Forest fire prevention early warning method and system based on multi-source sensor

Abstract

The invention provides a forest fire prevention early warning method and system based on a multi-source sensor, and the method comprises the steps: firstly obtaining a multi-source sensing monitoring data set of a target forest region, including data subsets of temperature, humidity, smoke concentration, visible light images and the like, and collecting each subset by at least two heterogeneous sensor nodes; performing space-time alignment processing on the multi-source sensing monitoring data set to generate a space-time distribution characteristic set containing various characteristics such as temperature gradient, calling a dynamic anomaly detection model to compare fire danger indexes, generating a fire danger grade quantitative score and an anomaly fluctuation index set, and performing data processing on the fire danger grade quantitative score and the anomaly fluctuation index set; the method comprises the steps of obtaining a fire hazard grade quantitative score, matching a multi-level early warning strategy based on the fire hazard grade quantitative score and a preset early warning threshold interval, generating a dynamic fire hazard early warning instruction, triggering fire extinguishing resource scheduling according to the dynamic fire hazard early warning instruction, generating a forest region escape path optimization strategy in combination with an abnormal fluctuation index, and achieving efficient and accurate forest fire prevention early warning and coping.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a forest fire early warning method and system based on multi-source sensors. Background Art

[0002] In the field of forest fire monitoring technology, traditional fire warning method research attempts to integrate multiple sensor data to improve warning performance, but most of these methods only stay at the simple superposition or preliminary correlation analysis at the data level, lacking in-depth mining and collaborative processing of multi-source heterogeneous data in the spatiotemporal dimension. Specifically, existing methods fail to effectively solve the differences in sampling frequency, timestamp, spatial location, etc. of different sensor data, resulting in poor data fusion effects and difficulty in accurately extracting key spatiotemporal distribution characteristics before a fire occurs. In addition, existing warning strategies are usually based on fixed threshold judgments, and are unable to adaptively adjust warning levels and response measures according to dynamic changes in the fire risk situation, limiting the flexibility and intelligence of the warning system.

[0003] In terms of fire risk assessment and early warning decision-making, existing technologies often separate the quantitative scoring of fire risk levels from the formulation of early warning strategies, lacking a closed-loop system that can directly convert fire risk assessment results into specific early warning instructions and emergency response strategies. At the same time, for fire-fighting resource scheduling and forest escape route planning, existing methods are mostly based on static plans or simple rules, and fail to fully consider the real-time impact of abnormal fluctuations in fire risk indicators on resource scheduling and escape route optimization, resulting in the inability to quickly and effectively allocate resources and evacuate personnel when a fire occurs. Summary of the invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, an embodiment of the present invention provides a forest fire early warning method based on a multi-source sensor, the method comprising: Acquire a multi-source sensor monitoring data set of a target forest area, wherein the multi-source sensor monitoring data set includes a temperature monitoring data subset, a humidity monitoring data subset, a smoke concentration monitoring data subset, and a visible light image data subset, wherein each data subset is collected and generated by at least two heterogeneous sensor nodes; Performing spatiotemporal alignment processing on the multi-source sensor monitoring data set to generate a spatiotemporal distribution feature set of the target forest area, wherein the spatiotemporal distribution feature set includes temperature gradient distribution features, humidity fluctuation correlation features, smoke diffusion trajectory features, and image texture change features; Calling a dynamic anomaly detection model to perform fire risk index comparison processing on the spatiotemporal distribution feature set to generate a fire risk level quantitative score and an abnormal fluctuation index set; Based on the fire risk level quantitative score and the preset warning threshold interval, a multi-level warning strategy is matched to generate a dynamic fire risk warning instruction for the target forest area; The fire-fighting resource scheduling operation is triggered according to the dynamic fire risk warning instruction, and the forest escape path optimization strategy is generated in combination with the abnormal fluctuation indicator set.

[0005] On the other hand, an embodiment of the present invention also provides a forest fire warning system based on multi-source sensors, including a processor and a machine-readable storage medium, wherein the machine-readable storage medium is connected to the processor, the machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0006] Based on the above aspects, the embodiment of the present application is based on the spatiotemporal alignment processing of the multi-source sensor monitoring data set, and the multi-dimensional features such as temperature gradient distribution, humidity fluctuation correlation, smoke diffusion trajectory and image texture change are integrated and modeled. The spatiotemporal distribution feature set is compared and processed by the dynamic anomaly detection model. It not only realizes the quantitative scoring of the fire risk level, but also accurately identifies the abnormal fluctuation index. On this basis, the strategy matching is combined with the preset multi-level warning threshold interval. The generated dynamic fire risk warning instruction can adaptively adjust the warning response mechanism according to different fire risk levels, which significantly improves the timeliness and accuracy of the warning. Furthermore, the fire extinguishing resource scheduling and the forest escape path optimization strategy are collaboratively designed, and the escape path is dynamically planned through the abnormal fluctuation index set, which effectively avoids the problem of separation of resource allocation and personnel evacuation strategy in the traditional warning system, thereby constructing a closed-loop forest fire warning system integrating fire risk monitoring, warning response, resource scheduling and escape guidance, which significantly improves the overall efficiency and intelligence level of forest fire prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 It is a schematic diagram of the execution flow of the forest fire early warning method based on multi-source sensors provided in an embodiment of the present invention.

[0008] Figure 2 It is a schematic diagram of exemplary hardware and software components of a forest fire warning system based on multi-source sensors provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] The present invention will be described in detail below with reference to the accompanying drawings. Figure 1 The present invention is a flowchart of a multi-source sensor-based forest fire early warning method provided by an embodiment of the present invention. The multi-source sensor-based forest fire early warning method is introduced in detail below.

[0010] Step S110: Acquire a multi-source sensor monitoring data set of the target forest area, wherein the multi-source sensor monitoring data set includes a temperature monitoring data subset, a humidity monitoring data subset, a smoke concentration monitoring data subset and a visible light image data subset, wherein each data subset is collected and generated by at least two heterogeneous sensor nodes.

[0011] The target forest area in this embodiment is set to be a mountain forest area with an area of ​​about 200 square kilometers. The vegetation types in this area are rich, including large areas of coniferous forests, broad-leaved forests and shrubs. In order to comprehensively monitor the conditions of the forest area, various types of sensor nodes are reasonably deployed throughout the forest area.

[0012] For example, in area A of the forest area, three temperature sensor nodes are arranged, named T1, T2 and T3. T1 uses thermistor technology to sense temperature, T2 works based on the principle of thermocouple, and T3 is a new temperature sensor based on semiconductor materials. These three sensors constitute a heterogeneous sensor node group, which continuously collects temperature data at their locations. For example, at 9 am on a certain day, the temperature value collected by T1 is 23°C, T2 collects 23.5°C, and T3 collects 23.2°C. As time goes by, a series of temperature data is continuously generated to form a temperature monitoring sequence.

[0013] For another example, in area B of the forest area, two humidity sensor nodes are set up, marked as H1 and H2. H1 is a capacitive humidity sensor, and H2 is a resistive humidity sensor, which are heterogeneous sensors. They monitor local humidity information in real time. For example, at a certain moment, H1 detects a humidity of 58% and H2 detects a humidity of 60%. Many such humidity data constitute a humidity monitoring data subset.

[0014] For another example, in area C of the forest area, two smoke concentration sensor nodes are installed, denoted as S1 and S2. S1 uses the principle of photoelectric sensing to detect smoke concentration, and S2 works based on the gas-sensitive characteristics of semiconductors, and is a heterogeneous sensor. They detect the smoke concentration in the surrounding air. For example, at a certain moment, S1 detects a smoke concentration of 0.08 mg / m³, and S2 detects 0.09 mg / m³. Many such data constitute a subset of smoke concentration monitoring data.

[0015] At the same time, multiple cameras are installed at different heights and locations in the forest area to collect visible light image data. These cameras have different resolutions and shooting angles. For example, the resolution of camera C1 is 2560×1440 and the shooting angle is 130°, and the resolution of camera C2 is 1920×1080 and the shooting angle is 110°. The image data collected by them together constitute the visible light image data subset. These cameras capture the forest area at set time intervals (such as 3 minutes). For example, at 10:03 am, C1 captured an image containing some hillside vegetation, and C2 captured an image near the valley.

[0016] Step S120: performing spatiotemporal alignment processing on the multi-source sensor monitoring data set to generate a spatiotemporal distribution feature set of the target forest area, wherein the spatiotemporal distribution feature set includes temperature gradient distribution features, humidity fluctuation correlation features, smoke diffusion trajectory features and image texture change features.

[0017] Step S121: extracting the temperature monitoring sequence of each sensor node in the temperature monitoring data subset, performing interpolation fitting processing based on the timestamp and spatial coordinates of the temperature monitoring sequence, and generating a continuous temperature field distribution map of the target forest area.

[0018] Take the temperature sensor nodes T1, T2 and T3 in area A as an example. T1 records the temperature every 10 minutes starting at 8 am, and obtains a temperature monitoring sequence, such as 22°C at 8 am, 22.2°C at 8:10 am, etc. T2 and T3 also record the temperature of their respective locations at the same time interval. Each sensor node has its corresponding spatial coordinates. Assume that the coordinates of T1 are (x1, y1, z1), the coordinates of T2 are (x2, y2, z2), and the coordinates of T3 are (x3, y3, z3). Interpolation fitting is performed based on these timestamps and spatial coordinates. For example, at 8:30, to determine the temperature of a point P between T1 and T2, first calculate the spatial distance from point P to T1 and T2. Assume that the distance from point P to T1 is d1 and the distance to T2 is d2. According to the distance ratio and the temperature values ​​of T1 and T2 at 8:30 (assuming T1 is 22.5℃ and T2 is 22.7℃), the temperature of point P is calculated as (d2×22.5+d1×22.7) / (d1+d2) by linear interpolation method. Calculate many such points in the forest area to generate a continuous temperature field distribution map of the entire target forest area at different times. The continuous temperature field distribution map can intuitively show the spatial distribution of the forest area temperature and its changes over time.

