Power transmission line adjacent area potential forest fire early warning grading method

Abstract

A power transmission line adjacent area potential forest fire early warning grading method comprises the steps that satellite image data of a selected target area are collected to serve as an initial data set; performing initialization processing on the initial data set; based on NDVI, NDWI, EVI, BSI, buildings and IBI, identifying and selecting five representative ground features of a water body, a forest, a city, a farmland and a grassland in the target area; rasterizing the selected target area, and calculating the forest coverage rate in each grid according to the area ratio of the five typical ground feature areas obtained through classification; a fire danger index model of the selected target area is established, and the fire danger occurrence probability is calculated according to the forest coverage rate, the soil humidity, the precipitation amount and the air temperature data of each grid; and according to the fire danger occurrence probability of each grid, dividing the fire danger grade of the grid, and obtaining a grading visualization result. According to the method, high-resolution ground coverage information can be provided, accurate identification of ground feature types is facilitated, and which areas have high forest fire risks can be accurately judged.

Description

Technical Field

[0001] The present invention relates to the technical field of prevention and prediction of wildfires on transmission lines, and particularly to a method for warning classification of potential wildfires in adjacent areas of transmission lines. Background Art

[0002] Forest fires are caused by weather conditions such as drought, high temperature, dryness, and improper human activities. There may be flammable vegetation such as thatch, pine trees, fir trees, shrubs, and arbor in the overhead transmission line corridors and their surrounding areas. Such vegetation is likely to cause fires under the above circumstances, posing potential risks to the safe operation of transmission lines. If the minimum clearance distance between various types of vegetation and the conductors within the transmission line is less than the specified value, the burning vegetation may cause the conductors to overheat and break down, and there may also be a hidden danger of line tripping. Therefore, it is necessary to use modern technical means to strengthen the identification and safety management of vegetation in adjacent areas of transmission lines, establish and improve a wildfire warning and monitoring system for adjacent areas of transmission lines, and timely grasp and predict the occurrence of wildfires, so as to promote the stable and reliable transmission of electricity.

[0003] In view of the serious problem of the current prevention and control of forest fires on transmission lines, it is urgent to carry out research on the method for predicting potential wildfires in adjacent areas of transmission corridors, and propose a method for warning classification of potential wildfires in adjacent areas of overhead transmission lines. Summary of the Invention

[0004] To solve the problem of calculating the probability of potential wildfires in the area where the transmission line is located, the present invention provides a method for warning classification of potential wildfires in adjacent areas of transmission lines, which can provide high-resolution ground cover information. By combining meteorological data such as temperature, humidity, wind speed, and topographic and geomorphic features and other factors, this method conducts comprehensive risk assessment, can divide warning levels according to different risk levels, and take corresponding preventive measures.

[0005] The technical solution adopted by the present invention is as follows:

[0006] A method for warning classification of potential wildfires in adjacent areas of transmission lines, comprising:

[0007] Step 1: Collect hyperspectral satellite image data of the selected target area as the initial data set;

[0008] Step 2: Perform initialization processing on the initial data set to remove cloud occlusion;

[0009] Step 3: Identify five representative land covers, namely water bodies, forests, cities, farmlands, and grasslands, in the selected target area based on the Normalized Difference Vegetation Index (NDVI), the Normalized Difference Water Index (NDWI), the Enhanced Vegetation Index (EVI) for vegetation coverage, the Bare Soil Index (BSI), and the Index of Built-up and Impervious Surfaces (IBI).

[0010] Step 4: rasterize the selected target area, and calculate the forest coverage rate in each grid according to the area proportion of the five typical land cover types obtained by classification in Step 3;

[0011] Step 5: establish a fire risk index model for the selected target area, calculate the fire occurrence probability according to the forest coverage rate, soil moisture, precipitation, and temperature data of each grid; divide the fire risk level of each grid according to the fire occurrence probability of each grid, and obtain the classified visualization result.

[0012] The said Step 2 includes the following steps:

[0013] S201: Read the original remote sensing hyperspectral satellite image in the initial dataset to obtain the original RGB color image matrix I RGB ; S202: Convert the original RGB color image matrix I RGB into the HSV image matrix I hsv based on hue, saturation, and value; S203: Set the hue, saturation, and value threshold matrices of the HSV space image to h threshold , s threshold , v threshold respectively, where h threshold (1), s threshold (1), v threshold (1) are the lower limits of the hue, saturation, and value thresholds respectively, and h threshold (2), s threshold (2), v threshold (2) are the upper limits of the hue, saturation, and value thresholds respectively;

[0014] S204: Create threshold masks h mask , s mask , v mask respectively as follows:

[0015] h mask =(I hsv (:,:,1)> = h threshold (1))&(I hsv (:,:,1)< = h threshold (2));

[0016] s mask =(I hsv (:,:,2)> = s threshold (1))&(I hsv (:,:,2)< = s threshold (2));

[0017] v mask =(I hsv (:,:,3)> = v threshold (1))&(Ihsv (:,:,3) <= v threshold (2));

[0018] Among them, I hsv is a three-dimensional matrix representing an image that has been converted from the original RGB color image to the HSV (Hue, Saturation, Value) color space. I hsv (:,:,1) represents the hue, which is the basic attribute of color, used to distinguish different types of colors, usually expressed in degrees, ranging from 0 to 360 degrees, and often normalized to an integer value between 0 and 1 or 0 and 255 in image processing; I hsv (:,:,2) represents the saturation, that is, the purity or intensity of the color, and the value range is also 0 to 1 or 0 to 255; I hsv (:,:,3) represents the value / brightness, that is, the brightness of the color, and the value range is also 0 to 1 or 0 to 255.

