A forest fire rescue path planning method and system

Abstract

The present application provides a kind of forest fire rescue path planning method and system, the present application relates to fire rescue technical field, the method includes obtaining forest fire area information, forest information, trapped personnel information;Based on each front verification path segment, forest fire area information determines the unmanned aerial vehicle front exploration flight path information of each front verification path segment;Based on the estimated traffic information of each extension planning path segment, the estimated traffic information of each front verification path segment is determined based on the unmanned aerial vehicle front exploration verification video of each front verification path segment;Based on the estimated traffic information of each front verification path segment, the estimated traffic information of each extension planning path segment is used to generate forest fire rescue path planning scheme, and forest fire rescue path planning scheme is sent to rescue command terminal, the method can efficiently and accurately identify the traffic risk of forest fire rescue path and determine the adaptive forest fire rescue path planning scheme.

Description

Technical Field

[0001] This invention relates to the field of fire rescue technology, specifically to a method and system for planning rescue routes in forest fires. Background Technology

[0002] Forest fire rescue route planning is a crucial aspect of forest fire emergency response, as the rationality of the rescue route directly affects the efficiency of evacuating trapped personnel and the safety of rescue workers. Traditional forest fire rescue route planning primarily relies on rescue commanders to formulate rescue plans based on topographic maps, forest road information, and fire monitoring results. This involves determining the direction and route of rescue by assessing the fire's extent, road distribution, and the location of trapped personnel. However, this method has significant limitations. Within a forest fire zone, the location of the fire line, the direction of smoke plume spread, and local access conditions continuously change with wind direction, terrain, and the state of combustion. Manual planning struggles to fully consider the dynamic impacts of the fire line approaching different areas, smoke plume pressing and accumulating in valleys, and changes in ventilation conditions on ridges on rescue routes. There is also a lack of systematic analysis of the access risks of different route segments during actual rescue operations. Often, route feasibility is judged solely based on existing road information and on-site experience. Similarly, reliable accessibility assessments are lacking for extended route segments not directly explored. This is especially true in areas with complex forest terrain, poor road continuity, or rapidly changing smoke and fire disturbance areas. Static maps and manual experience alone are insufficient to fully identify potential hazards along the route, such as approaching fire lines, smoke plume obstruction, road blockages, and terrain-restricted access. This limitation not only leads to insufficient basis for rescue route planning but also easily results in route interruptions, temporary reversals, or missed safe passage opportunities for rescue teams after entering high-risk areas. Insufficient analysis of the actual traffic conditions of the preliminary verification route sections and the estimated traffic conditions of the extended planned route sections will further weaken the reliability of the rescue route planning scheme, reduce the safety margin of forest fire rescue operations, and exacerbate the problems of delayed rescue of trapped personnel and ineffective consumption of rescue resources. In addition, traditional methods rely on manual judgment and on-site temporary adjustments, resulting in low route planning efficiency and difficulty in adapting to the actual needs of dynamic and precise rescue route planning under the conditions of rapid forest fire spread.

[0003] Therefore, how to efficiently and accurately identify the risks of forest fire rescue routes and determine suitable forest fire rescue route planning schemes is an urgent problem to be solved. Summary of the Invention

[0004] The main technical problem addressed by this invention is how to efficiently and accurately identify the risks of forest fire rescue routes and determine suitable forest fire rescue route planning schemes.

[0005] According to a first aspect, the present invention provides a forest fire rescue route planning method, comprising: acquiring forest fire area information, forest information, and trapped personnel information; determining multiple fire line approach areas, multiple smoke plume-pressed valley areas, and multiple ridge smoke-avoidance passage areas based on the forest fire area information, the forest information, and the trapped personnel information; determining multiple forward verification route segments and multiple extended planning route segments based on the multiple fire line approach areas, the multiple smoke plume-pressed valley areas, the multiple ridge smoke-avoidance passage areas, and the trapped personnel information; determining UAV forward flight path information for each forward verification route segment and the forest fire area information; and determining UAV forward flight path information for each forward verification route segment. The system retrieves reconnaissance route information and controls drones to acquire drone reconnaissance verification videos for each reconnaissance route segment. Based on these videos, it determines the estimated passability information for each reconnaissance route segment. Then, based on the multiple extended planning route segments, the forest fire area information, the drone reconnaissance verification videos for each reconnaissance route segment, and the estimated passability information for each reconnaissance route segment, it determines the estimated passability information for each extended planning route segment. Finally, it generates a forest fire rescue route planning scheme based on the estimated passability information for each reconnaissance route segment and the estimated passability information for each extended planning route segment, and sends the forest fire rescue route planning scheme to the rescue command terminal.

[0006] In one possible implementation, determining multiple forward verification path segments and multiple extended planning path segments based on the multiple fire line approach areas, the multiple smoke plume-pressed valley areas, the multiple ridge smoke-avoidance passage areas, and the trapped personnel information includes: determining multiple essential rescue route points and multiple rescue avoidance points based on the multiple fire line approach areas, the multiple smoke plume-pressed valley areas, the multiple ridge smoke-avoidance passage areas, and the trapped personnel information; clustering the multiple essential rescue route points to obtain multiple clusters; determining multiple candidate rescue path segments and smoke and fire disturbance information for each candidate rescue path segment based on the multiple clusters and the multiple rescue avoidance point information; and determining multiple forward verification path segments and multiple extended planning path segments based on the multiple candidate rescue path segment information and the smoke and fire disturbance information for each candidate rescue path segment.

[0007] In one possible implementation, generating a forest fire rescue route planning scheme based on the estimated accessibility information of each forward verification path segment and each extended planning path segment includes: constructing a rescue analysis graph, which includes multiple forward verification path segment nodes, multiple extended planning path segment nodes, and multiple edges between nodes. The node characteristics of the forward verification path segment nodes include the estimated accessibility information of the forward verification path segment, the node characteristics of the extended planning path segment nodes include the estimated accessibility information of the extended planning path segment, and the edges between nodes represent the positional relationships between different path segments; processing the rescue analysis graph based on a rescue analysis processing model to obtain multiple necessary rescue route segment information; and generating a forest fire rescue route planning scheme based on the multiple necessary rescue route segment information, the multiple necessary rescue point information, and the multiple rescue avoidance point information.

[0008] In one possible implementation, the input to the rescue analysis and processing model is the rescue analysis map, and the output of the rescue analysis and processing model is information on multiple necessary rescue route segments.

[0009] According to a second aspect, the present invention provides a forest fire rescue route planning system, comprising: an acquisition module for acquiring forest fire area information, forest information, and trapped personnel information; an area determination module for determining multiple fire line approach areas, multiple plume-pressed valley areas, and multiple ridge smoke-avoidance passage areas based on the forest fire area information, the forest information, and the trapped personnel information; a route segment determination module for determining multiple forward verification route segments and multiple extended planning route segments based on the multiple fire line approach areas, the multiple plume-pressed valley areas, the multiple ridge smoke-avoidance passage areas, and the trapped personnel information; a flight path determination module for determining UAV forward flight path information for each forward verification route segment based on each forward verification route segment and the forest fire area information; and a video acquisition module for determining video based on each forward verification route segment. The system includes: a drone forward flight path information module, which controls the drone to acquire drone forward verification video for each forward verification path segment; a first access information determination module, which determines the estimated access information for each forward verification path segment based on the drone forward verification video for each forward verification path segment; a second access information determination module, which determines the estimated access information for each extended planning path segment based on the multiple extended planning path segments, the forest fire area information, the drone forward verification video for each forward verification path segment, and the estimated access information for each forward verification path segment; and a scheme generation module, which generates a forest fire rescue route planning scheme based on the estimated access information for each forward verification path segment and the estimated access information for each extended planning path segment, and sends the forest fire rescue route planning scheme to the rescue command terminal.

[0010] In one possible implementation, the path segment determination module is further configured to: determine multiple essential rescue route information and multiple rescue avoidance point information based on the multiple fire line approach areas, the multiple smoke plume-pressed valley areas, the multiple ridge smoke avoidance and passage areas, and the trapped personnel information; cluster the multiple essential rescue route information to obtain multiple clusters; determine multiple candidate rescue path segment information and smoke and fire disturbance information for each candidate rescue path segment based on the multiple clusters and the multiple rescue avoidance point information; and determine multiple forward verification path segments and multiple extended planning path segments based on the multiple candidate rescue path segment information and the smoke and fire disturbance information for each candidate rescue path segment.

