Methods for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations

Abstract

This invention relates to the field of fire investigation technology and discloses a method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations. First, trace photos of the fire scene are collected and embedded with GPS coordinates, direction, and timestamp information to construct a trace dataset. Then, a spread direction recognition model is trained based on the trace dataset. Next, the trained spread direction recognition model identifies the true direction of flame spread in each trace photo to be examined. Finally, the direction vector of each trace photo is calculated based on the true direction. A continuous direction field is constructed, and each trajectory is integrated along the opposite direction of the direction field from each sampling point. The location of the origin of the fire is identified based on the endpoint of each trajectory. This invention significantly improves the efficiency of fire investigations, reduces the difficulty and risk of investigation in complex terrain environments, reduces subjective reliance, and improves the accuracy and reliability of tracing the origin of fires, providing a scientific basis and technical support for subsequent investigation of ignition sources and analysis of fire causes.

Description

Technical Field

[0001] This invention relates to the field of fire investigation technology, specifically to a method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations. Background Technology

[0002] In forest fire investigations, to determine the cause of a fire, the ignition point is often determined by analyzing the spread direction of typical flames at the forest site and considering various spread directions. Then, potential ignition sources are located at the ignition point to analyze and determine the possible causes of the fire. Currently, due to the large area, complex terrain, and danger of forest fire sites, relying solely on manual on-site investigation is often time-consuming and labor-intensive. Furthermore, the identification of typical forest traces currently relies mainly on experience. How to transform this experience into expert knowledge and develop a systematic method for rapid on-site investigation and identification has become an urgent problem to be solved in forest fire investigations. Summary of the Invention

[0003] To address the aforementioned problems, the present invention aims to provide a method for analyzing the direction of fire spread and tracing the ignition point in forest fire investigations. This method can effectively identify the direction of flame spread based on photographs of traces at the fire scene, and further obtain the location of the ignition point through inverse integration and clustering. It is accurate and efficient, providing strong support for investigating the ignition source and analyzing the cause of the fire. The technical solution is as follows:

[0004] A method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations includes the following steps:

[0005] Step 1: Construction of the image trace dataset: After the fire, collect photos of traces at the fire scene and embed GPS coordinates, direction and timestamp information to construct the image trace dataset;

[0006] Step 2: Spread direction identification: Train a spread direction identification model based on the trace dataset; then identify the true direction of flame spread in each trace photo to be detected based on the trained spread direction identification model;

[0007] Step 3: Origin point tracing: Based on the true direction and the coordinates of the sampling point corresponding to each trace photograph, calculate the direction vector of each trace photograph; construct a continuous direction field, starting from each sampling point, integrate each trajectory in the opposite direction of the direction field; use clustering to identify the dense region in space where the endpoint of each trajectory is located, and take the center of the largest cluster as the location of the ignition point.

[0008] The beneficial effects of this invention are:

[0009] This invention provides a method for analyzing the direction of fire spread and tracing the ignition point in forest fire investigations. It can automatically identify the direction of flame spread based on trace photos of the fire scene and accurately trace the ignition point through inverse integration and clustering algorithms.

[0010] Compared to traditional manual investigation methods, this invention significantly improves the efficiency of fire investigation and reduces the difficulty and risk of investigation in complex terrain environments. At the same time, by constructing a map trace dataset and training a spread direction recognition model, expert experience is transformed into a reusable intelligent method, reducing subjective dependence and improving the accuracy and reliability of tracing the source of the fire. This provides a scientific basis and technical support for subsequent investigation of ignition sources and analysis of the cause of the fire. Attached Figure Description

[0011] Figure 1 This is an example diagram for directional field simulation.

[0012] Figure 2 This is a simulation example diagram for tracing the ignition point. Detailed Implementation

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0014] This invention provides a method for analyzing the spread direction and tracing the origin of forest fires. It requires image acquisition equipment and a backend system for intelligent identification and analysis. The image acquisition equipment embeds GPS coordinates, direction, and timestamp information; the backend system embeds maps and CNN recognition algorithms to perform spread direction analysis and origin tracing. The specific implementation steps of this method are as follows:

[0015] Step 1: On-site modeling. Select a drone with surveying capabilities, determine the surveying area, set surveying parameters, and conduct surveying (mainly taking pictures and overlaying information such as altitude, latitude, longitude, and direction onto the images). After surveying, download the surveying data to the computer, use software to convert the surveying images into 3D format, import the 3D format images into the map in the backend system, and complete the on-site modeling in the backend system.

