Fire spread prediction and emergency warning method and system based on on-board computation

Abstract

The application relates to a fire spread prediction and emergency early warning method and system based on on-board calculation, and belongs to the technical field of remote sensing satellites. Fire point recognition, environment parameter extraction, lightweight dynamic spread simulation, multi-frame fire line space-time sequence compression coding are completed on board, and short messages of the Beidou navigation satellite system are used to instantaneously send the compressed codes to a ground terminal, so that an emergency closed loop of ''on-board shooting, on-board calculation, on-board prediction of future situation and on-board sending'' is realized. Therefore, the problems that in the prior art, forest fire spread prediction relies on ground station processing, time efficiency is poor, and in remote areas, communication links are lacked and early warning information cannot be received are solved.

Description

Technical Field

[0001] This invention relates to a method and system for predicting and providing emergency warnings of fire spread based on on-board computing, belonging to the field of remote sensing satellite technology. Background Technology

[0002] Forest fires are characterized by their suddenness, destructiveness, and rapid spread. Timely acquisition of fire situation data and early warning are crucial for fire fighting. Currently, forest fire monitoring and early warning mainly rely on two methods: remote sensing satellites and ground monitoring stations.

[0003] The existing remote sensing satellite forest fire monitoring process typically involves: a satellite carrying optical or infrared sensors takes images as it passes overhead, and the acquired image data is first stored in the satellite's onboard memory; when the satellite enters the receiving arc of a ground station, the data is transmitted to the ground receiving station via a data transmission channel; the ground data processing center performs radiometric calibration, geometric correction, and fire point identification on the images, and then runs a forest fire spread model for simulation and prediction; finally, the early warning information is sent to emergency departments and frontline firefighters via a ground communication network. This process has the following inherent defects: 1. Poor timeliness: Satellite data downlink relies on ground station receiving arcs. Low-orbit remote sensing satellites usually only have a few passing opportunities per day. From imaging to users obtaining early warning information, it often takes several hours or even longer. Meanwhile, forest fires can spread several kilometers in tens of minutes under high temperature and strong wind conditions. Traditional processes are difficult to meet emergency response needs.

[0004] 2. High dependence on communication: Early warning information ultimately needs to be delivered to frontline personnel through terrestrial communication networks (4G / 5G, satellite phones, etc.). However, forest fires often occur in remote mountainous areas where terrestrial network coverage is weak or even completely absent, meaning that even if the ground center generates early warning information, it cannot be transmitted to the on-site firefighting forces in a timely manner.

[0005] 3. Significant processing delays: Ground data processing relies on manual intervention or complex calculations. Even if the data is downloaded in a timely manner, subsequent processing steps may still cause additional delays.

[0006] In existing technologies, such as patent CN121214256B, a method for on-board disaster identification and lightweight downlink transmission of BeiDou short messages is proposed. However, this method only simplifies the transmission of static disaster-affected area boundaries and lacks the ability to predict the dynamic spread of future fires from the satellite. Existing ground-based forest fire spread prediction methods often involve massive computational loads, fail to consider the constraints of onboard computing resources, and do not address how to directly report multi-frame dynamic prediction results to frontline terminals in real time via narrowband satellite channels.

[0007] Therefore, there is an urgent need for a method and system that can simulate and predict forest fire spread on satellite and report the prediction results to frontline emergency personnel in real time via narrowband channels, so as to overcome the dual bottlenecks of timeliness and communication dependence of existing technologies. Summary of the Invention

[0008] This invention addresses the aforementioned technical problems by proposing a fire spread prediction and emergency early warning method and system based on on-board computing. The system completes fire point identification, environmental parameter extraction, lightweight dynamic spread simulation, and multi-frame fire line spatiotemporal sequence compression encoding on-board, and sends the data to the ground terminal in real time via BeiDou short message service, realizing an emergency closed loop of "on-board imaging, on-board computing, on-board prediction of future situation, and on-board transmission".

[0009] The technical solution adopted in this invention is as follows: A method for predicting fire spread and providing emergency warning based on satellite computing includes the following steps: S1. On-board fire point identification and physical parameter extraction: The satellite's mid-wave infrared remote sensing payload acquires raw images, which are then intelligently identified by the onboard AI computer to extract the precise location and radiation intensity of the fires. Pre-stored global land classification data and high-precision DEM data are retrieved, along with coarse-grained meteorological data acquired by onboard meteorological sensors, to retrieve wind speed and direction data. Based on the forest fire location information identified onboard, the satellite queries the corresponding forest area and extracts land classification data, DEM data, and retrieved wind speed and direction data within that area. S2. A cellular automata framework is used to divide the grid for calculating the fire propagation model and predict the fire position in the future. S3. After simulation at multiple time steps, the fire state distribution at the current moment is obtained, and the fire line boundary is extracted using morphological methods; S4. The Douglas-Puk algorithm is used to simplify the fireline polygons in each frame, key features are extracted from the simplified polygons, and multi-frame data is further compressed through inter-frame differential coding. S5. Encapsulate all compressed information into binary data packets according to the BeiDou short message protocol and send them directly via BeiDou short message. S6. After receiving the metadata, the ground terminal reconstructs the fire evolution sequence.

