay of the mth UAV performing intelligent disaster detection on the i-th sequence to be detected is expressed as:

[0070]

[0071] When a single drone receives multiple sequences to be detected, considering the serial processing of tasks, the calculation order of detection tasks is consistent with the order of unloading to the drone, and obeys the principle of first-come-first-served. The overall completion delay of intelligent disaster detection of the i-th sequence to be detected on the m-th drone can be calculated as:

[0072]

[0073] In order to balance the task completion delay and task deduction accuracy, the system overhead is defined as the product of the detection delay and accuracy of the disaster awareness task of all mobile data collection terminals, that is:

[0074]

[0075] The following introduces the problem model in the system model of this application.

[0076] Based on the above communication model and computing model, with the goal of minimizing the system overhead of multi-UAV-assisted multi-mobile terminals for disaster awareness, the offloading node in the system is selected as x i,m , spectrum allocation q i,f and super-resolution model selection ω i,k The equivariable model is the following optimization problem:

[0077]

[0078]

[0079]

[0080]

[0081]

[0082]

[0083] In the above optimization problem P1, the optimization variables include the unloading node variable x i,m , network spectrum allocation variable q i,f , and super-resolution model variables ω i,k Constraint C1 represents the binary decision allocation of spectrum in the network, constraint C2 represents the time delay limit for the detection of the sequence to be detected by the i-th acquisition terminal, constraint C3 represents the task detection accuracy value of the i-th acquisition terminal should be greater than the set threshold, and constraint C4 represents the signal to noise ratio of the i-th acquisition terminal. It should be greater than the minimum threshold γ0 defined by the transmission. The constraint C5 indicates that the binary offloading decision is executed between the sequence to be detected of the i-th collection terminal and the m-th drone. For the above problem, it involves selecting the offloading node x i,m , spectrum allocation q i,f And super-resolution model selection ω i,k The joint decision of is a mixed integer linear programming problem.

[0084] The following combination Figure 2The present invention describes an emergency disaster detection method, comprising: step 201, obtaining a sequence of disaster image sequences to be detected by each terminal in a target area, and unloading the sequence of disaster image sequences to be detected to a drone in the target area, wherein the drone provides computing power for detecting the sequence of disaster image sequences to be detected, and constructing an optimization model to balance the accuracy and delay of detecting the sequence of disaster image sequences to be detected;

[0085] Step 202: Based on the optimization model and according to the channel conditions between each terminal and each drone, as well as the existing computing capacity of each drone, a drone is assigned to each terminal, so as to achieve balanced clustering of the terminals and drones based on the channel conditions and the existing computing capacity.

[0086] In order to reduce inter-link interference and improve computing resource utilization, this embodiment allocates each terminal to a corresponding drone based on a clustering solution based on channel conditions and computing load balancing.

[0087] Step 203: Based on the optimization model, a bipartite graph of the terminals and the channels in the target area is established. A KM matching algorithm is used to allocate a channel to each terminal so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold. The sequence of disaster images to be detected on each terminal is offloaded to the UAV assigned to each terminal via the channel assigned to each terminal, thereby optimizing the channel assigned to each terminal.

[0088] Assume that a channel is multiplexed by at most two terminals at the same time, and each terminal can only occupy one channel when transmitting data, and consider the signal-to-interference-noise ratio of the current occupied channel. Constraints are imposed on the selection of multiplexing channels.

[0089] The mobile terminal access channel variable is q i,f When the i-th terminal requests to communicate with the UAV, it first monitors whether there are idle spectrum resources in the system. If there are idle channels, it directly accesses them. Let q i,f =1, indicating that the i-th terminal occupies the f-th channel.

[0090] Since the number of mobile terminal devices is much greater than the number of channels in the system, when there are no idle channel resources in the system, the mobile terminal needs to select a multiplexed channel.

[0091] The mapping relationship between mobile terminal devices and spectrum resources in the network is represented in the form of a bipartite graph, and multiple mobile terminal devices associated with the same spectrum resource perform spectrum reuse, such as Figure 3 After using the KM matching algorithm to select different spectrums for the terminal to obtain the corresponding rate, the uplink transmission rate of all terminals in the network is maximized and the final matching result is output as the target.i,f Optimization maximizes the system's uplink transmission rate and reduces transmission delay.

