An action management identification method based on earthquake field disaster investigation

Abstract

The application discloses an action management identification method based on a seismic site disaster investigation, and relates to the technical field of data processing.The method comprises the following steps: determining a seismic site region to be acted on, and planning a disaster search task of a plurality of unmanned aerial vehicles in the seismic site region to be acted on; then, the disaster search task is sent to the plurality of unmanned aerial vehicles, and the unmanned aerial vehicles are controlled to execute the disaster search task, so that an actual disaster investigation result corresponding to the disaster search task is obtained; then, according to the actual disaster investigation result, a disaster point is generated on an electronic map corresponding to the seismic site region to be acted on, so that a disaster investigation map is obtained; finally, the disaster investigation map is fed back to a device designated by a commander, so as to guide the disaster relief action of the commander, which not only improves the efficiency of the seismic site disaster investigation, but also forms comprehensive disaster data, so that the commander can comprehensively understand the disaster, auxiliary commanders quickly form a global and optimal action plan, and the disaster disposal efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and more specifically, to an action management identification method based on earthquake on-site disaster investigation. Background Technology

[0002] Earthquakes are extremely destructive natural disasters, often causing severe consequences such as casualties, building collapses, road closures, and frequent secondary disasters in a short period of time. After an earthquake, a rapid and accurate assessment of the disaster situation and the development of effective emergency response management strategies are crucial for saving lives, minimizing losses, and maintaining social order. Traditional earthquake emergency response management largely relies on experienced commanders making decisions based on limited and delayed information. Current technology typically involves rescue personnel conducting manual surveys at the earthquake site. After assessing the damage, rescue efforts are either initiated on-site or feedback is relayed to commanders for unified processing. Commanders must process large amounts of fragmented information, making it difficult to quickly formulate a comprehensive and optimal action plan. This can lead to delays and misallocations in the deployment of emergency resources (such as rescue teams, medical supplies, and engineering machinery), missing optimal rescue opportunities and impacting overall rescue efficiency. Summary of the Invention

[0003] This application aims to provide an action management identification method based on earthquake site disaster investigation, which addresses the problem of low rescue efficiency caused by existing technologies that rely on manual investigation of earthquake sites.

[0004] This application provides an action management identification method based on earthquake on-site disaster investigation, including:

[0005] Identify the earthquake-prone area requiring action and plan disaster search missions using multiple drones within that area.

[0006] The disaster search task is distributed to multiple drones, and the drones are controlled to execute the disaster search task to obtain the actual disaster investigation results corresponding to the disaster search task.

[0007] Based on the actual disaster investigation results corresponding to the disaster search task, disaster points are generated on the electronic map corresponding to the earthquake site area to be acted upon, and a disaster investigation map is obtained.

[0008] The disaster survey map is fed back to the device designated by the commander to guide the commander's disaster relief actions and complete the action management based on the earthquake site disaster survey.

[0009] In one possible implementation, determining the earthquake site area to be addressed and planning a multi-UAV disaster search mission corresponding to the earthquake site area includes:

[0010] By obtaining the closed areas identified by disaster relief personnel on the electronic map through human-computer interaction, the earthquake site area to be acted upon can be obtained;

[0011] Obtain information on power transmission lines and roads in the earthquake-prone area;

[0012] Based on the power transmission line information of the earthquake site area to be acted upon, a first disaster search task is generated to conduct a disaster investigation along the power transmission lines;

[0013] Based on the road information, a second disaster search task is generated to conduct a disaster investigation along the road.

[0014] With the goal of traversing the earthquake site area to be acted upon, a third disaster search task is generated to investigate the trapped personnel and collapsed buildings in the earthquake site area to be acted upon.

[0015] In one possible implementation, the step of distributing the disaster search task to multiple drones and controlling the drones to execute the disaster search task, and obtaining the actual disaster investigation results corresponding to the disaster search task, includes:

[0016] The first disaster search task is sent to the first UAV equipped with the power transmission line detection model, and the first UAV is controlled to perform target detection along the power transmission line to obtain the power transmission line investigation results.

[0017] The second disaster search task is sent to the second UAV equipped with a road detection model, and the second UAV is controlled to perform target detection along the road to obtain the road survey results;

[0018] The third disaster search task is sent to the third UAV carrying the personnel and housing detection model, and the third UAV is controlled to traverse the earthquake site area to be acted upon to obtain the personnel and housing survey results;

[0019] The results of the personnel survey, housing survey, road survey, and power line survey are collectively used as the actual disaster survey results corresponding to the disaster search task.

[0020] In one possible implementation, the transmission line detection model, road detection model, and personnel and building detection model are all set as YOLO models.

[0021] In one possible implementation, the mounting methods for the transmission line detection model, road detection model, and personnel and building detection model are the same, and all include:

[0022] A hybrid mapping strategy is used to initialize the model to be mounted, resulting in multiple parameter vectors; wherein, the model to be mounted is a transmission line detection model, a road detection model, or a personnel and building detection model;

[0023] Obtain the fitness corresponding to each parameter vector, and obtain the optimal parameter vector based on the fitness of the parameter vector;

[0024] Based on the optimal parameter vector, an adaptive search based on the uniformity of distribution is performed on the parameter vector to obtain the parameter vector after the adaptive search.

