A method and system for disaster monitoring based on deep learning of a UAV cluster cooperation

Abstract

The application discloses a kind of based on deep learning's unmanned aerial vehicle cluster coordination disaster monitoring method and system, method includes: step S1, the performance parameter data of each unmanned aerial vehicle in unmanned aerial vehicle cluster is obtained and is classified, and initial ability grade division is obtained;Step S2, disaster area is divided, based on the initial ability grade division result, using deep learning algorithm, initial task allocation is carried out to unmanned aerial vehicle cluster in the disaster area;Step S3, based on the disaster area, the real-time change information of each region's disaster environment is obtained, whether the initial task allocation is adjusted based on the real-time change information of the disaster environment;Step S4, based on the task of redistribution, the coordination disaster monitoring of unmanned aerial vehicle cluster is realized.The application is by establishing unmanned aerial vehicle ability database in advance, different performance index is classified and processed, and initial ability grade division result is determined.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle cooperative monitoring, in particular to a method and system for unmanned aerial vehicle cluster cooperative disaster monitoring based on deep learning. BACKGROUND

[0002] The application of unmanned aerial vehicle cluster in the field of disaster monitoring has vital value. It can quickly cover a large area and obtain real-time disaster information, providing indispensable support for rescue decision-making. With the frequent occurrence of natural disasters, how to efficiently utilize unmanned aerial vehicle cluster for cooperative monitoring has become an important direction to improve emergency response capability. Research in this field not only concerns technological innovation, but also directly relates to life and property safety, showing high social significance.

[0003] However, the current application of unmanned aerial vehicle cluster in disaster monitoring still faces many deficiencies. Existing methods often ignore the differences in individual capabilities of unmanned aerial vehicles, lack fine-grained management of resources and tasks within the cluster, resulting in task overload for some unmanned aerial vehicles and idle resources for others in complex disaster scenarios. In addition, in the face of dynamically changing disaster environments, existing solutions perform poorly in task coordination and real-time adjustment, making it difficult to adapt to multi-target demands in emergency situations.

[0004] Under this background, the core challenges faced by the research field gradually emerge. Unmanned aerial vehicle cluster needs to be reasonably divided according to the performance characteristics of different devices, but there is currently a lack of effective hierarchical mechanism to distinguish the roles and responsibilities of each unmanned aerial vehicle, which makes task allocation inefficient. As the complexity of disaster scenarios increases, the problem of inadequate hierarchy is further exacerbated, leading to chaotic internal coordination within the cluster, making it difficult to form overall strength, and limiting resource utilization and monitoring effectiveness.

[0005] Therefore, how to build a multi-level cooperative hierarchical mechanism based on deep learning to dynamically optimize task allocation and role adjustment of unmanned aerial vehicle cluster has become a key problem to improve disaster monitoring efficiency. SUMMARY

[0006] To solve the above technical problems, the present application proposes a method and system for unmanned aerial vehicle cluster cooperative disaster monitoring based on deep learning. The method specifically includes:

[0007] Step S1, obtain performance parameter data of each unmanned aerial vehicle in the unmanned aerial vehicle cluster and perform performance classification to obtain an initial ability level division;

[0008] Step S2, divide the disaster area, and based on the initial ability level division result, use a deep learning algorithm to perform initial task allocation of the unmanned aerial vehicle cluster in the disaster area;

[0009] Step S3, based on the disaster area, obtaining disaster environment real-time change information of each area, and determining whether to adjust the initial task allocation based on the disaster environment real-time change information;

[0010] Step S4, based on the re-allocated task, realizing cooperative disaster monitoring of the UAV cluster.

[0011] Optionally, in the step S1, the initial capability level division method comprises:

[0012] inputting the performance parameter data of the UAV into a BP whale neural network algorithm for classification of the initial capability of the UAV, and performing level division on the classification result according to a preset threshold value;

[0013] The BP whale neural network algorithm specifically comprises:

[0014] normalizing the performance parameters of the UAV to obtain preprocessed data;

[0015] initializing the improved whale algorithm and the BP neural network;

[0016] inputting the preprocessed data into the initialized improved whale algorithm, and obtaining an initial fitness value based on a weight threshold value obtained by the initialized BP neural network;

[0017] updating the fitness value based on the initial fitness value using the improved whale algorithm;

[0018] obtaining optimal weights and optimal threshold values based on the updated fitness value, and inputting the optimal weights and the optimal threshold values into the BP neural network for classification of the initial capability of the UAV.