[0010] The method also includes a data retransmission step. After the satellite enters the data transmission arc of the ground station or relay satellite, the complete high-precision simulation result data generated in step S3 is transmitted to the ground command center through the traditional data transmission channel for post-disaster review and precise command.

[0011] In the above method, step S2, which uses a cellular automata framework to divide the forest area to which the fire point belongs, is a regular grid, with each grid cell... It has the following attributes: state , where 0 indicates unburned, 1 indicates burning, and 2 indicates completely burned; Combustion progress This indicates the percentage of combustion completed and is only valid for combustion units. Flammability factor The value is obtained by looking up a table based on vegetation type and humidity, and it represents the flammability of the unit. Ignite cumulative progress It records the ignition contribution from neighboring combustion units.

[0012] The calculation process of the fire propagation model in step S2 is as follows: The fire spread in discrete time steps Progress, performing the following operations at each time step: (1) Combustion unit upgrade, For each combustion unit Its combustion rate increases: , Combustion duration Determined by the characteristics of the unit itself, ,

[0013] In the formula The theoretical propagation time of the unit, For an empirical constant, when If the cell is burnt out, its state changes to "burnt out"; otherwise, it is updated. B(i,j) ; (2) Calculation of ignition contribution, For each unburned unit Consider all combustion units within its 8-neighborhood. The neighborhood direction includes orthogonal and diagonal directions, from the combustion unit To the unburned unit The fire propagation time is obtained by simplifying the modified Rothermel formula: , In the formula, The Euclidean distance between the cell centers is [value], and the orthogonal direction is [value]. diagonal direction is ; The baseline propagation rate was determined using experimental data from a typical grassland. The flammability factor of the target unit reflects the inhibitory effect of vegetation and humidity on fire. The coordinates of the unburned units; The wind factor describes the effect of wind on the acceleration of the fire, and its calculation formula is as follows: , In the formula The directional angle from the combustion unit to the unburned unit. As a direction sensitivity index, Wind speed ( ), This is based on wind direction. This formula ensures that propagation is fastest with the wind and slowest against the wind. Each unburned cell accumulates the ignition contribution from all neighboring burning cells: , In the formula This is the change in the cumulative progress of ignition of unburned units. Represents the neighborhood of a cell; updates P ( u , v )=P'( u , v )+∆ P ( u , v ),in P ( u , v This indicates the cumulative ignition progress of the currently unburned units. P' ( u , v () indicates the cumulative ignition progress of the unburned unit before the change.

[0014] (3) Ignition determination, like Then the unit has a probability It is ignited, meaning its state changes to combustion. and reset , ).

[0015] Step S2 divides the mesh into multiple sub-blocks, performs parallel computation, and updates them simultaneously.

[0016] Step S3, which uses morphological methods to extract the fire line boundary, is as follows: a. Generate fire zone mask ,like This indicates burning or complete combustion; otherwise, it is 0. b. Identify boundary cells through 8-neighborhood convolution: If there are non-fire zone cells in the neighborhood of a fire zone cell, then the cell is a boundary point; c. Use the Marching Squares algorithm to extract continuous polygon contours from the boundary point set to obtain the fireline vector polygon. .

[0017] The process of simplifying the fireline polygon in each frame using the Douglas-Puk algorithm in step S4 is as follows: (1) The future was generated by on-board simulation. The evolution of the fire within minutes, in time steps Total output Framefire polygon: , in For the current front line, to For the front lines at various future moments. Each frame contains polygons. Composed of a sequence of vertices: , In the formula Indicates the first Frame Polygon The vertex coordinates, of which From 0 to Sequential numbering; Indicates the index of the time frame, with a value range of 100. ; Indicates the first Frame Polygon The total number of vertices contained therein; (2) Preprocess the polygons in each frame; (3) The Douglas-Peucker algorithm is used to compress the vertices of each frame of polygons, which greatly reduces the number of vertices while preserving shape features: Given a polygon boundary consisting of a sequence of points, let the first point be and the last point be . Calculate the distance from all intermediate points to the line. Find the vertical distance and the maximum distance. If the corresponding point is less than the preset tolerance Then delete all intermediate points and keep only the middle points. and Otherwise Divide the curve into two segments as a boundary, and process them recursively, point by point. to the straight line The distance is: ,

[0018] In the formula, tolerance Based on the ground distance setting, a balance is struck between compression ratio and accuracy. Simplified polygon vertex sequence: , in, Represents the simplified version of the first Frame Polygon The total number of vertices contained. .