[0025] A wave-like spiral chain cooperative search is performed on the parameter vector after the adaptive search to obtain the parameter vector after the wave-like spiral chain cooperative search.

[0026] A bidirectional mutation search is performed on the parameter vector after the wave-like spiral chain cooperative search to obtain the parameter vector after the bidirectional mutation search.

[0027] Determine if the current training iterations have reached the maximum number of training iterations. If so, obtain the optimal parameter vector again based on the parameter vector obtained after the bidirectional mutation search, and then obtain the target parameter vector. Otherwise, return to the step of obtaining the optimal parameter vector and proceed to the next training process.

[0028] The target parameter vector is used as the final parameter of the model to be mounted, and the model to be mounted is deployed on the drone according to the final parameter of the model to be mounted.

[0029] In one possible implementation, the hybrid mapping strategy is used to initialize the model to be mounted, resulting in multiple parameter vectors, including:

[0030] The model parameters of the model to be mounted are randomly initialized between the upper limit and the lower limit of the model parameters, and the initialized model parameters are encoded into a vector to obtain the basic vector;

[0031] Based on the aforementioned base vector, multiple other vectors are obtained using a chaotic mapping sequence.

[0032] Based on the other vectors, the upper limit of the model parameters, and the lower limit of the model parameters, a parameter vector is generated, and all other vectors are processed to obtain multiple parameter vectors.

[0033] In one possible implementation, obtaining the fitness corresponding to each parameter vector and obtaining the optimal parameter vector based on the fitness of the parameter vectors includes:

[0034] For any parameter vector, obtain the loss function value corresponding to the parameter vector, and add the loss function value to a preset non-zero constant term and take the reciprocal to obtain the fitness corresponding to the parameter vector;

[0035] The parameter vector with the highest fitness is taken as the optimal parameter vector.

[0036] In one possible implementation, the step of performing an adaptive search based on the distribution uniformity of the parameter vector, based on the optimal parameter vector, to obtain the parameter vector after the adaptive search, includes:

[0037] The parameter vectors are arranged in ascending order of fitness. Based on the arranged parameter vectors, and according to the fitness of the optimal parameter vector, the fitness of the parameter vectors, and the current number of training iterations, an adaptive balance search factor is obtained.

[0038] Obtain the Euclidean distance between adjacent parameter vectors, and determine the distribution uniformity based on the Euclidean distance between adjacent parameter vectors;

[0039] When the distribution uniformity is less than a preset uniformity threshold, the parameter vector is searched according to the optimal parameter vector and the adaptive balance search factor to obtain the parameter vector after adaptive search.

[0040] When the distribution uniformity is greater than or equal to a preset uniformity threshold, the parameter vector is searched according to the upper limit and lower limit of the model parameters to obtain the parameter vector after adaptive search.

[0041] In one possible implementation, performing a wave-like spiral chain cooperative search on the parameter vector after the adaptive search to obtain the parameter vector after the wave-like spiral chain cooperative search includes:

[0042] Determine the waveform adjustment factor based on the current number of training iterations;

[0043] If the volatility adjustment factor is greater than or equal to the preset adjustment threshold, then a spiral chain cooperative search is performed on the parameter vector after the adaptive search based on the optimal parameter vector to obtain the parameter vector after the volatility spiral chain cooperative search.

[0044] If the volatility adjustment factor is less than the preset adjustment threshold, a random parameter vector is determined from the parameter vector after the adaptive search, and a spiral chain cooperative search is performed based on the parameter vector after the adaptive search of the random parameter vector to obtain the parameter vector after the volatility spiral chain cooperative search.

[0045] In one possible implementation, performing a bidirectional mutation search on the parameter vector following the wave-like spiral chain cooperative search to obtain the parameter vector after the bidirectional mutation search includes:

[0046] The number of mutation dimensions, positive mutation factor, and negative mutation factor are obtained based on the current number of training iterations.

[0047] Based on the number of mutation dimensions, the dimension to be mutated of the parameter vector after the volatile spiral chain cooperative search is determined.

[0048] For any dimension to be mutated, a positive mutation factor is used to perform a positive mutation on the parameter vector after the volatile spiral chain cooperative search, or a negative mutation factor is used to perform a negative mutation on the parameter vector after the volatile spiral chain cooperative search, to obtain the parameter vector after the bidirectional mutation search.

[0049] Beneficial effects:

[0050] This application provides an action management identification method based on earthquake site disaster investigation. The method includes: determining the earthquake site area to be investigated and planning multiple UAVs to perform disaster search tasks in the corresponding area; then distributing the disaster search tasks to the multiple UAVs and controlling them to execute the search tasks, obtaining the actual disaster investigation results; next, generating disaster points on an electronic map corresponding to the earthquake site area based on the actual disaster investigation results, thus obtaining a disaster investigation map; finally, feeding the disaster investigation map back to the equipment designated by the command personnel to guide their disaster relief actions. This method not only improves the efficiency of earthquake site disaster investigation but also generates comprehensive disaster data, enabling command personnel to fully understand the disaster situation and quickly formulate a comprehensive and optimal action plan, thereby improving the efficiency of disaster response. Attached Figure Description

[0051] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating an action management identification method based on earthquake on-site disaster investigation, as proposed in one embodiment of this application.