[0019] Optionally, in the step S2, the initial task allocation specifically comprises:

[0020] based on the number of disaster processing to be allocated and the number of UAVs, using a chromosome coding method to realize chromosome coding population initialization, and calculating a fitness value based on the initialization result;

[0021] based on the fitness value calculation result, using an A* algorithm to generate an initial formation trajectory of the UAV, and completing the initial task allocation.

[0022] Optionally, the specific process of the step S3 comprises:

[0023] obtaining real-time environment data in a disaster monitoring area to generate an initial environment change data set;

[0024] preprocessing the initial environment change data set to obtain environment change preprocessing data;

[0025] calculate a fluctuation amplitude of the environmental change pre-processing data;

[0026] determine whether to adjust the initial task allocation based on the fluctuation amplitude.

[0027] The application further discloses a deep learning-based unmanned aerial vehicle cluster cooperative disaster monitoring system.

[0028] An unmanned aerial vehicle performance evaluation module is configured to obtain performance parameter data of each unmanned aerial vehicle in the unmanned aerial vehicle cluster and perform performance classification to obtain an initial ability level division.

[0029] An initial task allocation module is configured to divide a disaster area, and based on the initial ability level division result, use a deep learning algorithm to perform initial task allocation of the unmanned aerial vehicle cluster in the disaster area.

[0030] A task updating module is configured to obtain disaster environment real-time change information of each area based on the disaster area, and determine whether to adjust the initial task allocation based on the disaster environment real-time change information.

[0031] A cooperative disaster monitoring module is configured to perform cooperative disaster monitoring of the unmanned aerial vehicle cluster based on the re-allocated task.

[0032] Optionally, in the unmanned aerial vehicle performance evaluation module, the initial ability level division method comprises:

[0033] inputting the performance parameter data of the unmanned aerial vehicle into a BP whale neural network algorithm to classify the initial ability of the unmanned aerial vehicle, and performing level division on the classification result according to a preset threshold value;

[0034] The BP whale neural network algorithm specifically comprises:

[0035] normalizing the performance parameter of the unmanned aerial vehicle to obtain pre-processing data;

[0036] initializing the improved whale algorithm and the BP neural network;

[0037] inputting the pre-processing data into the initialized improved whale algorithm, and obtaining an initial fitness value based on a weight threshold value calculated by the initialized BP neural network;

[0038] updating the fitness value based on the initial fitness value using the improved whale algorithm;

[0039] obtaining optimal weights and optimal threshold values based on the updated fitness value, and inputting the optimal weights and the optimal threshold values into the BP neural network to classify the initial ability of the unmanned aerial vehicle.

[0040] Optionally, in the initial task allocation module, the initial task allocation is specifically:

[0041] Based on the number of disaster processing to be allocated and the number of unmanned aerial vehicles, a chromosome coding method is used to realize chromosome coding population initialization, and the fitness value is calculated based on the initialization result;

[0042] Based on the fitness value calculation result, an initial formation trajectory of the unmanned aerial vehicle is generated using the A* algorithm, and the initial task allocation is completed.

[0043] Optionally, the workflow of the task updating module is specifically:

[0044] Real-time environmental data in the disaster monitoring area is obtained to generate an initial environmental change data set;

[0045] The initial environmental change data set is preprocessed to obtain environmental change preprocessing data;

[0046] The fluctuation amplitude of the environmental change preprocessing data is calculated;

[0047] Based on the fluctuation amplitude, it is judged whether to adjust the initial task allocation.

[0048] Compared with the prior art, the beneficial effects of the present application are:

[0049] The present application divides the initial ability level of the unmanned aerial vehicle through the BP whale neural network algorithm, and combines the deep learning algorithm to perform initial task allocation, so that the task can be accurately allocated to the most suitable unmanned aerial vehicle according to the performance parameters of the unmanned aerial vehicle and the characteristics of the disaster area, and the efficiency and accuracy of task allocation are improved.

[0050] The present application can timely adjust the initial task allocation scheme based on real-time change information of the disaster environment. By monitoring the fluctuation amplitude of the environmental data, the abnormal area is quickly identified, and the monitoring resources are re-allocated to ensure that the task allocation is always adapted to the current disaster environment.