[0019] S205: Combine all threshold masks to obtain a mask matrix I mask :

[0020] I mask = h mask & s mask & v mask

[0021] S206: Use the mask matrix I mask to remove the cloud components in the HSV image matrix I hsv .

[0022] The said step 3 includes the following steps:

[0023] S301: Stretch the normalized difference vegetation index NDVI, water body feature index NDWI, improved vegetation cover index EVI, bare soil feature index BSI, and building and artificial surface feature index IBI from (-1, 1) to the [0, 255] gray level;

[0024] S302: Define five representative land cover fraction calculation functions:

[0025] C NDVI = (NDVI - 0.2) / (0.8 - 0.2)

[0026] C NDWI = (NDWI - 0.2) / (0.8 - 0.2)

[0027] C EVI = (EVI - 0.2) / (0.8 - 0.2)

[0028] C BSI = (BSI - 0.2) / (0.8 - 0.2)

[0029] C IBI = (IBI - 0.2) / (0.8 - 0.2)

[0030] Among them, C NDVI (Normalized Difference Vegetation Index), the normalized difference vegetation index) represents the standardized value of vegetation coverage, mapping the original NDVI value from -1 to 1 to between 0 and 1. The closer the value is to 1, the denser the vegetation, and it can more intuitively reflect the vegetation coverage rate. C NDWI (Normalized Difference Water Index, the normalized difference water index) represents the standardized value of water body coverage, mapping the original NDWI value from -1 to 1 to between 0 and 1. The closer the value is to 1, the more obvious the water body, and it can more intuitively reflect the distribution of the water body;

[0031] C EVI (Enhanced Vegetation Index, the improved vegetation index) represents the improved standardized value of vegetation coverage, mapping the original EVI value from -1 to 1 to between 0 and 1, and it can more accurately reflect the health status and coverage rate of the vegetation; C BSI (Bare Soil Index, the bare soil characteristic index) represents the standardized value of bare soil coverage, mapping the original BSI value from -1 to 1 to between 0 and 1, and it can more intuitively reflect the distribution of bare soil. The closer the value is to 1, the more obvious the bare soil; C IBI (Index-Based Built-Up Index, the building and artificial surface characteristic index) represents the standardized value of building and artificial surface coverage, mapping the original IBI value from -1 to 1 to between 0 and 1, and it can more intuitively reflect the distribution of buildings and artificial surfaces. The closer the value is to 1, the denser the buildings and artificial surfaces.

[0032] The above-mentioned calculation function of the standardized ground object coverage linearly converts the original index value to between 0 and 1, enabling different types of ground object characteristics to be compared and analyzed on the same scale, which is helpful for subsequent steps such as clustering analysis, classification, and fire risk assessment, and improves the accuracy and reliability of the model.

[0033] S303: Limit the values of the five representative ground object coverage calculation functions to between 0 and 1. If the calculation result of S302 is greater than 1, then assign the corresponding index value to 1;

[0034] S304: Given the number of clustering centers \(c\), the fuzziness index \(m\), the population size \(N\), the maximum number of iterations \(T\), the maximum velocity \(v\), max and the error value \(\varepsilon\);

[0035] S305: Calculate the fitness of a single particle in the population. The expression is:

[0036]

[0037] In Equation (1), \(fitness(i)\) represents the fitness value of a single particle in the population during the \(i\)-th iteration; \(J\) m is the objective function based on the normalized difference vegetation index; \(J\) m (U, V) represents the objective function of fuzzy C-means clustering; \(\alpha\) is the fitness correction factor, and its value range is \((0, 1]\).

[0038] S306: Generate the next generation of particles iteratively. The expressions for their velocity and position are:

[0039]

[0040] x id (t + 1)=x id (t)+βv id (t + 1) (3);

[0041] In Equations (2) and (3), \(w\) is the inertia weight, \(c1\) and \(c2\) are learning factors, \(rand()\) is a random function that varies within the range \((0, 1]\); \(v\) id (t) is the current velocity of the particle, \(v\) id (t + 1) is the updated velocity of the particle, \(x\) id (t) is the current position of the particle, \(x\) id (t + 1) is the updated position of the particle, \(\beta\) is the particle update speed weight, and its value range is \((0, 1]\); \(P\) id (t) is the individual optimal position at time \(t\); \(P\) gd (t) is the population optimal position at time \(t\).

[0042] S307: The design of the inertia weight \(w\) is shown in the following formula:

[0043]

[0044] In the above formula, \(k\) is a constant greater than 0, \(t\) is the number of iterations; \(e\) is the exponential function.