[0053] Figure 2 This is a schematic diagram of the structure of an action management identification device based on earthquake on-site disaster investigation, according to an embodiment of this application.

[0054] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application.

[0055] Explanation of reference numerals in the attached diagram: 201-Task planning module, 202-Disaster investigation module, 203-Disaster summary module, 204-Disaster feedback module, 301-Memory, 302-Processor, 303-Communication bus. Detailed Implementation

[0056] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0057] In related technologies, disaster relief personnel typically conduct manual surveys of the disaster situation at the earthquake site. After assessing the damage, they either carry out on-site rescue operations or report back to command personnel for unified processing. Command personnel need to process a large amount of fragmented information, making it difficult to quickly formulate a comprehensive and optimal action plan. This can lead to delays and misallocations in the allocation of emergency resources (such as rescue teams, medical supplies, and engineering machinery), missing the best rescue opportunities and affecting overall rescue efficiency.

[0058] In view of this, this application proposes an action management identification method based on earthquake on-site disaster investigation. It combines deep learning technology with multi-UAV collaborative task execution to investigate earthquake on-site disaster. Finally, the actual disaster investigation results are summarized into a disaster investigation map. This not only improves the efficiency of earthquake on-site disaster investigation, but also generates comprehensive disaster data, enabling commanders to fully understand the disaster situation and assisting them in quickly formulating a global and optimal action plan, thereby improving the efficiency of disaster response.

[0059] like Figure 1 As shown in the figure, this application provides an action management identification method based on earthquake on-site disaster investigation, including:

[0060] S101. Determine the earthquake site area to be addressed and plan disaster search missions for multiple UAVs in the earthquake site area to be addressed.

[0061] The earthquake site area to be acted upon can be a closed area determined by the command personnel on an electronic map through human-computer interaction. By planning the survey routes of multiple drones in the earthquake site area to be acted upon, disaster search tasks can be obtained.

[0062] S102. Distribute the disaster search task to multiple drones and control the drones to execute the disaster search task, and obtain the actual disaster investigation results corresponding to the disaster search task.

[0063] Different disaster search tasks can be assigned to different drones, enabling multiple drones to collaborate in conducting on-site earthquake disaster investigations and ultimately determine the actual disaster investigation results corresponding to different disaster search tasks.

[0064] S103. Based on the actual disaster investigation results corresponding to the disaster search task, generate disaster points on the electronic map corresponding to the earthquake site area to be acted upon, and obtain a disaster investigation map.

[0065] The actual disaster investigation results may include disaster photos, location information, and disaster classification labels, and generate points of interest on the electronic map. Each point of interest records a disaster photo, location information, and disaster classification label, thereby realizing the generation of disaster points on the electronic map corresponding to the earthquake site area to be acted upon.

[0066] S104. The disaster survey map is fed back to the device designated by the commander to guide the commander's disaster relief actions and complete the action management based on the earthquake site disaster survey.

[0067] By feeding back the disaster survey map to the device designated by the commander, comprehensive disaster data can be generated, enabling the commander to have a full understanding of the disaster situation, assisting the commander in quickly formulating a comprehensive and optimal action plan, and improving the efficiency of disaster response.

[0068] In one possible implementation, determining the earthquake site area to be addressed and planning a multi-UAV disaster search mission corresponding to the earthquake site area includes:

[0069] By obtaining closed areas identified by disaster relief personnel on electronic maps through human-computer interaction, the earthquake site areas awaiting action can be identified.

[0070] Obtain information on power transmission lines and roads in the earthquake-stricken area to be affected.

[0071] Based on the power transmission line information of the earthquake site area to be acted upon, a first disaster search task is generated to conduct a disaster investigation along the power transmission lines.

[0072] Information on power transmission lines in the earthquake-prone area can be obtained from the power grid data system or a third-party map. Then, a first disaster search task is generated to conduct a disaster investigation along the power transmission lines in the earthquake-prone area.

[0073] Based on the road information, a second disaster search task is generated to conduct a disaster investigation along the road.

[0074] Road information can be obtained from third-party maps, and then a second disaster search task can be generated to conduct a disaster investigation along roads within the earthquake site area to be acted upon.

[0075] With the goal of traversing the earthquake site area to be acted upon, a third disaster search task is generated to investigate the trapped personnel and collapsed buildings in the earthquake site area to be acted upon.

[0076] For trapped people and collapsed buildings, these are the key areas for earthquake relief and rescue. Therefore, it is necessary to generate search routes that can traverse the entire earthquake site area to be addressed, thus obtaining the third disaster search task. For example, an S-shaped search route can be generated, allowing drones to collect images of all areas in the earthquake site area to be addressed, thereby facilitating search and rescue operations.

[0077] It is worth noting that the embodiments of this application do not consider the issue of drone battery life, and the issues of drone return and battery life can be solved using existing technologies.

