Forest fire prediction method and system using remote sensing image

Abstract

The invention provides a forest fire prediction method and system using a remote sensing image, and relates to the technical field of data processing, and the method comprises the steps: defining a fitness function used for evaluating the advantages and disadvantages of each individual, and each individual represents a group of weight parameters; randomly generating an initial population meeting a coding mode, wherein each individual represents a group of weight parameters; the fitness of each individual is calculated according to a fitness function, and a new generation of population is generated through selection, crossover and mutation operation; judging whether a preset termination condition is met or not, if the termination condition is met, stopping iteration, and outputting a current final weight parameter combination; according to the final weight parameter combination, combined with land surface temperature, vegetation coverage rate and vegetation water content data, calculating a forest fire danger dynamic danger index; and classifying the fire danger grades according to the forest fire danger dynamic danger index. The accuracy of the prediction result can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a forest fire prediction method and system using remote sensing images. Background Art

[0002] Some traditional forest fire prediction methods rely on ground observations, meteorological data, and historical fire records, but these methods have problems such as difficulty in data acquisition, poor real-time performance, and limited prediction accuracy. For example, many existing prediction methods rely only on single parameters such as surface temperature or vegetation index (such as the Normalized Difference Vegetation Index NDVI). Although these parameters can reflect the forest fire risk to a certain extent, they cannot fully and accurately describe the complexity and variability of forest fires.

[0003] The occurrence and development of forest fires are affected by many factors, including surface temperature, vegetation coverage, vegetation type, moisture content of combustibles, meteorological conditions (such as wind speed, humidity, precipitation, etc.), and topography. Some existing methods ignore the interactions and influences between these factors, resulting in inaccurate and unreliable prediction results. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a forest fire prediction method and system using remote sensing images, which can improve the accuracy, reliability and real-time performance of the prediction results.

[0005] In order to solve the above technical problems, the technical solution of the present invention is as follows: In a first aspect, a forest fire prediction method using remote sensing images, the method comprising: Calculate the surface temperature based on the atmospheric transmittance, surface emissivity and radiation brightness value; Extract key data from preprocessed remote sensing images; use key data to calculate the normalized vegetation index; generate a new image based on the normalized vegetation index; calculate vegetation coverage based on the new image and a preset threshold; The fused image is reconstructed by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then undergoing inverse wavelet transformation. The number of clusters is determined according to the vegetation characteristics of the target area, and key vegetation features are extracted and standardized from the fused image. Cluster analysis is performed on the key vegetation features, and finally the distribution data of various types of vegetation are identified. According to the distribution data of various types of vegetation and combined with meteorological factors, the fuel moisture content model of various types of vegetation is constructed, and the moisture content of vegetation is calculated based on the fuel moisture content model; Set the population size of the genetic algorithm, which represents the number of individuals; determine the binary code used to represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic forest fire risk index; set the crossover rate, mutation rate and termination condition of the genetic algorithm; Define a fitness function for evaluating the quality of each individual, where each individual represents a set of weight parameters; Randomly generate an initial population that meets the encoding method, where each individual represents a set of weight parameters; calculate the fitness of each individual according to the fitness function, and generate a new generation of population through selection, crossover and mutation operations; determine whether the preset termination condition is met. If the termination condition is met, stop the iteration and output the current final weight parameter combination; According to the final combination of weight parameters, combined with the surface temperature, vegetation coverage and vegetation moisture content data, the dynamic forest fire risk index is calculated; Fire risk levels are classified according to the dynamic danger index of forest fire risk.

[0006] Furthermore, the surface temperature is calculated based on the atmospheric transmittance, surface emissivity and radiation brightness value, including: Obtain processed remote sensing image data containing thermal infrared bands, identify and select two adjacent thermal infrared bands; determine the central wavelength of the two adjacent thermal infrared bands, the difference in radiation brightness between the bands, the atmospheric transmittance and the surface emissivity; The atmospheric transmittance of two adjacent thermal infrared bands is estimated by using the atmospheric radiation transfer model and combining the atmospheric profile data of the target area. According to the land cover type, the surface emissivity of two adjacent thermal infrared bands is estimated based on the inversion method of remote sensing data; The pixel values ​​of the remote sensing image are converted into radiation brightness values; the surface temperature is calculated based on the split window algorithm, combining the atmospheric transmittance, surface emissivity and radiation brightness values.

[0007] Further, extract key data from the preprocessed remote sensing image; calculate the normalized vegetation index using the key data; generate a new image based on the normalized vegetation index; calculate the vegetation coverage rate based on the new image and a preset threshold, including: Extracting red band and near infrared band data from the processed remote sensing image data; Calculate the normalized vegetation index value based on the data of the red band and the near-infrared band; Assign the normalized vegetation index value to each pixel to generate a new normalized vegetation index image; A threshold is set, and the number of pixels in the normalized vegetation index image that are greater than the set threshold is counted, where the ratio of the number of pixels greater than the set threshold to the total number of pixels is the vegetation coverage rate.