[0051] The present application significantly improves the efficiency, flexibility and coordination ability of the unmanned aerial vehicle cluster in disaster monitoring through intelligent task allocation and dynamic adjustment mechanism, combined with advanced algorithms and sensor technology, which provides strong support for disaster emergency response and rescue work. BRIEF DESCRIPTION OF DRAWINGS

[0052] In order to more clearly illustrate the technical solutions of the present application, the following briefly introduces the drawings needed in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0053] Figure 1 is a method step diagram of a deep learning-based unmanned aerial vehicle cluster cooperative disaster monitoring method provided by an embodiment of the application. DETAILED DESCRIPTION

[0054] In order to make the above-mentioned objects, characteristics and advantages of the present application more obvious and easy to understand, the present application will be further described in detail below in combination with the drawings and specific embodiments.

[0055] Embodiment one:

[0056] A deep learning-based unmanned aerial vehicle cluster cooperative disaster monitoring method, as shown in the figure, the method comprises: Figure 1

[0057] Step S1, obtaining the performance parameter data of each unmanned aerial vehicle in the unmanned aerial vehicle cluster and performing performance classification to obtain an initial ability level division.

[0058] The initial ability level division method comprises:

[0059] The performance parameter data of the unmanned aerial vehicle is input into a BP whale neural network algorithm for classification of the initial ability of the unmanned aerial vehicle, and the classification results are divided into levels according to a preset threshold value;

[0060] The BP whale neural network algorithm specifically comprises:

[0061] The performance parameters of the unmanned aerial vehicle are normalized to obtain preprocessed data;

[0062] The improved whale algorithm and the BP neural network are initialized;

[0063] The preprocessed data is input into the initialized improved whale algorithm, and an initial fitness value is calculated based on the weight threshold value obtained by the initialized BP neural network;

[0064] The improved whale algorithm is used to update the fitness value based on the initial fitness value;

[0065] Based on the updated fitness value, the optimal weight value and the optimal threshold value are obtained, and the optimal weight value and the optimal threshold value are input into the BP neural network for classification of the initial ability of the unmanned aerial vehicle.

[0066] The improved whale algorithm specifically comprises:

[0067] Initialization parameters: population size N, maximum iteration T here, initial minimum weight w1, initial maximum weight w2, and current iteration number t=0; max

[0068] The fitness value of each individual is calculated, and the current optimal fitness value and its corresponding position vector are recorded.​​

[0069] The values of a, A, C, l, p and w of each individual are calculated and updated, and the value of a is calculated using the following value:

[0070] a = (2-2t / T max )(l-t 3 / T max 3 )

[0071] Where t is the current iteration number.

[0072] When |A|≥1, the position of the next generation is updated as follows:

[0073] X(t+1) = X rand -AD

[0074] Where x rand is a position vector randomly selected from the current whale group, and X is the position vector in which the individual is located. The updated A and C are as follows:

[0075] A = 2a·r1-a

[0076] C = 2r2

[0077] Where r1 and r2 are random numbers between 0 and 1.

[0078] When |A|<1, the position of the next generation is updated as follows:

[0079]

[0080] Where w(t) is:

[0081]

[0082] Where w(t) i is the weight of the i-th search agent at the t-th iteration, f(t) avg is the average fitness of the current population after the t-th iteration, f(t) min and f(t) max are the minimum fitness and the maximum fitness after the t-th iteration, respectively.

[0083] Step S2, divide the disaster area, based on the initial ability level division result, use a deep learning algorithm to perform initial task allocation for the unmanned aerial vehicle cluster in the disaster area.

[0084] In this embodiment, the M-H algorithm is used for multi-feature texture image segmentation for the division of the disaster area.

[0085] The initial task allocation specifically comprises: based on the number of disaster processing to be allocated and the number of unmanned aerial vehicles, a chromosome coding method is used to realize chromosome coding population initialization, and based on the initialization result, a fitness value is calculated; based on the fitness value calculation result, an A* algorithm is used to generate an initial formation trajectory of the unmanned aerial vehicle, and the initial task allocation is completed.