[0078] In one possible implementation, the step of distributing the disaster search task to multiple drones and controlling the drones to execute the disaster search task, and obtaining the actual disaster investigation results corresponding to the disaster search task, includes:

[0079] The first disaster search task is sent to the first UAV equipped with a power transmission line detection model, and the first UAV is controlled to perform target detection along the power transmission line to obtain the power transmission line investigation results.

[0080] It is worth noting that the transmission line detection model mainly detects the collapse of facilities on the transmission line, such as towers and transformers. The transmission line detection model can be pre-trained to enable it to identify normal facilities as well as collapsed facilities.

[0081] The second disaster search task is sent to the second UAV equipped with a road detection model, and the second UAV is controlled to perform target detection along the road to obtain road survey results.

[0082] It's worth noting that the road detection model primarily detects road fractures. This model can be pre-trained to identify both normal and fractured roads. However, both roads and power lines may have branching intersections. Therefore, it's possible to search only the main road or main power line to reduce the search volume. Alternatively, a return-to-home subtask can be set for each branching node, requiring the drone to return to the corresponding branching node after searching along the branch road to continue investigating the remaining unsearched areas.

[0083] The third disaster search task is sent to the third UAV equipped with a personnel and housing detection model, and the third UAV is controlled to traverse the earthquake site area to be acted upon to obtain the personnel and housing survey results.

[0084] It is worth noting that the personnel and building detection model mainly detects trapped personnel and collapsed buildings. The personnel and building detection model can be pre-trained to enable it to identify untrapped personnel, trapped personnel, normal buildings, and collapsed buildings.

[0085] The results of the personnel survey, housing survey, road survey, and power line survey are collectively used as the actual disaster survey results corresponding to the disaster search task.

[0086] When a drone identifies a target corresponding to its mission, it should record photos of the disaster, its location, and a disaster classification label so that a disaster point can be generated on an electronic map.

[0087] In one possible implementation, the transmission line detection model, road detection model, and personnel and building detection model are all set as YOLO models.

[0088] In one possible implementation, the mounting methods for the transmission line detection model, road detection model, and personnel and building detection model are the same, and all include:

[0089] A hybrid mapping strategy is used to initialize the model to be integrated, resulting in multiple parameter vectors. The model to be integrated can be a transmission line detection model, a road detection model, or a personnel and building detection model.

[0090] Optionally, the trainable parameters of a YOLO model are typically the connection weights between network layers. These connection weights can be initialized to determine multiple parameter vectors. For example, the model parameters can be randomly initialized between an upper and lower bound, and the randomly initialized model parameters can be encoded into vectors to obtain parameter vectors. Repeating this process multiple times can yield multiple parameter vectors.

[0091] Obtain the fitness of each parameter vector, and obtain the optimal parameter vector based on the fitness of the parameter vector.

[0092] Based on the optimal parameter vector, an adaptive search based on the uniformity of distribution is performed on the parameter vector to obtain the parameter vector after the adaptive search.

[0093] A wave-like spiral chain cooperative search is performed on the parameter vector after the adaptive search to obtain the parameter vector after the wave-like spiral chain cooperative search.

[0094] A bidirectional mutation search is performed on the parameter vector after the wave-like spiral chain cooperative search to obtain the parameter vector after the bidirectional mutation search.

[0095] Determine if the current training iterations have reached the maximum number of training iterations. If so, obtain the optimal parameter vector again based on the parameter vector obtained after the bidirectional mutation search, and then obtain the target parameter vector. Otherwise, return to the step of obtaining the optimal parameter vector and proceed to the next training process.

[0096] The target parameter vector is used as the final parameter of the model to be mounted, and the model to be mounted is deployed on the drone according to the final parameter of the model to be mounted.

[0097] In existing technologies, YOLO models are typically trained using gradient descent intelligent optimization algorithms or particle swarm optimization. However, both algorithms can get stuck in local optima during training, causing the trained YOLO model to fail to accurately complete the identification task. Therefore, this application provides a novel training algorithm that improves global search capabilities while ensuring fine-grained local search capabilities, thereby enhancing the target identification accuracy of the YOLO model and ultimately improving the accuracy of disaster investigation.

[0098] In one possible implementation, the hybrid mapping strategy is used to initialize the model to be mounted, resulting in multiple parameter vectors, including:

[0099] The model parameters of the model to be mounted are randomly initialized between the upper limit and the lower limit of the model parameters, and the initialized model parameters are encoded into a vector to obtain the basic vector;

[0100] Based on the aforementioned base vector, multiple other vectors are obtained using a chaotic mapping sequence:

[0101]

[0102] in, Indicates the first i The first of the other vectors j dimensional model parameters, and i When =1, This represents the j-th dimension of the model parameters of the basis vectors. j =1,2,…,M, where M represents the total dimension of the model parameters. Indicates the first i +1 other vectors j 3D model parameters, This represents the chaotic mapping parameter, which is set to 0.2. For the YOLO model, its model parameters can be the connection weights between different network layers, and these connection weights are located between [0,1]. Therefore, based on this, multiple other vectors can be obtained by using the chaotic mapping sequence.