[0011] In one possible implementation, the scheme generation module is further configured to: construct a rescue analysis map, the rescue analysis map including multiple forward verification path segment nodes, multiple extended planning path segment nodes, and multiple edges between nodes, the node features of the forward verification path segment nodes including the estimated passage information of the forward verification path segment, the node features of the extended planning path segment nodes including the estimated passage information of the extended planning path segment, and the edges between nodes representing the positional relationship between different path segments; process the rescue analysis map based on a rescue analysis processing model to obtain multiple necessary rescue route segment information; and generate a forest fire rescue route planning scheme based on the multiple necessary rescue route segment information, the multiple necessary rescue point information, and the multiple rescue avoidance point information.

[0012] In one possible implementation, the input to the rescue analysis and processing model is the rescue analysis map, and the output of the rescue analysis and processing model is information on multiple necessary rescue route segments.

[0013] According to a third aspect, embodiments of the present invention provide an electronic device, including: a processor; a memory; and a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the method as described above, the method including: acquiring forest fire area information, forest information, and trapped personnel information; determining multiple fire line approach zones, multiple plume-pressed valley zones, and multiple ridge smoke-avoidance passage zones based on the forest fire area information, the forest information, and the trapped personnel information; determining multiple forward verification path segments and multiple extended planning path segments based on the multiple fire line approach zones, the multiple plume-pressed valley zones, the multiple ridge smoke-avoidance passage zones, and the trapped personnel information; and determining each forward verification path segment based on each forward verification path segment and the forest fire area information. The system obtains advance flight path information for each advance verification path segment; based on the advance flight path information for each advance verification path segment, it controls the drone to acquire advance verification video for each advance verification path segment; based on the advance verification video for each advance verification path segment, it determines the estimated passability information for each advance verification path segment; based on the multiple extended planning path segments, the forest fire area information, the advance verification video for each advance verification path segment, and the estimated passability information for each advance verification path segment, it determines the estimated passability information for each extended planning path segment; based on the estimated passability information for each advance verification path segment and the estimated passability information for each extended planning path segment, it generates a forest fire rescue route planning scheme and sends the forest fire rescue route planning scheme to the rescue command terminal.

[0014] According to the fourth aspect, this embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the aforementioned forest fire rescue path planning method. The method includes: acquiring forest fire area information, forest information, and trapped personnel information; determining multiple fire line approach areas, multiple smoke plume-pressed valley areas, and multiple ridge smoke-avoidance passage areas based on the forest fire area information, the forest information, and the trapped personnel information; determining multiple forward verification path segments and multiple extended planning path segments based on the multiple fire line approach areas, the multiple smoke plume-pressed valley areas, the multiple ridge smoke-avoidance passage areas, and the trapped personnel information; and determining UAV forward reconnaissance for each forward verification path segment based on each forward verification path segment and the forest fire area information. Based on the UAV forward flight path information for each forward verification path segment, the UAV is controlled to acquire forward verification video for each forward verification path segment; based on the forward verification video for each forward verification path segment, the estimated passability information for each forward verification path segment is determined; based on the multiple extended planning path segments, the forest fire area information, the forward verification video for each forward verification path segment, and the estimated passability information for each forward verification path segment, the estimated passability information for each extended planning path segment is determined; based on the estimated passability information for each forward verification path segment and the estimated passability information for each extended planning path segment, a forest fire rescue route planning scheme is generated, and the forest fire rescue route planning scheme is sent to the rescue command terminal.

[0015] The present invention has at least the following beneficial effects:

[0016] 1. This invention addresses the complex and ever-changing conditions in rescue areas caused by the interplay of fire line advance, smoke plume spread, and terrain conditions within forest fire environments. It identifies different environmental impact units within a forest fire area, resulting in spatial divisions such as fire line approach zones, smoke plume-pressed valley zones, and ridge-based smoke-avoidance and passage zones. Compared to methods that only determine areas based on the fire's extent, this invention provides a more detailed description of the environmental differences within the fire-affected area, offering a more accurate basis for subsequent rescue path analysis.

[0017] 2. This invention introduces a UAV forward verification mechanism, which collects information on thermal convection interference, smoke imaging interference, and flight attitude stability at the corresponding locations of candidate forward verification path segments. This makes path evaluation no longer limited to existing map data and static environmental parameters, but can analyze different path segments in combination with real-time changes at the fire scene, thereby improving the reliability of path status judgment in complex fire environments.

[0018] 3. This invention combines regional environmental identification results with forward verification results to comprehensively analyze the environmental adaptability and traffic conditions of different path segments, and determines rescue path schemes accordingly. This avoids the limitations of traditional methods that select paths based solely on distance, terrain, or a single risk factor, enabling the generated rescue paths to better adapt to smoke obstruction, thermal airflow disturbances, and spatial environmental changes during forest fires, thereby improving the safety and effectiveness of the rescue planning process. Attached Figure Description

[0019] Figure 1 A flowchart illustrating a forest fire rescue route planning method provided in an embodiment of the present invention;

[0020] Figure 2 This is a schematic diagram of a ground fire monitoring device provided in an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of a UAV thermal imaging device provided in an embodiment of the present invention;

[0022] Figure 4 A schematic diagram of plume concentration distribution in a plume-pressed valley region provided in an embodiment of the present invention;

[0023] Figure 5 A schematic diagram illustrating the spatial distribution of multiple fire line approach zones, multiple smoke plume downdraft valley zones, and multiple ridge smoke avoidance and passage zones provided for embodiments of the present invention;

[0024] Figure 6 This is a flowchart illustrating the process of determining multiple forward verification path segments and multiple extended planning path segments, as provided in an embodiment of the present invention.

[0025] Figure 7 This is a schematic diagram illustrating the changes in the thermal convection interference intensity sequence and the smoke imaging interference intensity sequence of a forward verification path segment, provided in an embodiment of the present invention.

[0026] Figure 8 This is a schematic diagram illustrating the relationship between the visible distance change sequence of a forward verification path segment provided in an embodiment of the present invention.

[0027] Figure 9 This is a flowchart illustrating a method for generating a forest fire rescue route planning scheme, provided by an embodiment of the present invention.

[0028] Figure 10 This is a schematic diagram of a forest fire rescue route planning system provided in an embodiment of the present invention. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings. Similar elements in different embodiments are referred to by associated similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the invention. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present invention are not shown or described in the specification. This is to avoid obscuring the core parts of the invention with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0030] In this embodiment of the invention, the following are provided: Figure 1 The forest fire rescue route planning method shown in some embodiments, Figure 1 The forest fire rescue route planning method shown includes steps S1 to S8:

[0031] Step S1: Obtain information on the forest fire area, forest information, and information on trapped personnel.

[0032] Forest fire area information is obtained by detecting areas where forest fires have occurred using ground-based fire monitoring equipment and drone thermal imaging equipment. Figure 2 This is a schematic diagram of a ground fire monitoring device provided in an embodiment of the present invention. Figure 3 This is a schematic diagram of a drone thermal imaging device provided in an embodiment of the present invention.

[0033] Forest fire area information can include the coordinates of the fire center, the coordinate sequence of the fire edge outline, the temperature distribution of the fire, the local wind speed and direction, the vector of fire spread direction, the distribution sequence of flame radiation intensity, the direction of smoke plume movement, and the coverage area of ​​the smoke plume.

[0034] Forest information is obtained by retrieving data from the forestry geographic information system database. It describes the static natural environment data of the area where a forest fire occurred, including topography, vegetation, and existing access conditions. Forest information may include the distribution of ground elevation, ground slope, slope orientation, ridge boundaries, valley boundaries, vegetation cover density, forest road distribution, water body boundaries, and forest clearing areas in the forest fire area.

[0035] Information on trapped personnel is obtained through signal positioning of portable communication devices and scanning of infrared life detectors. It includes information on the location of trapped personnel and their rescue needs within the forest fire area. The information on trapped personnel may include the spatial coordinates of the trapped personnel, the number of trapped personnel, and the degree of obstruction at their location.

[0036] Step S2: Based on the forest fire area information, the forest information, and the trapped personnel information, determine multiple fire line approach areas, multiple smoke plume-pressed valley areas, and multiple ridge smoke-avoidance and passage areas.

[0037] In some embodiments, a region determination model can be used to identify multiple fire line approach zones, multiple smoke plume-pressed valley zones, and multiple ridge-based smoke-avoidance and passage zones. The region determination model is a Transformer model. The inputs to the region determination model are the forest fire area information, the forest information, and the information of trapped personnel; the outputs of the region determination model are the multiple fire line approach zones, multiple smoke plume-pressed valley zones, and multiple ridge-based smoke-avoidance and passage zones.