[0045] S308: The design of the learning factors \(c1\) and \(c2\) is shown in Equations (5) and (6):

[0046]

[0047] S309: Update and Iteration: For each particle, compare its fitness value with that of the previous generation. If it is greater than the fitness value of the previous generation, use it as the current individual optimal fitness value; otherwise, retain the current value.

[0048] S310: When the preset maximum number of iterations is reached, terminate the update and iteration and output the cluster centers; otherwise, jump to S307. S311: For the objective function Jm(U, V) of the fuzzy C - means clustering in Step 305, classify the generated cluster centers according to the following criteria:

[0049]

[0050] where: U is the membership matrix, V is the cluster center vector, xi is the i - th data to be clustered; v j represents the j - th cluster center; u ij the membership degree of the i - th data xi to be clustered i belonging to the j - th cluster center, c is the number of cluster centers, n is the number of data xi i and m is the fuzzy parameter, m = 2.

[0051] S312: Define the membership degree u i of the sample x j to the cluster center v ij as:

[0052]

[0053] In Equation (8), pi j is the relative membership weight of the sample x i to the cluster center v j , that is, the result after the reciprocal of the distance d i from the sample x j to the cluster center v ij is exponentiated; pi k is the relative membership weight of the sample x i to all cluster centers v k , that is, the result after the reciprocal of the distance d i from the sample x k to the cluster center v ik is exponentiated; di j is the distance between the sample x i and the cluster center v j , usually using the Euclidean distance di j = |xi - v j |, which reflects the geometric distance between the sample x i and the cluster center v j , and the smaller the distance, the closer the sample xi The higher the similarity with the clustering center v j ; di k is the distance from the sample xi to all the clustering centers v k The distance between them also uses the Euclidean distance di k = |xi - v k |, which is used to calculate the distance between the sample xi and all the clustering centers for normalization processing.

[0054] S313: The constraint condition of formula (7) is constructed as:

[0055]

[0056] where F represents the value of the extended objective function; λ represents the introduced Lagrange multiplier used to impose the constraint condition, and the value range of λ is (0, 1].

[0057] S314: Iteratively update the clustering center v j and the membership degree u ij , and the iteration termination condition is:

[0058]

[0059] where: k is an integer greater than 0, and ε is a preset error value; represents the membership degree of the i-th sample xi belonging to the j-th clustering center v j after the (k + 1)-th iteration, reflecting the membership degree value after one iteration update. represents the membership degree of the i-th sample xi belonging to the j-th clustering center v j in the current k-th iteration, reflecting the membership degree value in the current iteration.

[0060] S315: According to the clustering result, fill in the corresponding areas of water bodies, forests, cities, farmlands, and grasslands in the predetermined target area with "red (RGB value: 255, 0, 0), green (RGB value: 0, 255, 0), blue (RGB value: 0, 0, 255), orange (RGB value: 255, 128, 0), black (RGB value: 0, 0, 0)" respectively to form a visualization result of the ground object distribution in the predetermined target area.

[0061] In step 5, obtaining the fire risk occurrence probability of each 1 km 2 grid includes the following steps:

[0062] S501: Import data such as soil moisture, precipitation, and temperature corresponding to the recognition degree of 1 km in the National Tibetan Plateau Scientific Data Center for the corresponding time period;

[0063] S502: Normalize all variables so that their value ranges are between [0, 1]:

[0064]

[0065] Among them: X represents the original data of forest coverage rate F, soil humidity S, precipitation P, and temperature T, X min and X max are the minimum and maximum values of this variable respectively. For some variables, such as soil humidity S, a lower value means a higher fire risk, and the inverse-normalized value is taken:

[0066]

[0067] S503: Set the corresponding weights ω F 、ω S 、ω P 、ω T according to the importance of forest coverage rate F, soil humidity S, precipitation P, and temperature T data for fire risk, and the weight indices satisfy:

[0068] ω F +ω S +ω P +ω T =1(22);

[0069] S504: Use the method of linear combination to multiply and sum each normalized variable and its corresponding weight to obtain the fire risk index η:

[0070] η=ω F F norm +ω S S norm +ω P P norm +ω T T norm (23);

[0071] S505: Judge the fire risk level according to the value of the fire risk occurrence probability η of this grid:

[0072] η<ε (24).

[0073] Among them: ε is the preset probability criterion threshold.

[0074] S506: According to the fire risk occurrence probability result of step S505, divide each 1km 2 grid into 5 fire risk levels according to Table 1 below:

[0075] Table 1 Fire Risk Level Table

[0076] Fire risk level Flammability degree Hazard level Fire risk occurrence probability threshold ε Level 1 Cannot None ＜1.41％ Level 2 Difficultly Mild 1.41％～2.82％ Level 3 Can Moderate 2.82％～5.65％ Level 4 Easy High 5.65％～11.31％ Level 5 Extremely easy Extremely high >11.31％

[0077] S507: According to the obtained risk levels of "none, mild, moderate, high, extreme" for each grid in S506, fill them into "bright green (#67D43A, RGB value: 103, 212, 58) -> light sky blue (#98B5D1, RGB value: 152, 181, 209) -> blue-violet (#4D5BB8, RGB value: 77, 91, 184) -> fire red (#D02B2A, RGB value: 208, 43, 42) -> light magenta (#E688C7, RGB value: 230, 136, 199)" to form the rasterized fire risk level prediction visualization result of the selected target area.