[0103] Based on the other vectors, the upper bound of the model parameters, and the lower bound of the model parameters, the parameter vector is generated as follows:

[0104]

[0105]

[0106] in, Indicates the first i The first parameter vector j 3D model parameters, Indicates the first i The first intermediate vector j 3D model parameters, Indicates the first i The backward learning vector corresponding to the nth intermediate vector. j 3D model parameters, This represents the fitness of the intermediate vector. This represents the fitness corresponding to the back-learned vector. Indicates the first j The upper limit of model parameters corresponding to the dimensional model parameters. Indicates the first j The lower bound of the model parameters corresponding to the dimensional model parameters;

[0107] All other vectors are processed to obtain multiple parameter vectors.

[0108] This application employs a hybrid mapping strategy to initialize the model to be mounted, which enables the initial solution to be more evenly distributed in the solution space. Furthermore, it utilizes reverse learning to select the best solution, thereby further improving the training speed of the algorithm.

[0109] In one possible implementation, obtaining the fitness corresponding to each parameter vector and obtaining the optimal parameter vector based on the fitness of the parameter vectors includes:

[0110] For any parameter vector, obtain the loss function value corresponding to the parameter vector, and add the loss function value to a preset non-zero constant term (such as 0.0001) and take the reciprocal to obtain the fitness corresponding to the parameter vector;

[0111] The parameter vector with the highest fitness is taken as the optimal parameter vector.

[0112] The smaller the loss function value, the better the position of the parameter vector in the solution space. Therefore, the larger the fitness, the better the position of the parameter vector in the solution space. Thus, the parameter vector with the largest fitness is determined to be the optimal parameter vector.

[0113] During the training of the transmission line detection model, road detection model, and personnel and building detection model, the corresponding normal data and target data to be detected can be used for acquisition. For example, during the training of the transmission line detection model, image data with normal towers and / or transformers (i.e., normal data) and image data with collapsed towers and / or transformers (i.e., target data to be detected) can be used for training. Obtaining the loss function of the YOLO model is a relatively conventional technique, and will not be described in detail in the embodiments of this application.

[0114] In one possible implementation, the step of performing an adaptive search based on the distribution uniformity of the parameter vector, based on the optimal parameter vector, to obtain the parameter vector after the adaptive search, includes:

[0115] The parameter vectors are arranged in ascending order of fitness. Based on the arranged parameter vectors, and according to the fitness of the optimal parameter vector, the fitness of the parameter vectors, and the current number of training iterations, the adaptive balance search factor is obtained as follows:

[0116]

[0117] in, Indicates the first t During the training process, the first m The adaptive balance search factor corresponding to the permuted parameter vector. This represents the preset maximum value of the adaptive balance search factor, and is set to 0.9; Represents the natural constant. t This represents the current number of training iterations, and T represents the preset maximum number of training iterations. Indicates the first t During the training process, the first m A parameter vector Represents the optimal parameter vector. Represents the worst-case parameter vector. Represents the parameter vector fitness Represents the optimal parameter vector fitness Represents the worst parameter vector fitness Let lg represent the total number of parameter vectors, and lg represent the logarithmic function.

[0118] Obtain the Euclidean distance between adjacent parameter vectors, and determine the distribution uniformity based on the Euclidean distance between adjacent parameter vectors:

[0119]

[0120] in, Indicates uniformity of distribution. This represents the q-th Euclidean distance, i.e., the... q The parameter vector after the permutation and the first permutation q +1 Euclidean distance between the parameter vectors after permutation. This represents the average Euclidean distance. This represents the Euclidean distance between the parameter vector after the first permutation and the solution center. Let represent the Euclidean distance between the parameter vector after the NP-th permutation and the solution center. The solution center = (UB + LB) / 2. UB represents the upper bound vector, that is, the parameter of each dimension is the upper bound mean of all parameter vectors in the same dimension; LB represents the lower bound vector, that is, the parameter of each dimension is the lower bound mean of all parameter vectors in the same dimension.

[0121] A smaller uniformity of distribution indicates a more even distribution of parameter vectors in the solution space, a wider coverage area, and higher overall quality. In this case, a fast search can be performed around the optimal parameter vector. Conversely, a larger uniformity of distribution indicates a less even distribution of parameter vectors in the solution space, requiring an increased search range.

[0122] When the distribution uniformity is less than a preset uniformity threshold, the parameter vector is searched based on the optimal parameter vector and the adaptive balance search factor to obtain the parameter vector after adaptive search:

[0123]

[0124] in, Indicates the first m The parameter vector after adaptive search, This represents the first random parameter vector selected from the permuted parameter vectors. Let represent the first random number between (-b, b), where b represents the oscillation factor, and , This represents the inverse hyperbolic tangent function; or, b can be set as a constant term, such as 0.4, 0.5, 0.6, etc.

[0125] When the distribution uniformity is greater than or equal to a preset uniformity threshold, the parameter vector is searched according to the upper limit and lower limit of the model parameters to obtain the parameter vector after adaptive search:

[0126]

[0127] in, This represents the second random number between (0,1). Represents inertia weight, and , It represents pi (π).

[0128] This application embodiment enables adaptive control of population distribution by performing an adaptive search based on the uniformity of the distribution of the parameter vector. Furthermore, it controls the search by adaptively balancing the search factor and inertia weight. This improves the cooperative search capability and search accuracy under uniform distribution conditions, as well as the global search capability under uneven distribution conditions. It also avoids excessive global search in the later stages, thus comprehensively improving the training effect of the algorithm.