[0038] The Transformer model is a deep learning model based on the self-attention mechanism. Through self-attention layers, the Transformer model calculates the correlation between different positions in the input sequence, allowing data at each position to be updated in conjunction with data from other positions in the sequence. The Transformer model can process the input sequence in parallel and capture the dependencies between data points that are geographically distant.

[0039] Multiple fire approach zones are identified through a region determination model, representing several areas where the combustion front is continuously spreading towards surrounding unburned areas. These fire approach zones can include region boundary coordinates, combustion front location, combustion front advance direction, fire spread rate, and flame temperature distribution.

[0040] Multiple plume-pressing valley areas were identified using a regional determination model. These areas within a forest fire zone are characterized by plume deposition and accumulation, leading to significant deterioration in visibility and air quality due to the orientation of the topographic valleys. A plume-pressing valley area can include regional boundary coordinates, valley boundary extent, valley orientation, regional ground elevation distribution, plume concentration distribution, and regional average visibility.

[0041] Figure 4 This is a schematic diagram illustrating the plume concentration distribution in a subduction valley region, as provided in an embodiment of the present invention. (See attached diagram.) Figure 4 In the figure, the horizontal axis represents the spatial coordinate of the plume subduction valley area, the vertical axis represents the spatial coordinate of the plume subduction valley area, and different gray areas represent the spatial distribution of normalized plume concentration values.

[0042] Multiple ridge-based smoke-sheltered access zones were determined using a regional identification model. These zones are located near ridgelines within a forest fire area, where higher elevation allows them to avoid the downward pressure of the smoke plume and provide passage for people. Ridge-based smoke-sheltered access zones can include regional boundary coordinates, regional ground elevation distribution, ridgeline location, and road location distribution.

[0043] Figure 5This is a schematic diagram illustrating the spatial distribution of multiple fire line approach zones, multiple smoke plume depression valley zones, and multiple ridge smoke-avoidance passage zones, provided as an embodiment of the present invention. (See attached diagram.) Figure 5 The horizontal axis of the figure represents the spatial coordinates of the forest fire-affected area, and the vertical axis represents the spatial coordinates of the forest fire-affected area. Different marked areas represent the spatial distribution of the fire-approaching zone, the plume-pressed valley zone, and the ridge-based smoke-avoidance and passage zone, respectively. The fire-approaching zone is located near the direction of fire spread and indicates areas where the approaching fire line has a significant impact during fire spread. The plume-pressed valley zone is distributed along local valley topography and indicates areas where the plume sinks and accumulates due to topographic constraints. The ridge-based smoke-avoidance and passage zone is distributed along ridge topography and indicates relatively passable areas affected by topographic uplift. The figure also shows the boundaries of the forest fire-affected area, the location of the fire source, and the direction of plume spread, reflecting the spatial distribution relationships between these different regional units.

[0044] The Transformer model, through its self-attention mechanism, establishes a global connection between high-dimensional representations of forest fire area information, forest information, and trapped personnel information. The self-attention layer focuses on analyzing forest fire area information to characterize the nonlinear mapping between the fire spread direction vector and the terrain slope orientation, thereby calculating spatial grid nodes where the fire front is obstructed by terrain or accelerated by downwind. The feedforward neural network layer aggregates and delineates multiple fire front approach zones based on the calculated grid nodes, indicating that the fire front is continuously spreading towards surrounding unburned areas. The Transformer model's cross-attention module extracts the valley boundary range from the forest information and spatially overlaps and extrapolates this boundary range with the plume movement direction and plume coverage area from the forest fire area information. The model can identify low-lying areas where terrain orientation restricts plume airflow and causes significant plume sinking and accumulation, thus outputting multiple plume-pressed valley areas with significantly deteriorated visibility and air quality. When processing the spatial coordinates of trapped personnel in the trapped personnel information, the self-attention layer further searches for relatively high and flat ridge boundaries in the forest information. The model identifies spaces that can completely avoid the downdraft of smoke plumes and have the conditions for forest road distribution by comparing the distribution sequence of flame radiation intensity in the ridge boundary range and forest fire area information. Finally, it accurately outputs multiple ridge smoke-avoidance passage areas that can be easily passed by people.

[0045] Step S3: Based on the multiple fire line approach zones, the multiple smoke plume downdraft valley zones, the multiple ridge smoke avoidance and passage zones, and the information on trapped personnel, determine multiple forward verification path segments and multiple extended planning path segments.

[0046] In some embodiments, Figure 6This is a flowchart illustrating the process of determining multiple forward verification path segments and multiple extended planning path segments according to an embodiment of the present invention. The determination of multiple forward verification path segments and multiple extended planning path segments includes steps S31 to S34:

[0047] Step S31: Based on the multiple fire line approaching zones, the multiple smoke plume pressing valley zones, the multiple ridge smoke avoidance and passage zones, and the information of the trapped personnel, determine multiple necessary rescue route points and multiple rescue avoidance points.

[0048] In some embodiments, a location determination model can be used to determine information on multiple essential rescue routes and multiple rescue avoidance points. The location determination model is a deep neural network. The inputs to the location determination model are the multiple fire line approach zones, the multiple smoke plume-pressed valley zones, the multiple ridge smoke-avoidance passage zones, and the information on the trapped personnel. The outputs of the location determination model are the information on multiple essential rescue routes and multiple rescue avoidance points.

[0049] A deep neural network is a neural network model consisting of an input layer, multiple hidden layers, and an output layer. Deep neural networks use multiple layers of neurons to perform continuous non-linear transformations on the input data, gradually learning the complex correspondences between different parts of the input data, and then generating results corresponding to the processing task through the output layer.

[0050] The information on essential rescue routes is specific information about key locations that must be traversed to complete a rescue mission, determined by a location determination model. Each essential rescue route can include its spatial coordinates, ground elevation, fire impact level, smoke plume impact level, surface accessibility level, and area type.

[0051] The rescue avoidance point information is determined by a point determination model, which identifies multiple key points that need to be avoided during forest fire rescue operations. Each rescue avoidance point's information may include the point's spatial coordinates, ground elevation, distance from the fire front, plume density, visibility, and area type.

[0052] The fire approach zone indicates the location where the fire front may continue to approach during the rescue process; the smoke plume depression valley zone reveals the valley location where smoke tends to sink and accumulate, causing visibility deterioration; and the ridge smoke avoidance passage zone can define a passage space that is more suitable for avoiding the impact of the smoke plume.

[0053] A deep neural network uses nonlinear transformations across multiple hidden layers to project the spatial coordinates of multiple fire-approaching zones, smoke-plunging valleys, ridge-based smoke-avoidance zones, and trapped personnel information into a unified grid space. The input layer of the deep neural network encodes the spatial coordinates, number of trapped personnel, and degree of obstruction of their location from the trapped personnel information. The hidden layers combine the regional boundary coordinates, regional ground elevation distribution, ridgeline positions, and road locations within the multiple ridge-based smoke-avoidance zones to learn the spatial relationships between the trapped personnel's locations and passable areas. The model can identify potential rescue approach directions connecting the trapped personnel's locations to safe passable areas. The model can integrate the fire and smoke plume impact features corresponding to multiple fire-approaching zones and smoke-plunging valleys to comprehensively evaluate the passability risk at different approach locations and identify candidate safe coordinate points with high surface passability and low point fire impact levels. The model further aggregates the number of trapped personnel to calculate the spatial scale and capacity required for rescue operations, and then extracts and outputs information on multiple essential rescue points that meet the conditions for encampment and rescue deployment. The parallel computing branch of the deep neural network specifically handles the physical characteristics of high-risk elements in the spatial grid. The hidden layer assesses the degree of environmental thermal radiation attenuation by calculating the Euclidean distance from spatial coordinates to the advance direction of the combustion front in multiple fire-approaching zones. Based on the valley boundary range, plume concentration distribution, and regional average visibility in multiple plume-pressed valley areas, the deep neural network performs feature evaluation on the degree of plume impact at different spatial locations and reduces the passability evaluation weight of coordinate nodes located within high plume impact areas. The model summarizes the coordinate data of areas far from the combustion front but deeply embedded in the smoke deposition zone, and after classification and calculation by the output layer, generates information on multiple rescue avoidance points that need to be avoided on the rescue route.

[0054] Step S32: Cluster the multiple rescue route information to obtain multiple clusters.

[0055] In some embodiments, the K-means clustering algorithm can be used to cluster the information on the multiple essential rescue points to obtain multiple clusters.

[0056] K-means clustering is a clustering algorithm that groups samples based on the distance between their features. By setting the number of clusters, K-means clustering assigns input samples to the nearest cluster centers and updates the cluster centers multiple times to reduce the distance between samples within the same cluster and increase the differences between samples in different clusters.