[0086] In determining the chromosome length, flexible adjustment needs to be made according to the relationship between the number of unmanned aerial vehicles and the target number. If the number of unmanned aerial vehicles is less than the total number of targets, the chromosome length should be equal to the total number of targets to ensure that all target points are covered. When the number of unmanned aerial vehicles is equal to the total number of targets, both can be used as a reference for the length of the chromosome, and the length of the chromosome has flexibility and can be selected according to actual needs. However, if the number of unmanned aerial vehicles exceeds the total number of targets, the chromosome length should be set to the total number of unmanned aerial vehicles to ensure that each unmanned aerial vehicle has a corresponding representation in the chromosome. The chromosome is presented in the form of a one-dimensional array, and each array element represents a gene position. In this way, the relationship between the unmanned aerial vehicle and the target can be effectively represented.

[0087] The method for initializing the chromosome coding population specifically comprises:

[0088] The initial fitness value is calculated using a roulette strategy:

[0089]

[0090] Wherein, p i is the individual probability, f i is the individual fitness, and N is the population size.

[0091] The self-adaptive crossover probability and mutation probability formula avoids local optimal solution:

[0092]

[0093]

[0094] Wherein, f is the fitness value of the individual to be mutated, f' is the larger value of the two individuals to be crossed, and P cl is the crossover probability change parameter.

[0095] For the initial strategy, an eight-way search strategy is used to generate the initial trajectory of the unmanned aerial vehicle, and the optimal path is obtained.

[0096] Step S3, based on the disaster area, obtaining real-time change information of the disaster environment of each area, and determining whether to adjust the initial task allocation based on the real-time change information of the disaster environment.

[0097] Obtaining real-time environmental data in the disaster monitoring area, and generating an initial environmental change data set;

[0098] preprocessing the initial environment change data set to obtain environment change preprocessed data;

[0099] calculating fluctuation amplitudes of the environment change preprocessed data;

[0100] judging whether to adjust the initial task allocation based on the fluctuation amplitudes.

[0101] Real-time environment data is obtained from a disaster monitoring area through a sensor network, multi-dimensional information such as temperature, humidity, and wind speed is collected, and a preliminary environment change data set is obtained. According to the preliminary environment change data set, a data preprocessing method is used to remove noise and outliers, and a cleaned environment change feature set is determined. For the cleaned environment change feature set, the fluctuation amplitudes of each dimension of data are calculated, the quantization results of the fluctuation amplitudes are obtained using a standard deviation calculation method, and the distribution of the fluctuation amplitudes is judged. If the quantization results of the fluctuation amplitudes exceed a preset threshold range, the results are marked as an abnormal state, and the region position information corresponding to the abnormal state is obtained. Through the region position information corresponding to the abnormal state, a dynamic adjustment process is triggered, monitoring resources are re-allocated for the abnormal region, and an adjusted task allocation scheme is obtained. According to the adjusted task allocation scheme, the monitoring frequency and data collection strategy of the sensor network are updated, and updated real-time environment data is obtained. For the updated real-time environment data, the fluctuation amplitude calculation and threshold judgment process are executed in a loop, the environment change trend is continuously monitored, and the stability state of the monitoring area is determined.

[0102] Step S4, cooperative disaster monitoring of the UAV cluster is implemented based on the re-allocated tasks.

[0103] The real-time resource utilization state data is obtained from the unmanned aerial vehicle cluster through the trigger signal, the running load of each unmanned aerial vehicle is analyzed, and the distribution of resource utilization is obtained. According to the distribution of resource utilization, the individual with unbalanced load is identified, if the resource utilization of a certain unmanned aerial vehicle exceeds the preset threshold, it is marked as an object to be adjusted, and a preliminary adjustment list is determined. The marked object to be adjusted data is used in combination with the priority sorting rule to sort the unmanned aerial vehicles in the list, the priority sequence after sorting is judged, and a high-priority unmanned aerial vehicle list is obtained. The task data is extracted from the high-priority unmanned aerial vehicle list, the matching degree of the resource demand of the task and the current allocation is analyzed, if the matching degree is lower than the preset standard, a task adjustment instruction is generated, and a specific task adjustment scheme is determined. Through the task adjustment scheme, the task of the high-priority unmanned aerial vehicle is redistributed, the resource allocation strategy is combined, the adjusted resource utilization state is obtained, and whether the balanced state is reached is judged. According to the adjusted resource utilization state, the running data of the unmanned aerial vehicle cluster is continuously monitored, if there is still an unbalanced state, the priority sorting and task adjustment process are executed in a loop, and the final resource balanced distribution is obtained. The final resource balanced distribution data is obtained, the adjustment record and state monitoring information are stored, the adjusted unmanned aerial vehicle task allocation list is generated, and the whole process is completed.