[0019] Step S122: extracting the humidity monitoring sequence of each sensor node in the humidity monitoring data subset, calculating the humidity change rate of adjacent sensor nodes, and constructing a humidity gradient association matrix according to the spatial distance weight.

[0020] For the humidity sensor nodes H1 and H2 in area B, H1 records humidity every 15 minutes to form a humidity monitoring sequence, such as 55% at 9:00, 56% at 9:15, etc. H2 also records at 15-minute intervals. Calculate the humidity change rate of adjacent sensor nodes. Assume that at 9:30, H1 records humidity of 57%, and the previous record was 56% at 9:15. Then the humidity change rate of H1 in these 15 minutes is (57%-56%) / 15=0.067% / minute; H2 changes from 58% to 59% in the same time, and its humidity change rate is (59%-58%) / 15=0.067% / minute. Then construct the humidity gradient association matrix based on the spatial distance weight. Assume that the spatial distance between H1 and H2 is d, and define the distance weight as 1 / d. For the humidity change rate, assume that the change rate of H1 is r1, the change rate of H2 is r2, the matrix element M12 = r1 × (1 / d) + r2 × (1 / d), M21 = M12, M11 and M22 are respectively set to the product of the humidity change rate of H1 and H2 themselves and a small constant (such as 0.01), and so on to construct a complete humidity gradient association matrix, which reflects the change correlation of humidity between sensor nodes at different locations.

[0021] Step S123: performing diffusion direction tracking processing on the subset of smoke concentration monitoring data, and generating smoke diffusion trajectory characteristics based on the smoke concentration change gradient and wind speed monitoring data, wherein the smoke diffusion trajectory characteristics include diffusion speed, diffusion angle and concentration attenuation coefficient.

[0022] Take the smoke concentration sensor nodes S1 and S2 in area C as an example. Assume that S1 detects the smoke concentration every 20 minutes, and S2 also detects at intervals. At 10 o'clock, S1 detected a smoke concentration of 0.1mg / m³, and at 10:20 it detected 0.12mg / m³; S2 detected 0.11mg / m³ at 10 o'clock and 0.13mg / m³ at 10:20. Calculate the smoke concentration change gradient. Taking S1 as an example, within these 20 minutes, the concentration change gradient is (0.12-0.1) / 20=0.001mg / (m³·minute). At the same time, obtain the wind speed monitoring data of the area. Assume that during the period from 10 o'clock to 10:20, the average wind speed is 2m / s and the wind direction is due east. Generate smoke diffusion trajectory features based on these data. The diffusion speed is calculated based on the relationship between the smoke concentration gradient and wind speed. It is assumed that through a certain empirical relationship, the diffusion speed v=0.5×concentration gradient×wind speed, that is, v=0.5×0.001×2=0.001m / minute. The diffusion angle is determined to be due east according to the wind direction and is set to 0°. The concentration attenuation coefficient is calculated by comparing the smoke concentrations at different times and locations. Assuming that at 10:40, S1 detected 0.13mg / m³ and S2 detected 0.14mg / m³. According to the formula concentration attenuation coefficient = (S1 concentration at 10:40-S1 concentration at 10:20) / (S2 concentration at 10:40-S2 concentration at 10:20), the concentration attenuation coefficient is calculated, thus generating the smoke diffusion trajectory characteristics including the diffusion speed, diffusion angle and concentration attenuation coefficient.

[0023] Step S124: performing frame-by-frame analysis on the visible light image data subset, extracting the texture feature vector of each frame of the image, and comparing the similarity changes of the texture feature vectors based on a time series to generate image texture change features.

[0024] Step S1241: Divide the visible light image data subset into a plurality of image frame sequences at preset time intervals, and perform grayscale conversion and noise reduction processing on each image frame.

[0025] Assume that the visible light image data subset is composed of images taken by the camera every 5 minutes, and divide it into a sequence of image frames. For each image frame, grayscale processing is first performed, and the color image is converted into a grayscale image using the weighted average method. The specific calculation is that the grayscale value is equal to 0.299 multiplied by the R value of the image pixel plus 0.587 multiplied by the G value plus 0.114 multiplied by the B value. Take a pixel in a certain image frame as an example, its R value is 200, G value is 150, and B value is 100. Then the grayscale value of the pixel is equal to 0.299 multiplied by 200 plus 0.587 multiplied by 150 plus 0.114 multiplied by 100. The calculated grayscale value is about 144.35. After grayscale processing, noise reduction processing is performed, using the Gaussian filtering method, setting the Gaussian kernel size to 3 times 3, and the standard deviation to 1.5. For each pixel in the image frame, the filtered pixel value is calculated based on the weight distribution of the Gaussian kernel and the values ​​of the surrounding pixels. For example, for a certain pixel, the 8 surrounding pixels and itself form a 3 times 3 area. According to the Gaussian kernel weight, the values ​​of these 9 pixels are multiplied by the corresponding weights and then added together to obtain the noise reduction value of the pixel after Gaussian filtering.

[0026] Step S1242: using an edge detection algorithm to extract contour features in the processed image frame, and calculating the grayscale co-occurrence matrix of the pixels in the contour area.

[0027] The Canny edge detection algorithm is used to extract contour features, and the low threshold is set to 50 and the high threshold is set to 150. For the image frame after grayscale and noise reduction processing, the gradient is first calculated to obtain the gradient amplitude and direction of each pixel in the image. Then non-maximum suppression is performed to retain the pixel with the largest gradient amplitude and suppress other non-edge pixels. Then, double threshold detection is performed according to the set low threshold and high threshold. Pixels with gradient amplitudes greater than the high threshold are determined as strong edge pixels, and pixels with gradient amplitudes less than the low threshold are determined as non-edge pixels. Pixels between the two are determined to be edge pixels based on their connectivity with strong edge pixels. In this way, the contour features of the image are obtained.

[0028] For the pixels in the contour area, calculate their gray-level co-occurrence matrix. Assuming that the size of the gray-level co-occurrence matrix is ​​8 times 8, in the contour area, count the frequency of occurrence of different gray-level pairs in a certain direction. For example, in the 0-degree direction, starting from the upper left corner of the image, check the gray-levels of adjacent pixels in turn. If the gray-level of the current pixel is 3 and the gray-level of the adjacent pixel is 5, add 1 to the frequency at the corresponding position (3, 5) in the gray-level co-occurrence matrix. After traversing all the pixels in the contour area, the gray-level co-occurrence matrix is ​​obtained.

[0029] Step S1243: extracting contrast, energy value and homogeneity index from the gray level co-occurrence matrix to construct a texture feature vector of a single frame image.

[0030] Extract the contrast from the gray-level co-occurrence matrix. The calculation method is to multiply the square of the difference between the row and column indices of each element in the gray-level co-occurrence matrix by the element value, and then add all the results. For example, for an element P(i, j) in the gray-level co-occurrence matrix, the contrast is equal to the sum of the squares of all i and j, (ij) multiplied by P(i, j). Assuming that P(1, 2) in the gray-level co-occurrence matrix is ​​equal to 0.1, P(2, 3) is equal to 0.2, etc., the contrast value is calculated according to this formula.

[0031] The energy value is extracted by squaring each element value in the gray-level co-occurrence matrix and then adding them up, that is, the energy value is equal to the sum of the squares of all i and j, P(i, j).

[0032] The calculation of the homogeneity index is to divide the value of each element in the gray level co-occurrence matrix by 1 plus the absolute value of the difference between the row and column indices, and then add all the results. That is, the homogeneity is equal to the sum of P(i, j) divided by (1 plus |ij|) for all i and j.

[0033] The extracted contrast, energy value and homogeneity index are combined to construct the texture feature vector of a single frame image. For example, if the contrast calculated for a single frame image is 0.5, the energy value is 0.3, and the homogeneity is 0.4, then the texture feature vector of the frame image is (0.5, 0.3, 0.4).

[0034] Step S1244: Arrange the texture feature vectors in chronological order, calculate the cosine similarity between adjacent frames, and count the positions of mutation points where the similarity decrease rate exceeds a preset threshold.

[0035] Arrange the texture feature vectors of all image frames in the order of shooting time. For two adjacent frames, assume that the texture feature vector of the first frame is V1, and its elements are (v11, v12, v13), and the texture feature vector of the second frame is V2, and its elements are (v21, v22, v23). The calculation of cosine similarity is the dot product of V1 and V2 divided by the modulus of V1 multiplied by the modulus of V2. Among them, the modulus of V1 is the square root of v11 squared plus v12 squared plus v13 squared, and the modulus of V2 is calculated in a similar way. The dot product is v11 multiplied by v21 plus v12 multiplied by v22 plus v13 multiplied by v23. For example, V1 is (0.5, 0.3, 0.4), and V2 is (0.4, 0.35, 0.38). First, calculate that the modulus of V1 is approximately 0.707, and the modulus of V2 is approximately 0.614. The dot product is 0.5 times 0.4 plus 0.3 times 0.35 plus 0.4 times 0.38, which equals 0.447. The cosine similarity is approximately 0.447 divided by (0.707 times 0.614), which is approximately 1.02 (this is only an example calculation, and the actual calculation may need to be adjusted due to data range, etc.).