[0057] Multiple clusters are multiple point sets formed by dividing multiple essential rescue route information through similarity analysis using the K-means clustering algorithm. Each cluster includes multiple essential rescue route information that have high similarity in terms of spatial location, terrain conditions, and fire impact characteristics.

[0058] In some embodiments, the value of K can be determined by a preset relationship table between the number of clusters K and the total number of essential rescue points within the forest fire area. The larger the total number of essential rescue points within the forest fire area, the larger the value of K. The preset relationship table between the number of clusters K and the total number of essential rescue points is artificially constructed in advance.

[0059] As an example, specifically: the K-means clustering algorithm standardizes the spatial coordinates, ground elevation, fire impact level, smoke plume impact level, and surface accessibility level of multiple essential rescue route points. The algorithm sets multiple initial cluster centers for each essential rescue route point and calculates the overall similarity between each rescue route point and each cluster center in terms of spatial location, terrain conditions, smoke and fire impact, and accessibility. After each essential rescue route point is assigned to the cluster with the smallest overall difference, the K-means clustering algorithm redetermines the cluster centers based on all essential rescue route points within the current cluster, gradually bringing the cluster centers closer to the actual average state of the points within the cluster. Redetermining the cluster centers changes the assignment of some boundary points; the K-means clustering algorithm continuously adjusts the assignments and updates the cluster centers until the essential rescue route points no longer exhibit significant migration. After clustering, the essential rescue route points within the same cluster maintain a high degree of similarity in spatial location, terrain conditions, and fire impact characteristics, while clear differences emerge between different clusters, ultimately resulting in multiple clusters that can represent different rescue spatial groups.

[0060] By clustering multiple essential rescue routes into clusters, rescue routes with high similarity in spatial location, terrain conditions, and fire impact characteristics can be grouped into rescue units operating at the same pace, thereby achieving structured regional integration of dispersed rescue targets. This classification method can clearly delineate concentrated rescue operation areas within different geographical zones, avoiding disordered path planning for scattered essential routes. This facilitates the rapid connection of candidate rescue routes based on the overall spatial distribution of clusters, significantly improving the efficiency of rescue route segment generation and the rationality of overall planning.

[0061] Step S33: Based on the multiple clusters and the multiple rescue avoidance point information, determine multiple candidate rescue path segment information and smoke and fire disturbance information for each candidate rescue path segment.

[0062] In some embodiments, a candidate path segment determination model can be used to determine multiple candidate rescue path segment information and smoke and fire disturbance information for each candidate rescue path segment. The candidate path segment determination model is a deep neural network. The input to the candidate path segment determination model is the multiple clusters and the multiple rescue avoidance point information, and the output of the candidate path segment determination model is the multiple candidate rescue path segment information and the smoke and fire disturbance information for each candidate rescue path segment.

[0063] Multiple candidate rescue route segments are identified by a candidate route segment determination model, providing detailed information on alternative routes that connect different clusters and avoid rescue avoidance points. Each candidate rescue route segment can include the coordinate sequence of points along its path, route length, route accessibility level, terrain undulation, and slope variation sequence.

[0064] The smoke and fire disturbance information for each candidate rescue path segment is specific information about the degree of impact of smoke and fire along the path segment, determined by the candidate path segment determination model. This information may include the smoke concentration distribution and visibility values ​​along the path segment.

[0065] Multiple clusters organize information on essential rescue points with similar spatial locations, terrain conditions, and fire impact characteristics into different point sets, indicating which rescue spatial groups need to be connected by candidate rescue path segments and the direction of connection. Multiple rescue avoidance point information clearly marks locations unsuitable for traversal and reflects hazard constraints such as the distance from the fire front, the smoke plume concentration, and visibility. Multiple clusters determine the connection targets of the path segments corresponding to the candidate rescue path information, while multiple rescue avoidance point information restricts the available space of the path segments corresponding to the candidate rescue path information and provides a reference for nearby hazards in estimating the impact of smoke and fire along the route.

[0066] After receiving information from multiple clusters and rescue avoidance points through the input layer, the deep neural network first extracts features from the essential rescue points within the clusters, acquiring spatial distribution relationships, point connectivity relationships, and regional accessibility features between different clusters. The deep neural network further encodes the spatial coordinates of the points, their distance from the combustion front, plume concentration, and visibility from the multiple rescue avoidance point information, forming environmental risk features reflecting the impact of fire and plume. The hidden layer fuses the spatial connectivity features between clusters and environmental risk features through multi-layer nonlinear mapping, establishing an internal representation of candidate connectivity relationships between different clusters. It adjusts the path connection weights based on the correlation between candidate connection locations and rescue avoidance point information, ensuring that the generated candidate connections avoid high-risk areas. The deep neural network generates multiple candidate path trajectories based on the connectivity relationships between different clusters and further analyzes the continuity and safety of these candidate path trajectories by combining the path's location, path connection direction, and information from nearby rescue avoidance points. The model uses hidden layers to fuse features from the coordinate sequences of transit points, path length, path accessibility level, path terrain undulation, and path slope variation sequences in candidate path trajectories. This process evaluates the accessibility of different candidate path trajectories and filters them to form multiple candidate rescue path segments that connect different clusters and meet safety constraints. The deep neural network further infers the degree of impact of the smoke and fire environment on different candidate rescue path segments based on the spatial relationship between the coordinate sequences of transit points and multiple rescue avoidance points in the multiple candidate rescue path segments, combined with the smoke plume concentration and visibility at each point, thus determining the smoke and fire disturbance information for each candidate rescue path segment.

[0067] Step S34: Based on the information of the multiple candidate rescue path segments and the smoke and fire disturbance information of each candidate rescue path segment, determine multiple forward verification path segments and multiple extended planning path segments.

[0068] In some embodiments, a forward path segment determination model can be used to determine multiple forward verification path segments and multiple extended planning path segments. The forward path segment determination model is a deep neural network. The input to the forward path segment determination model is the information of the multiple candidate rescue path segments and the smoke and fire disturbance information of each candidate rescue path segment. The output of the forward path segment determination model is the multiple forward verification path segments and the multiple extended planning path segments.

[0069] Multiple forward-exploration verification path segments are determined by the forward-exploration path segment determination model from multiple candidate rescue path segments. These are the multiple path segments that require drones to enter the site to collect video and verify the passability.

[0070] Multiple extended planning path segments are a set of subsequent travel segments used to further extend and connect into the deeper areas after preliminary exploration and verification.

[0071] Information on multiple candidate rescue route segments presents the spatial orientation, length, terrain undulations, and accessibility of alternative routes, enabling the assessment of the connection points and extension value of each candidate rescue route segment within the rescue route. Smoke and fire disturbance information for each candidate rescue route segment reflects changes in smoke concentration and visibility along the route, revealing locations where on-site conditions change rapidly and accessibility is difficult to confirm based on existing information alone. The route structure and the degree of smoke and fire disturbance support the differentiation between advance verification needs and subsequent extended applications.

[0072] The deep neural network first extracts spatial features from multiple candidate rescue route segments, including the coordinate sequence of transit points, route length, route accessibility level, route terrain undulation, and route slope change sequence. Simultaneously, it extracts environmental state features from the smoke and fire disturbance information along each candidate rescue route segment, including the smoke concentration distribution and visibility values. Through nonlinear mapping between multiple layers of hidden nodes, the deep neural network fuses the spatial state features of the route with the smoke and fire disturbance state features, forming a comprehensive feature representation for evaluating the on-site confirmability and subsequent extension capability of candidate rescue route segments. The model further uses the forward propagation process to calculate the state differences between different candidate rescue route segments and adjusts the verification requirement weights of corresponding route segments based on the fluctuation of the route slope change sequence, the change in smoke concentration distribution along the route, and the stability of visibility values ​​along the route. This prioritizes candidate rescue route segments with significant on-site state changes and where static information cannot accurately reflect accessibility. The deep neural network simultaneously considers the spatial connectivity between multiple candidate rescue path segments to analyze their impact on subsequent route extensions, and calculates path extension evaluation results based on path continuity, traffic stability, and environmental disturbance levels. The output layer classifies candidate rescue path segments according to their verification requirement weights and path extension evaluation results. Candidate rescue path segments with high verification requirements are identified as multiple forward verification path segments, while those with high extension evaluations and the ability to form continuous rescue directions are identified as multiple extended planning path segments. Finally, the system outputs multiple forward verification path segments and multiple extended planning path segments.

[0073] Step S4: Determine the UAV forward flight path information for each forward verification path segment based on each forward verification path segment and the forest fire area information.