[0019] The key features described in step S4 include: (1) Firehead position: Along the current wind direction, the point at the foremost edge of the fire line. Calculate the projection of each vertex of the polygon onto the wind direction, and take the point with the largest projection as the firehead. , (2) Fire center: The centroid of the polygon or the center of the smallest circumscribed rectangle is used as the reference point. , (3) Spread speed and direction: The instantaneous spread speed is calculated based on the displacement of the fire head in adjacent frames. and direction of spread : , In the formula For the first The fire head coordinates of the frame, For the first The coordinates of the firehead in the frame; For the first The fire of the frame coordinate, For the first The fire of the frame coordinate; For the first The fire of the frame coordinate, For the first The fire of the frame coordinate; (4) Fire area: Calculate the area of ​​the polygon. It is used to assess the size of the fire.

[0020] The inter-frame differential coding mentioned above is based on the reference frame. For reference, subsequent frames By using affine transformation parameters, the data is divided into base frame data and differential frame data for compression.

[0021] The fire evolution sequence reconstruction process in step S6 is as follows: (1) Parse the reference frame: Reconstruct it based on the center coordinates and vertex offsets The latitude and longitude of the vertex; (2) Frame-by-frame recovery: Read the difference data to obtain the translation. Scaling Rotation ; reference frame Apply an affine transformation to each vertex: .

[0022] The on-board computing-based fire spread prediction and emergency early warning system includes an on-board remote sensing image acquisition module, an on-board fire point identification and location determination module, an on-board dynamic environment data inversion module, an on-board lightweight fire spread prediction module, an on-board fire line boundary extraction and simplification compression module, a data encapsulation and BeiDou short message transmission module, and a data ground receiving and reconstruction module. The onboard remote sensing image acquisition module is used to acquire raw mid-wave infrared remote sensing images; The onboard fire point identification and location determination module is used to intelligently identify fire points using an onboard AI computer and extract the precise location and radiation intensity of the fire points. The on-board dynamic environment data inversion module is used to invert wind speed and wind direction data from coarse-grained meteorological data acquired by meteorological sensors. The lightweight fire spread prediction module on the satellite utilizes static geographic data stored on the satellite and inverted dynamic environmental data. It employs a cellular automata framework to divide the grid and calculate the fire line propagation model, predicting the location of the fire line in the future. The on-board fire line boundary extraction and simplification compression module uses morphological methods to extract the fire line boundary from the fire field state distribution obtained after simulations at multiple time steps; it also simplifies the fire line polygons in each frame, extracts key features from the simplified polygons, and further compresses multi-frame data through inter-frame differential coding. The data encapsulation and BeiDou short message transmission module is used to encapsulate all compressed information into binary data packets according to the BeiDou short message protocol for direct BeiDou short message reporting. The data ground receiving and reconstruction module is used to reconstruct the fire scene evolution sequence after receiving the metadata packet.

[0023] The beneficial effects of this invention are: This invention overcomes the problems of poor timeliness caused by relying on ground station processing for forest fire spread prediction in existing technologies, and the inability to receive early warning information in remote areas due to lack of communication links. It provides an emergency method that can complete fire spread prediction on satellite and report it directly to the terminal via BeiDou short message.

[0024] The main advantages are: A lightweight onboard forest fire spread dynamic model is employed. This model is based on an improved cellular automata and compressed using a quantized neural network to adapt it to the low-power computing environment of onboard FPGAs or NPUs. The model outputs predicted fire line location vector data and fire spread rate for the next N hours (e.g., 1-6 hours).

[0025] The multi-frame fireline spatiotemporal joint compression coding method for narrowband channels innovatively employs the Douglas-Puk algorithm to geometrically thin the fireline vector in each frame, taking into account the extremely small single packet capacity (typically <229 bytes) of the BeiDou short message channel. Furthermore, it utilizes the affine transformation relationship between adjacent frames to perform inter-frame differential coding, compressing multiple frames of future firefield evolution data to within a few hundred bytes. Under the premise of ensuring that the loss of fireline geometric features is less than 30 meters, it achieves complete narrowband transmission of the future firefield spatiotemporal evolution sequence.

[0026] The generated multi-frame compressed metadata is encapsulated according to the BeiDou short message communication protocol and sent to the ground emergency terminal in the coverage area in real time during the non-data transmission arc segment via the BeiDou short message transmitter carried by the satellite. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is the original fireline polygon distribution map extracted in Embodiment 1 of the present invention; Figure 3 This is a comparison image of the simplified fireline polygon extracted in Embodiment 1 of the present invention. Detailed Implementation

[0028] The present invention will be further described below with reference to specific embodiments.