[0036] After calculating the cosine similarity between adjacent frames, the position of the mutation point where the similarity decrease rate exceeds the preset threshold is counted. Assuming the preset threshold is 0.1, if the cosine similarity of two adjacent frames decreases from 0.8 to 0.6, and the time interval is 10 minutes (assuming the interval between two frames), then the similarity decrease rate is (0.8-0.6) divided by 10, which is equal to 0.02 per minute. If the decrease rate exceeds 0.1, the position of the second frame is recorded as the mutation point.

[0037] Step S1245: Associating the mutation point position with the corresponding image frame coordinates to generate image texture change features reflecting the mutation of vegetation status.

[0038] The statistically obtained mutation point position is associated with the coordinates of the corresponding image frame in the entire image sequence. For example, if the mutation point position corresponds to the 10th frame image, and its coordinates in the image sequence are (x, y), then (10, x, y) is recorded to form an image texture change feature that reflects the mutation of the vegetation state. Multiple such records are combined to form a multi-dimensional image texture change feature set, which is used for subsequent analysis of the mutation of the vegetation state.

[0039] Step S125: integrating the continuous temperature field distribution map, the humidity gradient association matrix, the smoke diffusion trajectory characteristics and the image texture change characteristics into the spatiotemporal distribution feature set.

[0040] The previously generated continuous temperature field distribution map, humidity gradient correlation matrix, smoke diffusion trajectory characteristics and image texture change characteristics are correlated and integrated. For example, with time as the correlation clue, the temperature information corresponding to the continuous temperature field distribution map at the same time, the humidity change correlation reflected by the humidity gradient correlation matrix, the diffusion parameters in the smoke diffusion trajectory characteristics and the mutation point information in the image texture change characteristics are integrated to form a comprehensive spatiotemporal distribution feature set, which comprehensively reflects the various key characteristics of the target forest area at different times and their interrelationships.

[0041] Step S130: calling the dynamic anomaly detection model to perform fire risk index comparison processing on the spatiotemporal distribution feature set to generate a fire risk level quantitative score and an abnormal fluctuation index set.

[0042] Step S131: extract the maximum temperature value, temperature change slope and regional temperature difference extreme value from the continuous temperature field distribution diagram of the spatiotemporal distribution feature set, perform dynamic threshold comparison on the maximum temperature value, temperature change slope and regional temperature difference extreme value with the historical temperature reference interval stored in the dynamic anomaly detection model, and generate a temperature anomaly score.

[0043] Extract relevant features from the continuous temperature field distribution map generated previously. For example, in the time period from 8 am to 4 pm on a certain day, by analyzing the continuous temperature field distribution map, find the highest temperature value in this time period, assuming it is 35°C. Calculate the temperature change slope, select the hour from 10 am to 11 am, assuming that the temperature at 10 am is 28°C and the temperature at 11 am is 30°C, the temperature change slope = (30-28) / 1 = 2°C / hour. Calculate the regional temperature difference extreme value, in two specific areas A and B in the forest area, assuming that the average temperature of area A is 32°C and the average temperature of area B is 26°C, the regional temperature difference extreme value = 32-26 = 6°C. Then compare these values ​​with the historical temperature reference interval stored in the dynamic anomaly detection model for dynamic thresholds. Assume that in the historical temperature reference interval, the normal range of the highest temperature value is 20°C-30°C, the normal range of the temperature change slope is 0-1°C / hour, and the normal range of the regional temperature difference extreme value is 0-5°C. For the maximum temperature value, the proportion outside the normal range is (35-30) / (35-20)=0.33; for the temperature change slope, the proportion outside the normal range is (2-1) / (2-0)=0.5; for the regional temperature difference extreme value, the proportion outside the normal range is (6-5) / (6-0)=0.17. Through a certain weighted calculation, assuming that the weight of the maximum temperature value is 0.5, the weight of the temperature change slope is 0.3, and the weight of the regional temperature difference extreme value is 0.2, the temperature anomaly score = 0.5×0.33+0.3×0.5+0.2×0.17=0.371.

[0044] Step S132: extracting the humidity change rate of adjacent sensor nodes from the humidity gradient association matrix, determining the humidity fluctuation deviation according to the humidity attenuation threshold predefined in the dynamic anomaly detection model, and weighting the humidity fluctuation deviation in combination with the current ambient humidity value.

[0045] Extract the humidity change rate of adjacent sensor nodes from the humidity gradient association matrix. Take the previously constructed humidity gradient association matrix as an example. Assume that the humidity change rates of H1 and H2 in the humidity gradient association matrix are 0.067% / minute and 0.067% / minute respectively. According to the humidity attenuation threshold predefined in the dynamic anomaly detection model, assuming the threshold is 0.05% / minute, calculate the humidity fluctuation deviation. For H1, the humidity fluctuation deviation = (0.067-0.05) / 0.067 = 0.254; for H2, the humidity fluctuation deviation = (0.067-0.05) / 0.067 = 0.254. Then, the weight correction is performed in combination with the current ambient humidity value. Assuming that the current ambient humidity value is 60%, a humidity-related weight function is set, such as weight = 1-(60-50) / 100 = 0.9 (assuming that the weight changes linearly between 50% and 70%). The corrected humidity fluctuation deviation = 0.254×0.9 = 0.229.

[0046] Step S133: extracting diffusion velocity and concentration attenuation coefficient from the smoke diffusion trajectory characteristics, inputting the diffusion velocity and concentration attenuation coefficient into the smoke diffusion history feature library in the dynamic anomaly detection model for trajectory similarity matching, and generating a smoke diffusion anomaly index.

[0047] Extract the diffusion velocity and concentration attenuation coefficient from the smoke diffusion trajectory characteristics. Assume that the previously calculated diffusion velocity is 0.001m / min and the concentration attenuation coefficient is 0.8. Enter these two values ​​into the smoke diffusion history feature library in the dynamic anomaly detection model. The history feature library stores a large amount of smoke diffusion velocity and concentration attenuation coefficient data under different conditions. For example, there is a set of data in the history feature library with a diffusion velocity of 0.0008m / min and a concentration attenuation coefficient of 0.75. Perform trajectory similarity matching by calculating the Euclidean distance. Assume that the Euclidean distance formula is d=√[(0.001-0.0008)²+(0.8-0.75)²]. In actual calculation, an alternative method of square root and square root can be used, such as by looking up a pre-calculated correspondence table, etc., assuming that the distance value is 0.05. Set a threshold, such as 0.1. If the distance value is less than the threshold, the similarity is high, and the smoke diffusion anomaly index is set to 0.2 (assuming that the relationship between the similarity and the anomaly index is pre-set); if the distance value is greater than the threshold, the similarity is low, and the smoke diffusion anomaly index is set to 0.8. Here it is assumed that the obtained smoke diffusion anomaly index is 0.2.

[0048] Step S134: extracting high-frequency component energy values ​​from the image texture change features, performing frequency domain matching on the high-frequency component energy values ​​and the reference energy spectrum of the vegetation burning feature library in the dynamic anomaly detection model, and generating an image anomaly matching degree.

[0049] Extract the high-frequency component energy value from the image texture change feature. Take the previously generated image texture change feature as an example. By performing Fourier transform and other methods on the image (the specific transformation process is not elaborated here), assume that the high-frequency component energy value of a certain frame of image is 0.6. This value is matched with the reference energy spectrum of the vegetation burning feature library in the dynamic anomaly detection model in the frequency domain. Assume that there is a reference energy spectrum value of 0.5 in the vegetation burning feature library. By calculating the difference ratio between the two, such as (0.6-0.5) / 0.6=0.167, a conversion relationship is set to convert the difference ratio into image anomaly matching degree. Assuming that the conversion relationship is image anomaly matching degree = 1-0.167=0.833, the image anomaly matching degree is generated.

[0050] Step S135: Input the temperature anomaly score, the corrected humidity fluctuation deviation, the smoke diffusion anomaly index and the image anomaly matching degree into the multi-index fusion layer of the dynamic anomaly detection model, perform linear weighted calculation according to the preset fire risk weight coefficient, generate the quantitative score of the fire risk level, and integrate the parts of the temperature anomaly score, humidity fluctuation deviation, smoke diffusion anomaly index and image anomaly matching degree that exceed the corresponding sub-indicator threshold into the abnormal fluctuation index set.

[0051] Step S1351: normalize the temperature anomaly score to map it to a dimensionless value in the range of 0-1.

[0052] Assuming that the temperature anomaly score ranges from 0 to 1, the normalization method is used for standardization. Let the original temperature anomaly score be T_score and the standardized temperature anomaly score be T_score_norm. The calculation method is that T_score_norm is equal to T_score minus the minimum value in the temperature anomaly score, and then divided by the maximum value minus the minimum value in the temperature anomaly score. For example, if the minimum value of the temperature anomaly score is 0.1 and the maximum value is 0.8, and a certain original temperature anomaly score is 0.4, then T_score_norm is equal to (0.4-0.1) divided by (0.8-0.1), which is about 0.429, which is mapped to the interval of 0-1 and becomes a dimensionless value.

[0053] Step S1352: Calculate the difference between the corrected humidity fluctuation deviation and the preset humidity reference value, and divide it by the upper limit of the humidity monitoring range to convert it into a dimensionless deviation coefficient.