[0129] In one possible implementation, performing a wave-like spiral chain cooperative search on the parameter vector after the adaptive search to obtain the parameter vector after the wave-like spiral chain cooperative search includes:

[0130] Based on the current number of training iterations, the waveform adjustment factor is determined as follows:

[0131]

[0132]

[0133] in, Indicates the waveform adjustment factor. Represents nonlinear control parameters. Represents pi;

[0134] If the volatility adjustment factor is greater than or equal to a preset adjustment threshold, then a spiral chain cooperative search is performed on the parameter vector after the adaptive search based on the optimal parameter vector to obtain the parameter vector after the volatility spiral chain cooperative search:

[0135]

[0136] in, Indicates the first t During the training process, the first n The parameter vector after adaptive search, Indicates the first n The parameter vector following a wave-like spiral chain cooperative search. Indicates the first t During the training process, the first n +1 parameter vectors after adaptive search; when n equals NP, This represents the parameter vector after a random adaptive search. Represents a third random number between (0, 1); This represents the spiral cooperative search coefficient. ; This represents the fourth random number between (0, 1);

[0137] If the volatility adjustment factor is less than the preset adjustment threshold, a random parameter vector is determined from the parameter vector after the adaptive search, and a spiral chain cooperative search is performed based on the parameter vector after the adaptive search of the random parameter vector to obtain the parameter vector after the volatility spiral chain cooperative search.

[0138]

[0139] in, This represents the second randomly generated parameter vector.

[0140] This invention employs a fluctuating spiral chain cooperative search on the parameter vector following the adaptive search. Different spiral chain cooperative search methods are selected through a fluctuation adjustment factor. This fluctuation adjustment factor increases with the number of training iterations, exhibiting ups and downs during the increase. Therefore, training with different spiral chain cooperative search methods effectively explores the space between different solutions, improving the ability to escape local optima while increasing the ability to find better solutions, thereby enhancing the global search capability. It can be seen that in the early stages of the algorithm, random parameter vectors are mainly used for spiral chain cooperative search, providing stronger global search capability while searching the unknown space. In the later stages, either random parameter vectors or optimal parameter vectors are used for spiral chain cooperative search, which can improve local search capability to some extent while maintaining solution diversity. In the later stages of the algorithm, only optimal parameter vectors are used for spiral chain cooperative search to ensure solution quality.

[0141] In one possible implementation, performing a bidirectional mutation search on the parameter vector following the wave-like spiral chain cooperative search to obtain the parameter vector after the bidirectional mutation search includes:

[0142] The number of mutation dimensions, positive mutation factor, and negative mutation factor are obtained based on the current training iterations:

[0143]

[0144]

[0145]

[0146] in, Indicates the number of mutation dimensions. This represents the floor function. This represents the total dimension of the model parameters. This represents the fifth random number between (0,1). Indicates the first t During the training process, the first k The parameter vector after the first wave of the wave-like spiral chain cooperative search. j 3D model parameters The corresponding positive mutation factor, Indicates the first t During the training process, the first k The parameter vector after the first wave of the wave-like spiral chain cooperative search. j 3D model parameters The corresponding reverse mutation factor, This indicates the total mutation order, which can be set to 2, 3, or 4;

[0147] Based on the number of mutation dimensions, the dimension to be mutated of the parameter vector after the volatile spiral chain cooperative search is determined.

[0148] For example, based on the number of mutation dimensions The parameter vector can be randomly selected from the parameters after the wave-like spiral chain cooperative search. The model parameters are used to obtain the dimension to be mutated of the parameter vector after the wave-like spiral chain cooperative search.

[0149] For any dimension to be mutated, a positive mutation factor is used to perform a positive mutation on the parameter vector after the wave-like spiral chain cooperative search, or a negative mutation factor is used to perform a negative mutation on the parameter vector after the wave-like spiral chain cooperative search, resulting in the parameter vector after the bidirectional mutation search:

[0150]

[0151] or

[0152]

[0153] in, Indicates the first t During the training process, the first k The dimension to be mutated in the parameter vector following a wave-like spiral chain cooperative search. Indicates the first k The parameter vector after a bidirectional mutation search is the dimension to be mutated. For dimensions that do not need to be mutated, the model parameters remain unchanged. This represents the mutation control coefficient, and is set to 0.4; Indicates the mutation order. The first parameter vector represents the optimal parameter vector.j 3D model parameters, Indicates a positive mutation factor. This indicates a reverse mutation factor.

[0154] Optionally, simulated annealing can be used to control the parameter vectors after the bidirectional mutation search. This increases the diversity of the parameter vectors and guides them to evolve in a better direction, which helps accelerate the convergence speed of the algorithm and improves the stability of the solution. Furthermore, after each change in the parameter vectors, out-of-bounds handling can be performed, and the optimal parameter vectors can be maintained using the out-of-bounds parameter vectors. For example, out-of-bounds model parameters can be randomly generated within the corresponding upper and lower limits or set to the nearest boundary value.