[0074] In some embodiments, a flight path information determination model can be used to determine the UAV forward flight path information for each forward verification path segment. The flight path information determination model is a deep neural network. The input to the flight path information determination model is each forward verification path segment and the forest fire area information; the output of the flight path information determination model is the UAV forward flight path information for each forward verification path segment.

[0075] The UAV forward flight path information for each forward verification path segment is used to control the UAV to complete the on-site reconnaissance flight along the corresponding forward verification path segment. The UAV forward flight path information for each forward verification path segment may include the UAV's takeoff position coordinates, flight path entry position coordinates, flight waypoint coordinate sequence, flight altitude sequence, flight speed sequence, shooting orientation sequence, and flight path end position coordinates, etc.

[0076] The forward verification path segment determines the location and extension direction of the ground route that needs to be explored by the drone on-site, which can limit the shooting coverage area and the path-related areas that the drone needs to continuously observe. Forest fire area information can reflect the fire edge, local wind field, fire spread and smoke plume movement, which can constrain the high-temperature areas, windward smoke plume areas and rapidly approaching fire directions that the drone needs to avoid.

[0077] Deep neural networks can project the ground trajectory of each forward verification path segment into a three-dimensional flight space above the forest fire area, and set continuous observation positions along the forward verification path segment. The drone's takeoff position coordinates are determined based on a location far from the fire edge, with safe takeoff space, and able to approach the forward verification path segment. The flight path entry position coordinates fall above the starting area of ​​the forward verification path segment and avoid the main movement direction of the smoke plume. The deep neural network raises the flight altitude of nearby high-temperature areas according to the fire temperature distribution and flame radiation intensity distribution sequence, and adjusts the lateral spacing of adjacent flight waypoints according to local wind speed and direction to prevent the drone from staying in strong crosswind areas for a long time. When the smoke plume coverage is large or the smoke plume movement direction intersects with the flight path direction, the deep neural network will change the flight waypoint coordinate sequence and reduce the flight speed, so that visible light cameras and infrared thermal imaging equipment can obtain more stable continuous images. The shooting orientation sequence always points to the ground passage space corresponding to the forward verification path segment, and the shooting orientation is deflected in advance at turning positions to reduce the observation gap. The deep neural network determines the coordinates of the end position of the flight path near the end of the forward verification path segment, and checks whether the flight altitude sequence, flight speed sequence and waypoint spacing meet the requirements for continuous shooting, and finally generates the UAV forward flight path information for each forward verification path segment.

[0078] In some embodiments, determining the UAV forward flight path information for each forward verification path segment based on each forward verification path segment and the forest fire area information includes steps S41 to S43:

[0079] Step S41: Based on each forward verification path segment and the forest fire area information, determine the UAV forward thermal convection interference intensity sequence and smoke imaging interference intensity sequence for each forward verification path segment.

[0080] In some embodiments, a deep neural network can be used to determine the UAV forward thermal convection interference intensity sequence and smoke imaging interference intensity sequence for each forward verification path segment.

[0081] The UAV forward thermal convection disturbance intensity sequence for each forward verification path segment is determined by a deep neural network. It is a sequence of the degree of impact of hot airflow on the UAV on the forward verification path segment. The UAV forward thermal convection disturbance intensity sequence for each forward verification path segment includes the thermal convection disturbance values ​​at different locations on the path segment.

[0082] The smoke imaging interference intensity sequence for each forward verification path segment is determined by a deep neural network, representing the degree to which smoke affects the image clarity of the imaging device. The UAV forward smoke imaging interference intensity sequence for each forward verification path segment includes smoke imaging interference values ​​at different locations along the path segment. Figure 7 This is a schematic diagram illustrating the changes in the thermal convection interference intensity sequence and the smoke imaging interference intensity sequence of a forward verification path segment, provided as an embodiment of the present invention. (Refer to the attached diagram.) Figure 7 In the figure, the horizontal axis represents the spatial location or sampling point number of the forward verification path segment, and the vertical axis represents the normalized interference intensity value. The solid line in the figure represents the sequence of thermal convection interference intensity during UAV forward exploration, and the dashed line represents the sequence of smoke imaging interference intensity. The two curves reflect the changes in the degree of thermal convection disturbance and smoke imaging impact on the UAV at different path locations, respectively. The interference intensity at different locations in the figure shows a continuous variation trend, and the range of interference intensity variation is represented by a reference threshold line.

[0083] The primary neurons of the deep neural network can spatially superimpose wind direction and speed with local topography in forest fire area information to deduce the heat transfer vector field within the fire zone. Subsequently, the network's deep convolutional layers precisely locate and anchor the trajectory of each forward verification path segment within this heat transfer vector field, and simulate the evolution of thermal convection vortices generated when surface heat sources radiate upwards into three-dimensional space using nonlinear activation functions. In this physical simulation, the deep neural network can identify airflow disturbance peaks formed at different height levels above the forward verification path segment under specific wind speeds and fire spread pressures, and assess the degree of spatial fragmentation caused by this vertically and horizontally interwoven shear thermal airflow. The network concatenates and normalizes these disturbance intensity values ​​according to the spatial coordinates of the path nodes, thus outputting a sequence of UAV forward thermal convection interference intensity for each forward verification path segment. Simultaneously, another hidden layer of the deep neural network performs three-dimensional spatial diffusion calculations on the plume coverage and spread vector in the forest fire area information. The model simulates the suspended concentration distribution of tiny carbon particles in the space above the target path segment and the offset and accumulation effect caused by wind direction. It can calculate the transmittance loss of environmental particulate matter to visible and infrared light at different height domains layer by layer. The deep network accurately captures dense smoke areas that are prone to diffuse reflection and severe light absorption, and arranges the absolute image quality loss values ​​caused by dense smoke obstruction along the direction of the path nodes. Finally, the output layer processes and outputs a smoke imaging interference intensity sequence.

[0084] Step S42: Based on the UAV forward thermal convection interference intensity sequence, smoke imaging interference intensity sequence, and forest fire area information for each forward verification path segment, determine the UAV flight attitude stability control tolerance level, thermal radiation safe flight space domain, and effective observation space domain for imaging clarity for each forward verification path segment.

[0085] In some embodiments, deep neural networks can be used to determine the UAV flight attitude stability control tolerance level, thermal radiation safe flight space domain, and imaging clarity effective observation space domain for each forward verification path segment.

[0086] The UAV flight attitude stability control tolerance level for each forward verification path segment is determined by a deep neural network. It characterizes the maximum allowable attitude deviation margin for the UAV to maintain stable operation of the sensing equipment on the corresponding path segment. The UAV flight attitude stability control tolerance level for each forward verification path segment includes the allowable attitude tilt angle deviation level and the allowable upper limit of angular acceleration at different locations on the path segment.

[0087] The thermal radiation safe flight space domain for each forward verification path segment is determined by a deep neural network, which defines the safe physical boundary of the UAV from being burned by high-temperature heat sources.

[0088] The effective observation spatial domain for imaging clarity of each forward verification path segment is determined by a deep neural network, ensuring that the camera equipment carried by the UAV can penetrate the smoke for effective data collection.

[0089] The deep neural network first extracts dimensional features from the UAV's forward thermal convection interference intensity sequence for each forward verification path segment. The model identifies abrupt change nodes and peak features in the thermal convection values ​​within the sequence and establishes a mapping relationship with the learned UAV mechanical wind resistance and attitude adjustment limits. This allows it to calculate the maximum attitude fluctuation limit that the UAV can withstand to maintain stable flight at the corresponding location, thereby determining the UAV's flight attitude stability control tolerance level for each forward verification path segment. Next, the deep neural network performs cross-feature calculations on the UAV's forward thermal convection interference intensity sequence for each forward verification path segment and the fire temperature distribution and flame radiation intensity distribution from the forest fire area information. By extrapolating the attenuation process of thermal radiation in space, the model calculates the three-dimensional spatial boundary that satisfies the UAV's hardware safe operating temperature limits, eliminating high-temperature areas exceeding the safe temperature range, thus determining the UAV's thermal radiation safe flight spatial domain for each forward verification path segment. Finally, the deep neural network analyzes the spatial gradient changes of the UAV's forward smoke imaging interference intensity sequence for each forward verification path segment. The model calculates the attenuation rate of visible and infrared light under different smoke concentrations, and then deduces the maximum allowable observation distance and altitude limits of the UAV relative to the ground under the premise of ensuring qualified image quality. The model extracts the spatial regions that meet the image clarity requirements and finally determines the effective observation spatial domain of UAV imaging clarity for each forward verification path segment.