[0008] Furthermore, the rectified panchromatic image and multispectral image are decomposed and fused by wavelet, and then the fused image is reconstructed by inverse wavelet transform, including: Perform wavelet decomposition on the corrected panchromatic image and multispectral image to obtain wavelet coefficients of different scales and directions corresponding to the panchromatic image and the multispectral image, respectively. The wavelet coefficients include approximate coefficients and detail coefficients. According to the preset fusion rules, the wavelet coefficients of different scales and directions corresponding to the panchromatic image and the multispectral image are fused to obtain the fused wavelet coefficients; The fused wavelet coefficients are subjected to inverse wavelet transform to reconstruct the fused image.

[0009] Furthermore, the number of clusters is determined according to the vegetation characteristics of the target area, key vegetation features are extracted and standardized from the fused image, cluster analysis is performed on the key vegetation features, and finally the distribution data of various types of vegetation are identified, including: According to the vegetation type and distribution characteristics of the target area, determine the corresponding cluster number K value; Select key features reflecting vegetation characteristics from the fused image, including spectral reflectance and vegetation index; Standardize the key features to obtain standardized data; use the selected key features and the standardized data to apply the K-means clustering algorithm; initialize the cluster center, iteratively assign each pixel to the corresponding cluster center, and update the position of the cluster center until a preset number of iterations is reached to obtain a clustering result; Each clustering result is marked and matched with the actual vegetation type to obtain the distribution data of each type of vegetation.

[0010] Furthermore, based on the distribution data of various types of vegetation and combined with meteorological factors, a fuel moisture content model for each type of vegetation is constructed, and the moisture content of vegetation is calculated based on the fuel moisture content model, including: Obtain distribution data of various types of vegetation, as well as meteorological data related to vegetation moisture content; Obtain the correlation between the moisture content of fuels of different vegetation types and meteorological factors, and build a neural network model based on the correlation and data characteristics; Select features related to moisture content of combustibles from meteorological data as input data for the neural network model; Using historical data to train a neural network model to obtain a trained neural network model; The trained neural network model is verified and optimized through cross-validation and grid search methods to obtain the final neural network model; Preprocessing the real-time meteorological data to obtain preprocessed meteorological data; The preprocessed meteorological data are input into the final neural network model to obtain the predicted values ​​of moisture content of various types of vegetation.

[0011] Furthermore, the fitness function The calculation formula is: ; in, Represents the index of the sample point, which is used to traverse all sample points and calculate the difference between the predicted value and the true value; Represents the total number of samples, which is a positive integer, indicating that there are a total of samples; represents the weight parameter; Represents the total number of time steps in the time series, which is a positive integer and represents the length of the entire time series; represents the index of the time step; Indicates The predicted value of samples; Indicates The true value of samples; The exponent representing the error norm; Indicated in The actual data value observed at each time step; Indicated in The predicted value at each time step; , and They respectively represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk.

[0012] Furthermore, the calculation formula of the dynamic risk index of forest fire risk is: ; in, It represents the dynamic danger index of forest fire risk; represents the surface temperature; Indicates vegetation coverage; Indicates the moisture content of vegetation; and represents weight; and Respectively represent the minimum and maximum values ​​of the surface temperature; Indicates the maximum value of vegetation coverage; Indicates the minimum value of vegetation moisture content; Represents the base of natural logarithms.

[0013] In the second aspect, a forest fire prediction system using remote sensing images includes: The acquisition module is used to calculate the surface temperature based on the atmospheric transmittance, surface emissivity and radiation brightness value; The classification module is used to extract key data from the preprocessed remote sensing image; calculate the normalized vegetation index using the key data; generate a new image based on the normalized vegetation index; calculate the vegetation coverage rate based on the new image and the preset threshold; reconstruct the fused image by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then performing inverse wavelet transformation; determine the number of clusters based on the vegetation characteristics of the target area, extract and standardize key vegetation features from the fused image, perform cluster analysis on the key vegetation features, and finally identify the distribution data of various types of vegetation; A construction module is used to construct a fuel moisture content model for each type of vegetation based on the distribution data of each type of vegetation and in combination with meteorological factors, and calculate the moisture content of vegetation based on the fuel moisture content model; set the population size of the genetic algorithm, where the population size represents the number of individuals; determine the binary code used to represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk; set the crossover rate, mutation rate and termination condition of the genetic algorithm; define a fitness function for evaluating the quality of each individual, where each individual represents a set of weight parameters; randomly generate an initial population that meets the encoding method, where each individual represents a set of weight parameters; calculate the fitness of each individual based on the fitness function, and generate a new generation of populations through selection, crossover and mutation operations; determine whether the preset termination condition is met, and if the termination condition is met, stop the iteration and output the current final weight parameter combination; The calculation module is used to calculate the dynamic danger index of forest fire risk based on the final combination of weight parameters, combined with surface temperature, vegetation coverage and vegetation moisture content data; and classify the fire risk level according to the dynamic danger index of forest fire risk.

[0014] According to a third aspect, a computing device includes: one or more processors; The storage device is used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described.

[0015] In a fourth aspect, a computer-readable storage medium stores a program, and when the program is executed by a processor, the method described is implemented.

[0016] The above scheme of the present invention includes at least the following beneficial effects.

[0017] The present invention makes comprehensive use of multiple key parameters such as surface temperature, vegetation coverage and vegetation moisture content. These parameters are accurately calculated through remote sensing image data and can fully reflect the complexity and variability of forest fire risk. By optimizing the weight parameters through genetic algorithms, the dynamic danger index of forest fire risk is accurately calculated, which significantly improves the accuracy of prediction.