[0092] In step 204, multiple super-resolution network models on the drone are selected based on the optimization model, the sequence of disaster images to be detected is reconstructed, and disaster detection is performed using the reconstructed sequence of disaster images to be detected. The reconstructed image quality is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold, thereby achieving a balance between disaster detection accuracy and overall delay.

[0093] After channel selection is complete, a suitable super-resolution model is selected for different tasks based on the current latency requirements to restore high-quality images, ensuring detection accuracy while minimizing the latency required to complete the detection task. Detection is performed using the restored disaster image sequence to be detected, using the target detection model YOLO (You Only Look Once) or R-CNN (Region Convolutional Neural Network). This embodiment does not limit the detection method.

[0094] Restoring images of the same resolution using different super-resolution models requires different computational resources and image restoration quality. Higher image quality leads to greater detection accuracy. Therefore, the solution in this embodiment is to select the optimal super-resolution network model based on the accuracy and latency constraints of the detection task to restore high-quality images and improve decision accuracy. This approach also considers the computational requirements of the super-resolution model and minimizes the time required to restore high-resolution images.

[0095] This embodiment first clusters terminal nodes based on channel conditions and the amount of computing tasks carried by drones to reduce node selection delay. Secondly, the KM matching algorithm is used to optimize the allocation of spectrum resources in the network. Finally, based on relevant constraints such as delay and accuracy, an appropriate super-resolution model is loaded for the sequence to be detected for each terminal to restore high-quality disaster visual information. While ensuring the accuracy of disaster detection, the completion delay of the detection task is minimized, thereby improving the completion efficiency of collaborative detection tasks and reducing system overhead.

[0096] Based on the above embodiment, in this embodiment, the step of allocating a drone to each terminal based on the optimization model according to the channel conditions between each terminal and each drone, and the existing computing capacity of each drone includes:

[0097] Obtaining a first distance between each terminal of each unassigned drone and each drone, and a second distance between each terminal and each drone assigned to the terminal, and using the first distance and the second distance as the channel condition;

[0098] This embodiment takes into account the impact of the distance between terminals and drones, and between terminals on the link, as well as the amount of computing tasks carried by drones, and proposes a clustering solution based on channel conditions and computing load balancing.

[0099] First, consider a line-of-sight channel between the terminal and the drone. According to Shannon's formula, channel transmission distance is an important factor affecting the transmission rate. Therefore, based on the position between the terminal and the drone, the Euclidean distance from the terminal I to the M drones in the network is calculated:

[0100]

[0101] Secondly, consider the load balancing problem of the computing resources of drones in the network, that is, to offload the computing tasks to each drone as evenly as possible to ensure the efficient use of computing resources. The computing amount of the detection task of the i-th terminal is expressed as β i The clustering basis is represented as the channel transmission distance between the i-th terminal and the m-th UAV and the existing computational load β on the m-th UAV. m The weighted sum of is:

[0102] η i,m =λ d l i,m +λ com β m (1-10)

[0103] The first distance, the second distance, and the existing calculation amount corresponding to each drone are weighted and added together, and the drone with the smallest addition result is allocated to the terminal of the unassigned drone.

[0104] Since the interference generated by the farther terminal is smaller, the distance factor between the user terminal nodes is considered to reduce the impact of inter-cluster interference. The distance between the i-th node and the rest of the nodes assigned to the drone is expressed as δ i,j =-||d i -d j || indicates that the above formula can be rewritten as:

[0105] η i,m =λ d l i,m +λ com β m +λ δ δ i,j (1-11)

[0106] Among them, λ d ,λ com and λ δ They represent the weight of the distance between the UAV and the terminal, the weight of the distance between the terminals, and the weight of the computational effort on the UAV.

[0107] Optionally, δ i,j It is the negative value of the maximum value of the distance between the i-th node and each of the remaining nodes assigned under the drone, and the maximum value corresponds to the j-th node assigned under the drone.

[0108] According to the above clustering scheme, it is determined to which drone node the sequence to be detected on the terminal should be offloaded to perform the disaster detection task.