[0016] Step 2: On-site photography. Images of fire scene traces are captured using image acquisition devices, and GPS coordinates, direction, and timestamp information are embedded. For larger fire scenes, multiple image acquisition devices can simultaneously capture images of the fire scene traces. The captured images are wirelessly transmitted to the backend system in real time to build an image trace dataset.

[0017] After receiving the trace photos from the image acquisition device, the backend system creates a trace dataset:

[0018] Training set: Photos of traces left by combustible materials at forest fire sites are classified according to first-person perspective, categorizing the traces into 0-degree, 45-degree, 90-degree, 135-degree, and 180-degree directions. Specifically, 0-degree refers to traces moving from left to right, 45-degree from lower left to upper right, 90-degree from bottom to top, 135-degree from lower right to upper left, and 180-degree from right to left. To improve recognition accuracy, a negative sample class is added, containing vegetation photos from various seasons and scenes, as well as traces where the direction of fire spread is not clearly defined.

[0019] Test set: Select trace photos of various degrees, such as 0 degrees, 45 degrees, 90 degrees, 135 degrees and 180 degrees, as the test set. The photos should not be repeated with the training set photos and the number of photos should not be less than 20% of the number of photos in the test set.

[0020] Step 3: Identification of fire spread direction and tracing of ignition point: After receiving photos of fire scene traces from the image acquisition device, the backend system constructs a trace dataset, trains a fire spread direction identification model based on the trace dataset, completes the fire spread direction identification, and then traces the ignition point based on the identified fire spread direction.

[0021] Step 31: Identify the direction of spread

[0022] Step 311: Establish a propagation direction identification model

[0023] 1) Build a CNN network model, using the ResNet-18CNN model, and set the model parameters including:

[0024] Set the input image size (inputSize);

[0025] Set the batchSize of input samples for each training iteration;

[0026] Set the number of times to fully traverse the entire training dataset, numEpochs.

[0027] Set the learning rate lr;

[0028] Set the ImageNet mean and standard deviation to normalize the image.

[0029] 2) Training

[0030] The sample photos from the training set are input into the CNN network model for training, resulting in the corresponding propagation direction recognition model.

[0031] 3) Testing and Evaluation

[0032] The propagation direction recognition model is used to identify each trace image in the test set, and the recognition accuracy is calculated. If the accuracy reaches a%, the test is completed; if the accuracy requirement is not met, return to step 2) to readjust the parameters of the CNN network model. The parameter adjustment strategies mainly include: for small datasets, the batch size can be reduced or numEpochs can be increased; for overfitting, data augmentation (rotation, flipping, cropping, etc., to artificially expand the data) or numEpochs can be reduced.

[0033] Step 312: Use the established propagation direction recognition model to identify the trace photos uploaded by the image acquisition device to obtain the recognition direction. The specific recognition process is as follows:

[0034] 1) The trained propagation direction recognition model is used to identify the trace photos uploaded by the image acquisition device to obtain the direction angle b of flame propagation in the photos.

[0035] 2) Extract the directional information D from the trace photos uploaded by the image acquisition device, and calculate it with b to obtain the true azimuth angle α of the flame spread:

[0036] α=D+(90-b)(1)

[0037] Step 32: Tracing the source of the fire

[0038] The specific steps are as follows:

[0039] Step 321: Assume the image acquisition device transmits N trace images, and the latitude and longitude coordinates of the sampling point corresponding to each trace image are:

[0040] H i =(E i N i ), i = 0, 1, 2, ..., N-1 (2)

[0041] Using the first point (E0, N0) as the origin, the coordinates are converted to a plane coordinate system (local Cartesian coordinate system, unit: meters) using the following formula:

[0042]

[0043] Where R is the Earth's radius, with a value of R = 6,378,137 meters, thus obtaining the sampling point P corresponding to each trace photograph. i The relative coordinates of the observation point:

[0044] P i =(x i ·y i (4)

[0045] The identified true azimuth angle (the true azimuth angle of the flame spread at the i-th sampling point is α) is then used to determine the true azimuth angle. i Convert the coordinates to the direction vector at that point and calculate the direction vector d. i :

[0046] d i =(u i v i (5)

[0047] The unit direction vector corresponding to each point represents the direction of flame spread away from the fire source; if the true azimuth angle α of the flame spread is known... i (North is 0 degrees, rotating clockwise), then:

[0048]

[0049] Step 322: Construct a continuous direction field D(x,y) using the inverse distance weighted interpolation (IDW) method:

[0050]

[0051] Where p is the weight parameter (usually p = 2), ∈ = 0.00001 (a constant to avoid division by zero).