[0029] Example 1: A method for fire spread prediction and emergency early warning based on on-board computing, including steps (such as...) Figure 1 )as follows: S1. On-board fire point identification and physical parameter extraction: The satellite's mid-wave infrared remote sensing payload acquires the original images. After intelligent fire point identification by the onboard AI computer (using existing technology), the precise location and radiation intensity of the fire point are extracted. The satellite calls the global land classification data and high-precision DEM data stored in advance, and combines them with the coarse-grained meteorological data obtained by the onboard meteorological sensors to invert the wind speed and wind direction data (using existing technology such as reference [1] Zhong Junjie. Research on the authenticity verification method of sea surface wind field of tropical cyclone remote sensing by satellite microwave scatterometer [D]. Nanjing University of Information Science and Technology, 2024. DOI:10.27248 / d.cnki.gnjqc.2024.000659.). Based on the forest fire point location information identified on the satellite, the forest area to which it belongs is queried. Based on this area, the land classification data, DEM data and wind speed and wind direction data obtained by inversion are extracted, which are the input data required for forest fire spread simulation.

[0030] S2. A cellular automata framework is used to divide the mesh for calculating the fire propagation model and predict the fire line location over a future period: Based on real-time fire location data, environmental parameters, and pre-stored geographic information acquired from the satellite, a lightweight physical model of forest fire spread is run to predict the location of the fire line in the near future. Addressing the limitations of onboard computing resources and the high real-time requirements, we have made a lightweight modification of the traditional Rothermel physical model for onboard edge computing. Using a cellular automata framework combined with lookup tables and parallel processing, we have achieved fast and efficient fire spread simulation.

[0031] (a) Input data preparation Static geographic data: Raster data of the study area, including digital elevation model (DEM), vegetation type map, and combustible moisture map, are read from the onboard solid-state storage. This data has been pre-cropped and resampled to match the resolution of the onboard remote sensing imagery.

[0032] Dynamic environmental parameters: Wind speed and wind direction data are obtained by inverting coarse-grained meteorological data from onboard meteorological sensors.

[0033] Initial fire point position: The set of coordinates of the current combustion unit provided by the on-board real-time fire point identification algorithm.

[0034] (ii) Grid State Definition The forest area where the fire occurred was divided into... A regular grid, each grid cell It has the following attributes: a. state , where 0 indicates unburned, 1 indicates burning, and 2 indicates completely burned; b. Combustion progress This indicates the percentage of combustion completed and is only valid for combustion units. c. Flammability factor The value is obtained by looking up a table based on vegetation type and humidity, and it represents the flammability of the unit. d. Ignite the cumulative progress It records the ignition contribution from neighboring combustion units.

[0035] (III) Fire propagation model The fire spread in discrete time steps Advance, performing the following operations at each time step: (1) Combustion unit update For each combustion unit Its combustion rate increases: , Combustion duration Determined by the characteristics of the unit itself: , In the formula The theoretical propagation time of the unit, This is an empirical constant (ranging from 5 to 40 depending on vegetation type, representing the ratio of the duration of combustion to the duration of spread). When At that time, the unit's state changes to "burned out". Otherwise update .

[0036] (2) Calculation of ignition contribution For each unburned unit Consider all combustion units within its 8-neighborhood. (Neighborhood directions include orthogonal and diagonal). From the combustion unit To the unburned unit The fire propagation time is obtained by simplifying the modified Rothermel formula: , In the formula, The Euclidean distance between the center of the element (orthogonal direction is) diagonal direction is ); The baseline propagation rate was determined using experimental data from a typical grassland. The flammability factor of the target unit reflects the inhibitory effect of vegetation and humidity on fire. Let be the coordinates of the combustion unit; The wind factor describes the effect of wind on the acceleration of the fire, and its calculation formula is as follows: , In the formula The directional angle from the combustion unit to the unburned unit. The direction sensitivity index (usually taken as 1.5~2.0, fitted from experimental data) is used. Wind speed ( ), This represents the wind direction. This formula ensures that propagation is fastest with the wind and slowest against the wind.

[0037] Each unburned cell accumulates the ignition contribution from all neighboring burning cells: , In the formula Indicates the cumulative ignition progress of unburned units. This is the change in the cumulative progress of ignition of unburned units. Represents the neighborhood of a cell; updates P ( u , v )= P' ( u , v )+∆ P ( u ,v ),in P ( u , v This indicates the cumulative ignition progress of the currently unburned units. P' ( u , v () indicates the cumulative ignition progress of the unburned unit before the change.