[0155] This application embodiment performs a bidirectional mutation search on the parameter vector after the wave-like spiral chain cooperative search, which has a powerful global search capability and helps the algorithm escape local optima.

[0156] This application provides an action management identification method based on earthquake site disaster investigation. The method includes: determining the earthquake site area to be investigated and planning multiple UAVs to perform disaster search tasks in the corresponding area; then distributing the disaster search tasks to the multiple UAVs and controlling them to execute the search tasks, obtaining the actual disaster investigation results; next, generating disaster points on an electronic map corresponding to the earthquake site area based on the actual disaster investigation results, thus obtaining a disaster investigation map; finally, feeding the disaster investigation map back to the equipment designated by the command personnel to guide their disaster relief actions. This method not only improves the efficiency of earthquake site disaster investigation but also generates comprehensive disaster data, enabling command personnel to fully understand the disaster situation and quickly formulate a comprehensive and optimal action plan, thereby improving the efficiency of disaster response.

[0157] like Figure 2 As shown, based on the same inventive concept, another embodiment of this application provides an action management identification device based on earthquake site disaster investigation, comprising:

[0158] The mission planning module 201 is used to determine the earthquake site area to be acted upon and to plan the disaster search missions of multiple UAVs in the earthquake site area to be acted upon.

[0159] The disaster investigation module 202 is used to distribute the disaster search task to multiple drones, control the drones to execute the disaster search task, and obtain the actual disaster investigation results corresponding to the disaster search task.

[0160] The disaster summary module 203 is used to generate disaster points on the electronic map corresponding to the earthquake site area to be acted upon, based on the actual disaster investigation results corresponding to the disaster search task, and obtain a disaster investigation map.

[0161] The disaster feedback module 204 is used to feed back the disaster survey map to the device designated by the commander to guide the commander's disaster relief actions and complete the action management based on the earthquake site disaster survey.

[0162] The action management identification device based on earthquake on-site disaster investigation provided in this application embodiment can execute the above-mentioned method and technical solution. Its principle and beneficial effects are similar, and will not be described again here.

[0163] like Figure 3 As shown, based on the same inventive concept, another embodiment of this application provides an electronic device, which includes a memory 301 and a processor 302. The memory 301 and the processor 302 communicate with each other via a communication bus 303.

[0164] The memory 301 is used to store code instructions.

[0165] The processor 302 is used to run the code instructions, causing the electronic device to execute the action management and identification method based on earthquake on-site disaster investigation provided in the embodiments of this application.

[0166] The aforementioned communication bus 303 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 303 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one thick line is used in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used for communication between the aforementioned terminal and other devices. The memory 301 can include random access memory (RAM), or it can include non-volatile memory, such as at least one disk storage device. Optionally, the memory 301 can also be at least one storage device located remotely from the aforementioned processor 302.

[0167] The processor 302 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0168] In addition, to achieve the above objectives, embodiments of this application also propose a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the action management identification method based on earthquake on-site disaster investigation provided in embodiments of this application.

[0169] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0170] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0171] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0172] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0173] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.

[0174] Finally, it should be noted that in the embodiments of this application, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0175] This application uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for action management identification based on earthquake on-site disaster investigation, characterized in that, include: Identify the earthquake-prone area requiring action and plan disaster search missions using multiple drones within that area. The disaster search task is distributed to multiple drones, and the drones are controlled to execute the disaster search task to obtain the actual disaster investigation results corresponding to the disaster search task. Based on the actual disaster investigation results corresponding to the disaster search task, disaster points are generated on the electronic map corresponding to the earthquake site area to be acted upon, and a disaster investigation map is obtained. The disaster survey map is fed back to the device designated by the commander to guide the commander's disaster relief actions and complete the action management based on the earthquake site disaster survey; The process of determining the earthquake site area to be addressed and planning a multi-UAV disaster search mission corresponding to that area includes: By obtaining the closed areas identified by disaster relief personnel on the electronic map through human-computer interaction, the earthquake site area to be acted upon can be obtained; Obtain information on power transmission lines and roads in the earthquake-prone area; Based on the power transmission line information of the earthquake site area to be acted upon, a first disaster search task is generated to conduct a disaster investigation along the power transmission lines; Based on the road information, a second disaster search task is generated to conduct a disaster investigation along the road. With the goal of traversing the earthquake site area to be acted upon, a third disaster search task is generated to investigate the disaster situation of trapped personnel and collapsed buildings in the earthquake site area to be acted upon. The step of distributing the disaster search task to multiple drones, controlling the drones to execute the disaster search task, and obtaining the actual disaster investigation results corresponding to the disaster search task includes: The first disaster search task is sent to the first UAV equipped with the power transmission line detection model, and the first UAV is controlled to perform target detection along the power transmission line to obtain the power transmission line investigation results. The second disaster search task is sent to the second UAV equipped with a road detection model, and the second UAV is controlled to perform target detection along the road to obtain the road survey results; The third disaster search task is sent to the third UAV carrying the personnel and housing detection model, and the third UAV is controlled to traverse the earthquake site area to be acted upon to obtain the personnel and housing survey results; The results of the personnel and housing surveys, road surveys, and power line surveys are collectively used as the actual disaster survey results corresponding to the disaster search task. The mounting methods for the transmission line detection model, road detection model, and personnel and building detection model are the same, and all include: A hybrid mapping strategy is used to initialize the model to be mounted, resulting in multiple parameter vectors; wherein, the model to be mounted is a transmission line detection model, a road detection model, or a personnel and building detection model; Obtain the fitness corresponding to each parameter vector, and obtain the optimal parameter vector based on the fitness of the parameter vector; Based on the optimal parameter vector, an adaptive search based on the uniformity of distribution is performed on the parameter vector to obtain the parameter vector after the adaptive search. A wave-like spiral chain cooperative search is performed on the parameter vector after the adaptive search to obtain the parameter vector after the wave-like spiral chain cooperative search. A bidirectional mutation search is performed on the parameter vector after the wave-like spiral chain cooperative search to obtain the parameter vector after the bidirectional mutation search. Determine if the current training iterations have reached the maximum number of training iterations. If so, obtain the optimal parameter vector again based on the parameter vector obtained after the bidirectional mutation search, and then obtain the target parameter vector. Otherwise, return to the step of obtaining the optimal parameter vector and proceed to the next training process. The target parameter vector is used as the final parameter of the model to be mounted, and the model to be mounted is deployed on the drone according to the final parameter of the model to be mounted.