[0018] The present invention uses real-time remote sensing image data, which can timely reflect the changes in forest fire risk conditions. The iterative process of the genetic algorithm can converge quickly, so that the prediction results can be quickly updated to meet the real-time requirements. Through the fusion of panchromatic images and multispectral images, as well as the fine processing of vegetation classification, the method can more accurately identify different types of vegetation and combustibles, providing a more accurate data basis for building a combustible moisture content model.

[0019] The present invention takes into account the impact of meteorological factors on forest fire risk and constructs a combustible moisture content model for various types of vegetation to make the prediction results more in line with the actual situation. As an intelligent optimization algorithm, the genetic algorithm has strong adaptability and robustness, can handle complex optimization problems, and ensure the stability and reliability of the prediction results. The present invention provides support for the prevention and control of forest fires by calculating the dynamic danger index of forest fire risk and classifying the fire risk level. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flow chart of a forest fire prediction method using remote sensing images provided by an embodiment of the present invention.

[0021] Figure 2 It is a schematic diagram of a forest fire prediction system using remote sensing images provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0022] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0023] like Figure 1 As shown, an embodiment of the present invention provides a forest fire prediction method using remote sensing images, the method comprising the following steps: Step 11, calculating the surface temperature according to the atmospheric transmittance, surface emissivity and radiation brightness value; Step 12, extracting key data from the preprocessed remote sensing image; calculating the normalized vegetation index using the key data; generating a new image based on the normalized vegetation index; calculating the vegetation coverage rate based on the new image and a preset threshold; Step 13, reconstructing the fused image by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then performing inverse wavelet transform; determining the number of clusters according to the vegetation characteristics of the target area, extracting and standardizing key vegetation features from the fused image, performing cluster analysis on the key vegetation features, and finally identifying the distribution data of various types of vegetation; Step 14, constructing a fuel moisture content model for each type of vegetation based on the distribution data of each type of vegetation and in combination with meteorological factors, and calculating the moisture content of the vegetation based on the fuel moisture content model; Step 15, setting the population size of the genetic algorithm, where the population size represents the number of individuals; determining the binary code used to represent the weights of the surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk; setting the crossover rate, mutation rate and termination condition of the genetic algorithm; Step 16, defining a fitness function for evaluating the quality of each individual, where each individual represents a set of weight parameters; Step 17, randomly generate an initial population that meets the encoding method, each individual represents a set of weight parameters; calculate the fitness of each individual according to the fitness function, and generate a new generation of population through selection, crossover and mutation operations; determine whether the preset termination condition is met, if the termination condition is met, stop iteration, and output the current final weight parameter combination; Step 18, calculating the forest fire risk dynamic hazard index based on the final weight parameter combination, combined with the surface temperature, vegetation coverage and vegetation moisture content data; Step 19, classify the fire risk level according to the dynamic forest fire risk index.

[0024] In an embodiment of the present invention, the surface temperature is one of the important factors in assessing forest fire risk. By calculating the surface temperature through remote sensing image data, the temperature condition of the forest surface can be monitored in real time, providing key information for judging the fire risk level. The normalized difference vegetation index (NDVI) is an important indicator reflecting the growth status of vegetation. By calculating the NDVI, the vegetation coverage rate can be accurately determined, and then the density of the forest and the distribution of combustibles can be assessed, providing an important basis for fire risk prediction. Panchromatic images have the characteristics of high resolution, while multispectral images can provide rich spectral information. Through geometric correction and image fusion, the advantages of the two can be combined to obtain images that are both clear and rich in information. Furthermore, through vegetation classification, different types of vegetation can be finely divided, providing more accurate data support for constructing a combustible moisture content model. Vegetation moisture content is one of the key factors in assessing forest fire risk. By combining meteorological factors and the distribution data of various types of vegetation, a combustible moisture content model is constructed, and the moisture content of vegetation can be accurately calculated.

[0025] Genetic algorithm is an intelligent optimization algorithm. By setting a reasonable population size, encoding method, crossover rate, mutation rate and termination condition, the algorithm can be ensured to run efficiently and solve accurately. The fitness function is the standard for evaluating the quality of each individual. By defining a reasonable fitness function, the genetic algorithm can be guided to evolve towards the optimal solution, ensuring that the optimization results of the weight parameters meet the actual needs. Through the iterative optimization of the genetic algorithm, the optimal weight combination of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic hazard index of forest fire risk can be found. This improves the accuracy and reliability of fire risk prediction. According to the optimized weight parameters and the real-time acquired surface temperature, vegetation coverage and vegetation moisture content data, the dynamic hazard index of forest fire risk can be calculated. By classifying the dynamic hazard index of forest fire risk, the fire risk level can be divided into different levels, which is convenient for decision-making departments to take corresponding preventive measures and emergency responses according to the fire risk level, which helps to improve the prevention and control efficiency of forest fires and reduce fire losses.