[0078] The technical effects of a method for early warning and grading of potential wildfires in the adjacent area of a transmission line according to the present invention are as follows:

[0079] 1) The present invention solves the problem of calculating the probability of potential wildfires in the area where the transmission line is located, can provide high-resolution ground coverage information, helps to accurately identify land cover types, such as vegetation coverage rate, land use type, etc. Through the classification and analysis of different land cover types, it can accurately judge which areas have a higher wildfire risk.

[0080] 2) The method of using satellite high-resolution remote sensing images in the present invention is not restricted by geographical conditions, can achieve continuous monitoring of remote areas. By using regularly updated satellite data, it can timely detect environmental changes and potential wildfire signs, improve the response speed of the early warning system; by combining factors such as meteorological data (such as temperature, humidity, wind speed, etc.) and topographical features, and conducting comprehensive risk assessment, it can divide the early warning levels according to different risk levels and take corresponding preventive measures.

[0081] 3) The data and analysis results provided by the present invention can be directly used to formulate emergency response plans and resource scheduling schemes, which helps power companies and other relevant departments to quickly respond to wildfire events and reduce losses.

[0082] 4) The improved particle swarm fuzzy clustering algorithm adopted by the present invention solves the problem of difficult selection of initialization parameters. The improved particle swarm fuzzy clustering algorithm can jump out of the local optimal solution and obtain the global optimal solution by dynamically and randomly changing the learning factors c1 and c2. The design of the non-linear inertia weight w can balance global and local searches and avoid premature convergence. The improved particle swarm fuzzy clustering algorithm improves the accuracy of transformer fault diagnosis results. Description of the Drawings

[0083] The present invention will be further described below in conjunction with the drawings and examples;

[0084] Figure 1 It is a flow chart of a method for early warning and grading of potential wildfires in the adjacent area of a transmission line according to the present invention.

[0085] Figure 2 This is the satellite remote sensing image of the target area in the embodiment of the present invention.

[0086] Figure 3 This is the comparison chart before and after the processing of the present invention.

[0087] Figure 4 This is the prediction grading result chart of the present invention.

[0088] Figure 5 This is the selected target area chart in the embodiment of the present invention. Detailed implementation manners

[0089] The present invention will be further described below in conjunction with the accompanying drawings and specific implementation manners.

[0090] As Figure 1 shown in the process flow of a potential wildfire warning grading method for the adjacent area of an overhead transmission line, the specific steps of the present invention are as follows:

[0091] Step S1: Collect hyperspectral satellite image data of the selected target area:

[0092] For example, high-resolution images of Xiaogan area in Hubei Province obtained by the "Natural Resources Satellite Remote Sensing Cloud Service Platform using the China-Brazil Earth Resources Satellite 04A (CB04A)" are used as the initial data set.

[0093] Step S2: Initialize the collected data set:

[0094] In the said step S2, the initialization processing of the data set includes the following sub-steps:

[0095] S201: Read the original remote sensing hyperspectral image in the initial data set to obtain the original RGB color image matrix I RGB ;

[0096] S202: Convert the original RGB color image matrix I RGB to the HSV color space matrix I based on hue, saturation, and brightness hsv ;

[0097] S203: Set the hue, saturation, and brightness threshold matrices of the HSV space image as h threshold , s threshold , v threshold , where h threshold (1), s threshold (1), v threshold (1) are the lower limits of the hue, saturation, and brightness thresholds respectively, and h threshold (2), s threshold (2), v threshold(2) are the upper limits of hue, saturation, and brightness thresholds respectively;

[0098] S204: Create threshold masks h mask , s mask , v mask are respectively:

[0099] h mask = (I hsv (:,:,1) >= h threshold (1)) & (I hsv (:,:,1) <= h threshold (2));

[0100] s mask = (I hsv (:,:,2) >= s threshold (1)) & (I hsv (:,:,2) <= s threshold (2));

[0101] v mask = (I hsv (:,:,3) >= v threshold (1)) & (I hsv (:,:,3) <= v threshold (2));

[0102] S205: Combine all threshold masks to obtain the mask matrix I mask :

[0103] I mask = h mask & s mask & v mask

[0104] S206: Use the mask matrix I mask to remove the cloud components in the HSV image matrix I hsv . As Figure 2 shown in the comparison diagram before and after processing, the left side is before processing and the right side is after processing.

[0105] Step S3: Based on the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Enhanced Vegetation Index (EVI), Bare Soil Index (BSI), and Index-Based Built-Up Index (IBI), identify five representative land cover types, namely water bodies, forests, cities, farmlands, and grasslands, within the selected target area. The following sub-steps are included:

[0106] S301: Stretch the Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), Enhanced Vegetation Index (EVI), Bare Soil Index (BSI), and Index-Based Built-Up Index (IBI) from (-1, 1) to the [0, 255] gray level.