[0104] Embodiment two:

[0105] A deep learning-based unmanned aerial vehicle cluster collaborative disaster monitoring system, the system comprising:

[0106] An unmanned aerial vehicle performance evaluation module for obtaining performance parameter data of each unmanned aerial vehicle in the unmanned aerial vehicle cluster and performing performance classification to obtain an initial ability level division;

[0107] The initial ability level division method comprises:

[0108] The performance parameter data of the unmanned aerial vehicle is input into the BP whale neural network algorithm for classification of the initial ability of the unmanned aerial vehicle, and the classification results are divided into levels according to a preset threshold;

[0109] The BP whale neural network algorithm specifically comprises:

[0110] The performance parameters of the unmanned aerial vehicle are normalized to obtain preprocessed data;

[0111] The improved whale algorithm and the BP neural network are initialized;

[0112] The preprocessed data is input into the initialized improved whale algorithm, and an initial fitness value is calculated based on the weight threshold value obtained by the initialized BP neural network;

[0113] The improved whale algorithm is used to update the fitness value based on the initial fitness value.

[0114] Based on the updated fitness value, the optimal weight value and the optimal threshold value are obtained, and the optimal weight value and the optimal threshold value are input into the BP neural network to classify the initial ability of the unmanned aerial vehicle.

[0115] The improved whale algorithm is specifically:

[0116] Initialize parameters: population size N, maximum iteration T max , initial minimum weight w1, initial maximum weight w2, and current iteration number t=0;

[0117] Calculate the fitness value of each individual, and record the current optimal fitness value and its corresponding position vector;

[0118] Calculate and update the values of a, A, C, l, p and w of each individual, and use the following value to calculate the value of a:

[0119] a=(2-2t / T max )(l-t 3 / T max 3 )

[0120] Wherein, t is the current iteration number.

[0121] When |A|≥1, update the position of the next generation:

[0122] X(t+1)=X rand -AD

[0123] Wherein, xrand is a position vector randomly selected from the current whale group, and X is the position vector where the individual is located. The updated A and C are:

[0124] A=2 a ·r1-a

[0125] C=2r2

[0126] Wherein, r1 and r2 are random numbers between 0 and 1.

[0127] When |A|<1, update the position of the next generation:

[0128]

[0129] Wherein, w(t) is:

[0130]

[0131] Wherein, w(t) i is the weight of the i-th search agent at the t-th iteration, f(t) avgThe average fitness of the population after the current tth iteration is f(t) min And f(t) max The minimum fitness and the maximum fitness after the tth iteration, respectively.

[0132] An initial task allocation module is configured to divide a disaster area, and based on the initial ability level division result, an initial task allocation is performed for the UAV cluster in the disaster area using a deep learning algorithm.

[0133] In this embodiment, the M-H algorithm is used for multi-feature texture image segmentation for dividing the disaster area.

[0134] The initial task allocation is specifically as follows: based on the number of disaster processing to be allocated and the number of UAVs, a chromosome coding population initialization is realized using a chromosome coding method, and a fitness value is calculated based on the initialization result; based on the fitness value calculation result, an initial formation trajectory of the UAV is generated using an A* algorithm, and the initial task allocation is completed.

[0135] When determining the chromosome length, it needs to be flexibly adjusted according to the relationship between the number of UAVs and the target number. If the number of UAVs is less than the total number of targets, the chromosome length should be equal to the total number of targets to ensure that all target points are covered. When the number of UAVs is equal to the total number of targets, both can be used as a reference for the chromosome length, and the chromosome length has flexibility and can be selected according to actual needs. However, if the number of UAVs exceeds the total number of targets, the chromosome length should be set to the total number of UAVs to ensure that each UAV has a corresponding representation in the chromosome. The chromosome is presented in the form of a one-dimensional array, and each array element represents a gene position. In this way, the relationship between the UAV and the target can be effectively represented.

[0136] The method for initializing the chromosome coding population is specifically as follows:

[0137] The initial fitness is calculated using a roulette strategy:

[0138]

[0139] Where p i is the individual probability, f i is the individual fitness, and N is the population size.