[0109] Based on the above embodiment, the step of using the KM matching algorithm to allocate a channel to each terminal in this embodiment includes:

[0110] When the number of the terminals is less than the number of channels in the network, obtaining a SINR for each terminal on each channel for each drone, and obtaining a link uplink transmission rate for each terminal on each channel for each drone based on the SINR;

[0111] Add up the link uplink transmission rates of each terminal for each UAV on each channel to obtain the sum of the link uplink transmission rates of all terminals;

[0112] Taking the sum of the link uplink transmission rates of all mobile terminals in the system as an indicator, The transmission quality constraint maximizes the overall communication performance, that is:

[0113]

[0114] stC1,C4 (1-12a)

[0115] The terminal and the channel are taken as the matching parties, the link uplink transmission rate of each terminal for each drone on each channel is taken as the matching edge weight, and the KM matching algorithm is used to allocate the channel to each terminal, so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal for each drone on the allocated channel is greater than the first preset threshold.

[0116] This embodiment solves the aforementioned optimization problem using the KM algorithm within the maximum weighted sum matching algorithm. The terminal node and the spectrum resources in the network represent the two matching parties. The matching edge weight represents the link transmission rate. After the terminal selects different spectrums to obtain the corresponding rates, the final matching result is output with the goal of maximizing the sum of the uplink transmission rates of all terminals in the network. By optimizing the spectrum allocation variables, the system's uplink transmission rate is maximized and transmission latency is reduced.

[0117] Based on the above embodiment, the step of obtaining the SINR of each terminal for each drone on each channel in this embodiment includes:

[0118] When any terminal has been assigned to any drone, the SINR of the terminal for the drone on each channel is obtained based on the interference of the terminal assigned to the drone on each channel and the interference of the terminals assigned to other drones in the target area on each channel.

[0119] In this embodiment, for clustered drones and terminal devices, the interference term of the uplink transmission link SINR can be rewritten as the sum of intra-cluster interference and inter-cluster interference, that is, the interference generated by the link between the nth terminal and the mth drone on channel f to the terminal i, and the interference generated by the link between the jth terminal and the sth drone on channel f to the terminal i:

[0120]

[0121] Based on the above embodiment, the step of obtaining the SINR of each terminal for each drone on each channel in this embodiment includes:

[0122] When any terminal has been assigned to any drone, obtain the number of terminals assigned to the drone;

[0123] When the number of terminals is greater than the number of channels, the SINR of the terminal for the drone on each channel is obtained according to the interference of the terminal assigned to the drone on the terminal on each channel.

[0124] For F channels in the system, if the number of terminals in a single cluster is less than the number of channels, there is no multiplexing, that is, the terminals in the cluster use orthogonal spectrum.

[0125] If the number of terminals in a single cluster is greater than the number of channels, there is intra-cluster reuse, and the interference caused by intra-cluster terminals is much greater than that caused by inter-cluster terminals. Ignoring the impact of inter-cluster interference, only considering the interference caused by mobile terminals in the same cluster reusing the same channel, the above formula can be rewritten as:

[0126]

[0127] Based on the above embodiments, in this embodiment, the steps of selecting multiple super-resolution network models on the drone based on the optimization model and reconstructing the sequence of disaster images to be detected include:

[0128] Acquire a transmission delay for unloading the sequence of disaster images to be detected and a calculation delay for performing emergency disaster detection using the sequence of disaster images to be detected;

[0129] Obtaining the reconstruction delay and image quality of reconstructing the sequence of disaster images to be detected using each super-resolution network model;

[0130] The sum of the transmission delay, calculation delay and reconstruction delay is used as the overall delay of disaster detection corresponding to the terminal;

[0131] A super-resolution network model whose image quality is greater than a second preset threshold and whose overall delay in disaster detection is less than a third preset threshold is selected to reconstruct the sequence of disaster images to be detected.

[0132] For example, four super-resolution network models were used to perform super-resolution restoration on images at three resolutions, yielding the corresponding image quality and reconstruction time for 8K, 4K, and 2K images after super-resolution. Tables 1 through 3 show the time required for super-resolution reconstruction and the quality of the restored images for the received images at different resolutions.