[0054] Assume that the corrected humidity fluctuation deviation is H_dev, the preset humidity base value is H_base, and the upper limit of the humidity monitoring range is H_max. The dimensionless deviation coefficient H_coeff is calculated as H_coeff equals (H_dev-H_base) divided by H_max. For example, the preset humidity base value is 50%, the upper limit of the humidity monitoring range is 100%, and the corrected humidity fluctuation deviation is 0.254, then H_coeff is equal to (0.254-0.5) divided by 1, which is -0.246, converted to the dimensionless deviation coefficient.

[0055] Step S1353: performing logarithmic transformation on the smoke diffusion anomaly index, compressing the value fluctuation range and normalizing it to a diffusion risk coefficient within the range of 0-1.

[0056] Suppose the smoke diffusion anomaly index is S_index, and first perform a logarithmic transformation on it, and let the transformed value be S_log. S_log is equal to the logarithm of S_index plus 1 to the base of 10 (the addition of 1 is to avoid the appearance of 0 in the logarithm). For example, if the smoke diffusion anomaly index is 0.2, then S_log is equal to the logarithm of (0.2+1) to the base of 10, which is about 0.08. Then S_log is normalized, and the normalized diffusion risk coefficient is S_coeff. S_coeff is equal to S_log minus the minimum value in S_log, and then divided by the maximum value in S_log minus the minimum value. Assuming that the minimum value of S_log is 0.05 and the maximum value is 0.15, then S_coeff is equal to (0.08-0.05) divided by (0.15-0.05), which is equal to 0.3. It is normalized to the interval of 0-1 to become the diffusion risk coefficient.

[0057] Step S1354: Calculate the ratio of the image anomaly matching degree to a preset energy spectrum threshold to generate a dimensionless image anomaly coefficient.

[0058] Assume that the image anomaly matching degree is I_match and the preset energy spectrum threshold is I_threshold. The dimensionless image anomaly coefficient I_coeff is calculated as I_coeff equals I_match divided by I_threshold. For example, if the image anomaly matching degree is 0.833 and the preset energy spectrum threshold is 1, then I_coeff equals 0.833 divided by 1 equals 0.833, generating a dimensionless image anomaly coefficient.

[0059] Step S1355: Extract the fire risk weight coefficient set stored in the multi-index fusion layer of the dynamic anomaly detection model, wherein the fire risk weight coefficient set includes temperature weight, humidity weight, smoke weight and image weight, and the weight values ​​of temperature weight, humidity weight, smoke weight and image weight are normalized to ensure that the sum is 1.

[0060] Obtain the fire risk weight coefficient set from the multi-index fusion layer of the dynamic anomaly detection model, assuming that the temperature weight is W_T, the humidity weight is W_H, the smoke weight is W_S, and the image weight is W_I. These weight values ​​have been normalized, for example, W_T is 0.4, W_H is 0.2, W_S is 0.3, and W_I is 0.1, and their sum 0.4+0.2+0.3+0.1 is equal to 1.

[0061] Step S1356: Multiply the dimensionless temperature anomaly score, the dimensionless deviation coefficient, the diffusion risk coefficient and the image anomaly coefficient by the corresponding fire risk weight coefficients respectively to generate weighted temperature risk value, humidity risk value, smoke risk value and image risk value.

[0062] The weighted temperature risk value T_risk is equal to the dimensionless temperature anomaly score T_score_norm multiplied by the temperature weight W_T. For example, if T_score_norm is 0.429 and W_T is 0.4, then T_risk equals 0.429 multiplied by 0.4, which equals 0.1716.

[0063] The weighted humidity risk value H_risk is equal to the dimensionless deviation coefficient H_coeff multiplied by the humidity weight W_H. If H_coeff is -0.246 and W_H is 0.2, then H_risk is -0.246 multiplied by 0.2, which equals -0.0492.

[0064] The weighted smoke risk value S_risk is equal to the diffusion risk coefficient S_coeff multiplied by the smoke weight W_S. If S_coeff is 0.3 and W_S is 0.3, then S_risk is equal to 0.3 multiplied by 0.3, which is equal to 0.09.

[0065] The weighted image risk value I_risk is equal to the image abnormality coefficient I_coeff multiplied by the image weight W_I. When I_coeff is 0.833 and W_I is 0.1, I_risk is equal to 0.833 multiplied by 0.1, which is equal to 0.0833.

[0066] Step S1357: The weighted temperature risk value, humidity risk value, smoke risk value and image risk value are cumulatively summed to generate the quantitative score of the fire risk level.

[0067] The quantitative score of the fire risk level, Risk_score, is equal to the weighted temperature risk value T_risk plus the humidity risk value H_risk plus the smoke risk value S_risk plus the image risk value I_risk. That is, Risk_score is equal to 0.1716+(-0.0492)+0.09+0.0833, which is equal to 0.3357, generating a quantitative score of the fire risk level. At the same time, the temperature anomaly score, humidity fluctuation deviation, smoke diffusion anomaly index, and image anomaly matching that exceed the corresponding sub-indicator threshold are integrated into an abnormal fluctuation index set. Assume that the temperature anomaly score threshold is 0.3, the humidity fluctuation deviation threshold is 0.2, the smoke diffusion anomaly index threshold is 0.3, and the image anomaly matching threshold is 0.8. The temperature anomaly score of 0.429 exceeds the threshold of 0.3, and the excess is 0.429-0.3, which equals 0.129; the humidity fluctuation deviation of 0.254 exceeds the threshold of 0.2, and the excess is 0.254-0.2, which equals 0.054; the smoke diffusion anomaly index of 0.2 does not exceed the threshold of 0.3, and there is no excess value; the image anomaly matching degree of 0.833 exceeds the threshold of 0.8, and the excess is 0.833-0.8, which equals 0.033. These excess parts are integrated into an abnormal fluctuation index set, which includes values ​​such as 0.129, 0.054, and 0.033, reflecting the degree of abnormal fluctuation of each monitoring indicator relative to the normal range.

[0068] Step S140: Based on the fire risk level quantitative score and the preset warning threshold interval, a multi-level warning strategy is matched to generate a dynamic fire risk warning instruction for the target forest area.

[0069] Step S141: When the fire risk level quantitative score is within the first threshold range, a primary warning instruction is generated, and the primary warning instruction triggers the sensor node sampling frequency increase operation and the drone cruise path planning.

[0070] Assume that the preset first threshold interval is 0-0.4. The fire risk level quantitative score 0.3375 calculated above is within this interval. At this time, a primary warning instruction is generated.

[0071] For the sensor node sampling frequency increase operation, take the temperature sensor nodes T1, T2, and T3 as an example. They originally collected temperature data every 10 minutes, and now the sampling frequency is increased to every 5 minutes. This can obtain more intensive temperature data, so as to monitor temperature changes more timely.

[0072] For drone cruise path planning, it is assumed that the existing drone initial cruise paths in the forest area fly in a fixed grid pattern, covering most of the forest area, but the coverage frequency of key areas is low. Now re-plan the drone cruise path based on the primary warning. For example, tilt the drone cruise path toward areas with abnormal temperature, abnormal humidity, and abnormal smoke concentration, and increase the frequency of inspections in these potential fire risk areas. It can be set that the coverage of the drone cruise path is doubled within a certain radius (such as a radius of 2 kilometers) with these abnormal areas as the center, to ensure that changes in these areas can be more closely monitored.

[0073] Step S142: When the fire risk level quantitative score is within the second threshold interval, an intermediate warning instruction is generated, and the intermediate warning instruction triggers the pre-start operation of the fire extinguishing equipment in the key area and the broadcast notification of personnel evacuation.

[0074] Assuming that the preset second threshold interval is 0.4-0.6, if the fire risk level quantitative score is within this interval, for example, the score reaches 0.45, then a medium-level warning instruction is generated.

[0075] For the pre-start operation of fire-fighting equipment in key areas, the key areas must be determined first. According to the multi-source sensor monitoring data obtained earlier, such as the temperature abnormality area, the area pointed by the smoke diffusion path, etc., it is determined as the key area. Assume that there are five key areas determined in this way in the forest area, marked as Z1, Z2, Z3, Z4, and Z5 respectively. In each key area, fire-fighting equipment such as fire hydrants and fire extinguishers are equipped. Now pre-start operations are performed on the fire-fighting equipment in these key areas, such as checking the water pressure of the fire hydrant to ensure that the water pressure is within the normal range and can discharge water to extinguish the fire at any time; check whether the pressure pointer of the fire extinguisher is in the green normal area to ensure that it can be used normally.

[0076] For personnel evacuation broadcast notifications, multiple broadcast points are set up in the forest area, covering every corner of the forest area. When a medium-level warning instruction is issued, an evacuation notice is issued to people who may be affected in the forest area through these broadcast points. The notification content includes information such as the current fire risk situation, the scope of the dangerous area, and the recommended evacuation direction. For example, the broadcast notification: "Please note that the current forest fire risk level has been upgraded to medium. There are fire hazards near Z1, Z2 and other areas. People nearby are requested to evacuate to safe areas in an orderly manner according to the direction of the sign." Step S143: When the fire risk level quantitative score is within the third threshold interval, an advanced warning instruction is generated, and the advanced warning instruction triggers the coordinated response of fire extinguishing resources in the entire area and the opening operation of the emergency channel.

[0077] Assuming that the preset third threshold interval is above 0.6, if the fire risk level quantitative score reaches 0.7, an advanced warning instruction is generated.

[0078] The coordinated response of fire-fighting resources in the entire region means mobilizing all available fire-fighting resources in the forest area. This includes fire trucks, large-scale fire-fighting equipment, and fire-fighting forces that can be supported in surrounding areas. For example, there are three fire trucks in the forest area, parked in different locations, numbered F1, F2, and F3. After receiving the advanced warning command, F1 drives to the area with abnormal temperature and high smoke concentration, F2 goes to the direction where the fire may spread to block it, and F3 is ready to support other vehicles as a backup force. At the same time, contact the fire departments in the surrounding areas to request support. For example, the fire brigades in the surrounding cities may send five fire trucks to assist.