[0107] S302: Define five representative land cover fraction calculation functions:

[0108] C NDVI = (NDVI - 0.2) / (0.8 - 0.2)

[0109] C NDWI = (NDWI - 0.2) / (0.8 - 0.2)

[0110] C EVI = (EVI - 0.2) / (0.8 - 0.2)

[0111] C BSI = (BSI - 0.2) / (0.8 - 0.2)

[0112] C IBI = (IBI - 0.2) / (0.8 - 0.2)

[0113] S303: Limit the values of the five representative land cover type calculation functions between 0 and 1. If the calculation result in step S302 is greater than 1, assign the corresponding index value to 1.

[0114] S304: Given the number of clustering centers c, the fuzzy exponent m, the population size N, the maximum number of iterations T, the maximum speed v max and the error value ε;

[0115] S305: Calculate the fitness of a single particle in the population, and the expression is:

[0116]

[0117] In formula (1), J m is the objective function based on the normalized difference vegetation index, and is the α fitness correction factor, with a value range of (0, 1];

[0118] S306: Iteratively generate the next generation of particles, and the expressions for their speed and position are:

[0119]

[0120] x id (t + 1) = x id (t) + βv id (t + 1) (3);

[0121] In formulas (2) and (3), w is the inertia weight, c1 and c2 are the learning factors, rand() is a random function that varies within the range (0, 1]; v id (t) is the current speed of the particle, v id (t + 1) is the updated speed of the particle, x id (t) is the current position of the particle, x id (t + 1) is the updated position of the particle, P id is the individual optimal position, P gd is the population optimal position, β is the particle update speed weight, with a value range of (0, 1];

[0122] S307: The design of the inertia weight w is shown in the following formula:

[0123]

[0124] In the above formula, k is a constant greater than 0, and t is the number of iterations;

[0125] S308: The design of the learning factors c1 and c2 is shown in formulas (5) and (6):

[0126]

[0127] S309: Update the iteration: For each particle, compare its fitness value with the fitness value of the previous generation. If it is greater than the fitness value of the previous generation, then use it as the current individual optimal fitness value; otherwise, retain the current value.

[0128] S310: When the preset maximum number of iterations is reached, terminate the update iteration and output the cluster centers; otherwise, jump to step S307;

[0129] S311: Classify the generated cluster centers according to the following criteria:

[0130]

[0131] where U is the membership matrix, V is the cluster center vector, and x i is the i-th data to be clustered; v j represents the j-th cluster center; u ij the membership degree of the i-th data x to be clustered i belonging to the j-th cluster center, c is the number of cluster centers, n is the number of data x i , m is the fuzzy parameter, m = 2;

[0132] S312: Define the membership degree u i of the sample x j to the cluster center v ij as:

[0133]

[0134] S313: The constraint condition of formula (7) is constructed as:

[0135]

[0136] where the value range of λ is (0, 1].

[0137] S314: Iteratively update the cluster center v j and the membership degree u ij , and the iteration termination condition is:

[0138]

[0139] where: k is an integer greater than 0, and ε is the preset error value;

[0140] S315: According to the clustering results, fill in the corresponding areas of water bodies, forests, cities, farmlands, and grasslands in the predetermined target area with "red (RGB value: 255, 0, 0), green (RGB value: 0, 255, 0), blue (RGB value: 0, 0, 255), orange (RGB value: 255, 128, 0), black (RGB value: 0, 0, 0)" to form a visualization result of the ground object distribution in the predetermined target area.

[0141] Step S4: Divide the selected target area into 1 km 2Perform rasterization in units, including:

[0142] S401: Determine the origin coordinates (X0, Y0): Select the lower left corner of the target image as the origin.

[0143] S402: Set the raster size: total length 1000 meters, total height 1000 meters.

[0144] S403: Calculate the number of pixel rows and columns:

[0145]

[0146] Among them, represents the ceiling function, ensuring that even if the last raster is not fully filled, it will be included.

[0147] S404: Generate the center points of the cells:

[0148] The center point of each cell can be calculated by the following formula:

[0149] X center = X0 + i × total raster length + total raster length / 2 (13);

[0150] Y center = Y0 + j × total raster height + Cell Height / 2 (14);

[0151] Among them, i is the column index (starting from 0), and j is the row index (starting from 0).

[0152] S405: Define the boundaries of each raster: The four vertices of each raster can be determined by adjusting the position of the center point. For example, for the cell in the i-th column and j-th row, its lower left vertex is:

[0153] X left_bottom = X0 + i × total raster length (15);

[0154] Y left_bottom = Y0 + j × total raster height (16);

[0155] The upper right vertex is:

[0156] X right_top = X left_bottom + total raster length (17);

[0157] Y right_top = Y left_bottom + total raster height (18);

[0158] S406: Crop to the target area according to the calculated vertices.

[0159] Calculate the forest coverage rate in each grid according to the proportion of the areas of the five typical land cover types classified in step S2;

[0160] The calculation formula for the forest coverage rate F:

[0161]

[0162] According to the operation in step S4, where the area of each grid is 1 km 2 .