[0140] The self-adaptive crossover probability and mutation probability formulas avoid local optimal solutions:

[0141]

[0142]

[0143] Wherein, f is the fitness value of the individual to be mutated, f' is the value of the larger fitness in the two individuals to be crossed, P is the crossover probability change parameter. cl is the crossover probability change parameter.

[0144] For the initial strategy, the application uses an eight-way search strategy to generate an initial trajectory for the trajectory of the unmanned aerial vehicle, and obtains an optimal path.

[0145] The task updating module is configured to obtain real-time disaster environment change information of each region based on the disaster region, and determine whether to adjust the initial task allocation based on the real-time disaster environment change information.

[0146] Obtain real-time environment data in the disaster monitoring region, and generate an initial environment change data set.

[0147] Preprocess the initial environment change data set to obtain environment change preprocessing data.

[0148] Calculate the fluctuation amplitude of the environment change preprocessing data.

[0149] Determine whether to adjust the initial task allocation based on the fluctuation amplitude.

[0150] Through the sensor network, real-time environment data is obtained from the disaster monitoring region, and multi-dimensional information such as temperature, humidity, and wind speed is collected to obtain a preliminary environment change data set. According to the preliminary environment change data set, a data preprocessing method is used to remove noise and outliers to determine a cleaned environment change feature set. For the cleaned environment change feature set, the fluctuation amplitude of each dimension of data is calculated, the quantization result of the fluctuation amplitude is obtained by using a standard deviation calculation method, and the distribution of the fluctuation amplitude is determined. If the quantization result of the fluctuation amplitude exceeds a preset threshold range, the result is marked as an abnormal state, and the region position information corresponding to the abnormal state is obtained. Through the region position information corresponding to the abnormal state, a dynamic adjustment process is triggered, and monitoring resources are re-allocated for the abnormal region to obtain an adjusted task allocation scheme. According to the adjusted task allocation scheme, the monitoring frequency and data collection strategy of the sensor network are updated, and updated real-time environment data is obtained. For the updated real-time environment data, the fluctuation amplitude calculation and threshold judgment process are executed in a loop to continuously monitor the environment change trend and determine the stability state of the monitoring region.

[0151] The collaborative disaster monitoring module is configured to implement collaborative disaster monitoring of the unmanned aerial vehicle cluster based on the re-allocated task.

[0152] The real-time resource utilization state data is obtained from the unmanned aerial vehicle cluster through the trigger signal, the operation load of each unmanned aerial vehicle is analyzed, and the distribution of resource utilization is obtained. According to the distribution of resource utilization, the individual with unbalanced load is identified. If the resource utilization of a certain unmanned aerial vehicle exceeds the preset threshold, the unmanned aerial vehicle is marked as an object to be adjusted, and a preliminary adjustment list is determined. The marked object to be adjusted data is used in combination with the priority sorting rule to sort the unmanned aerial vehicles in the list, the priority sequence after sorting is judged, and a high-priority unmanned aerial vehicle list is obtained. Task data is extracted from the high-priority unmanned aerial vehicle list, the matching degree of the resource demand of the task and the current allocation is analyzed, and if the matching degree is lower than the preset standard, a task adjustment instruction is generated, and a specific task adjustment scheme is determined. Through the task adjustment scheme, the task of the high-priority unmanned aerial vehicle is redistributed, the resource allocation strategy is combined, the adjusted resource utilization state is obtained, and whether the balanced state is reached is judged. According to the adjusted resource utilization state, the operation data of the unmanned aerial vehicle cluster is continuously monitored, if there is still an unbalanced state, the priority sorting and task adjustment process is circularly executed, and the final resource balanced distribution is obtained. The final resource balanced distribution data is obtained, the adjustment record and state monitoring information are stored, the adjusted unmanned aerial vehicle task allocation list is generated, and the whole process is completed.