[0090] Step S43: Based on the UAV flight attitude stability control tolerance level, thermal radiation safe flight space domain, and imaging clarity effective observation space domain of each forward verification path segment, determine the UAV forward flight path information for each forward verification path segment.

[0091] In some embodiments, a deep neural network can be used to determine the UAV forward flight path information for each forward verification path segment.

[0092] The deep neural network utilizes hidden layers to perform geometric intersection calculations on the safe flight space domain of UAV thermal radiation and the effective observation space domain of UAV imaging clarity for each forward verification path segment. Within the feasible three-dimensional flight channel enclosed by upper and lower curved surfaces, the deep neural network arranges discrete grid points. Subsequently, the model transforms the UAV flight attitude stability control tolerance level for each forward verification path segment into a kinematic penalty weight matrix and attaches it to these grids. When solving the state transition equations between adjacent grid nodes, the model automatically identifies and avoids transition directions that could cause drastic changes in UAV attitude or trigger dynamic system anomalies. The neural network performs dynamic topology planning along the extension direction of the path to be explored, extracting points-by-point connections with smooth curvature and reasonable observation angles from the ground target within this safe space envelope. With multiple forward propagation and backward verification convergences of the deep network's hidden layers, the model determines the physical locations of collision-free and visually unobstructed coherent points in three-dimensional space, and simultaneously matches each spatial node with a throughput rate that meets the stability tolerance level and the corresponding airborne gimbal pitch angle. Finally, the deep neural network transforms this optimal topology connection that meets safety and quality requirements into specific trajectory data, thereby outputting the UAV forward flight path information for each forward verification path segment.

[0093] Step S5: Based on the UAV forward flight path information for each forward verification path segment, control the UAV to acquire the UAV forward verification video for each forward verification path segment.

[0094] The drone forward verification video for each forward verification path segment is obtained by continuously filming the corresponding forward verification path segment using visible light camera and infrared thermal imaging equipment carried by the drone while the drone flies according to the forward flight path information for each forward verification path segment.

[0095] The drone forward verification video of the forward verification path section is used to record the ground passage space, obstacle occupancy range, vegetation burning range, smoke plume coverage range, visibility changes along the route, and temperature changes along the route for the corresponding forward verification path section.

[0096] Step S6: Determine the estimated passage information for each forward verification path segment based on the UAV forward verification video of each forward verification path segment.

[0097] In some embodiments, a first access information determination model can be used to determine the estimated access information for each forward verification path segment. The first access information determination model is a gated loop unit. The input to the first access information determination model is the UAV forward verification video for each forward verification path segment, and the output of the first access information determination model is the estimated access information for each forward verification path segment.

[0098] A gated recurrent unit (ROU) is a type of recurrent neural network with update and reset gates. The update gate controls the proportion of historical states retained in the current computation, while the reset gate controls the degree to which historical states influence the current input. A gated ROU can process continuous data sequentially and preserve the relationships between different time points within the continuous data.

[0099] The estimated passage information for each forward verification path segment is determined by the first passage information determination model. This information includes the expected passage capacity of each forward verification path segment under the current smoke and fire conditions. The estimated passage information for each forward verification path segment may include the effective passage width, obstacle occupancy length, smoke coverage ratio, high-temperature zone distribution, visibility distance change sequence, and continuous passage time of the path segment.

[0100] Figure 8 This is a schematic diagram illustrating the relationship between the visible distance change sequence of a forward verification path segment, provided as an embodiment of the present invention. (Refer to...) Figure 8 , Figure 8 This diagram illustrates the dynamic change of estimated visibility distance over time for each forward exploration verification path segment under different smoke concentration levels. The horizontal axis represents the video frame time or UAV advance time for each forward exploration verification path segment, in seconds (s); the vertical axis represents the estimated visibility distance, in meters (m). Different curves in the diagram represent the changes in visibility distance under different smoke concentration levels, and the visibility-limited intervals within different advance periods are marked.

[0101] The advance verification video of the UAV on the advance verification route section recorded continuous frames of images in chronological order, including the ground passage space, the range of obstacles occupying the road, and the range of vegetation burning on the target road section. The video data completely preserved the dynamic process of smoke diffusion and the real image details of the ground road conditions in the real fire scene. The temporal visual information provides an intuitive observation basis for the model to capture the changes in the visible distance in the harsh fire scene environment and to estimate the time window that can be continuously passed.

[0102] The gated recurrent unit (GRU) demonstrates powerful time-series tracking capabilities when processing continuous video image frames, relying on the recurrent network topology. The GRU receives continuous frame data arranged in time sequence from the UAV forward verification video of each forward verification path segment. The model's update gate is responsible for preserving and transmitting the fire spread characteristics and smoke concentration evolution trends accumulated in past time steps. When analyzing image information frame by frame, the GRU can cumulatively correlate and calculate the obstacle-occupied area, vegetation burning area, and smoke coverage ratio extracted from the image along the time axis. As the video sequence progresses, the GRU can deduce the diffusion rate of the high-temperature ground section and the decay curve of the visible distance change sequence. The state vector within the model integrates the dynamic deterioration trend of road conditions over a sustained period to calculate the dynamic narrowing of the effective passage width. The recurrent neuron nodes further comprehensively calculate the critical time point when smoke obscures the road and the time difference when high-temperature flames block the road surface, predicting the limit window duration for which rescue teams can safely move forward on the corresponding road segment. The output node of the gated loop unit summarizes the predicted physical impedance and spatiotemporal constraint indicators from various dimensions, and accurately outputs the estimated passage information for each forward verification path segment, including the effective passage width and continuous passage time of the path segment.

[0103] Step S7: Based on the multiple extended planning path segments, the forest fire area information, the UAV forward verification video of each forward verification path segment, and the estimated passage information of each forward verification path segment, determine the estimated passage information of each extended planning path segment.

[0104] In some embodiments, a second access information determination model can be used to determine the estimated access information for each extended planning path segment. The second access information determination model is a Transformer model. The inputs to the second access information determination model are the plurality of extended planning path segments, the forest fire area information, the UAV forward verification video of each forward verification path segment, and the estimated access information for each forward verification path segment. The output of the second access information determination model is the estimated access information for each extended planning path segment.

[0105] The estimated traffic information for each extended planned path segment is predicted by the second traffic information determination model, which provides information on the expected traffic capacity of each extended planned path segment under the current smoke and fire conditions. The estimated traffic information for each extended planned path segment may include the effective traffic width, obstacle occupancy length, smoke coverage ratio, high-temperature zone distribution, visibility distance change sequence, and continuous traffic time of the path segment.

[0106] Forest fire area information provides the macro-environmental background of the fire site. The drone forward verification videos of the forward verification path segments and the estimated traffic information of each forward verification path segment provide the actual road condition change patterns of locally known road segments under the current real fire environment. The verified traffic status evolution patterns of local road segments, combined with global fire site environmental parameters, can serve as reliable inference reference samples, supporting the model to reasonably extrapolate and predict the traffic status of surrounding extended road segments that have not been directly explored by drones.

[0107] The Transformer model utilizes a self-attention mechanism to simultaneously capture the trajectory of multiple extended planning path segments, as well as macroscopic wind direction and fire spread vectors from forest fire area information. The Transformer model's cross-attention layer uses the smoke diffusion dynamics shown in the drone-based advance verification video of each advance verification path segment and the estimated passability information of each advance verification path segment as key factual evidence. During computation, the Transformer model establishes a similarity mapping matrix between the extended planning path segments and the advance verification path segments in terms of spatial distance and terrain features. When the attention mechanism detects an extended planning path segment that is leeward and has continuous terrain with a known high-risk advance verification path segment, the self-attention layer transfers the evolutionary patterns of rapidly increasing smoke coverage and sharply decreasing visibility in the known path segment to the extended planning path segment. The feedforward neural network layer performs nonlinear correction and compensation on the transferred state indicators based on spatial distance and terrain differences. The Transformer model, through cross-regional attention weight allocation, can fully absorb the intensity of vegetation burning spread reflected in local field videos, accurately predicting the potential increase in the length of obstacles occupying the extended path and the spread trend of high-temperature sections. The model outputs structured traffic indicators for each extended road segment over a future period, deriving estimated traffic information for each extended planned road segment, covering the effective passage width and continuous passage time of the extended planned road segment.

[0108] Step S8: Generate a forest fire rescue route planning scheme based on the estimated passage information of each forward verification route segment and the estimated passage information of each extended planning route segment, and send the forest fire rescue route planning scheme to the rescue command terminal.