[0026] In a preferred embodiment of the present invention, the above step 11 calculates the surface temperature according to the atmospheric transmittance, the surface emissivity and the radiation brightness value. The following is the detailed implementation process of this step: Obtain initial remote sensing image data, which comes from satellite or aerial remote sensing platforms, such as Landsat, Sentinel, SPOT and other satellites, as well as remote sensing equipment carried by aircraft such as drones. Select appropriate remote sensing platforms and sensors to obtain remote sensing image data for a specific time and specific area. Eliminate or reduce radiation errors in remote sensing images caused by sensor response, atmospheric absorption and scattering, and sun angle.

[0027] According to the characteristics of the sensor, such as spectral response curve, gain and offset, the image is corrected to restore the real radiation information of the object. The atmospheric correction model (such as 6S, MODTRAN, etc.) and meteorological data (such as atmospheric visibility, water vapor content, etc.) are used to eliminate the influence of the atmosphere on the remote sensing signal and obtain the real information such as surface reflectivity or radiation temperature. Considering the influence of solar altitude angle and azimuth angle on surface reflectivity, the solar angle correction is performed to ensure the temporal and spatial consistency of the image data. Correct the geometric deformation in the remote sensing image to match the image with the actual position and shape of the object.

[0028] On the remote sensing image and reference map (such as topographic map, Google Earth, etc.), identify and mark a set of clear and easily identifiable common feature points, such as road intersections, river bends, building corners, etc. Ensure that the control points are evenly distributed and cover the entire image area to improve the accuracy of the correction, and record the coordinate information of each control point on the image and reference map. Use a polynomial fitting method (such as a first-order polynomial) to establish a relationship model between the image coordinates and the reference map coordinates; input the coordinate information of the control points into the polynomial fitting algorithm, and solve the polynomial coefficients by the least squares method to obtain a geometric correction model, which can describe the position of each pixel in the image in the actual geographic space. Apply the established geometric correction model to the entire remote sensing image and calculate the new position of each pixel in the reference map coordinate system. Since the new position may not be an integer coordinate, resampling is required. Select the nearest neighbor method as the resampling strategy, that is, find the closest original pixel value as the new pixel value based on the new position. Traverse each pixel in the image, apply the correction model and resampling method, and obtain the geometrically corrected image.

[0029] For example, for a first-degree polynomial (linear) fit, assume that the relationship between each point (x, y) on the image and the corresponding point (X, Y) on the reference map can be expressed as: ; Where, (x, y) is the coordinate on the image, (X, Y) is the coordinate on the reference map; , , , , and are the linear transformation parameters that need to be solved. To solve these parameters, at least 3 non-collinear control points are required. For each control point, a system of equations with 6 unknowns can be obtained. If there are n control points (n ≥ 3), a system of equations with 2n equations can be obtained: ; This system of equations can be solved by the least squares method or other linear algebra methods (such as matrix inversion). After solving, we can get , , , , and The specific values ​​of these 6 parameters. Once these parameters are solved, they can be used to transform any point (x, y) on the image to the corresponding point (X, Y) on the reference map: ; This is the specific process of using a linear polynomial fitting method to establish a relationship model between image coordinates and reference map coordinates.

[0030] Select the map projection method according to the research purpose, regional characteristics and subsequent application requirements. For example, UTM (Universal Transverse Mercator) projection can be selected for large areas, and longitude and latitude projection (such as WGS84) can be selected for global areas. Use the projection transformation tool in the geographic information system (GIS) software or remote sensing processing software to transform the geometrically corrected image data from the original coordinate system (such as the sensor coordinate system) to the target projection coordinate system (such as UTM or longitude and latitude projection); input image data, original coordinate system information, target projection method and other parameters, and perform projection transformation operations. Through the above specific steps, high-quality remote sensing image data with accurate geometric and radiation characteristics can be obtained.

[0031] In a preferred embodiment of the present invention, the processed remote sensing image data containing the thermal infrared band is obtained, two adjacent thermal infrared bands are identified and selected; the central wavelength of the two adjacent thermal infrared bands, the difference in radiant brightness between the bands, the atmospheric transmittance and the surface emissivity are determined, specifically including: Acquire remote sensing image data containing thermal infrared bands from remote sensing satellites or aerial platforms. These data are stored in digital form and contain information from multiple bands. Use remote sensing image processing software (such as ENVI, ERDAS Imagine, QGIS, etc.) to open the acquired remote sensing image data. In the software interface, view the band list of the remote sensing image, including the wavelength range, resolution and other key information of each band. According to the wavelength range of the band, identify which bands belong to the thermal infrared band. Usually, the wavelength range of the thermal infrared band is between 8 and 14 microns. Among the identified thermal infrared bands, select two adjacent bands for subsequent analysis. Consider factors such as the sensitivity, resolution and data quality of the band when selecting.

[0032] Open the metadata file or related documents of the remote sensing image data, which contain detailed parameter information of the bands; in the metadata or documents, find the central wavelength information of the two selected adjacent thermal infrared bands and record it. Select a representative surface area on the remote sensing image and ensure that the area is clearly visible in both bands. Use image processing software to extract the radiation brightness values ​​of the selected area in the two thermal infrared bands, which involves operations such as image correction, cropping, and pixel value extraction. Perform mathematical operations on the extracted radiation brightness values ​​to calculate the difference in radiation brightness between the two bands in the same surface area.