[0133] Table 1 Time required to restore 8K images and quality of reconstructed images

[0134] Model FSRCNN CARN SRResNet RCAN Super-resolution reconstruction time (S) 0.09 0.22 1.1 6.4 Image quality (dB) 32.66 33.18 33.5 33.76

[0135] Table 2 Time required to restore 4K images and quality of reconstructed images

[0136] Model FSRCNN CARN SRResNet RCAN Super-resolution reconstruction time (S) 0.09 0.22 1.1 6.4 Image quality (dB) 26.91 27.34 27.65 27.89

[0137] Table 3 Time required to restore 2K image and quality of reconstructed image

[0138] Model FSRCNN CARN SRResNet RCAN Super-resolution reconstruction time (S) 0.09 0.22 1.1 6.4 Image quality (dB) 25.61 25.95 26.19 26.39

[0139] Based on the above experimental data, the corresponding relationship between the recovery time of the super-resolution model and the visual data quality can be obtained. For each disaster image sequence to be detected in the system, the transmission delay required for offloading and the computational delay required for detection are first calculated:

[0140]

[0141] When the UAV performs super-resolution reconstruction on the sequence to be detected, the reconstruction delay is T i k ,when When the delay constraint is met, the super-resolution model that meets the delay constraint is obtained. In the super-resolution network model that meets the delay constraint, a network that meets the accuracy constraint is selected. When When it is established, the model that meets both the delay and accuracy constraints is selected, and then the detection accuracy and delay are minimized. The appropriate super-resolution model is selected to restore high-quality disaster visual data for detection, so as to achieve the purpose of improving detection accuracy.

[0142] In this embodiment, the optimization variables include the unloading node variable x i,m Selection of network spectrum allocation variable q i,f , and super-resolution model selection variable ωi,k The analysis of these sub-problems is carried out, and the above variables are jointly optimized under the constraints of communication quality, disaster awareness task completion delay, detection accuracy, etc., in order to minimize the system overhead. Figure 4 As shown in the figure, first, the information such as the drone position, terminal position, and task attributes are input, the weight coefficient of the clustering scheme is initialized, and the number of iterations is set; secondly, a clustering scheme based on channel conditions and computational load balancing is used, and each drone in the network is used as a cluster head. The mobile terminal nodes are clustered considering the impact of the distance between the terminal and the drone, and between the terminals on the link, as well as the amount of computing tasks carried by the drone; thirdly, the overall uplink transmission rate is defined as the edge weight and the KM matching algorithm is used to allocate spectrum resources to minimize transmission delay and improve spectrum utilization; finally, according to relevant constraints such as delay and accuracy, a suitable super-resolution model is loaded for the sequence to be detected for each terminal to restore high-quality disaster visual information, improve the accuracy of disaster detection, and minimize the completion delay of the detection task.

[0143] The pseudo code of the computation offloading matching mechanism based on the super-resolution network provided in this embodiment is as follows:

[0144]

[0145]

[0146] The solution proposed in this example was simulated using the Python 3.6 software platform, using an Intel i5-10210U CPU @ 1.60GHz processor. Three drones were deployed in the air to receive data collected from the ground and assist 15 ground terminal collection nodes in intelligent detection. The terminal node transmit power was a maximum of 400mW, the CPU speed of the MEC carried by the drones was 500 cycles / s, the noise power in the network was -100dBm, the transmission bandwidth was 2MHz, and the channel gain was determined based on the positions of the drones and the terminal nodes. Unless otherwise specified, all simulation parameters in this example were set according to Table 4.

[0147] Table 4 Simulation parameters

[0148]

[0149]

[0150] This embodiment uses a multi-node computing offloading matching mechanism assisted by a super-resolution network to perform task deduction of the edge collaborative architecture. First, consider the impact of different clustering algorithms and compare the following two solutions after receiving the image:

[0151] (1) Distance-based clustering: Considering the distance between the terminal node and the drone, the terminal node and the drone node are clustered.

[0152] (2) Clustering based on distance and task computation: While considering the distance between the terminal node and the UAV, the distance effect between users is introduced, and clustering is performed considering the balanced task computation on the UAV.

[0153] Figure 5 and Figure 6 The horizontal axis represents the X-axis coordinate of the node in the two-dimensional coordinate system, with a value between [0,12]m. The vertical axis represents the Y-axis coordinate of the node in the two-dimensional coordinate system, with a value between [0,15]m. Points of the same shape are classified into the same cluster. Figure 5 For the first option, Figure 6 This is the second solution. By comparison, it can be seen that the solution that only considers distance factors for clustering is likely to cause multiple nodes to offload detection tasks to one drone, causing the drone's server to be overloaded, reducing system resource utilization and increasing overhead.