[0052] Let the direction of any point (x, y) be D(x, y), and let the coordinates of the existing i points be P. i The direction d of the point and the other party i Calculate the direction D(x,y) of any point coordinates (x,y), thus forming a set D(x,y) called the direction field.

[0053] Step 323: Integrate in reverse, from each sampling point P i Start by following the integral trajectory of -D(x,y).

[0054] Numerical integration is performed using the fourth-order Runge-Kutta method (RK4), and the steps are as follows:

[0055] Step a: Initialize parameters

[0056] Step size h (adjusted according to accuracy requirements);

[0057] Maximum number of steps T (to avoid infinite loops).

[0058] Termination condition: directional field modulus ||r n+1 -r n || < 0.01 (close to the fire source) or n ≥ T;

[0059] Step b: Single-step RK4 calculation

[0060] Input r0 = Pi For the current position r at step n (n≥1) n =(x n y n ):

[0061] 1) Calculate the slopes k1, k2, k3, and k4 using the formula:

[0062] k1=-D(r n (9)

[0063]

[0064]

[0065] k4=-D(r n +hk3)(12)

[0066] 2) Update position r n+1 ,

[0067]

[0068] Where h is the step size. k1, k2, k3, and k4 are the intermediate slopes (process values); r n =(x n y n ) is the current position coordinate of the step.

[0069] Termination check condition: If the termination condition is met, integration stops, and the position is P. i The endpoint coordinates obtained by streamline integration under the sampling point conditions.

[0070] Step c: Traverse all observation points and perform RK4 calculations for each. Since there are N sampling points, N streamline endpoints are ultimately formed, and the set of streamline endpoints is Q = {q1, q2, ..., q N}

[0071] Step 324: Calculate the location of the ignition point

[0072] The DBSCAN (Density-Based Spatial Clustering of Applications with Noise) density clustering method is used to identify dense regions of streamline endpoints in space, and the center of the largest cluster is used as the fire source estimate. This method can remove noise or outliers from the streamline endpoint set, avoiding increased source tracing errors. The specific steps are as follows:

[0073] Step 1) Parameter settings:

[0074] Set the neighborhood radius ε, which is the distance threshold used to determine whether two points belong to the same cluster; set the minimum number of points min. pts This is used to limit the minimum number of neighborhood points required to form a cluster.

[0075] Step 2) Calculate the core points:

[0076] For each point q in the set of streamline endpoints Q i Calculate the number of points in its neighborhood:

[0077] N ε (q i )={q j |||q i -q j ||≤ε} (14)

[0078] Where, q j For q i Points within the neighborhood. If |N ε (q i )|≥min pts Then the point q i For a given core point, points within its neighborhood that satisfy the above conditions are considered members of that core point; otherwise, the point q... i If the number of neighborhood points is insufficient, it may belong to the neighborhood of another core point, or it may be noise. Traverse Q to eventually form a set of core points S = {s1, s2, ..., s...} z (Z≤N).

[0079] Step 3) Cluster expansion:

[0080] From a certain core point s i Starting from this point, recursively merge all density-reachable points to form a cluster. The specific implementation method is as follows:

[0081] a) Create a new cluster C1, and randomly select an unvisited core point s. i , will s i Add to cluster C1.

[0082] b) Traversal calculations s i Each member q in the neighborhood c Does it satisfy |N ε (q c )|≥min pts If satisfied, then q c All members in the neighborhood are merged into cluster C1, and visited points are marked. If this condition is not met, the member is considered non-core, belonging to the boundary point or noise of the cluster.

[0083] c) Process the next unmarked core point s j This is then labeled as a new cluster C2. The above steps are repeated until cluster C2 is finally formed.

[0084] d) Proceed sequentially downwards until all core points in S have been visited. If a point is not included in any cluster, it is marked as noise and will not participate in the ignition point tracing calculation.