[0079] In terms of emergency channel opening operation, multiple emergency channels are preset in the forest area to ensure that personnel and fire-fighting equipment can pass quickly. These emergency channels are usually closed or restricted to protect the ecological environment of the forest area. When a high-level warning instruction is issued, these emergency channels are quickly opened. For example, emergency channel Y1 was originally equipped with roadblocks and door locks. Now a special person is arranged to open the roadblocks and door locks, and set up obvious signs at the entrance of the channel to guide personnel and vehicles to pass quickly. At the same time, ensure that there are no obstacles in the emergency channel and the road is unobstructed, so that fire fighting and personnel evacuation can be carried out efficiently in an emergency.

[0080] Step S144: dynamically adjusting the boundary value of the warning threshold interval according to the priority ranking of each sub-indicator in the abnormal fluctuation indicator set, and updating the triggering condition of the dynamic fire risk warning instruction based on the adjusted threshold interval.

[0081] Step S1441: Load a preset sub-indicator priority list from the dynamic anomaly detection model, wherein the list is arranged in descending order of fire impact as follows: temperature anomaly score, smoke diffusion anomaly index, image anomaly matching degree, and humidity fluctuation deviation degree.

[0082] The preset sub-indicator priority list is obtained from the dynamic anomaly detection model, which clearly shows that the temperature anomaly score has the highest impact on the occurrence of fire, followed by the smoke diffusion anomaly index, and then the image anomaly matching degree, and the humidity fluctuation deviation has the lowest impact. For example, in multiple simulated fire scenes and actual fire case analysis, it was found that abnormal temperature increases are often an important precursor to fire, so the temperature anomaly score is ranked first; smoke diffusion can directly reflect the spread trend of the fire, so the smoke diffusion anomaly index is ranked second; image texture changes can show the relationship between vegetation state changes and fires to a certain extent, and are ranked third; although humidity has an impact on fire, its impact is slightly lower than the first three, so the humidity fluctuation deviation is ranked fourth.

[0083] Step S1442: extract the current value of each sub-indicator in the abnormal fluctuation indicator set, and calculate the deviation percentage of each sub-indicator relative to the historical benchmark value.

[0084] Assume that the temperature anomaly score in the abnormal fluctuation index set exceeds 0.071, the smoke diffusion anomaly index exceeds none (recorded as 0), the image anomaly matching exceeds 0.033, and the humidity fluctuation deviation exceeds 0.054. At the same time, assume that the historical benchmark values ​​of the temperature anomaly score are 0.05, the historical benchmark values ​​of the smoke diffusion anomaly index are 0.02, the historical benchmark values ​​of the image anomaly matching are 0.02, and the historical benchmark values ​​of the humidity fluctuation deviation are 0.03.

[0085] Calculate the percentage deviation of the temperature anomaly score from the historical baseline value: (0.071-0.05) / 0.05×100%=42% Smoke diffusion abnormality index deviation percentage: (0-0.02) / 0.02×100%=-100% (because the current value is 0, which is less than the reference value) Image abnormality matching deviation percentage: (0.033-0.02) / 0.02×100%=65% Humidity fluctuation deviation percentage: (0.054-0.03) / 0.03×100%=80% Step S1443: According to the sub-indicator priority list, a threshold adjustment weight is assigned to the temperature anomaly score with the highest priority, and the adjustment weights of the remaining sub-indicators are decreased in order of priority.

[0086] According to the sub-indicator priority list, a higher threshold adjustment weight is assigned to the temperature anomaly score, which is set to 0.4. Since the smoke diffusion anomaly index has the second highest priority, it is assigned a weight of 0.3; the image anomaly matching weight is set to 0.2; and the humidity fluctuation deviation weight is set to 0.1. This weight allocation reflects the different importance of each sub-indicator to the adjustment of the warning threshold interval. The temperature anomaly score has the greatest impact on the threshold interval adjustment, while the humidity fluctuation deviation has the least impact.

[0087] Step S1444: multiply the offset percentage of each sub-indicator by the corresponding threshold adjustment weight to generate the boundary value adjustment amount of each sub-indicator to the warning threshold interval.

[0088] Boundary value adjustment of temperature anomaly score: 42%×0.4=0.168 Smoke diffusion abnormality index boundary value adjustment amount: -100%×0.3=-0.3 Image abnormality matching boundary value adjustment amount: 65%×0.2=0.13 Humidity fluctuation deviation boundary value adjustment amount: 80%×0.1=0.08 Step S1445: superimpose the boundary value adjustment amount onto the original boundary value of the warning threshold interval according to the sign, wherein the upper limit of the first threshold interval is increased by the adjustment amount, the upper and lower limits of the second threshold interval are synchronously offset, and the lower limit of the third threshold interval is reduced by the adjustment amount.

[0089] Assume that the original first threshold interval is 0-0.4, the second threshold interval is 0.4-0.6, and the third threshold interval is above 0.6.

[0090] The upper limit of the first threshold interval is adjusted to: 0.4+0.168=0.568, and the adjusted first threshold interval becomes 0-0.568.

[0091] The lower limit of the second threshold interval is adjusted to: 0.4-0.3+0.13+0.08=0.31, and the upper limit is adjusted to: 0.6-0.3+0.13+0.08=0.51. The adjusted second threshold interval becomes 0.31-0.51.

[0092] The lower limit of the third threshold interval is adjusted to: 0.6-0.3+0.13+0.08-0.168=0.342, and the adjusted third threshold interval becomes above 0.342.

[0093] Step S1446: perform boundary value range check on the adjusted warning threshold interval to ensure that the intervals do not overlap and are arranged in ascending order of values, and use the checked intervals as input conditions for subsequent warning strategy matching.

[0094] Check the adjusted warning threshold intervals. The first threshold interval is 0-0.568, the second threshold interval is 0.31-0.51, and the third threshold interval is above 0.342. It is found that the lower limit of the second threshold interval 0.31 is less than the upper limit of the first threshold interval 0.568, and there is an overlap problem. The intervals need to be adjusted and the adjustment amount reallocated. For example, take out 0.1 from the adjustment amount of 0.168 for the temperature anomaly score and evenly distribute it to the other three sub-indicator adjustments (0.033 for the smoke diffusion anomaly index, 0.033 for the image anomaly matching degree, and 0.034 for the humidity fluctuation deviation).

[0095] After readjustment, the upper limit of the first threshold interval is: 0.4+0.068=0.468 The lower limit of the second threshold interval is: 0.4-0.267+0.163+0.114=0.41 The upper limit is: 0.6-0.267+0.163+0.114=0.61 The lower limit of the third threshold interval is: 0.6-0.267+0.163+0.114-0.068=0.542 After checking again, the first threshold interval is 0-0.468, the second threshold interval is 0.41-0.61, and the third threshold interval is above 0.542. The intervals do not overlap and are arranged in ascending order. These verified intervals are used as input conditions for subsequent early warning strategy matching, so that the early warning strategy can more accurately trigger the corresponding level of early warning instructions according to the real-time abnormal fluctuation situation.

[0096] Step S150: triggering a fire-fighting resource scheduling operation according to the dynamic fire risk warning instruction, and generating a forest escape path optimization strategy in combination with the abnormal fluctuation indicator set.

[0097] Step S151: parsing the level identifier of the dynamic fire risk warning instruction, and matching the corresponding fire fighting equipment list and personnel deployment plan from the fire fighting resource database.

[0098] Assume that the dynamic fire risk warning instruction received is an intermediate warning. Find the fire equipment list and personnel configuration plan corresponding to the intermediate warning from the fire extinguishing resource database. The fire extinguishing resource database records the resource information required for different warning levels. For the intermediate warning, the list may include: 5 fire extinguishers, 2 fire hydrants, and a team of (3 people) professional firefighters are configured in the key areas Z1, Z2, and Z3 respectively. 3 fire extinguishers, 1 fire hydrant, and a team of (2 people) professional firefighters are configured in the Z4 and Z5 areas respectively. These equipment and personnel configurations are determined based on the analysis of the possible fire scale and response needs under different fire risk levels. For example, the fire in the key areas Z1, Z2, and Z3 is expected to be large, so more fire extinguishing equipment and personnel are configured.

[0099] Step S152: Based on the diffusion angle and speed in the smoke diffusion trajectory characteristics, the fire spread prediction range is calculated, and a fire extinguishing resource deployment coordinate set is generated in combination with geographic information system data.

[0100] Take the diffusion speed of 0.001m / min and the diffusion angle of 0° (due east) in the smoke diffusion trajectory characteristics as an example. Assuming that the current time is 12 o'clock, based on the smoke diffusion data of the past period of time (such as 30 minutes), predict the fire spread range in the next hour (until 13 o'clock). Taking the current smoke source location as the starting point, since the diffusion speed is 0.001m / min, the smoke will spread 60×0.001=0.06m in the due east direction within one hour. Considering that there may be a certain degree of randomness and uncertainty in the spread of the fire, a certain angle (such as 15°) is extended on both sides of the diffusion direction to form a fan-shaped area as the fire spread prediction range.