[0154] The clustering scheme that balances the task computation and channel conditions takes into account both communication link interference and computation load balancing, effectively solving the problem of excessive computation load caused by offloading multiple computation tasks to the same UAV and reducing system overhead. Figure 7 As shown in the figure, the horizontal axis represents the transmit power of the ground collection node, ranging from [0 to 90] mW, and the vertical axis represents the system overhead at different power levels. As can be seen, the greater the power level, the greater the gain achieved by the clustering scheme, which balances computational effort with channel conditions. When the power reaches 80 mW, the gain reaches 12.1%.

[0155] Next, consider the system gains brought about by different carrier allocation methods. The comparison schemes are as follows:

[0156] (1) Transmitting high-quality original images based on KM matching: The terminal transmits high-quality original images and uses the KM algorithm for carrier allocation.

[0157] (2) Transmitting high-quality original images based on random matching: The terminal transmits high-quality original images and performs random matching carrier allocation.

[0158] (3) 8K super-resolution recovery based on KM matching transmission of low-quality images: The terminal uses low-quality images, uses the KM algorithm for carrier allocation, and uses the super-resolution network to improve the image quality to 8K for disaster detection.

[0159] (4) 8K super-resolution recovery based on random matching transmission of low-quality images: The terminal receives low-quality images and performs random matching carrier allocation. The image quality is improved to 8K through the super-resolution network for disaster detection.

[0160] Figure 8 The horizontal axis represents the different transmission powers adopted by the terminal node, with values ​​ranging from [0,400] mW, and the vertical axis represents the system overhead. Taking the transmission of the original image plus random matching as a benchmark, the system overheads of the original image plus KM matching, the super-resolution restoration of the 8K image plus KM matching, and the super-resolution restoration of the 8K image plus random matching are compared. The results show that when the transmission power is 350 mW, the carrier allocation mechanism based on KM matching proposed in the present invention aims to maximize the uplink transmission rate, effectively improving the spectrum utilization and reducing the system overhead caused by intelligent task detection. On the original image transmission curve, the reduction is about 24.1%, and on the transmission curve of the super-resolution restoration of the 8K image, the reduction is about 16.7%. Therefore, under the conditions of different transmission powers, the algorithm proposed in this embodiment can effectively reduce the system overhead.

[0161] Finally, consider the gain effect of selecting a suitable super-resolution model when receiving low-quality disaster images. The comparison scheme is as follows:

[0162] (1) Transmit high-quality original images for detection based on KM matching: Transmit high-quality original images and perform intelligent detection tasks directly on the drone side.

[0163] (2) Low quality transmission based on KM matching Figure 8 Detection after K super-resolution restoration: Transmit low-quality images, use the super-resolution model to improve the image quality to 8K, and then perform intelligent detection tasks.

[0164] (3) Low quality transmission based on KM matching Figure 4 Detection after K super-resolution restoration: Transmit low-quality images, use the super-resolution model to improve the image quality to 4K, and then perform intelligent detection tasks.

[0165] (4) Low quality transmission based on KM matching Figure 2 Detection after K super-resolution restoration: Transmit low-quality images, use the super-resolution model to improve the image quality to 2K, and then perform intelligent detection tasks.

[0166] Figure 9The horizontal axis represents the different transmission powers of the terminal node, with values ​​ranging from [0,400] mW, and the vertical axis represents the system overhead. Taking the transmission of the original image and the direct intelligent task deduction on the drone end as the benchmark, the different system overheads of the intelligent task deduction schemes after transmitting low-quality images and super-resolving and restoring 8K, 4K, and 2K images are compared. The results show that the deduction task offloading mechanism based on the super-resolution network proposed in the present invention effectively reduces the transmission delay based on the idea of ​​reducing the amount of transmitted data. At the same time, the super-resolution model is used to enhance the quality of low-quality disaster visual data, thereby improving the accuracy of disaster detection. When the transmission power is 10 mW, the comparison between the transmission of the original image and the super-resolved restoration of the 2K image curve is reduced by about 56.1%. When the power is 200 mW, the comparison between the transmission of the original image and the super-resolved restoration of the 8K image curve is reduced by about 15.2%. Therefore, under the conditions of different transmission powers, the algorithm proposed in this embodiment can effectively reduce system overhead.