[0163] Step S5: Establish a fire risk index model for the selected target area, and calculate the fire risk probability according to the forest coverage rate F, soil moisture S, precipitation P, and air temperature T data of each grid; specifically as follows:

[0164] According to the forest coverage rate F, soil moisture S, precipitation P, and air temperature T data of each 1 km 2 grid, calculate the fire risk probability; including:

[0165] S501: Import the soil moisture, precipitation, air temperature and other data corresponding to the 1 km recognition rate of the National Tibetan Plateau Data Center for the corresponding time period;

[0166] S502: Normalize all variables so that their value ranges are between [0,1]:

[0167]

[0168] where X represents the original data of the forest coverage rate F, soil moisture S, precipitation P, and air temperature T, X min and X max are the minimum and maximum values of this variable respectively. For some variables, such as soil moisture S, a lower value means a higher fire risk, and take the anti-normalized value:

[0169]

[0170] S503: Set the corresponding weights ω F , ω S , ω P , ω T according to the importance of the forest coverage rate F, soil moisture S, precipitation P, and air temperature T data to the fire risk, and the weight index satisfies:

[0171] ω F +ω S +ω P +ω T =1(22);

[0172] S504: Using the method of linear combination, multiply each normalized variable by its corresponding weight and sum them up to obtain the fire risk index η:

[0173] η = ω F F norm + ω S S norm + ω P P norm + ω T T norm (23);

[0174] S505: According to the value of the fire risk occurrence probability η of the grid, judge the fire risk level:

[0175] η < ε (24).

[0176] Where: ε is the preset probability criterion threshold.

[0177] S506: According to the fire risk occurrence probability result of step S5, divide each 1km 2 grid into 5 fire risk levels according to Table 1 below:

[0178] Table 1 Fire Risk Level Table

[0179] Fire risk level Flammability degree Hazard level Fire risk occurrence probability threshold ε Level 1 Cannot None ＜1.41％ Level 2 Difficultly Mild 1.41％～2.82％ Level 3 Can Moderate 2.82％～5.65％ Level 4 Easy High 5.65％～11.31％ Level 5 Extremely easy Extremely high >11.31％

[0180] Step S6: Obtain the visualized result of fire risk classification according to the fire risk occurrence probability of each 1km 2 grid.

[0181] In the said step S6, obtaining the visualized result of the classification of each 1km 2 grid includes the following steps:

[0182] S601: According to the "none, mild, moderate, high, extreme" risk levels of each grid obtained in step S5, fill in the corresponding "bright green (#67D43A, RGB value: 103, 212, 58) -> light sky blue (#98B5D1, RGB value: 152, 181, 209) -> blue-violet (#4D5BB8, RGB value: 77, 91, 184) -> fire red (#D02B2A, RGB value: 208, 43, 42) -> light magenta (#E688C7, RGB value: 230, 136, 199)", and form the visualized result of the prediction of the fire risk level of the rasterized selected target area.

[0183] Figure 2The embodiment of the present invention uses the high-resolution image of Xiaogan area in Hubei Province obtained by the China-Brazil Earth Resources Satellite 04A (CB04A) through the Natural Resources Satellite Remote Sensing Cloud Service Platform. It can be seen that there is 5% cloud cover in the hyperspectral satellite image of the selected target area, and the ground object features are not obvious;

[0184] Figure 3 This is the comparison chart before and after the processing of the present invention. It can be seen that the ground object features are more obvious after processing, which is beneficial for subsequent analysis and processing

[0185] Figure 4 This is the prediction classification result chart of the present invention. It can be intuitively seen from the image that there are different fire risk levels from level 1 to level 5 in the target area. Among them, the area where water bodies exist obviously does not have the conditions for a fire, so it is marked as level 1, which corresponds exactly to the satellite image.

[0186] Figure 5 This is the selected target area chart of the embodiment of the present invention, used for comparison with Figure 4 It can be seen that typical ground object areas such as water bodies and forests are compared with the markings in Figure 4 the label.

Claims

1. A potential wildfire warning classification method for the area adjacent to a transmission line, characterized in that include: Step 1: Collect satellite image data of the selected target area as the initial data set; Step 2: Initialize the initial data set; Step 3: Based on the Normalized Difference Vegetation Index (NDVI), the Water Body Characteristic Index (NDWI), the Improved Vegetation Cover Index (EVI), the Bare Soil Characteristic Index (BSI), and the Building and Artificial Surface Characteristic Index (IBI), five representative landforms in the selected target area, namely, water bodies, forests, cities, farmlands, and grasslands, are identified; Step 4: Grid the selected target area, and calculate the forest coverage rate in each grid according to the area proportions of the five typical landforms classified in step 3; Step 5: Establish a fire risk index model for the selected target area, and calculate the probability of fire risk occurrence based on the forest coverage, soil moisture, precipitation, and temperature data of each grid; divide the fire risk level of each grid according to the probability of fire risk occurrence of each grid, and obtain graded visualization results.