[0101] Combined with GIS data, the GIS data contains detailed information on the terrain, roads, buildings, etc. in the forest area. Assuming that in the GIS data, the coordinates of the smoke source are determined to be (x0, y0), multiple coordinate points on the boundary of the sector area are calculated based on the fan angle and distance of the fire spread prediction range. For example, in the east direction, the coordinates at 0.06m away from the smoke source are (x0+0.06×cos(0°), y0+0.06×sin(0°)), and the coordinates at 0.06m away from the smoke source in the 15° direction are calculated by trigonometric functions (x0+0.06×cos(15°), y0+0.06×sin(15°)), etc. These coordinate points are combined into a set of fire-fighting resource deployment coordinates, which are the approximate locations where fire-fighting equipment and personnel should be deployed to more effectively control the spread of fire.

[0102] Step S153: extracting the coordinates of the temperature anomaly area and the smoke diffusion direction in the abnormal fluctuation index set, and calling the path planning algorithm to mark the high-risk avoidance area in the digital terrain map.

[0103] Step S1531: extracting a temperature anomaly area coordinate set and a smoke diffusion direction vector from the abnormal fluctuation index set, and converting the temperature anomaly area coordinate set into a geographic coordinate system in a digital topographic map.

[0104] Assume that the temperature anomaly area in the abnormal fluctuation index set is composed of multiple point coordinates, such as {(x1, y1), (x2, y2), (x3, y3)}. The smoke diffusion direction vector is assumed to be (1, 0) indicating the due east direction. The digital topographic map has its own specific geographic coordinate system, and the coordinates of the temperature anomaly area are converted from the current coordinate system to the geographic coordinate system of the digital topographic map. For example, it is known that the conversion relationship between the current coordinate system and the digital topographic map geographic coordinate system is: new x coordinate = original x coordinate × scale + offset x, new y coordinate = original y coordinate × scale + offset y. Assume that the scale is 1:1000, the offset x=100, and the offset y=200. Then (x1, y1) is converted to ((x1×1000+100), (y1×1000+200)), and so on to convert all the coordinate points of the temperature anomaly area.

[0105] Step S1532: Generate a smoke diffusion direction correction vector by superimposing and calculating the smoke diffusion direction vector and the wind direction data in the digital topographic map.

[0106] Assume that the wind direction data in the digital topographic map shows that the wind direction is 30° east by north, which is represented by a vector (cos(30°), sin(30°)). The smoke diffusion direction vector is (1, 0). The two are superimposed and calculated, for example, using the vector addition principle, the smoke diffusion direction correction vector = (1+cos(30°), 0+sin(30°)) = (1+0.866, 0+0.5) = (1.866, 0.5). This correction vector comprehensively considers the smoke diffusion direction and the actual wind direction, and more accurately reflects the diffusion trend of smoke in the actual environment.

[0107] Step S1533: Based on the temperature anomaly area coordinate set, a polygonal area whose temperature exceeds a preset threshold is delineated in the digital topographic map as an initial high-risk area.

[0108] Assuming that the preset temperature threshold is 30℃, query the temperature value of the corresponding position in the continuous temperature field distribution map according to the coordinate set of the temperature anomaly area. For example, the temperature corresponding to the coordinate point (x1, y1) is 32℃, the temperature corresponding to (x2, y2) is 31℃, and the temperature corresponding to (x3, y3) is 33℃. By connecting these coordinate points whose temperatures exceed the threshold, a polygonal area is formed as the initial high-risk area. If the distribution of these points is relatively scattered, it may be necessary to screen and connect the points according to certain rules. For example, the idea of ​​Delaunay triangulation algorithm is adopted to first triangulate all temperature anomaly points, and then select the edges of the triangles containing the points whose temperatures exceed the threshold to construct polygons. In actual operation, for each triangle, check whether the temperatures of its three vertices are all above 30℃. If so, retain the edges of the triangle; if the temperatures of some vertices do not exceed the threshold, select the edges according to the conditions of the adjacent triangles, and finally form a closed polygonal area, which is the initial high-risk area, marking the area with abnormal temperature and may have an important impact on the development of fire.

[0109] Step S1534: input the smoke diffusion direction correction vector into the obstacle generation module in the path planning algorithm, and extend the preset distance along the diffusion direction to generate a fan-shaped diffusion prediction area.

[0110] The smoke diffusion direction correction vector (1.866, 0.5) calculated above is input into the obstacle generation module in the path planning algorithm. The preset distance is set to 100 meters (this distance is set based on the actual scale of the forest area and the experience of the fire impact range). With the smoke source location as the center, extend 100 meters along the direction of the smoke diffusion direction correction vector. Since smoke diffusion has a certain fan-shaped range, take the correction vector as the center line and expand 15° on both sides (also set based on experience, considering the uncertainty of smoke diffusion).

[0111] Calculate the coordinates of the points on the boundary of the sector area. First, calculate the coordinates of the point 100 meters away from the smoke source in the direction of the center line. Assuming that the coordinates of the smoke source are (x0, y0), the coordinates of the point are calculated based on the modulus and direction of the vector as (x0+100×1.866 / √(1.866²+0.5²), y0+100×0.5 / √(1.866²+0.5²)). Then calculate the coordinates of the points 100 meters away from the smoke source in the direction of the boundaries on both sides. For the left boundary, the angle is the center line direction angle minus 15°, and the right boundary angle is the center line direction angle plus 15°. The corresponding coordinates are also calculated based on trigonometric functions. By connecting these boundary points, a sector diffusion prediction area is formed. The sector diffusion prediction area simulates the range of possible smoke diffusion and serves as an area that needs to be avoided during path planning.

[0112] Step S1535: performing spatial superposition operation on the initial high-risk area and the fan-shaped diffusion prediction area, merging overlapping areas and eliminating isolated areas, and generating a comprehensive high-risk avoidance area.

[0113] The initial high-risk area (polygonal area) and the fan-shaped diffusion prediction area are superimposed in the spatial coordinate system of the digital topographic map. For each part of the two areas, check whether they overlap. For each boundary point of the initial high-risk area, it can be determined whether it is within the fan-shaped diffusion prediction area, and vice versa, by point-by-point checking. For the overlapping parts, merge them into one area. For example, if a part of the initial high-risk area overlaps with a part of the fan-shaped diffusion prediction area, merge the two parts into a unified area, using a polygon merging algorithm, such as by recombining the boundaries of the overlapping parts to form a new polygon.

[0114] At the same time, check whether there are isolated areas in the merged area. Isolated areas refer to small areas that are not connected to other main areas. These areas may have little impact on the overall escape path in actual fire scenarios. Determine whether it is an isolated area by analyzing the connectivity of the area. For example, use the breadth-first search algorithm (BFS) or the depth-first search algorithm (DFS). Start from a point in the area and mark all the points connected to it. If there is a set of unmarked points belonging to the area, then the set of points constitutes an isolated area. These isolated areas are eliminated, and finally a comprehensive high-risk avoidance area is generated. The comprehensive high-risk avoidance area comprehensively considers the impact of temperature anomalies and smoke diffusion, and more accurately identifies the dangerous areas that need to be avoided when escaping.

[0115] Step S1536: Mark the comprehensive high-risk avoidance area as a red warning layer in the digital topographic map, and transmit the layer boundary coordinates to the escape path optimization module.

[0116] On the digital topographic map, the generated comprehensive high-risk avoidance area is marked in red to form a red warning layer. Digital topographic maps usually have graphics drawing and layer management functions. By calling the corresponding functions or tools, the polygonal boundaries of the comprehensive high-risk avoidance area are filled or stroked in red, so that it is clearly marked in red on the map for easy and intuitive viewing.

[0117] Then, the boundary coordinates of the red warning layer are extracted, which accurately define the scope of the comprehensive high-risk avoidance area. These boundary coordinates are transmitted to the escape path optimization module, which will avoid the dangerous area when planning the escape path based on these coordinate information. For example, the escape path optimization module may use a path planning algorithm such as the A* algorithm and the Dijkstra algorithm. When generating a path, the area defined by the boundary coordinates of the red warning layer is regarded as an inaccessible area, thereby planning a safe escape path.

[0118] Step S154: Generate multiple candidate escape paths based on the high-risk avoidance area and the safe exit location, and perform comprehensive scoring and sorting based on path length, slope and fire coverage probability.

[0119] The location of the known safety exit is clearly marked with coordinates in the digital terrain map. Assume that the coordinates of the safety exit are (xs, ys). With the safety exit as the end point, on the digital terrain map, starting from different locations in the forest area, avoiding high-risk avoidance areas, generate multiple candidate escape paths. You can use a path planning algorithm, such as the A* algorithm. When searching for a path, this algorithm will consider the estimated distance from the starting point to the end point (such as the Euclidean distance) and the distance that has been traveled. By continuously expanding nodes, find a path from the starting point to the safety exit and avoid high-risk avoidance areas.

[0120] For each candidate escape path generated, calculate its path length. The path length can be obtained by accumulating the distances between adjacent points on the path. Assuming that the path consists of a point set {(x1, y1), (x2, y2), ..., (xn, yn)}, the path length L = ∑(i=1ton-1)√((xi+1-xi)²+(yi+1-yi)²).

[0121] Calculate the slope of the area through which the path passes. The slope can be calculated using the elevation data of the area in the digital topographic map. Assume that the coordinates of a point on the path are (x, y), and obtain the elevation value h(x, y) of the point by querying the elevation data of the digital topographic map. For two adjacent points (xi, yi) and (xi+1, yi+1), the slope S=|h(xi+1, yi+1)-h(xi, yi)| / √((xi+1-xi)²+(yi+1-yi)²). Calculate the average slope of the entire path as the slope index of the path.

[0122] The fire coverage probability is evaluated and calculated based on the positional relationship between the fire spread prediction range and the candidate escape path. For example, if 20% of the path length is within the fire spread prediction range, the fire coverage probability is 20%.