[0085] Through clustering expansion, a collective C = {C1, C2, ..., C} of all possible clusters can be formed. m (m≤Z).

[0086] Step 4): Select the largest cluster in C and calculate the mean value of all points in that cluster as the ignition point location S. DBSCAN :

[0087]

[0088] Among them, C max It is the point set of the largest cluster, |C max | represents C max The cardinality of a set is the number of points in the largest cluster.

[0089] For example, the set of points of the largest cluster obtained by DBSCAN clustering has the following form:

[0090] C = {P1, P2, ..., P | C |}, P i =(x i ,y i (16)

[0091] Step 5: Transfer each point P i The coordinates of the fire point are overlaid onto the corresponding modeled map using directional arrows. This direction represents the direction of the direction vector at that point, thus generating a visualized spread trend map. The location of the fire point S is then... DBSCAN The coordinates are overlaid onto the corresponding pre-modeled map and marked with special icons to generate a visual schematic diagram of the fire ignition point.

[0092] Since the final calculated coordinates are planar coordinates, they need to be converted to latitude and longitude coordinates before being overlaid on the map. The conversion method is as follows: Given planar coordinates (x, y), convert them back to latitude and longitude coordinates (Ec, Nc):

[0093]

[0094] (x,y): Planar coordinates (unit: meters), where x is the east direction and y is the north direction.

[0095] (E0,N0): Longitude and latitude of the reference point (unit: degrees); R is the Earth's radius, with the same value as above.

[0096] Example of specific identification and ignition point determination process:

[0097] 1) The image acquisition device uploads 10 photos (observation points), labeled P1, P2, ..., P... 10 Each sampling point corresponds to latitude and longitude coordinates, direction information D, direction angle b, true azimuth angle α, and actual plane coordinates (x, y). i y i ) and direction vector d i As shown in Table 1.

[0098] Table 1 Sampling Point Data Table

[0099]

[0100] 2) Orientation field construction

[0101] Suppose we need to calculate the orientation field at point (10,5): First, calculate the weight w at that point using the formula. i (x, y) are as follows:

[0102] Taking the sampling point P1 = (0,0) as an example:

[0103]

[0104] Similarly, calculate the other weights w2, w3, ..., w 10 .

[0105] The direction field of sampling point P1 is:

[0106]

[0107] The answer can be obtained by substituting the values ​​step by step.

[0108] Since the sampling points are discrete, while the direction field D(x,y) is continuous and represents the propagation direction of each point (x,y) on the planar region, it is impossible to traverse all points on the plane in actual calculations. The direction field of the region can be calculated by setting the step size and traversal range in the program. A simulation example is shown below. Figure 1 As shown.

[0109] 3) RK4 integration

[0110] From observation point P i Set off,

[0111] Step a: Initialize parameters

[0112] • Step size h = 0.5 meters (adjustable according to accuracy requirements)

[0113] • Maximum number of steps T = 100 (to avoid infinite loops)

[0114] Termination condition: Directional field modulus ||P1-(10,5)|| < 0.01 (near the ignition source)

[0115] Step b: Single-step RK4 calculation

[0116] Input r0 = P i For the current position r at step n (n≥1) n =(x n y n ):

[0117] Calculate the slopes k1, k2, k3, and k4 using the formula.

[0118] Step c: Update position r n+1

[0119] Terminate the check if ||D(x) is satisfied. n+1 y n+1 If || < 0.01 or n ≥ T, then the integration stops, and the position is P. i The endpoint coordinates obtained by streamline integration under the observation point conditions.

[0120] 4) Calculation of the location of the ignition point

[0121] Repeat the RK4 integration for all observation points to obtain the coordinates of the endpoints of multiple streamlines, Q = {q1, q2, ..., q 10 Then perform DBSCAN clustering.

[0122] 1) Parameter settings

[0123] • Neighborhood radius ε: Selected based on data distribution (e.g., ε = 3 meters).

[0124] ·Minimum points min pts : Usually set to 2-5 (e.g., min) pts =3).

[0125] 2) Calculate the neighborhood

[0126] Calculate the core points, for each point q i Calculate the number of points within its neighborhood:

[0127] N ε (q i )={q j |||q i- q j ||≤ε}(20)

[0128] If |N ε (q i )|≥min pts Then the point q i A point q is considered a core point; points within the domain that satisfy the above conditions are members of that core point. Otherwise, the point q is considered a core point. i The domain has insufficient points, which may belong to another core domain or be noise.