2. The action management identification method based on earthquake on-site disaster investigation according to claim 1, characterized in that, The transmission line detection model, road detection model, and personnel and building detection model are all set to YOLO models.

3. The action management identification method based on earthquake on-site disaster investigation according to claim 2, characterized in that, The hybrid mapping strategy is used to initialize the model to be mounted, resulting in multiple parameter vectors, including: The model parameters of the model to be mounted are randomly initialized between the upper limit and the lower limit of the model parameters, and the initialized model parameters are encoded into a vector to obtain the basic vector; Based on the aforementioned base vector, multiple other vectors are obtained using a chaotic mapping sequence. Based on the other vectors, the upper limit of the model parameters, and the lower limit of the model parameters, a parameter vector is generated, and all other vectors are processed to obtain multiple parameter vectors.

4. The action management identification method based on earthquake on-site disaster investigation according to claim 2, characterized in that, The step of obtaining the fitness corresponding to each parameter vector and obtaining the optimal parameter vector based on the fitness of the parameter vectors includes: For any parameter vector, obtain the loss function value corresponding to the parameter vector, and add the loss function value to a preset non-zero constant term and take the reciprocal to obtain the fitness corresponding to the parameter vector; The parameter vector with the highest fitness is taken as the optimal parameter vector.

5. The action management identification method based on earthquake on-site disaster investigation according to claim 2, characterized in that, The step of performing an adaptive search based on the distribution uniformity of the parameter vector, based on the optimal parameter vector, to obtain the parameter vector after the adaptive search, includes: The parameter vectors are arranged in ascending order of fitness. Based on the arranged parameter vectors, and according to the fitness of the optimal parameter vector, the fitness of the parameter vectors, and the current number of training iterations, an adaptive balance search factor is obtained. Obtain the Euclidean distance between adjacent parameter vectors, and determine the distribution uniformity based on the Euclidean distance between adjacent parameter vectors; When the distribution uniformity is less than a preset uniformity threshold, the parameter vector is searched according to the optimal parameter vector and the adaptive balance search factor to obtain the parameter vector after adaptive search. When the distribution uniformity is greater than or equal to a preset uniformity threshold, the parameter vector is searched according to the upper limit and lower limit of the model parameters to obtain the parameter vector after adaptive search.

6. The action management identification method based on earthquake on-site disaster investigation according to claim 5, characterized in that, The step of performing a wave-like spiral chain cooperative search on the parameter vector after the adaptive search to obtain the parameter vector after the wave-like spiral chain cooperative search includes: Determine the volatility adjustment factor based on the current number of training iterations; If the volatility adjustment factor is greater than or equal to the preset adjustment threshold, then a spiral chain cooperative search is performed on the parameter vector after the adaptive search based on the optimal parameter vector to obtain the parameter vector after the volatility spiral chain cooperative search. If the volatility adjustment factor is less than the preset adjustment threshold, a random parameter vector is determined from the parameter vector after the adaptive search, and a spiral chain cooperative search is performed based on the parameter vector after the adaptive search of the random parameter vector to obtain the parameter vector after the volatility spiral chain cooperative search.

7. The action management identification method based on earthquake on-site disaster investigation according to claim 6, characterized in that, The step of performing a bidirectional mutation search on the parameter vector after the wave-like spiral chain cooperative search to obtain the parameter vector after the bidirectional mutation search includes: The number of mutation dimensions, positive mutation factor, and negative mutation factor are obtained based on the current number of training iterations. Based on the number of mutation dimensions, the dimension to be mutated of the parameter vector after the volatile spiral chain cooperative search is determined. For any dimension to be mutated, a positive mutation factor is used to perform a positive mutation on the parameter vector after the volatile spiral chain cooperative search, or a negative mutation factor is used to perform a negative mutation on the parameter vector after the volatile spiral chain cooperative search, to obtain the parameter vector after the bidirectional mutation search.

Images