[0038] (3) Ignition determination like Then the unit has a probability It is ignited (i.e., its state changes to combustion). and reset , ).

[0039] To adapt to the spaceborne computing environment, the present invention has made the following optimizations to the above model: a. Lookup table replaces real-time calculation: Flammability factor... Wind factors The combination of wind speed and direction angle is pre-calculated into a two-dimensional lookup table, using wind speed and direction angle as indexes, to avoid floating-point operations; b. Integer arithmetic: Convert the propagation time into an integer representation of the "inverse rate" format, and use fixed-point numbers for cumulative updates; c. Parallel processing: Utilizing the parallel capabilities of onboard multi-core CPUs or FPGAs, the mesh is divided into multiple sub-blocks for simultaneous updates, and boundary dependencies are handled through message passing; d. Event-driven update: Only the burning cell and its neighborhood are updated, avoiding traversing the entire mesh and significantly reducing the amount of computation.

[0040] S3. After simulation at multiple time steps, the fire state distribution at the current moment is obtained, and the fire line boundary is extracted using morphological methods: go through After simulation at several time steps, the fire state distribution at the current moment is obtained. Morphological methods are used to extract the fire line boundary. a. Generate fire zone mask ,like This indicates burning or complete combustion; otherwise, it is 0. b. Identify boundary cells through 8-neighborhood convolution: If there are non-fire zone cells in the neighborhood of a fire zone cell, then the cell is a boundary point; c. Use the Marching Squares algorithm to extract continuous polygon contours from the boundary point set to obtain the fireline vector polygon. .

[0041] S4. The Douglas-Puk algorithm is used to simplify the fireline polygons in each frame. Key features are extracted from the simplified polygons, and multi-frame data is further compressed through inter-frame differential coding. In response to the limitations of spaceborne computing resources and narrow channel bandwidth, this invention uses the Douglas-Puk algorithm to simplify the fireline polygon in each frame and further compresses multi-frame data through inter-frame differential coding to ensure that the future fire evolution situation is completely transmitted within a few hundred bytes.

[0042] Input data, future fireline polygon sequence: Assume that the on-board simulation generated the future... The evolution of the fire within minutes, in time steps Total output Framefire polygon: , in For the current front line, to For the front lines at various future moments. Each frame contains polygons. Composed of a sequence of vertices: , In the formula Indicates the first Frame Polygon The vertex coordinates, of which From 0 to Sequential numbering; Indicates the index of the time frame, with a value range of 100. ; Indicates the first Frame Polygon The total number of vertices included may vary in different frames due to the complexity of the fire scene shape, which can change over time.

[0043] Meta-information: Current simulation timestamp Wind speed ,wind direction Total fire area (Each frame) etc.

[0044] Polygon preprocessing: The following preprocessing steps are performed on each frame of polygons to ensure data quality: a. Closure check: If the first and last vertices do not coincide, the coordinates of the endpoint are automatically added to form a closed loop.

[0045] b. Uniform orientation: Ensure that all polygon vertices are arranged in a counterclockwise direction (adjusted by calculating the signed area) to facilitate subsequent deformation processing.

[0046] c. Noise filtering: Remove isolated small polygons with an area smaller than the threshold, as these are considered false alarm areas.

[0047] (3) Simplification of Douglas-Puk algorithm The Douglas-Peucker algorithm is used to compress the vertices of each frame of polygons, significantly reducing the number of vertices while preserving shape features. This algorithm recursively removes redundant points, retaining key points that contribute significantly to the shape.

[0048] Given a polygon boundary consisting of a sequence of points, let the first point be... The last point is Calculate the distance from all intermediate points to the line. Find the vertical distance and the maximum distance. and its corresponding points .like Less than the preset tolerance Then delete all intermediate points and keep only the middle points. and Otherwise Divide the curve into two segments, using the boundary as the boundary, and process each segment recursively. to the straight line The distance is: , In the formula, tolerance The compression ratio and accuracy are balanced based on the ground distance setting.

[0049] Simplified polygon vertex sequence: , in, Represents the simplified version of the first Frame Polygon The total number of vertices contained. .

[0050] (4) Key feature extraction: The following key features are extracted from the simplified polygon for differential coding and emergency decision-making: Firehead position: The point at the very front of the fire line along the current wind direction. Calculate the projection of each vertex of the polygon onto the wind direction, and take the point with the largest projection as the firehead. .wind direction The quantity is known.

[0051] a. Fire center: The centroid of a polygon or the center of its smallest circumscribed rectangle is used as a reference point. .

[0052] b. Spread speed and direction: The instantaneous spread speed is calculated based on the fire head displacement in adjacent frames. and direction of spread : , c. Fire area: Calculate the area of ​​the polygon. (Unit: km²), used to assess the size of a fire.