2. According to claim 1, a potential wildfire early warning classification method for the area adjacent to a power transmission line is characterized by: The step 2 comprises the following steps: S201: Read the original remote sensing hyperspectral satellite image in the initial data set to obtain the original RGB color image matrix I RGB ; S202: The original RGB color image matrix I RGB Convert to HSV image matrix based on hue, saturation, and brightness I hsv ; S203: Set the HSV space image hue, saturation, and brightness threshold matrices to h respectively. threshold 、s threshold 、v threshold , where h threshold (1)s threshold (1) v threshold (1) are the lower limits of hue, saturation and brightness thresholds, respectively, h threshold (2) s threshold (2) v threshold (2) are the upper thresholds of hue, saturation, and brightness respectively; S204: Create a threshold mask h mask 、s mask 、v mask They are: h mask =(I hsv (:,:,1)>=h threshold (1))&(I hsv (:,:,1)<=h threshold (2)); s mask =(I hsv (:,:,2)>=s threshold (1))&(I hsv (:,:,2)<=s threshold (2)); v mask =(I hsv (:,:,3)>=v threshold (1))&(I hsv (:,:,3)<=v threshold (2)); Where: I hsv (:,:,1) represents hue, which is the basic attribute of color. It is used to distinguish different types of colors. It is expressed as an angle ranging from 0 to 360 degrees. In image processing, it is often standardized to an integer value between 0 and 1 or between 0 and 255. hsv (:,:,2) represents saturation, that is, the purity or intensity of the color, and the value range is also 0 to 1 or 0 to 255; I hsv (:,:,3) represents brightness / brightness, that is, the brightness of the color, and the value range is also 0 to 1 or 0 to 255; S205: Merge all threshold masks to obtain mask matrix I mask : I mask =h mask &s mask &v mask S206: Use mask matrix I mask Remove HSV image matrix I hsv The cloud composition.

3. According to claim 1, a potential wildfire early warning classification method for the area adjacent to a power transmission line is characterized by: The step 3 comprises the following steps: S301: stretching the normalized difference vegetation index NDVI, the water body characteristic index NDWI, the improved vegetation cover index EVI, the bare soil characteristic index BSI, and the building and artificial surface characteristic index IBI from (-1, 1) to [0, 255] grayscale levels; S302: Define five representative ground feature coverage calculation functions: C NDVI =(NDVI-0.2) / (0.8-0.2) C NDWI =(NDWI-0.2) / (0.8-0.2) C EVI =(EVI-0.2) / (0.8-0.2) C BSI (BSI-0.2) / (0.8-0.2) C IBI =(IBI-0.2) / (0.8-0.2) Among them: Normalized difference water index C NDWI Represents the standardized value of water body coverage, mapping the original NDWI value from -1 to 1 to between 0 and 1. The closer the value is to 1, the more obvious the water body is, so as to more intuitively reflect the distribution of water bodies; Improved vegetation index C EVI An improved standardized value representing vegetation coverage, mapping the original EVI value from -1 to 1 to between 0 and 1, which can more accurately reflect the health status and coverage of vegetation; Bare soil characteristic index C BSI Indicates the standardized value of bare soil coverage. Mapping the original BSI value from -1 to 1 to between 0 and 1 can more intuitively reflect the distribution of bare soil. The closer the value is to 1, the more obvious the bare soil is. Building and artificial surface characteristics index C IBI Represents the standardized value of building and artificial surface coverage. Mapping the original IBI value from -1 to 1 to between 0 and 1 can more intuitively reflect the distribution of buildings and artificial surfaces. The closer the value is to 1, the denser the buildings and artificial surfaces are. S303: Limit the values ​​of the five representative land cover calculation functions to between 0 and 1. If the calculation result of S302 is greater than 1, assign a value of 1 to the corresponding indicator.

4. A method for grading potential wildfire warnings in areas adjacent to power transmission lines according to claim 3, characterized in that: Also includes: S304: Given the number of cluster centers c, fuzzy index m, population size N, maximum number of iterations T, and maximum speed v max and error value ε; S305: Calculate the fitness of a single particle in the population. The expression is: In formula (1), fitness(i) represents the fitness value of a single particle in the population in the i-th iteration process; J m (U, V) represents the objective function of fuzzy C-means clustering; α is the fitness correction factor; S306: Iterate to generate the next generation of particles, whose speed and position are expressed as: x id (t+1)=x id (t)+βv id (t+1) (3); In formula (2) and formula (3), w is the inertia weight, c1 and c2 are learning factors, rand() is a random function that varies in the range (0,1]; v id (t) is the current velocity of the particle, v id (t+1) is the updated velocity of the particle, x id (t) is the current position of the particle, x id (t+1) is the updated position of the particle, β is the particle update speed weight; P(t) is the optimal position of the individual at time t; P ( (t) is the optimal position of the population at time t; S307: The inertia weight w is designed as follows: In the above formula, k is a constant greater than 0, t is the number of iterations; e is an exponential function; S308: The design of learning factors c1 and c2 is shown in equation (5) and equation (6): S309: Update iteration: For each particle, compare its fitness value with the fitness value of the previous generation. If it is greater than the fitness value of the previous generation, it is used as the optimal fitness value of the individual at this time, otherwise the current value is retained; S310: When the preset maximum number of iterations is reached, the update iteration is terminated and the cluster center is output, otherwise jump to S307; S311: The objective function J(U,V) of the fuzzy fuzzy C-means clustering in step 305 is to classify the generated cluster centers according to the following criteria: Where: U is the membership matrix, V is the cluster center vector, x is the i-th data to be clustered; v j represents the jth cluster center; u ij The i-th data x to be clustered i The membership degree belonging to the jth cluster center, c is the number of cluster centers, n is the data x i The number of, m is the fuzzy parameter; S312: Define sample x i For cluster center v j The membership degree u ij for: In formula (8), p j is the sample x i For cluster center v j The relative membership weight of sample x i To cluster center v j The distance d ij The result of the reciprocal of after exponentiation; p + is the sample x i For all cluster centers v k The relative membership weight of sample x i To cluster center v k The distance d ik The result of the reciprocal of after exponentiation; d j is the sample x i With cluster center v j The distance between them is usually the Euclidean distance d j =|xv j |, reflects the sample x i With cluster center v j The geometric distance between them is smaller, the closer the sample x i With cluster center v j The higher the similarity, the higher the + is the sample x to all cluster centers v k The distance between them is also the Euclidean distance d + =|xv + |, used to calculate the distance between sample x and all cluster centers for normalization; S313: The constraints of construct (7) are: Where F represents the value of the expanded objective function; λ represents the introduced Lagrange multiplier, which is used to impose constraints; S314: Iteratively update cluster center v j and membership u ij , the iteration termination condition is: Where: k is an integer greater than 0, ε is a preset error value; Indicates that the i-th sample x belongs to the j-th cluster center v after the k+1th iteration j The membership of reflects the membership value after one iteration update; The table indicates that the i-th sample x belongs to the j-th cluster center v after the current k-th iteration j The membership of reflects the membership value in the current iteration; S315: According to the clustering results, the corresponding areas of water bodies, forests, cities, farmlands, and grasslands in the predetermined target area are filled with "red (RGB value: 255, 0, 0), green (RGB value: 0, 255, 0), blue (RGB value: 0, 0, 255), orange (RGB value: 255, 128, 0), and black (RGB value: 0, 0, 0)" to form a visualization result of the distribution of objects in the predetermined target area.