[0123] Comprehensive scoring is performed based on path length, slope, and fire coverage probability. Assume that the path length weight is 0.4, the slope weight is 0.3, and the fire coverage probability weight is 0.3. For a candidate escape path, the path length is L1, the average slope is S1, and the fire coverage probability is P1. Comprehensive score Score = 0.4×(1-L1 / Lmax)+0.3×(1-S1 / Smax)+0.3×(1-P1), where Lmax is the longest path length among all candidate paths, and Smax is the maximum slope value among all candidate paths. In this way, all candidate escape paths are scored and sorted from high to low according to the score.

[0124] Step S155: The candidate escape path with the highest score is superimposed and displayed with the real-time fire spread prediction range to generate a dynamically updated forest escape path optimization strategy.

[0125] For example, step S1551: rendering and displaying the topological structure and path score details of the candidate escape paths in the digital terrain map interface.

[0126] On the digital topographic map interface, the candidate escape paths after scoring are displayed graphically. For each candidate escape path, its topological structure is drawn, that is, by connecting the points on the path to form lines to indicate the direction of the path. At the same time, the score details of each path are marked on the map interface, such as displaying "Path 1: Score 85 points" next to the path, so that the pros and cons of different paths can be easily viewed and compared. In this way, the operator can intuitively see the situation of different candidate escape paths.

[0127] Step S1552: Convert the real-time fire spread prediction range into a thermal map layer and perform transparency overlay with the topological structure.

[0128] Convert the real-time updated fire spread prediction range data into a heat map layer. The heat map can intuitively display the probability distribution of fire spread. For example, areas with high fire spread probability are represented by red, and areas with low probability are represented by yellow or green. According to the degree of fire probability at different locations within the fire spread prediction range, assign corresponding color values ​​to each location. For example, for areas with a fire coverage probability of 80%-100%, red is assigned; for areas with a fire coverage probability of 50%-80%, orange is assigned, and so on.

[0129] Then, the transparency of the heat map layer is superimposed on the topological structure of the candidate escape paths. In the layer management function of the digital terrain map, the transparency of the heat map layer is set, for example, to 50%, so that the heat map can show the general situation of the spread of the fire without completely covering the topological structure of the candidate escape paths below. In this way, the operator can see the relationship between the escape path and the spread of the fire at the same time, and intuitively judge which paths are safer and which paths may be threatened by the fire.

[0130] Step S1553: When it is detected that the overlap between the predicted fire spread range and a certain escape route exceeds a preset safety threshold, a route replanning instruction is triggered.

[0131] Preset a safety threshold, such as 30%. For each candidate escape path, calculate its overlap with the real-time fire spread prediction range. The overlap can be calculated by comparing the points on the path with the points within the fire spread prediction range. For example, if 35% of the points on the path are within the fire spread prediction range, the overlap is 35%. When it is detected that the overlap between an escape path and the fire spread prediction range exceeds 30%, it means that there is a greater safety risk for the path, and the path replanning instruction is triggered. The path replanning instruction will start the path planning algorithm to re-search for a new escape path that avoids the current fire spread area to ensure the safety of the escape path.

[0132] Step S1554: Update the fire spread prediction range according to the latest sensor data, and re-execute the path scoring sorting and overlay display operations.

[0133] As time goes by, sensors continuously collect new data, such as smoke density sensors detecting new changes in smoke density, and temperature sensors detecting changes in temperature. Based on these latest sensor data, the fire spread prediction range is recalculated. For example, an increase in smoke density may mean that the fire spreads faster, thus expanding the fire spread prediction range; an abnormal increase in temperature may make it easier for the fire to spread to certain areas, and the direction and size of the prediction range will be adjusted accordingly.

[0134] After recalculating the predicted range of fire spread, the path length, slope and fire coverage probability of all candidate escape paths are calculated again, and the scores are ranked according to the previous comprehensive scoring formula. Then, the candidate escape path with the highest score is displayed with the heat map layer converted from the updated fire spread prediction range, and the above operations of rendering the candidate escape path topology structure and path scoring details in the digital terrain map interface, and overlaying the heat map layer with the topology structure transparency are repeated to ensure that the forest escape path optimization strategy can be dynamically updated according to the real-time situation, and provide the safest and most reasonable escape path guidance for people in the forest area.

[0135] Step S1555: associate the path change record in the dynamic update process with the timestamp and store it in the policy execution log database.

[0136] After each triggering of the path replanning instruction and the generation of a new escape path, the detailed information of the path change is recorded. This includes the identification of the original escape path, the identification of the new escape path, the reason for the change (such as the change in the fire spread causing the overlap to exceed the threshold), etc. At the same time, the current timestamp is obtained to record the exact time when the path change occurred. For example, the record is "Original path ID: P1, New path ID: P2, Change reason: The overlap between the fire spread and the path reached 35%, exceeding the threshold of 30%, Time: 2024-10-05 15:30:00".

[0137] Store these path change records and corresponding timestamps in the strategy execution log database. The strategy execution log database can use a relational database, such as MySQL, to create a corresponding table structure to store these records. The table structure can include fields such as path change ID (auto-increment primary key), original path ID, new path ID, change reason, timestamp, etc. By storing these records, the dynamic update process of the forest escape path optimization strategy can be traced and analyzed in detail, which helps to summarize experience and further optimize the escape strategy.

[0138] Figure 2 The schematic diagram shows exemplary hardware and software components of a forest fire warning system 100 based on multi-source sensors that can implement the concept of the present application provided by some embodiments of the present application. For example, the processor 120 can be used in the forest fire warning system 100 based on multi-source sensors and used to perform the functions in the present application.

[0139] The forest fire early warning system 100 based on multi-source sensors can be a general server or a special-purpose server, both of which can be used to implement the forest fire early warning method based on multi-source sensors of the present application. Although only one server is shown in the present application, for convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0140] For example, the forest fire early warning system 100 based on multi-source sensors may include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and storage media 140 in different forms, such as a disk, ROM, or RAM, or any combination thereof. Exemplarily, the forest fire early warning system 100 based on multi-source sensors may also include program instructions stored in ROM, RAM, or other types of non-temporary storage media, or any combination thereof. The method of the present application may be implemented according to these program instructions. The forest fire early warning system 100 based on multi-source sensors also includes an input / output (I / O) interface 150 between the computer and other input / output devices.

[0141] For ease of explanation, only one processor is described in the forest fire warning system 100 based on multi-source sensors. However, it should be noted that the forest fire warning system 100 based on multi-source sensors in the present application may also include multiple processors, so the steps performed by one processor described in the present application may also be performed jointly or individually by multiple processors. For example, if the processor of the forest fire warning system 100 based on multi-source sensors executes step A and step B, it should be understood that step A and step B may also be performed jointly by two different processors or individually in one processor. For example, the first processor executes step A, the second processor executes step B, or the first processor and the second processor execute steps A and B together.

[0142] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer executable instructions are preset. When a processor executes the computer executable instructions, the above-mentioned forest fire warning method based on multi-source sensors is implemented.

[0143] It should be noted that in order to simplify the description of the present invention and thus help understand one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A forest fire early warning method based on multi-source sensors, characterized in that: The method comprises: Acquire a multi-source sensor monitoring data set of a target forest area, wherein the multi-source sensor monitoring data set includes a temperature monitoring data subset, a humidity monitoring data subset, a smoke concentration monitoring data subset, and a visible light image data subset, wherein each data subset is collected and generated by at least two heterogeneous sensor nodes; Performing spatiotemporal alignment processing on the multi-source sensor monitoring data set to generate a spatiotemporal distribution feature set of the target forest area, wherein the spatiotemporal distribution feature set includes temperature gradient distribution features, humidity fluctuation correlation features, smoke diffusion trajectory features, and image texture change features; Calling a dynamic anomaly detection model to perform fire risk index comparison processing on the spatiotemporal distribution feature set to generate a fire risk level quantitative score and an abnormal fluctuation index set; Based on the fire risk level quantitative score and the preset warning threshold interval, a multi-level warning strategy is matched to generate a dynamic fire risk warning instruction for the target forest area; The fire-fighting resource scheduling operation is triggered according to the dynamic fire risk warning instruction, and the forest escape path optimization strategy is generated in combination with the abnormal fluctuation indicator set.

2. The forest fire early warning method based on multi-source sensors according to claim 1 is characterized in that: The performing of spatiotemporal alignment processing on the multi-source sensor monitoring data set to generate a spatiotemporal distribution feature set of the target forest area includes: Extracting the temperature monitoring sequence of each sensor node in the temperature monitoring data subset, performing interpolation fitting processing based on the timestamp and spatial coordinates of the temperature monitoring sequence, and generating a continuous temperature field distribution map of the target forest area; Extracting the humidity monitoring sequence of each sensor node in the humidity monitoring data subset, calculating the humidity change rate of adjacent sensor nodes, and constructing a humidity gradient association matrix according to the spatial distance weight; Perform diffusion direction tracking processing on the subset of smoke concentration monitoring data, and generate smoke diffusion trajectory features based on the smoke concentration change gradient and wind speed monitoring data, wherein the smoke diffusion trajectory features include diffusion speed, diffusion angle and concentration attenuation coefficient; Performing frame-by-frame analysis on the subset of visible light image data, extracting texture feature vectors of each frame of image, and comparing similarity changes of the texture feature vectors based on time series to generate image texture change features; The continuous temperature field distribution map, the humidity gradient association matrix, the smoke diffusion trajectory characteristics and the image texture change characteristics are associated and integrated into the spatiotemporal distribution feature set.