[0109] In some embodiments, Figure 9 This is a flowchart illustrating a method for generating a forest fire rescue route planning scheme according to an embodiment of the present invention. The method for generating the forest fire rescue route planning scheme includes steps S81 to S83:

[0110] Step S81: Construct a rescue analysis map. The rescue analysis map includes multiple forward verification path segment nodes, multiple extended planning path segment nodes, and multiple edges between nodes. The node features of the forward verification path segment nodes include the estimated passage information of the forward verification path segment. The node features of the extended planning path segment nodes include the estimated passage information of the extended planning path segment. The edges between nodes represent the positional relationship between different path segments.

[0111] The rescue analysis graph is a graph data consisting of multiple forward verification path segment nodes, multiple extended planning path segment nodes, and edges between multiple nodes. The rescue analysis graph is used to represent the expected traffic status of different path segments and the spatial connectivity between different path segments.

[0112] Multiple forward verification path segment nodes are nodes used in the rescue analysis map to represent multiple forward verification path segments. Each forward verification path segment node corresponds to one forward verification path segment, and the node characteristics of each forward verification path segment node are the estimated passage information of the corresponding forward verification path segment.

[0113] Multiple extended planning path segment nodes are nodes used in the rescue analysis map to represent multiple extended planning path segments. Each extended planning path segment node corresponds to one extended planning path segment, and the node characteristics of each extended planning path segment node are the estimated passage information of the corresponding extended planning path segment.

[0114] The edges between multiple nodes represent the graph connections used to link nodes on different path segments in the rescue analysis graph. The edges between multiple nodes also represent the positional relationships between different path segments.

[0115] The positional relationship between different path segments includes the distance between the endpoints of the path segments and the angle of the connection direction of the path segments.

[0116] Step S82: Based on the rescue analysis and processing model, the rescue analysis map is processed to obtain information on multiple necessary rescue route segments.

[0117] In some embodiments, the rescue analysis and processing model is a graph neural network. The input to the rescue analysis and processing model is the rescue analysis map, and the output of the rescue analysis and processing model is information on multiple necessary rescue route segments.

[0118] Graph Neural Networks (GNNs) are deep learning models used to process graph data. GNNs transmit node content through the connections between nodes in the graph and aggregate the content of neighboring nodes to update the representation of each node. GNNs can simultaneously process node content, edge attributes, and graph connection structure, and output node selection results and graph analysis results.

[0119] Information on multiple essential rescue route segments is used to identify several route segments that need to be included in the final forest fire rescue route. This information can include the spatial coordinate sequence of the essential rescue route segments, the direction of travel for the essential rescue route segments, the estimated passability information for the essential rescue route segments, and the safety level for travel for the essential rescue route segments.

[0120] The rescue analysis graph uses a graph structure to represent the connections and positional relationships between path segments, visually displaying the preliminary verification path segments and extended planning path segments, along with their geographic topological relationships. Nodes in the preliminary verification path segments and extended planning path segments each carry their respective estimated accessibility information, while edges between nodes represent the positional relationships between different path segments. This structured representation facilitates the understanding and processing of complex firefield road network relationships by graph neural networks. Through node features and edge information, graph neural networks can better capture the connectivity and influence relationships between the accessibility of different path segments. For example, node features provide the expected accessibility of each path segment under the current smoke and fire environment, while edges reflect the connections and extensions of these accessibility features under specific spatial geographic topology. The construction of the rescue analysis graph effectively organizes the path segments and their spatial relationships, facilitating processing by graph neural networks. This organizational method reduces computational complexity and improves the model's inference efficiency.

[0121] Because the rescue analysis map provides structured node and edge information, graph neural networks can better process this information, avoiding the problems of data sparsity and computational dimensionality disasters encountered by traditional path planning algorithms in complex firefield geographic grids. This structural advantage makes graph neural networks more advantageous in handling complex spatial road network relationships. By comprehensively analyzing the characteristics of nodes in the forward verification path segments, the nodes in the extended planning path segments, and the positional relationships between nodes, graph neural networks can filter out multiple essential rescue route segments that can both ensure the safety of rescue passage and efficiently connect various areas.

[0122] Graph Neural Networks (GNNs) propagate features and perform depth aggregation along the edges between multiple nodes in the rescue analysis graph using a message passing mechanism. First, the GNN extracts estimated passage information from each forward verification path segment node and extended planning path segment node, including effective passage width, obstacle occupancy length, smoke coverage ratio, high-temperature zone distribution, visibility distance change sequence, and continuous passage time of the path segment. These parameters are then converted into initial feature vectors for the corresponding nodes. Based on the edge information between nodes in the rescue analysis graph, the GNN encodes the spatial connectivity between adjacent path segment nodes and adjusts the information transmission strength between nodes according to the positional relationship, endpoint distance, and connection direction angle between different path segments. Path segment nodes with small endpoint distances and smooth, continuous connection direction angles maintain strong correlations during feature updates, allowing the passage status of adjacent path segments to influence each other. Path segment nodes with large endpoint distances or obvious reversals in the connection direction have reduced correlations, thus minimizing the interference of abnormal path connections on the overall route judgment. After multiple rounds of message passing, the features of each path segment node not only include its own estimated passage status but also integrate the continuous passage capability of adjacent path segments and local route connectivity. The graph neural network further performs correlation analysis on the updated node features to identify path segment node combinations that can maintain effective passage width, minimize smoke and fire disturbance, and possess continuous connectivity. It then extracts continuous rescue routes along the connections between nodes. The output layer organizes the spatial coordinate sequence of path segments, the direction of passage for essential rescue routes, and the safety level of passage for essential rescue routes based on the selected node connections, ultimately outputting information on multiple essential rescue route segments.

[0123] Step S83: Generate a forest fire rescue route planning scheme based on the information of the multiple necessary rescue route segments, the information of the multiple necessary rescue points, and the information of the multiple rescue avoidance points.

[0124] In some embodiments, a planning scheme determination model can be used to generate forest fire rescue route planning schemes. The planning scheme determination model is a deep neural network. The inputs to the planning scheme determination model are the information on multiple necessary rescue route segments, multiple necessary rescue points, and multiple rescue avoidance points; the output of the planning scheme determination model is the forest fire rescue route planning scheme.

[0125] A forest fire rescue route planning scheme is a route execution plan used to guide rescue personnel from the rescue starting point to the location of trapped personnel. A forest fire rescue route planning scheme may include a global route coordinate sequence, time limits for passing through each essential rescue route segment, execution order of each essential rescue route segment, direction of travel for each essential rescue route segment, and estimated rescue completion time.

[0126] Information on essential rescue routes provides a spatially continuous sequence of coordinates ensuring high safety, establishing the main connecting parts of the rescue path. Information on essential rescue points indicates the coordinates and access requirements of nodes that the rescue operation must reach sequentially, clarifying the locations that must be included in the route planning. Information on rescue avoidance points marks the coordinate areas in the fire scene where there is a threat of high concentrations of smoke or flames, limiting the areas that are prohibited from entering during route planning.

[0127] The deep neural network first encodes the spatial coordinate sequences, travel directions, and safety levels of multiple essential rescue route segments. It then establishes route connections based on the spatial adjacency of these segments. Through multiple rounds of feature propagation, the model analyzes the continuity between these essential rescue route segments, identifying the initial route structure and mapping multiple essential rescue point information to the corresponding route structure to determine the coverage of the rescue target location. For essential rescue point information not covered by the current route structure, the model evaluates candidate connection locations by combining the spatial relationships of adjacent essential rescue route segments, their safety levels, and path extension directions, generating route connections that supplement the target location. The deep neural network further integrates the distance between the point and the combustion front, the plume concentration, and the visibility of the point from multiple rescue avoidance point information, imposing risk constraints on the newly added route connections. This ensures that the final route structure avoids high-risk areas while maintaining continuous connectivity between multiple essential rescue point information. After optimizing the route structure, the deep neural network determines the direction of travel on essential rescue routes based on the spatial connections between adjacent routes. It then determines the execution order of these routes by considering the rescue entry point, the location of trapped personnel, and the connection sequence between each route. The model further extracts smoke coverage, high-temperature zone distribution, and continuous travel time from the estimated travel information of each route segment. It analyzes the continued passability of each route segment, determines the limited passage time for each segment, and calculates the estimated rescue completion time based on the path length. Finally, the deep neural network organizes the global route coordinate sequence, the limited passage time for each essential rescue route segment, the execution order of each segment, the direction of travel, and the estimated rescue completion time, generating a forest fire rescue route planning scheme.