5. According to claim 1, a potential wildfire early warning classification method for areas adjacent to power transmission lines is characterized by: Step S4: Select the target area by 1km 2 Rasterization is performed for units, including: S401: Determine the origin coordinates (X0, Y0): select the lower left corner of the target image as the origin; S402: Setting the grid size: total length 1000 meters, total height 1000 meters; S403: Calculate the number of pixel rows and columns: in, represents the rounding up function, ensuring that even if the last grid is not completely filled, it will be included; S404: Generate cell center point: The center point of each cell can be calculated by the following formula: X center =X0+i×total grid length+total grid length / 2 (13); Y center =Y0+j×total grid height+Cell Height / 2 (14); Among them, i is the column index and j is the row index; S405: Define the boundaries of each grid: The four vertices of each grid can be determined by adjusting the position of the center point; for example, for a cell in the i-th column and the j-th row, its lower left corner vertex is: X left_bottom =X0+i×total length of grid (15); Y left_bottom =Y0+j×total grid height (16); The top right vertex is: X right_top =X left_bottom + total length of grid (17); Y right_top =Y left_bottom + grid_total_height(18); S406: Clipping to the target area according to the calculated vertices.

6. A method for grading potential wildfire warnings in areas adjacent to power transmission lines according to claim 5, characterized in that: According to the area ratio of the five typical landforms obtained by classification, the forest coverage rate in each grid is calculated; the calculation formula of forest coverage rate F is:

7. A method for grading potential wildfire warnings in areas adjacent to power transmission lines according to claim 1, characterized in that: In step 5, obtain each 1km 2 The fire risk probability of the grid includes the following steps: S501: Importing data such as soil moisture, precipitation, and temperature for a time period corresponding to 1km recognition; S502: Normalize all variables so that their value range is between [0, 1]: Among them: X represents the original data of forest coverage rate F, soil moisture S, precipitation P, and temperature T, X min and X max are the minimum and maximum values ​​of the variable, respectively. For some variables, such as soil moisture S, lower values ​​mean higher fire risk. Take the inverse normalized value: S503: According to the importance of forest coverage F, soil moisture S, precipitation P, and temperature T data to fire risk, set the corresponding weight ω F ,ω S ,ω P ,ω T , and the weight index satisfies: oh F +oh S +oh P +oh T =1(22); S504: Using a linear combination method, each normalized variable and its corresponding weight are multiplied and summed to obtain the fire risk index η: the = the F F no[m +oh S S no[m +oh P P no[m +oh T T no[m (23); S505: Determine the fire risk level according to the fire risk occurrence probability η of the grid: η<ε (24); Where: ε is the preset probability criterion threshold; S506: According to the fire risk probability result of step S505, 2 The grid is divided into 5 fire risk levels.

8. A method for grading potential wildfire warnings in areas adjacent to power transmission lines according to claim 1, characterized in that: According to the danger level of "none, mild, moderate, high, extreme" for each grid obtained in S506, fill in "bright green (#67D43A, RGB value: 103, 212, 58) -> light sky blue (#98B5D1, RGB value: 152, 181, 209) -> blue-purple (#4D5BB8, RGB value: 77, 91, 184) -> fiery red (#D02B2A, RGB value: 208, 43, 42) -> light magenta (#E688C7, RGB value: 230, 136, 199)" to form the rasterized fire danger level prediction visualization result of the selected target area.

Images