[0053] (5) Multi-frame differential coding (based on affine transformation) To efficiently transmit future multi-frame Fireline, an inter-frame differential coding strategy is adopted: using the reference frame... For reference, subsequent frames By using affine transformation parameters, the amount of data is significantly compressed.

[0054] a. Reference Frame Encoding: Reference Frame The coordinates of the polygon vertices are encoded relative to the center of the fire. Using the origin as the coordinate system, convert the vertex coordinates to offsets relative to the center, and represent them as integers. For example: , In the formula, This is the quantization step size. Each offset is stored as a 2-byte signed integer (range -32768 to 32767, corresponding to ±327.68 km, meeting the fire field range requirements). Number of vertices. Stored using 1 byte.

[0055] b. Affine Transformation Parameter Calculation: For the first... Frame, calculate the optimal affine transformation from the reference frame to this frame, such that the transformed reference polygon As close as possible The affine transformation model is as follows: , In the formula, This is a scaling factor (reflecting the expansion / contraction of the fire). This is a rotation matrix (reflecting the rotation of the fire field); It is a translation vector (reflecting the overall movement of the fire).

[0056] c. Packaging of difference data: For each frame Package the following data: Table 1 Differential Data Block Structure .

[0057] S5. Encapsulate all compressed information into binary data packets according to the BeiDou short message protocol and send them directly via BeiDou short message: All compressed information is encapsulated into binary data packets according to the BeiDou short message protocol. The data packet structure is designed as follows: Table 2 Fire Data Structure , Table 3. Baseline Frame Data Block Structure .

[0058] S6. After receiving the metadata data, the ground terminal reconstructs the fire evolution sequence: a. Parse the reference frame: Reconstruct it based on the center coordinates and vertex offsets. The latitude and longitude of the vertex.

[0059] b. Frame-by-frame recovery: Read the difference data to obtain the translation. Scaling Rotation ; reference frame Apply an affine transformation to each vertex: .

[0060] S7. Data Retransmission: After the satellite enters the data transmission arc of the ground station or relay satellite, the complete high-precision simulation result data (including the original fire line grid map and high-resolution image) generated in step 3 is transmitted to the ground command center through the traditional data transmission channel for post-disaster review and precise command.

[0061] This embodiment demonstrates the entire process from satellite simulation to direct reporting from BeiDou, based on real forest fire simulation data: The on-board simulation employs an improved cellular automata model with a time step of [missing information]. The simulation measures the fire's development over a total of 3 hours (10800 seconds), outputting one frame of fireline polygons every 3600 seconds. Since the onboard processing results need to be compared with ground verification data, the three frames of polygons output from the ground model (corresponding to the simulation time) are used directly here. s, s, s) is the raw output of the on-board simulation. The number of polygon vertices in the three frames are 22, 48, and 106, respectively, and the fire areas are 0.663 km², 2.839 km², and 6.177 km², respectively. Figure 2 The distribution of the original fireline polygons is shown.

[0062] Douglas-Puk algorithm simplifies and sets tolerance. Meters (corresponding to ground distance), the Douglas-Puk algorithm is applied to simplify the polygons in each frame. Table 4 shows a comparison of the number of vertices before and after simplification, and the polygon distribution is as follows: Figure 3 .

[0063] Table 4 Comparison of Vertex Count Before and After Simplification .

[0064] Key feature extraction: Key parameters such as the fire head position, fire center, and spread rate were calculated for each frame, and the results are shown in Table 5. The fire head is defined as the farthest point along the current wind direction, which is measured in real time by the onboard meteorological sensor (in this example, wind speed and direction recorded by the ground model are used).

[0065] Table 5 Key Feature Information .

[0066] Multi-frame differential coding (based on affine transformation). Starting with the first frame ( Using the base frame as the reference frame, calculate the second frame ( ) and the third frame ( The affine transformation parameters relative to the reference frame are shown in Table 6.

[0067] Table 6 Affine Transformation Parameter Information .

[0068] Fire situation metadata is packaged. All information is encapsulated into binary data packets according to the protocol. The length of each byte in the data packet is shown in Table 7. As can be seen from Table 7, the byte length of the fire situation data packet is much smaller than the typical capacity of a single BeiDou short message card (229 bytes), which meets the transmission requirements.

[0069] Table 7 Data packet byte length .