[0167] This embodiment supports the collaborative deduction of disaster detection tasks in emergency scenarios through real-time scheduling of network transmission and computing resources, effectively improving disaster awareness efficiency. In an edge architecture with multiple acquisition terminals and multiple drones, this embodiment first proposes a clustering scheme based on channel conditions and balanced task computing load. When the power reaches 80mW, a gain of 12.1% is achieved. Secondly, a spectrum resource allocation scheme based on the KM matching algorithm is proposed, aiming to maximize the uplink transmission rate. This effectively improves spectrum utilization and reduces the system overhead caused by intelligent task detection by approximately 24.1%. Finally, a scheme comparing the transmission of the original image with the super-resolution restoration of a low-quality image shows a reduction of approximately 56.1% when the transmission power is 10mW for the original image and the super-resolution restoration of the 2K image. At 200mW, a reduction of approximately 15.2% is achieved when the transmission power is 200mW for the original image and the super-resolution restoration of the 8K image.

[0168] The emergency disaster situation detection device provided by the present invention is described below. The emergency disaster situation detection device described below and the emergency disaster situation detection method described above can be referenced to each other.

[0169] like Figure 10 As shown, the apparatus includes a construction module 1001, a first allocation module 1002, a second allocation module 1003 and a selection module 1004, wherein:

[0170] The construction module 1001 is used to obtain a sequence of disaster image sequences to be detected by each terminal in the target area, and unload the sequence of disaster image sequences to be detected to a drone in the target area. The drone provides computing power for detecting the sequence of disaster image sequences to be detected, and constructs an optimization model to balance the accuracy and delay of detecting the sequence of disaster image sequences to be detected.

[0171] The first allocation module 1002 is used to allocate a drone to each terminal based on the channel conditions between each terminal and each drone in the target area and the existing computing capacity of each drone;

[0172] The second allocation module 1003 is configured to allocate a channel of the network in the target area to each terminal using a KM matching algorithm, so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold, and to offload the disaster image sequence to be detected on each terminal to the UAV allocated to each terminal through the channel allocated to each terminal;

[0173] The selection module 1004 is used to select the super-resolution network model on the drone for each terminal to reconstruct the disaster image sequence to be detected, use the reconstructed disaster image sequence to be detected to perform disaster detection, and the reconstructed image quality is greater than the second preset threshold and the overall disaster detection delay corresponding to the terminal is less than the third preset threshold.

[0174] This embodiment first clusters terminal nodes based on channel conditions and the amount of computing tasks carried by drones to reduce node selection delay. Secondly, the KM matching algorithm is used to optimize the allocation of spectrum resources in the network. Finally, based on relevant constraints such as delay and accuracy, an appropriate super-resolution model is loaded for the sequence to be detected for each terminal to restore high-quality disaster visual information. While ensuring the accuracy of disaster detection, the completion delay of the detection task is minimized, thereby improving the completion efficiency of collaborative detection tasks and reducing system overhead.

[0175] Figure 11 An example of a physical structure diagram of an electronic device is shown below. Figure 11As shown, the electronic device may include: a processor (processor) 1110, a communication interface (Communications Interface) 1120, a memory (memory) 1130 and a communication bus 1140, wherein the processor 1110, the communication interface 1120, and the memory 1130 communicate with each other through the communication bus 1140. The processor 1110 can call the logic instructions in the memory 1130 to execute the emergency disaster detection method, which includes: allocating a drone to each terminal based on the channel conditions between each terminal and each drone in the target area, and the existing computing power of each drone; using the KM matching algorithm to allocate a channel to each terminal so that the link uplink transmission rate of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold, and unloading the disaster image sequence to be detected on each terminal to the drone allocated to each terminal through the channel allocated to each terminal; selecting a super-resolution network model on the drone for each terminal to reconstruct the disaster image sequence to be detected, and using the reconstructed disaster image sequence to be detected to perform disaster detection, so that the reconstructed image quality is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold.

[0176] In addition, the logic instructions in the above-mentioned memory 1130 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the 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 enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0177] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the emergency disaster detection method provided by the above methods, which includes: allocating a drone to each terminal based on the channel conditions between each terminal and each drone in the target area, and the existing computing power of each drone; using the KM matching algorithm to allocate a channel to each terminal so that the link uplink transmission rate of all terminals is maximum and the SINR of each terminal on the allocated channel is greater than a first preset threshold, and unloading the disaster image sequence to be detected on each terminal to the drone allocated to each terminal through the channel allocated to each terminal; selecting a super-resolution network model on the drone for each terminal to reconstruct the disaster image sequence to be detected, and using the reconstructed disaster image sequence to be detected to perform disaster detection, so that the reconstructed image quality is greater than the second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold.