3. The forest fire early warning method based on multi-source sensors according to claim 2 is characterized in that: The calling of the dynamic anomaly detection model to perform fire risk index comparison processing on the spatiotemporal distribution feature set to generate a fire risk level quantitative score and an abnormal fluctuation index set includes: Extracting the highest temperature value, the temperature change slope and the regional temperature difference extreme value from the continuous temperature field distribution diagram of the spatiotemporal distribution feature set, performing dynamic threshold comparison between the highest temperature value, the temperature change slope and the regional temperature difference extreme value and the historical temperature reference interval stored in the dynamic anomaly detection model, and generating a temperature anomaly score; Extracting the humidity change rate of adjacent sensor nodes from the humidity gradient association matrix, determining the humidity fluctuation deviation according to the humidity attenuation threshold predefined in the dynamic anomaly detection model, and weighting the humidity fluctuation deviation in combination with the current ambient humidity value; Extracting diffusion velocity and concentration attenuation coefficient from the smoke diffusion trajectory characteristics, inputting the diffusion velocity and concentration attenuation coefficient into the smoke diffusion history feature library in the dynamic anomaly detection model for trajectory similarity matching, and generating a smoke diffusion anomaly index; Extracting a high-frequency component energy value from the image texture change feature, performing frequency domain matching on the high-frequency component energy value and a reference energy spectrum of a vegetation burning feature library in the dynamic anomaly detection model, and generating an image anomaly matching degree; The temperature anomaly score, the corrected humidity fluctuation deviation, the smoke diffusion anomaly index and the image anomaly matching degree are input into the multi-index fusion layer of the dynamic anomaly detection model, and a linear weighted calculation is performed according to the preset fire risk weight coefficient to generate the quantitative score of the fire risk level, and the parts of the temperature anomaly score, the humidity fluctuation deviation, the smoke diffusion anomaly index and the image anomaly matching degree that exceed the corresponding sub-indicator threshold are integrated into the abnormal fluctuation index set.

4. The forest fire early warning method based on multi-source sensors according to claim 3 is characterized in that: The temperature anomaly score, the corrected humidity fluctuation deviation, the smoke diffusion anomaly index and the image anomaly matching degree are input into the multi-index fusion layer of the dynamic anomaly detection model, and a linear weighted calculation is performed according to a preset fire risk weight coefficient to generate the fire risk level quantitative score, including: The temperature anomaly score is normalized to be mapped to a dimensionless value in the range of 0-1; The difference between the corrected humidity fluctuation deviation and the preset humidity reference value is calculated, and the result is divided by the upper limit of the humidity monitoring range to convert into a dimensionless deviation coefficient; Performing logarithmic transformation on the smoke diffusion anomaly index, compressing the value fluctuation range and normalizing it to a diffusion risk coefficient within the range of 0-1; Calculate the ratio of the image abnormality matching degree to a preset energy spectrum threshold to generate a dimensionless image abnormality coefficient; Extracting a fire risk weight coefficient set stored in the multi-index fusion layer of the dynamic anomaly detection model, wherein the fire risk weight coefficient set includes temperature weight, humidity weight, smoke weight and image weight, and the weight values ​​of the temperature weight, humidity weight, smoke weight and image weight are normalized to ensure that the sum is 1; The dimensionless temperature anomaly score, the dimensionless deviation coefficient, the diffusion risk coefficient and the image anomaly coefficient are respectively multiplied by the corresponding fire risk weight coefficient to generate a weighted temperature risk value, a humidity risk value, a smoke risk value and an image risk value; The weighted temperature risk value, humidity risk value, smoke risk value and image risk value are cumulatively summed to generate the fire risk level quantitative score.

5. The forest fire early warning method based on multi-source sensors according to claim 1 is characterized in that: The method of matching a multi-level warning strategy based on the fire risk level quantitative score with a preset warning threshold interval to generate a dynamic fire risk warning instruction for the target forest area includes: When the fire risk level quantitative score is within the first threshold interval, a primary warning instruction is generated, and the primary warning instruction triggers the sensor node sampling frequency increase operation and the drone cruise path planning; When the fire risk level quantitative score is within the second threshold interval, an intermediate warning instruction is generated, and the intermediate warning instruction triggers the pre-start operation of the fire extinguishing equipment in the key area and the broadcast notification of personnel evacuation; When the fire risk level quantitative score is within the third threshold interval, an advanced warning instruction is generated, and the advanced warning instruction triggers the joint response of fire extinguishing resources in the whole area and the opening operation of the emergency channel; According to the priority ranking of each sub-indicator in the abnormal fluctuation indicator set, the boundary value of the warning threshold interval is dynamically adjusted, and the triggering condition of the dynamic fire risk warning instruction is updated based on the adjusted threshold interval.

6. The forest fire early warning method based on multi-source sensors according to claim 5 is characterized in that: The dynamically adjusting the boundary value of the warning threshold interval according to the priority ranking of each sub-indicator in the abnormal fluctuation indicator set includes: Loading a preset sub-indicator priority list from the dynamic anomaly detection model, wherein the list is arranged in descending order of fire impact, namely, temperature anomaly score, smoke diffusion anomaly index, image anomaly matching degree, and humidity fluctuation deviation degree; Extracting the current value of each sub-indicator in the abnormal fluctuation indicator set, and calculating the deviation percentage of each sub-indicator relative to the historical benchmark value; According to the sub-indicator priority list, a threshold adjustment weight is assigned to the temperature anomaly score with the highest priority, and the adjustment weights of the remaining sub-indicators are decreased in order of priority; Multiply the offset percentage of each sub-indicator by the corresponding threshold adjustment weight to generate the boundary value adjustment amount of each sub-indicator to the warning threshold interval; The boundary value adjustment amount is superimposed on the original boundary value of the warning threshold interval according to the sign, wherein the upper limit of the first threshold interval is increased by the adjustment amount, the upper and lower limits of the second threshold interval are synchronously offset, and the lower limit of the third threshold interval is reduced by the adjustment amount; The adjusted warning threshold interval is checked for boundary value range to ensure that the intervals do not overlap and are arranged in ascending order of values, and the checked intervals are used as input conditions for subsequent warning strategy matching.

7. The forest fire early warning method based on multi-source sensors according to claim 1 is characterized in that: The triggering of the fire extinguishing resource scheduling operation according to the dynamic fire risk warning instruction and the generation of the forest escape path optimization strategy in combination with the abnormal fluctuation index set include: Parse the level identification of the dynamic fire risk warning instruction, and match the corresponding fire fighting equipment list and personnel deployment plan from the fire fighting resource database; Based on the diffusion angle and speed in the smoke diffusion trajectory characteristics, the fire spread prediction range is calculated, and a fire extinguishing resource deployment coordinate set is generated in combination with geographic information system data; Extract the coordinates of the temperature abnormality area and the smoke diffusion direction in the abnormal fluctuation index set, and call the path planning algorithm to mark the high-risk avoidance area in the digital terrain map; Generate multiple candidate escape paths according to the high-risk avoidance area and the safe exit location, and rank them by comprehensive scores based on path length, slope and fire coverage probability; The candidate escape routes with the highest scores are superimposed with the real-time fire spread prediction range to generate a dynamically updated forest escape route optimization strategy.

8. The forest fire early warning method based on multi-source sensors according to claim 7 is characterized in that: The extracting the coordinates of the temperature abnormal area and the smoke diffusion direction in the abnormal fluctuation index set, and calling the path planning algorithm to mark the high-risk avoidance area in the digital terrain map, includes: Extracting a temperature anomaly region coordinate set and a smoke diffusion direction vector from the abnormal fluctuation index set, and converting the temperature anomaly region coordinate set into a geographic coordinate system in a digital topographic map; Generate a smoke diffusion direction correction vector based on the smoke diffusion direction vector and the wind direction data in the digital topographic map; Based on the coordinate set of the temperature anomaly area, a polygonal area whose temperature exceeds a preset threshold is delineated in the digital topographic map as an initial high-risk area; Inputting the smoke diffusion direction correction vector into the obstacle generation module in the path planning algorithm, and extending the preset distance along the diffusion direction to generate a fan-shaped diffusion prediction area; Perform spatial superposition operation on the initial high-risk area and the fan-shaped diffusion prediction area, merge overlapping areas and remove isolated areas to generate a comprehensive high-risk avoidance area; The comprehensive high-risk avoidance area is marked as a red warning layer in the digital topographic map, and the layer boundary coordinates are transmitted to the escape path optimization module.

9. The forest fire early warning method based on multi-source sensors according to claim 2 is characterized in that: The frame-by-frame analysis processing is performed on the visible light image data subset to extract the texture feature vector of each frame image, and the similarity change of the texture feature vector is compared based on the time series to generate the image texture change feature, including: Dividing the visible light image data subset into a plurality of image frame sequences at preset time intervals, and performing grayscale and noise reduction processing on each image frame; The edge detection algorithm is used to extract the contour features in the processed image frame, and the gray level co-occurrence matrix of the pixels in the contour area is calculated; Extracting contrast, energy value and homogeneity index from the gray-level co-occurrence matrix to construct a texture feature vector of a single-frame image; Arrange the texture feature vectors in chronological order, calculate the cosine similarity between adjacent frames, and count the locations of mutation points where the similarity decrease rate exceeds a preset threshold; The mutation point position is associated with the corresponding image frame coordinates to generate image texture change features reflecting the mutation of vegetation status.

10. A forest fire early warning system based on multi-source sensors, characterized in that: It includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the forest fire warning method based on multi-source sensors as described in any one of claims 1 to 9.

Images