[0129] 3) Clustering expansion

[0130] From the core point q i Starting from the point, recursively merge density-reachable points:

[0131] Create a new cluster C1, and randomly select an unvisited core point q. i traverse and calculate q i Each member q c Does it satisfy |N ε (q c )|≥min pts If satisfied, then q c All members are merged into C1.

[0132] For other core points, if a member in that point is not included in the task cluster, a new cluster C2 is created, and the above process is repeated to form cluster C2.

[0133] Continue this process sequentially until all core points have been visited. Non-core points not included in any cluster: If a point is not included in any cluster, it is marked as noise.

[0134] The average value of all points in the largest cluster is taken as the ignition point location S. DBSCAN An example of a fire ignition point tracing simulation is shown below. Figure 2 As shown.

Claims

1. A method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations, characterized in that, Includes the following steps: Step 1: Construction of the image trace dataset: After the fire, collect photos of traces at the fire scene and embed GPS coordinates, direction and timestamp information to construct the image trace dataset; Step 2: Spread direction identification: Train a spread direction identification model based on the trace dataset; then identify the true direction of flame spread in each trace photo to be detected based on the trained spread direction identification model; Step 3: Origin point tracing: Based on the true direction and the coordinates of the sampling points corresponding to each trace photograph, calculate the direction vector of each trace photograph; construct a continuous direction field, starting from each sampling point, integrate each trajectory in the opposite direction of the direction field; use clustering to identify the dense region in space where the endpoint of each trajectory is located, and take the center of the largest cluster as the location of the ignition point. Step 3 is as follows: Step 3.1: Represent the latitude and longitude coordinates of the sampling point corresponding to each trace photograph as follows: ,i=0,1,2,…,N-1; in, The latitude and longitude coordinates of the sampling point and These are the longitude and latitude of the sampling points, respectively. With the first sampling point Using the origin as an example, the coordinates are converted to planar coordinates using the following formula: ； Where R is the Earth's radius. These are the sampling points corresponding to the i-th trace image. The x and y coordinates are represented as: ； The flame spread at the identified i-th sampling point is determined by the true azimuth angle. Converted to the direction vector at that sampling point : ； in, and They are vectors Components in the x and y directions; ； ； Step 3.2: For any point (x, y), construct a continuous direction field D(x, y) using inverse distance weighted interpolation: ； ； Where p is the weight parameter; A constant set; The number of trace photographs. The weights are the direction vectors corresponding to any point (x, y). Step 3.3: From each sampling point The process begins with inverse integration, specifically numerical integration using the fourth-order Runge-Kutta method, as follows: Step a: Initialize parameters Initialize the step size h and the maximum number of steps T; The termination condition is set as follows: directional field mode length. or ; It is the first The current position coordinates of the step; Step b: Single-step 4th-order Runge-Kutta calculation Calculate the intermediate slopes k1, k2, k3, k4, and iterate to obtain the current position coordinates at the (n+1)th step. : ； ； ； ； ； in, , , and These are the intermediate slopes; The formula for calculating the direction field is shown below; Termination check condition: If the termination condition is met, integration stops, and the current position is... for The coordinates of the streamline endpoint obtained by streamline integration under sampling point conditions; Step c: Traverse all sampling points and perform 4th-order Runge-Kutta calculations for each. With N sampling points, N streamline endpoints are ultimately formed, and the set of streamline endpoints is Q = { , ,…, }; Step 3.4: Calculate the location of the ignition point Step 1) Parameter settings: Neighborhood radius Distance threshold for determining whether two points belong to the same cluster; Minimum points min pts The minimum number of neighborhood points required to form a cluster; Step 2) Calculate the core points: For each point in the streamline endpoint set Q Calculate the number of points in its neighborhood: ； in, for Points within the neighborhood; If point Points within the neighborhood min pts Then point If a point is a core point, then points within its neighborhood that satisfy this condition are considered members of that core point; otherwise, the points... The number of neighboring points is insufficient; it belongs to the neighborhood of another core point; or it is noise. Traversing the set of streamline endpoints Q, we eventually form the set of core points S = { , ,…, }, ; Step 3) Cluster expansion: From the core point Starting from this point, recursively merge all density-reachable points to form a cluster; the specific implementation method is as follows: a) Create a new cluster C1, randomly selecting an unvisited core point. The core point Join cluster C1; b) Traversal calculation of core points Each member of the neighborhood Does it meet the requirements? min pts If satisfied, then the member will be... All members of the neighborhood are filled into cluster C1, and the core point is... Mark visited core points; if not satisfied, then the member... It is a non-core point, belonging to the boundary point or noise of cluster C1; c) Process the next unvisited core point s j And mark it as the new cluster C2, repeat step b), and finally complete the element filling of cluster C2; d) Continue sequentially until all core points in set S have been visited; if a core point is not included in any cluster, it is marked as noise and will not participate in the ignition point tracing calculation. Through cluster expansion, a collective C = { is formed for all possible clusters}. , ,…, }, ; Step 4) Determine the location of the fire ignition point: Take the largest cluster from the set C, and calculate the mean value of all points in that cluster as the location of the ignition point. : ； in, It is the point set of the largest cluster. Represents a point set The cardinality of a set is the number of points in the largest cluster.