[0178] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the emergency disaster detection method provided by the above-mentioned methods, the method comprising: allocating a drone to each terminal based on the channel conditions between each terminal and each drone in the target area, and the existing computing power of each drone; using the KM matching algorithm to allocate a channel to each terminal so that the link uplink transmission rate of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold, and the sequence of disaster images to be detected on each terminal is unloaded to the drone allocated to each terminal through the channel allocated to each terminal; selecting a super-resolution network model on the drone for each terminal to reconstruct the sequence of disaster images to be detected, and using the reconstructed sequence of disaster images to be detected to perform disaster detection, so that the quality of the reconstructed image is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold.

[0179] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e., they may be located in one location or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of the present embodiment. Persons of ordinary skill in the art will be able to understand and implement the present invention without inventive effort.

[0180] 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, or of course, by hardware. Based on this understanding, the above technical solution, in essence, 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, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or certain parts of the embodiments.

[0181] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for detecting emergency disasters, characterized in that: include: Each terminal in the target area acquires a sequence of images of the disaster condition to be detected and offloads the sequence to a drone in the target area. The drone provides computing power for detecting the sequence of images of the disaster condition to be detected and constructs an optimization model to balance the accuracy and latency of detecting the sequence of images of the disaster condition to be detected. Based on the optimization model, according to the channel conditions between each terminal and each drone, and the existing computing capacity of each drone, a drone is assigned to each terminal, so as to achieve balanced clustering of the terminals and drones based on the channel conditions and the existing computing capacity; The step of assigning a drone to each terminal based on the optimization model according to the channel condition between each terminal and each drone, and the existing computing capacity of each drone includes: obtaining a first distance between each terminal of an unassigned drone and each drone, and a second distance between each terminal and the terminal assigned to each drone, and using the first distance and the second distance as the channel condition; weighting the first distance, the second distance, and the existing computing capacity corresponding to each drone, and adding them together, and assigning the drone with the smallest sum to the terminal of the unassigned drone; Based on the optimization model, a bipartite graph of the terminals and the channels in the target area is established, and a KM matching algorithm is used to allocate channels to each terminal so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold. The sequence of disaster images to be detected on each terminal is unloaded to the drone allocated to each terminal through the channel allocated to each terminal, thereby optimizing the channel allocated to each terminal. The step of allocating channels to each terminal using the KM matching algorithm includes: when the number of terminals is less than the number of channels in the network, obtaining the SINR of each terminal for each drone on each channel, and obtaining the link uplink transmission rate of each terminal for each drone on each channel based on the SINR; adding the link uplink transmission rates of each terminal for each drone on each channel to obtain the sum of the link uplink transmission rates of all terminals; using the terminal and the channel as the two matching parties, using the link uplink transmission rate of each terminal for each drone on each channel as the matching edge weight, and using the KM matching algorithm to allocate the channel to each terminal so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal for each drone on the allocated channel is greater than the first preset threshold. Based on the optimization model, multiple super-resolution network models on the drone are selected to reconstruct the disaster image sequence to be detected, and the reconstructed disaster image sequence to be detected is used for disaster detection. The reconstructed image quality is greater than a second preset threshold and the overall disaster detection delay corresponding to the terminal is less than a third preset threshold, so as to achieve a balance between disaster detection accuracy and overall disaster detection delay; the step of selecting multiple super-resolution network models on the drone based on the optimization model to reconstruct the disaster image sequence to be detected includes: obtaining the transmission delay of unloading the disaster image sequence to be detected and the calculation delay of using the disaster image sequence to be detected for emergency disaster detection; obtaining the reconstruction delay and image quality of reconstructing the disaster image sequence to be detected using each super-resolution network model; taking the sum of the transmission delay, calculation delay and reconstruction delay as the overall disaster detection delay corresponding to the terminal; selecting the super-resolution network model whose image quality is greater than the second preset threshold and whose overall disaster detection delay is less than the third preset threshold, and reconstructing the disaster image sequence to be detected.