[0070] Example 2: A fire spread prediction and emergency early warning system based on on-board computing, including an on-board remote sensing image acquisition module, an on-board fire point identification and location determination module, an on-board dynamic environment data inversion module, an on-board lightweight fire spread prediction module, an on-board fire line boundary extraction and simplified compression module, a data encapsulation and BeiDou short message transmission module, and a data ground receiving and reconstruction module. The onboard remote sensing image acquisition module is used to acquire raw mid-wave infrared remote sensing images; The onboard fire point identification and location determination module is used to intelligently identify fire points using an onboard AI computer and extract the precise location and radiation intensity of the fire points. The on-board dynamic environment data inversion module is used to invert wind speed and wind direction data from coarse-grained meteorological data acquired by meteorological sensors. The lightweight fire spread prediction module on the satellite utilizes static geographic data stored on the satellite and inverted dynamic environmental data. It employs a cellular automata framework to divide the grid and calculate the fire line propagation model, predicting the location of the fire line in the future. The on-board fire line boundary extraction and simplification compression module uses morphological methods to extract the fire line boundary from the fire field state distribution obtained after simulations at multiple time steps; it also simplifies the fire line polygons in each frame, extracts key features from the simplified polygons, and further compresses multi-frame data through inter-frame differential coding. The data encapsulation and BeiDou short message transmission module is used to encapsulate all compressed information into binary data packets according to the BeiDou short message protocol for direct BeiDou short message reporting. The data ground receiving and reconstruction module is used to reconstruct the fire scene evolution sequence after receiving the metadata packet.

[0071] The above is a further description of the present invention in conjunction with specific embodiments, and the scope of protection of the present invention is not limited thereto.

Claims

1. A method for fire spread prediction and emergency early warning based on on-board computing, characterized in that, The steps include the following: S1. On-board fire point identification and physical parameter extraction: The satellite's mid-wave infrared remote sensing payload acquires raw images, which are then intelligently identified by the onboard AI computer to extract the precise location and radiation intensity of the fires. Pre-stored global land classification data and high-precision DEM data are retrieved, along with coarse-grained meteorological data acquired by onboard meteorological sensors, to retrieve wind speed and direction data. Based on the forest fire location information identified onboard, the satellite queries the corresponding forest area and extracts land classification data, DEM data, and retrieved wind speed and direction data within that area. S2. A cellular automata framework is used to divide the grid for calculating the fire propagation model and predict the fire position in the future. The calculation process for the fire propagation model is as follows: The fire spread in discrete time steps Progress, performing the following operations at each time step: (1) Combustion unit upgrade, For each combustion unit Its combustion rate increases: ， Combustion duration Determined by the characteristics of the unit itself, ， In the formula The theoretical propagation time of the unit, For an empirical constant, when If the cell is burnt out, its state changes to "burnt out"; otherwise, it is updated. ; (2) Calculation of ignition contribution, For each unburned unit Consider all combustion units within its 8-neighborhood. The neighborhood direction includes orthogonal and diagonal directions, from the combustion unit To the unburned unit The fire propagation time is obtained by simplifying the modified Rothermel formula: ， In the formula, The Euclidean distance between the cell centers is [value], and the orthogonal direction is [value]. diagonal direction is ; The baseline propagation rate was determined using experimental data from a typical grassland. The flammability factor of the target unit reflects the inhibitory effect of vegetation and humidity on fire. The coordinates of the unburned units; The wind factor describes the effect of wind on the acceleration of the fire, and its calculation formula is as follows: ， In the formula The directional angle from the combustion unit to the unburned unit. As a direction sensitivity index, For wind speed, , Wind direction; Each unburned cell accumulates the ignition contribution from all neighboring burning cells: ， In the formula This is the change in the cumulative progress of ignition of unburned units. Represents the neighborhood of a cell; update P(u,v)=P'(u,v)+∆P(u,v), where P(u,v) represents the cumulative ignition progress of the current unburned cell, and P'(u,v) represents the cumulative ignition progress of the unburned cell before the change; (3) Ignition determination, like Then the unit has a probability It is ignited, meaning its state changes to combustion. and reset , ; S3. After simulation at multiple time steps, the fire state distribution at the current moment is obtained, and the fire line boundary is extracted using morphological methods; S4. The Douglas-Puk algorithm is used to simplify the fire line polygon in each frame, key features are extracted from the simplified polygon, and multi-frame data is further compressed through inter-frame differential coding; the key features include: fire head position, fire center, spread speed and direction, and fire area. S5. Encapsulate all compressed information into binary data packets according to the BeiDou short message protocol and send them directly via BeiDou short message. S6. After receiving the metadata, the ground terminal reconstructs the fire evolution sequence.

2. The fire spread prediction and emergency early warning method based on on-board computing according to claim 1, characterized in that, It also includes a data retransmission step. After the satellite enters the data transmission arc of the ground station or relay satellite, the complete high-precision simulation result data generated in step S3 is transmitted to the ground command center through the traditional data transmission channel for post-disaster review and precise command.