2. The method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations according to claim 1, characterized in that, Step 1, constructing the map trace dataset, specifically includes: Constructing a training set: Photos of traces showing a clear direction of spread after the burning of combustibles at a forest fire scene are classified and used as positive samples. Specifically, traces in the photos are categorized according to the first-person perspective: 0-degree direction, 45-degree direction, 90-degree direction, 135-degree direction, and 180-degree direction. The 0-degree direction indicates traces from left to right, the 45-degree direction from lower left to upper right, the 90-degree direction from bottom to top, the 135-degree direction from lower right to upper left, and the 180-degree direction from right to left. A negative sample set is also added, including vegetation photos from various scenes in different seasons of the corresponding area, as well as photos of traces with unclear directions of fire spread. Construct a test set: Select trace photos of various directions of spread, including 0 degrees, 45 degrees, 90 degrees, 135 degrees and 180 degrees, as the test set. The test set should not be duplicated from the trace photos in the training set and should be no less than 20% of the total number of trace photos in the test set.

3. The method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations according to claim 2, characterized in that, Step 2 is as follows: Step 2.1: Establish a propagation direction identification model: The ResNet-18 CNN network model was used, and the model parameters were set, including the input image size, the number of input samples in each training iteration, the number of times the entire training dataset was traversed, the learning rate, and the ImageNet mean and standard deviation used for image normalization. The trace photos in the training set are input into the CNN network model for training to obtain the corresponding propagation direction recognition model; the propagation direction recognition model is used to identify each trace photo in the test set, and the recognition accuracy is calculated. If the accuracy reaches a%, the test is completed; otherwise, the corresponding model parameters are adjusted until training is completed. Step 2.2: Use the trained propagation direction recognition model to identify each trace photograph to be detected, obtain the direction angle b of flame propagation in the photograph, extract the direction information D from the corresponding trace photograph, and calculate the true azimuth angle α of flame propagation in each trace photograph: α = D + (90 - b).

4. The method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations according to claim 1, characterized in that, Before step 1, there is also step 0: using a drone to survey the area to be investigated, overlaying altitude, latitude, longitude and direction information onto the captured images, and then converting the obtained survey images into 3D format and importing them into the map of the backend system to complete the on-site modeling of the area to be investigated. Step 3 is followed by step 4: The coordinates of each sampling point are overlaid onto the corresponding modeled map with directional arrows pointing in the direction of the sampling point's direction vector, thus generating a visualized spread trend map; the coordinates of the ignition point are overlaid onto the corresponding modeled map and marked with special icons, thus generating a visualized ignition point diagram.

5. The method for analyzing the direction of fire spread and tracing the origin of fires in forest fire investigations according to claim 1, characterized in that, In step 4, when the coordinates of each sampling point are overlaid onto the corresponding modeled map using directional arrows, First, convert the planar coordinates to latitude and longitude coordinates, and then overlay them onto the map. The conversion method is as follows: Given any point (x, y) in a plane coordinate system, convert it back to latitude and longitude coordinates (Ec, Nc): ； Where (x, y) are planar coordinates, x is the east direction, and y is the north direction; the origin is ( , Let R be the Earth's radius, and ) be the reference point.

Images