2. The emergency disaster detection method according to claim 1, characterized in that: The step of obtaining the SINR of each terminal for each drone on each channel includes: When any terminal has been assigned to any drone, the SINR of the terminal for the drone on each channel is obtained based on the interference of the terminal assigned to the drone on each channel and the interference of the terminals assigned to other drones in the target area on each channel.

3. The emergency disaster detection method according to claim 1, characterized in that: The step of obtaining the SINR of each terminal for each drone on each channel includes: When any terminal has been assigned to any drone, obtain the number of terminals assigned to the drone; When the number of terminals is greater than the number of channels, the SINR of the terminal for the drone on each channel is obtained according to the interference of the terminal assigned to the drone on the terminal on each channel.

4. An emergency disaster detection device, characterized in that: include: A construction module is configured to obtain a sequence of disaster image sequences to be detected by each terminal in the target area, and offload the sequence of disaster image sequences to be detected to a drone in the target area. The drone provides computing power for detecting the sequence of disaster image sequences to be detected, and construct an optimization model to balance the accuracy and delay of detecting the sequence of disaster image sequences to be detected; A first allocation module is configured to allocate a drone to each terminal based on the optimization model according to the channel conditions between each terminal and each drone, and the existing computing capacity of each drone, so as to achieve balanced clustering of the terminals and drones based on the channel conditions and the existing computing capacity; The method of allocating a drone to each terminal based on the optimization model according to the channel condition between each terminal and each drone, and the existing computing capacity of each drone, includes: obtaining a first distance between each terminal of an unassigned drone and each drone, and a second distance between each terminal of an unassigned drone and each drone, and using the first distance and the second distance as the channel condition; weighting and adding the first distance, the second distance, and the existing computing capacity corresponding to each drone, and allocating the drone with the smallest sum to the terminal of the unassigned drone; The second allocation module is used to establish a bipartite graph of the terminals and the channels in the target area based on the optimization model, and use the KM matching algorithm to allocate channels to each terminal so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal on the allocated channel is greater than a first preset threshold, and the sequence of disaster images to be detected on each terminal is unloaded to the drone allocated to each terminal through the channel allocated to each terminal, so as to optimize the channel allocated to each terminal; the use of the KM matching algorithm to allocate channels to each terminal includes: when the number of the terminals is less than the number of channels in the network, obtaining the number of each terminal for each drone on each channel; The method comprises the following steps: calculating the SINR on each channel, obtaining the link uplink transmission rate of each terminal for each drone on each channel according to the SINR; adding the link uplink transmission rates of each terminal for each drone on each channel to obtain the sum of the link uplink transmission rates of all terminals; taking the terminal and the channel as the two matching parties, taking the link uplink transmission rate of each terminal for each drone on each channel as the matching edge weight, and using the KM matching algorithm to allocate the channel to each terminal, so that the sum of the link uplink transmission rates of all terminals is maximized and the SINR of each terminal for each drone on the allocated channel is greater than the first preset threshold; A selection module is configured to select multiple super-resolution network models on the drone based on the optimization model, reconstruct the sequence of disaster images to be detected, and use the reconstructed sequence of disaster images to be detected for disaster detection. The quality of the reconstructed image is greater than a second preset threshold and the overall delay of disaster detection corresponding to the terminal is less than a third preset threshold, thereby achieving a balance between disaster detection accuracy and overall delay of disaster detection. The selection of multiple super-resolution network models on the drone based on the optimization model and reconstruction of the sequence of disaster images to be detected includes: obtaining the transmission delay of unloading the sequence of disaster images to be detected and the calculation delay of using the sequence of disaster images to be detected for emergency disaster detection; obtaining the reconstruction delay and image quality of reconstructing the sequence of disaster images to be detected using each super-resolution network model; taking the sum of the transmission delay, calculation delay and reconstruction delay as the overall delay of disaster detection corresponding to the terminal; selecting a super-resolution network model whose image quality is greater than the second preset threshold and whose overall delay of disaster detection is less than the third preset threshold, and reconstructing the sequence of disaster images to be detected.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the emergency disaster detection method according to any one of claims 1 to 3 is implemented.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the emergency disaster detection method according to any one of claims 1 to 3 is implemented.

7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the emergency disaster detection method according to any one of claims 1 to 3 is implemented.

Images