[0033] According to the atmospheric conditions of the target area and the characteristics of remote sensing image data, select an atmospheric radiation transfer model (such as MODTRAN); collect atmospheric profile data of the target area through meteorological observation stations, meteorological satellites or reanalysis data, including the distribution information of temperature, humidity, pressure, etc. with height; input the collected atmospheric profile data into the selected atmospheric radiation transfer model, and set relevant parameters (such as band information, observation angle, etc.); run the atmospheric radiation transfer model (such as MODTRAN) to estimate the atmospheric transmittance of the selected two thermal infrared bands. Use remote sensing image data to determine the surface cover type of the target area; according to the surface cover type and the characteristics of remote sensing data, select a suitable estimation method to calculate the surface emissivity. For example, for vegetation-covered areas, the empirical relationship between vegetation index and surface emissivity can be used for estimation, and field measurements can also be performed to obtain the accurate value of surface emissivity. Alternatively, use an existing surface emissivity database for query and estimation.

[0034] In a preferred embodiment of the present invention, the atmospheric radiation transmission model is used in combination with the atmospheric profile data of the target area to estimate the atmospheric transmittance of two adjacent thermal infrared bands, specifically including: Analyze the climate characteristics, seasonal changes, weather conditions, etc. of the target area, and understand the factors such as water vapor, aerosols, and clouds in the atmosphere that may affect remote sensing image data. Check the metadata of remote sensing image data in detail to understand key information such as image acquisition time, satellite platform, sensor type, and band settings. According to the atmospheric conditions of the target area and the characteristics of remote sensing image data, select the most suitable model from the commonly used atmospheric radiation transfer models (such as MODTRAN). MODTRAN is a commonly used and powerful atmospheric radiation transfer model that is suitable for a variety of atmospheric conditions and observation geometries. Determine the type of atmospheric profile data required, which usually includes the distribution information of temperature, humidity, pressure, etc. with height. Find and obtain atmospheric profile data through meteorological observation stations, meteorological satellites (such as NASA's AIRS, European meteorological satellite's IASI, etc.), reanalysis data (such as NCEP / NCAR, ERA-Interim, etc.). Screen the acquired data to ensure the timeliness, accuracy, and completeness of the data, and organize the screened data into the format required by MODTRAN. In the atmospheric radiation transfer model, relevant parameters are set, including the wavelength range of the two adjacent thermal infrared bands, observation angle (zenith angle, azimuth angle), surface type, etc.

[0035] Input the sorted atmospheric profile data into MODTRAN, start the atmospheric radiation transfer model, and let it calculate according to the input atmospheric profile data and set parameters. MODTRAN will simulate the atmospheric transmission process of radiation and take into account the effects of absorption, scattering and reflection in the atmosphere. After MODTRAN is completed, extract the atmospheric transmittance of two adjacent thermal infrared bands from the output results.

[0036] In a preferred embodiment of the present invention, according to the surface cover type, based on the inversion method of remote sensing data, the surface emissivity of two adjacent thermal infrared bands is estimated, specifically including: Create an empty list (or array) to store the surface emissivity value of each pixel, set the identifiers of two adjacent thermal infrared bands (such as band1 and band2); for each pixel in the surface cover type map, perform the following steps and add the results to the list: Create an empty list named es to store the surface emissivity value of each pixel. Prepare the surface cover type map lp and remote sensing image data ra. For each pixel in lp (assuming the pixel index is i, j), perform the following steps: Get the surface coverage type of pixel (i, j), recorded as ce; according to ce, find the corresponding surface emissivity value in the lookup table eb, recorded as ed1 and ed2 respectively.

[0037] Alternatively, using NDVI estimation, obtain the reflectance values ​​of pixel (i, j) in two thermal infrared bands (assuming band 1 and band 2) from ra, recorded as rd1 and rd2 respectively. If emissivity data is used, the corresponding emissivity value is

[0038] Calculate NDVI: ; Use the known relationship between NDVI and surface emissivity or an estimation formula. For example, for vegetation-covered areas, the following empirical formula can be used: ed1=a×NDVI+b; ed2=c×NDVI+d; Among them, a, b, c and d are empirical coefficients, and the estimated ed1 and ed2 are added to the es list as a tuple (or list), that is, [ed1, ed2].

[0039] Traverse the es list and output the surface emissivity value corresponding to each pixel. For example, the output can be in text format, with one line per pixel, containing the pixel index (or position) and the corresponding surface emissivity value.

[0040] In a preferred embodiment of the present invention, the pixel values ​​of the remote sensing image are converted into radiance values; the surface temperature is calculated according to the split window algorithm by combining the atmospheric transmittance, the surface emissivity and the radiance value, which specifically includes: Get remote sensing image data files, including the image data itself and related metadata or header files. The metadata contains information such as image quantization method, radiometric calibration parameters, etc. Open the metadata file and look for radiometric calibration parameters, such as gain and offset values, or lookup tables (LUTs), which are used to convert image pixel values ​​(usually digital quantization values, such as DN values) into radiometric brightness values.

[0041] According to the radiometric calibration parameters in the metadata, the pixel value is converted to the radiometric brightness value using the corresponding formula. For example, for linear calibration, the formula is as follows: Radiance=Gain×DN+Offset; Apply this formula to each pixel in the image to obtain the corresponding radiation brightness value.