[0128] Based on the same inventive concept Figure 10 This is a schematic diagram of a forest fire rescue route planning system provided in an embodiment of the present invention. The forest fire rescue route planning system includes:

[0129] Module 91 is used to acquire information on forest fire areas, forest information, and information on trapped personnel.

[0130] The area determination module 92 is used to determine multiple fire line approach areas, multiple smoke plume pressure valley areas, and multiple ridge smoke avoidance and passage areas based on the forest fire area information, the forest information, and the trapped personnel information;

[0131] The path segment determination module 93 is used to determine multiple forward verification path segments and multiple extended planning path segments based on the multiple fire line approach areas, the multiple smoke plume downpressure valley areas, the multiple ridge smoke avoidance and passage areas, and the information of trapped personnel.

[0132] The route determination module 94 is used to determine the UAV forward route information for each forward verification path segment based on each forward verification path segment and the forest fire area information;

[0133] The video acquisition module 95 is used to control the UAV to acquire the UAV forward verification video for each forward verification path segment based on the UAV forward flight path information for each forward verification path segment.

[0134] The first passage information determination module 96 is used to determine the estimated passage information of each forward verification path segment based on the UAV forward verification video of each forward verification path segment.

[0135] The second traffic information determination module 97 is used to determine the estimated traffic information of each extended planning path segment based on the multiple extended planning path segments, the forest fire area information, the UAV forward verification video of each forward verification path segment, and the estimated traffic information of each forward verification path segment.

[0136] The scheme generation module 98 is used to generate a forest fire rescue route planning scheme based on the estimated passage information of each forward verification route segment and the estimated passage information of each extended planning route segment, and send the forest fire rescue route planning scheme to the rescue command terminal.

[0137] It should be noted that, in order to simplify the descriptions disclosed herein and thus aid in the understanding of one or more embodiments of the invention, the foregoing description of embodiments of this specification may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this method of disclosure does not imply that the subject matter of this specification requires more features than those mentioned in the claims. In fact, the embodiments contain fewer features than all the features of a single embodiment disclosed above.

[0138] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A method for planning forest fire rescue routes, characterized in that, include: Obtain information on forest fire areas, forest conditions, and trapped individuals; Based on the forest fire area information, the forest information, and the information on trapped personnel, multiple fire line approach areas, multiple smoke plume-pressed valley areas, and multiple ridge smoke-avoidance and passage areas were identified. Based on the multiple fire line approach areas, the multiple smoke plume downdraft valley areas, the multiple ridge smoke avoidance and passage areas, and the information on trapped personnel, multiple forward verification path segments and multiple extended planning path segments were determined; The UAV forward flight path information for each forward verification path segment is determined based on the forest fire area information. Based on the UAV forward flight path information for each forward verification path segment, control the UAV to acquire the UAV forward verification video for each forward verification path segment. Based on the UAV forward verification video of each forward verification path segment, the estimated passage information of each forward verification path segment is determined. Based on the multiple extended planning path segments, the forest fire area information, the UAV forward verification video of each forward verification path segment, and the estimated passage information of each forward verification path segment, the estimated passage information of each extended planning path segment is determined. Based on the estimated passage information of each forward verification path segment and the estimated passage information of each extended planning path segment, a forest fire rescue path planning scheme is generated, and the forest fire rescue path planning scheme is sent to the rescue command terminal.

2. The forest fire rescue route planning method as described in claim 1, characterized in that, The determination of multiple forward verification path segments and multiple extended planning path segments based on the multiple fire line approach zones, the multiple smoke plume-pressed valley zones, the multiple ridge smoke-avoidance and passage zones, and the information on trapped personnel includes: Based on the multiple fire line approaching areas, the multiple smoke plume pressing valley areas, the multiple ridge smoke avoidance and passage areas, and the information of trapped personnel, multiple rescue essential route points and multiple rescue avoidance points are determined; Multiple clusters were obtained by clustering based on the information of the multiple essential rescue points; Based on the information of the multiple clusters and the multiple rescue avoidance points, multiple candidate rescue path segments and smoke and fire disturbance information for each candidate rescue path segment are determined. Based on the information of the multiple candidate rescue path segments and the smoke and fire disturbance information of each candidate rescue path segment, multiple forward verification path segments and multiple extended planning path segments are determined.

3. The forest fire rescue route planning method as described in claim 2, characterized in that, The process of generating a forest fire rescue route planning scheme based on the estimated accessibility information of each forward-tested route segment and the estimated accessibility information of each extended planned route segment includes: A rescue analysis graph is constructed, which includes multiple forward verification path segment nodes, multiple extended planning path segment nodes, and multiple edges between nodes. The node characteristics of the forward verification path segment nodes include the estimated passage information of the forward verification path segment, the node characteristics of the extended planning path segment nodes include the estimated passage information of the extended planning path segment, and the edges between nodes represent the positional relationship between different path segments. Based on the rescue analysis and processing model, the rescue analysis map is processed to obtain information on multiple necessary rescue route segments; Based on the information of the multiple necessary rescue route segments, the multiple necessary rescue points, and the multiple rescue avoidance points, a forest fire rescue route planning scheme is generated.

4. The forest fire rescue route planning method as described in claim 3, characterized in that, The input to the rescue analysis and processing model is the rescue analysis map, and the output of the rescue analysis and processing model is information on multiple necessary rescue route segments.

5. A forest fire rescue route planning system, characterized in that, include: The acquisition module is used to acquire information on forest fire areas, forest information, and information on trapped people. The area determination module is used to determine multiple fire line approach areas, multiple smoke plume pressure valley areas, and multiple ridge smoke avoidance and passage areas based on the forest fire area information, the forest information, and the trapped personnel information; The path segment determination module is used to determine multiple forward verification path segments and multiple extended planning path segments based on the multiple fire line approach areas, the multiple smoke plume downdraft valley areas, the multiple ridge smoke avoidance and passage areas, and the information of trapped personnel. The route determination module is used to determine the UAV forward flight route information for each forward verification path segment based on each forward verification path segment and the forest fire area information; The video acquisition module is used to control the UAV to acquire the UAV forward verification video for each forward verification path segment based on the UAV forward flight path information for each forward verification path segment. The first passage information determination module is used to determine the estimated passage information of each forward verification path segment based on the UAV forward verification video of each forward verification path segment. The second traffic information determination module is used to determine the estimated traffic information of each extended planning path segment based on the multiple extended planning path segments, the forest fire area information, the UAV forward verification video of each forward verification path segment, and the estimated traffic information of each forward verification path segment. The scheme generation module is used to generate a forest fire rescue route planning scheme based on the estimated passage information of each forward verification route segment and the estimated passage information of each extended planning route segment, and send the forest fire rescue route planning scheme to the rescue command terminal.

6. The forest fire rescue route planning system as described in claim 5, characterized in that, The path segment determination module is also used for: Based on the multiple fire line approaching areas, the multiple smoke plume pressing valley areas, the multiple ridge smoke avoidance and passage areas, and the information of trapped personnel, multiple rescue essential route points and multiple rescue avoidance points are determined; Multiple clusters were obtained by clustering based on the information of the multiple essential rescue points; Based on the information of the multiple clusters and the multiple rescue avoidance points, multiple candidate rescue path segments and smoke and fire disturbance information for each candidate rescue path segment are determined. Based on the information of the multiple candidate rescue path segments and the smoke and fire disturbance information of each candidate rescue path segment, multiple forward verification path segments and multiple extended planning path segments are determined.

7. The forest fire rescue route planning system as described in claim 6, characterized in that, The scheme generation module is also used for: A rescue analysis graph is constructed, which includes multiple forward verification path segment nodes, multiple extended planning path segment nodes, and multiple edges between nodes. The node characteristics of the forward verification path segment nodes include the estimated passage information of the forward verification path segment, the node characteristics of the extended planning path segment nodes include the estimated passage information of the extended planning path segment, and the edges between nodes represent the positional relationship between different path segments. Based on the rescue analysis and processing model, the rescue analysis map is processed to obtain information on multiple necessary rescue route segments; Based on the information of the multiple necessary rescue route segments, the multiple necessary rescue points, and the multiple rescue avoidance points, a forest fire rescue route planning scheme is generated.

8. The forest fire rescue route planning system as described in claim 7, characterized in that, The input to the rescue analysis and processing model is the rescue analysis map, and the output of the rescue analysis and processing model is information on multiple necessary rescue route segments.

9. An electronic device, characterized in that, include: processor; Memory; And a computer program; wherein the computer program is stored in the memory and configured to be executed by the processor to implement the forest fire rescue route planning method as described in any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the forest fire rescue route planning method as described in any one of claims 1 to 4.

Images