[0153] The above-described embodiments are only descriptions of the preferred modes of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A method for collaborative disaster monitoring by unmanned aerial vehicle (UAV) swarms based on deep learning, characterized in that, The method includes: Step S1: Obtain the performance parameter data of each drone in the drone cluster and classify its performance to obtain the initial capability level classification; Step S2: Divide the disaster area. Based on the initial capability level division results, use a deep learning algorithm to perform initial task allocation for the UAV cluster in the disaster area. Step S3: Based on the disaster area, obtain real-time disaster environment change information for each area, and determine whether to adjust the initial task allocation based on the real-time disaster environment change information; Step S4: Implement collaborative disaster monitoring of the UAV swarm based on the reassigned tasks; In step S1, the initial capability level classification method includes: The performance parameter data of the UAV is input into the BP whale neural network algorithm to classify the initial capabilities of the UAV, and the classification results are divided into levels according to a preset threshold. The BP whale neural network algorithm is specifically as follows: The performance parameters of the UAV are normalized to obtain preprocessed data. The improved whale algorithm and BP neural network are initialized; The preprocessed data is input into the improved whale algorithm after initialization, and the initial fitness value is calculated based on the weight threshold obtained by the BP neural network after initialization. The improved whale algorithm is used to update the fitness value based on the initial fitness value. The optimal weights and thresholds are obtained based on the updated fitness values, and then the optimal weights and thresholds are input into the BP neural network to classify the initial capabilities of the UAV.

2. The method for collaborative disaster monitoring by unmanned aerial vehicle (UAV) swarms based on deep learning according to claim 1, characterized in that, In step S2, the initial task allocation is specifically as follows: Based on the number of disasters to be assigned and the number of drones, a chromosome encoding method is used to initialize the chromosome-encoded population, and the fitness value is calculated based on the initialization results. Based on the fitness value calculation results, A The algorithm generates the initial formation trajectory of the drone and completes the initial task assignment.

3. The method for collaborative disaster monitoring by unmanned aerial vehicle (UAV) swarms based on deep learning according to claim 1, characterized in that, The specific process of step S3 is as follows: Acquire real-time environmental data within the disaster monitoring area and generate an initial environmental change dataset; The initial environmental change dataset is preprocessed to obtain preprocessed environmental change data; Calculate the fluctuation amplitude of the environmental change preprocessing data; Whether to adjust the initial task allocation is determined based on the fluctuation amplitude.

4. A deep learning-based UAV swarm collaborative disaster monitoring system, the system being used to implement the UAV swarm collaborative disaster monitoring method according to any one of claims 1-3, characterized in that, The system includes: The drone performance evaluation module is used to acquire performance parameter data of each drone in the drone cluster and classify its performance to obtain an initial capability level classification. The initial task allocation module is used to divide the disaster area and, based on the initial capability level division results, use a deep learning algorithm to perform initial task allocation for the UAV cluster in the disaster area; The task update module is used to obtain real-time disaster environment change information for each disaster area, and determine whether to adjust the initial task allocation based on the real-time disaster environment change information. The collaborative disaster monitoring module is used to achieve collaborative disaster monitoring of UAV swarms based on reassigned tasks.

5. The deep learning-based UAV swarm collaborative disaster monitoring system according to claim 4, characterized in that, In the UAV performance evaluation module, the initial capability level classification method includes: The performance parameter data of the UAV is input into the BP whale neural network algorithm to classify the initial capabilities of the UAV, and the classification results are divided into levels according to a preset threshold. The BP whale neural network algorithm is specifically as follows: The performance parameters of the UAV are normalized to obtain preprocessed data. The improved whale algorithm and BP neural network are initialized; The preprocessed data is input into the improved whale algorithm after initialization, and the initial fitness value is calculated based on the weight threshold obtained by the BP neural network after initialization. The improved whale algorithm is used to update the fitness value based on the initial fitness value. The optimal weights and thresholds are obtained based on the updated fitness values, and then the optimal weights and thresholds are input into the BP neural network to classify the initial capabilities of the UAV.

6. The deep learning-based UAV swarm collaborative disaster monitoring system according to claim 5, characterized in that, In the initial task allocation module, the initial task allocation specifically involves: Based on the number of disasters to be assigned and the number of drones, a chromosome encoding method is used to initialize the chromosome-encoded population, and the fitness value is calculated based on the initialization results. Based on the fitness value calculation results, A The algorithm generates the initial formation trajectory of the drone and completes the initial task assignment.

7. The deep learning-based UAV swarm collaborative disaster monitoring system according to claim 4, characterized in that, The workflow of the task update module is as follows: Acquire real-time environmental data within the disaster monitoring area and generate an initial environmental change dataset; The initial environmental change dataset is preprocessed to obtain preprocessed environmental change data; Calculate the fluctuation amplitude of the environmental change preprocessing data; Whether to adjust the initial task allocation is determined based on the fluctuation amplitude.

Images