[0042] According to the characteristics of remote sensing data and application requirements, select the two-channel algorithm, which uses the radiation brightness values ​​of two adjacent thermal infrared bands to estimate the surface temperature; find the specific formula and required parameters of the two-channel algorithm. If the atmospheric transmittance is not a known parameter, it needs to be estimated based on the atmospheric conditions (such as humidity, temperature, air pressure, etc.) when the remote sensing data is acquired. The atmospheric radiation transfer model (such as MODTRAN) can be used to estimate the atmospheric transmittance. The surface emissivity can be estimated using indices such as NDVI (as described above); Based on the calculated radiance value, estimated atmospheric transmittance and surface emissivity Calculate the surface temperature value of each pixel; in, is the surface temperature (unit: °C, based on and The unit is determined by and are constants related to the characteristics of remote sensing data (obtained from metadata or related documents), and are the radiance values ​​of the two thermal infrared bands, and is the atmospheric transmittance of the corresponding band, and is the surface emissivity of the corresponding band.

[0043] In the embodiment of the present invention, by selecting the thermal infrared band that is sensitive to the surface temperature, an accurate data basis is provided for the subsequent surface temperature calculation. Determining the key parameters helps to accurately calculate the surface temperature in the subsequent steps and reduce errors. Atmospheric transmittance is a key parameter in the surface temperature calculation. It affects the amount of surface radiation received by the remote sensor. By accurately estimating the atmospheric transmittance, the influence of the atmosphere on the surface temperature calculation can be reduced, and the accuracy of the calculation results can be improved. The surface emissivity is another key parameter in the surface temperature calculation. It reflects the influence of different surface coverage types on radiation emission. By accurately estimating the surface emissivity, the accuracy of the surface temperature calculation can be further improved, making the result closer to the actual situation. The split window algorithm is a mature and effective surface temperature calculation method. It uses the information of two bands to reduce the influence of the atmosphere and surface coverage types on the temperature calculation. By comprehensively applying the previous estimation results and the split window algorithm, a more accurate surface temperature distribution can be obtained.

[0044] In a preferred embodiment of the present invention, the above step 12, extracting key data from the preprocessed remote sensing image; calculating the normalized vegetation index using the key data; generating a new image according to the normalized vegetation index; and calculating the vegetation coverage rate according to the new image and a preset threshold, includes: Step 121, extracting data of the red band and the near infrared band from the processed remote sensing image data, specifically includes: checking the metadata or attribute information of the remote sensing image, which includes the band information of the image, such as the band number, name, wavelength range, etc.; in the metadata, identifying the red band (usually located in the visible light range, with a wavelength of about 620-750 nanometers) and the near infrared band (usually located in the near infrared range, with a wavelength of about 750-1100 nanometers) according to the band number or name; extracting data of the red band and the near infrared band from the remote sensing image data according to the identified band identifiers, which involves selecting the band, setting extraction parameters (such as resolution, area, etc.), and performing the extraction operation.

[0045] Step 122, calculating the normalized vegetation index value according to the data of the red band and the near infrared band, specifically includes: opening or loading the red band and the near infrared band data extracted in step 121 in the remote sensing data processing software; using the calculation formula of the normalized vegetation index (NDVI), that is, NDVI = (NIR-Red) / (NIR+Red), where NIR represents the reflectance value of the near infrared band, and Red represents the reflectance value of the red band, to generate the NDVI value of each pixel.

[0046] Step 123, assigning the normalized vegetation index value to each pixel to generate a new normalized vegetation index image, specifically including: creating a new image file or dataset for storing the NDVI image, setting the image parameters such as size, resolution and format to ensure consistency with the original remote sensing image data; assigning the NDVI value calculated in step 122 to the corresponding pixel in the new image pixel by pixel, which usually involves traversing the NDVI result dataset and writing each value to the corresponding position of the new image.

[0047] Step 124, set a threshold, count the number of pixels in the normalized vegetation index image that are greater than the set threshold, wherein the ratio of the number of pixels greater than the set threshold to the total number of pixels is the vegetation coverage rate, specifically including: according to the research purpose and regional characteristics, set an NDVI threshold, which is used to distinguish between vegetation-covered and non-vegetation-covered areas. Generally, a higher NDVI value indicates denser vegetation coverage. In the remote sensing data processing software, open or load the NDVI image generated in step 123; use the statistical tools provided by the software to traverse each pixel in the NDVI image and count the number of pixels greater than the set threshold, which involves setting statistical parameters (such as thresholds, statistical ranges, etc.) and performing statistical operations; use the number of pixels greater than the threshold obtained by statistics and the total number of pixels in the NDVI image to calculate the vegetation coverage rate, and the calculation formula is: vegetation coverage rate = (number of pixels greater than the threshold / total number of pixels) × 100%.