3. The fire spread prediction and emergency early warning method based on on-board computing according to claim 1, characterized in that, Step S2, which uses a cellular automata framework to divide the forest area to which the fire point belongs, is called the grid division. A regular grid, each grid cell It has the following attributes: state , where 0 indicates unburned, 1 indicates burning, and 2 indicates completely burned; Combustion progress This indicates the percentage of combustion completed and is only valid for combustion units. Flammability factor The value is obtained by looking up a table based on vegetation type and humidity, and it represents the flammability of the unit. Ignite cumulative progress It records the ignition contribution from neighboring combustion units.

4. The fire spread prediction and emergency early warning method based on on-board computing according to claim 1, characterized in that, Step S2 divides the mesh into multiple sub-blocks, performs parallel computation, and updates them simultaneously.

5. The fire spread prediction and emergency early warning method based on on-board computing according to claim 1, characterized in that, Step S3, which uses morphological methods to extract the fire line boundary, is as follows: a. Generate fire zone mask ,like This indicates burning or complete combustion; otherwise, it is 0. b. Identify boundary cells through 8-neighborhood convolution: If there are non-fire zone cells in the neighborhood of a fire zone cell, then the cell is a boundary point; c. Use the Marching Squares algorithm to extract the continuous polygonal contours from the boundary point set to obtain the fireline vector polygon P.

6. The fire spread prediction and emergency early warning method based on on-board computing according to claim 1, characterized in that, The process of simplifying the fireline polygon in each frame using the Douglas-Puk algorithm in step S4 is as follows: (1) The on-board simulation generated the fire evolution within the next T minutes, according to the time step. Total output Framefire polygon: ， in For the current front line, to For the front lines at various future moments, each frame contains polygons. Composed of a sequence of vertices: ， In the formula Indicates the first Frame Polygon The vertex coordinates, of which From 0 to Sequential numbering; Indicates the index of the time frame, with a value range of 100. ; Indicates the first Frame Polygon The total number of vertices contained therein; (2) Preprocess the polygons in each frame; (3) The Douglas-Puk algorithm is used to compress the vertices of each frame of polygons, which greatly reduces the number of vertices while preserving shape features: Given a polygon boundary consisting of a sequence of points, let the first point be... The last point is Calculate the distance from all intermediate points to the line. Find the vertical distance and the maximum distance. and its corresponding point C, if Less than the preset tolerance Then delete all intermediate points and keep only the middle points. and ; Otherwise, divide the curve into two segments with point C as the boundary, and process them recursively, from point C to the line. The distance is: ， In the formula, tolerance Based on the ground distance setting, a balance is struck between compression ratio and accuracy. Simplified polygon vertex sequence: ， in, Represents the simplified version of the first Frame Polygon The total number of vertices contained. .

7. The fire spread prediction and emergency early warning method based on on-board computing according to claim 1, characterized in that, The fire evolution sequence reconstruction process in step S6 is as follows: (1) Parse the reference frame: Reconstruct the frame based on the center coordinates and vertex offsets. The latitude and longitude of the vertex; (2) Frame-by-frame recovery: Read the difference data to obtain the translation. Scaling Rotation ; reference frame Apply an affine transformation to each vertex: 。 8. A system for implementing the fire spread prediction and emergency early warning method based on on-board computing as described in claim 1, characterized in that, It includes an onboard remote sensing image acquisition module, an onboard fire point identification and location determination module, an onboard dynamic environment data inversion module, an onboard lightweight fire spread prediction module, an onboard fire line boundary extraction and simplification compression module, a data encapsulation and BeiDou short message transmission module, and a data ground receiving and reconstruction module. The onboard remote sensing image acquisition module is used to acquire raw mid-wave infrared remote sensing images; The onboard fire point identification and location determination module is used to intelligently identify fire points using an onboard AI computer and extract the precise location and radiation intensity of the fire points. The on-board dynamic environment data inversion module is used to invert wind speed and wind direction data from coarse-grained meteorological data acquired by meteorological sensors. The lightweight fire spread prediction module on the satellite utilizes static geographic data stored on the satellite and inverted dynamic environmental data. It employs a cellular automata framework to divide the grid and calculate the fire line propagation model, predicting the location of the fire line in the future. The on-board fire line boundary extraction and simplification compression module uses morphological methods to extract the fire line boundary from the fire field state distribution obtained after simulations at multiple time steps; it also simplifies the fire line polygons in each frame, extracts key features from the simplified polygons, and further compresses multi-frame data through inter-frame differential coding. The data encapsulation and BeiDou short message transmission module is used to encapsulate all compressed information into binary data packets according to the BeiDou short message protocol for direct BeiDou short message reporting. The data ground receiving and reconstruction module is used to reconstruct the fire scene evolution sequence after receiving the metadata packet.

Images