[0048] In an embodiment of the present invention, NDVI calculation is based on the spectral reflectance characteristics of the red band and the near-infrared band, and these two bands are very sensitive to the chlorophyll content and leaf structure of the vegetation. Therefore, NDVI can accurately reflect the growth and coverage of vegetation. Through the automated processing flow (such as steps 121 to 124), remote sensing image data can be processed quickly and in large quantities to generate NDVI images and count vegetation coverage, which greatly improves work efficiency. NDVI value can be used as one of the important indicators for evaluating the quality of the ecological environment. A high NDVI value usually indicates lush vegetation and a good ecological environment; while a low NDVI value may indicate sparse vegetation and a degraded ecological environment. By comparing NDVI images and vegetation coverage data at different time points, the changing trend of vegetation coverage can be monitored, and ecological and environmental problems such as land degradation and deforestation can be discovered in a timely manner.

[0049] In the embodiment of the present invention, the process of performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then reconstructing the fused image through inverse wavelet transformation, further includes: Visit the official website of the remote sensing data provider or use a dedicated remote sensing data download tool; select and download panchromatic and multispectral images according to the research area, time range and resolution requirements; ensure that the downloaded image data contains necessary metadata, such as projection information, geographic coordinates, etc.; open remote sensing image processing software, such as ENVI, ERDAS Imagine, QGIS or GDAL, etc.; select an appropriate geometric correction method according to the type and degree of image distortion. For slight distortion, nearest neighbor interpolation or bilinear interpolation can be used; for complex distortion, polynomial fitting or more advanced correction methods may be required; obtain accurate geographic coordinate data of the image corresponding area from reliable geographic data sources (such as GPS measurement data, high-precision maps, etc.), ensure that the coordinate system of the reference data is consistent with the coordinate system of the image, or perform necessary coordinate system conversion.

[0050] In the remote sensing image processing software, open the panchromatic image file, check the metadata of the image, make sure you understand the projection information, resolution, etc. of the image, and select multiple obvious landmarks on the panchromatic image as control points. These landmarks should be easily recognizable features on the image, such as road intersections, building edges, etc. For each control point, enter its corresponding precise geographic coordinates. This can be done manually or using the geolocation tool provided by the software; select the previously determined geometric correction method in the software, use the control points and correction algorithm to perform geometric correction on the panchromatic image, the software will automatically calculate the transformation parameters and adjust the image to the correct geographic location, and you can adjust the correction parameters as needed, such as the order of the polynomial, interpolation method, etc., to obtain better correction results, save the corrected panchromatic image as a new file, select the appropriate image format and compression options, and ensure that the saved file contains the necessary metadata, such as the corrected projection information, geographic coordinates, etc.; in the remote sensing image processing software, open the multispectral image file, check the metadata of the image, make sure you understand the resolution, projection information, etc. of each band, if there is a significant difference in resolution or features between the multispectral image and the panchromatic image, you may need to add or adjust the control points to adapt to the characteristics of the multispectral image. Multispectral images are geometrically corrected using the same correction algorithm and control points (or adjusted control points) as panchromatic images.

[0051] In a preferred embodiment of the present invention, the rectified panchromatic image and multispectral image are subjected to wavelet decomposition and fusion, and then subjected to inverse wavelet transformation to reconstruct a fused image, including: The corrected panchromatic image and multispectral image are decomposed by wavelet to obtain the wavelet coefficients of different scales and directions corresponding to the panchromatic image and multispectral image respectively. The wavelet coefficients include approximate coefficients and detail coefficients. Specifically, Haar wavelet is selected according to the image characteristics and processing requirements; Haar wavelet has compact support and orthogonality, and the corresponding wavelet transform library (such as PyWavelets) is called, and Haar wavelet is specified as the basis function. Set the number of wavelet decomposition layers. The choice of decomposition layers depends on the resolution of the image and the desired level of detail. Generally speaking, images with higher resolution can choose more decomposition layers to capture finer details. Use image processing libraries (such as OpenCV, PIL, etc.) to read panchromatic image data and convert it into a format suitable for wavelet transform (such as grayscale image). Call the decomposition function in the wavelet transform library, pass in the panchromatic image data and the selected Haar wavelet basis function, specify the number of decomposition layers, and start wavelet decomposition. The decomposition result includes a series of wavelet coefficients, which are divided into approximate coefficients (low-frequency part) and detail coefficients (high-frequency part). The approximate coefficients represent the general outline and overall structure of the image, while the detail coefficients capture the details and edge information of the image.

[0052] Use the image processing library to read multispectral image data, which usually contains multiple bands (such as red, green, blue, etc.). For each band of the multispectral image, repeat the same wavelet decomposition process as the panchromatic image, that is, call the decomposition function of the wavelet transform library for each band, pass in the band data and Haar wavelet basis function, and specify the same number of decomposition layers. Each band will get a set of independent wavelet coefficients, including approximate coefficients and detail coefficients.

[0053] According to the preset fusion rules, the wavelet coefficients of different scales and directions corresponding to the panchromatic image and the multispectral image are fused to obtain the fused wavelet coefficients, which specifically include: Analyze the characteristics of panchromatic and multispectral images, including their resolution, spectral information, contrast, texture, etc. Determine the fusion goal, such as emphasizing spatial details, retaining spectral information, enhancing specific features, or improving the overall visual effect. Based on the fusion goal, select an appropriate fusion strategy, such as maximum fusion and minimum fusion rules. Select an appropriate wavelet basis and decomposition layer number, perform wavelet decomposition on the panchromatic image, and obtain wavelet coefficients at different scales and directions. Perform wavelet decomposition on each band of the multispectral image to obtain their respective wavelet coefficients. Create an empty array or data structure for the fused wavelet coefficients. For each pixel (or each area) at each scale and direction: Compare the wavelet coefficients at corresponding positions in the panchromatic image and the multispectral image, determine the fusion coefficient at that position according to the selected fusion strategy (such as maximum fusion or minimum fusion), store the fused coefficients in the initialized array, repeat the above fusion process to ensure that each band of the multispectral image is fused with the panchromatic image, and store the fused wavelet coefficients in an appropriate form for subsequent processing.

[0054] Perform inverse wavelet transform on the fused wavelet coefficients to reconstruct the fused image, which includes: Determine the wavelet basis and the number of decomposition levels, and ensure that the same wavelet basis function and number of decomposition levels are used as those used in decomposition. For the fused wavelet coefficients of the panchromatic image: Initialize an empty array or image container to store the image data after inverse transformation; call the inverse wavelet transform function, passing in the fused approximation coefficients and detail coefficients, as well as the wavelet basis and decomposition layer number and other parameters. Store the result of the inverse transform in the previously initialized array, which will be the spatial domain representation of the fused panchromatic image. For each band of the multispectral image: Repeat the above inverse transformation process to perform inverse transformation on the wavelet coefficients after fusion of each band, ensure that each band uses the same wavelet basis and decomposition level for inverse transformation, and store the results of inverse transformation of each band into their respective arrays.

[0055] Initialize a multidimensional array or image object that can hold all band data, combine the results of the inverse transformation of each band into the multispectral image container according to the band order of the original multispectral image, and check and ensure that the format and type of the combined multispectral image data are consistent with the original image for subsequent processing or analysis. According to application requirements, select an appropriate image format (such as TIFF, JPEG, PNG, etc.) to save the fused multispectral image, and save the fused multispectral image to the specified file path or database.

[0056] In the embodiment of the present invention, the panchromatic image usually has a high spatial resolution and can clearly show the details and edge information of the object. Through wavelet decomposition and fusion, this information can be effectively transferred to the multispectral image, and the spatial resolution of the fused image can be improved. The multispectral image contains rich spectral information and can reflect the characteristics of different bands of the object. In the wavelet fusion process, through reasonable fusion rules, it can be ensured that this information is effectively retained in the fused image, thereby enhancing the recognition and analysis capabilities of the image. The panchromatic image and the multispectral image after geometric correction are consistent in space. Through wavelet decomposition and fusion, the geometric distortion and spectral distortion between the images can be further reduced. The wavelet transform has good denoising performance. In the decomposition process, high-frequency noise is usually allocated to the high-frequency coefficients. Through reasonable fusion rules, these noise components can be effectively suppressed and the signal-to-noise ratio of the fused image can be improved; wavelet decomposition can decompose the image into wavelet coefficients of different scales and directions, and these coefficients represent different structures and detail information of the image. Through fusion, the detail information in the panchromatic image can be effectively transferred to the multispectral image, and the detail expression capability of the fused image can be enhanced.

[0057] In a preferred embodiment of the present invention, the number of clusters is determined according to the vegetation characteristics of the target area, key vegetation features are extracted and standardized from the fused image, cluster analysis is performed on the key vegetation features, and finally the distribution data of various types of vegetation are identified, including: According to the vegetation type and distribution characteristics of the target area, the corresponding cluster number K value is determined, including: analyzing the vegetation type and distribution characteristics of the target area, which includes understanding the vegetation type, growth environment, distribution pattern, etc. in the area. Combining the existing vegetation classification system, remote sensing image interpretation experience and historical data of the area, the range of the cluster number K is preliminarily determined. Through multiple experiments, different K values ​​are used for clustering, and the stability and accuracy of the clustering results are evaluated. According to the evaluation results, the K value is gradually adjusted until the optimal number of clusters is found.

[0058] Select key features that reflect vegetation characteristics from the fused image, including spectral reflectance and vegetation index, including: analyzing the reflectance differences of different vegetation types in each band of the image. The reflectance characteristics of vegetation in different bands are different. Analyze the relationship between various vegetation indices (such as NDVI, EVI, SAVI, etc.) and vegetation types. Vegetation indices are usually calculated through specific band combinations and can reflect vegetation growth conditions, coverage and other information. Analyze the texture information in the image, such as the roughness, uniformity, and contrast of vegetation. Texture features can provide information about vegetation structure and help distinguish different types of vegetation. Calculate statistics such as correlation and variance between features to evaluate the distinguishing ability of features. The selected features can clearly distinguish different vegetation types, and the features can show stable distinguishing ability under different conditions (such as different seasons, different lighting conditions, etc.).

[0059] For example, for spectral reflectance, you can choose those bands or band combinations that have the largest reflectance differences between different vegetation types. For vegetation indices, choose indices that are closely related to vegetation growth conditions and have obvious differences between different vegetation types. For texture features, choose texture metrics that can reflect differences in vegetation structure and are stable.

[0060] Use remote sensing image processing software (such as ENVI, ERDAS, etc.) to process the fused image according to the selected feature type and calculation method. For spectral reflectance features, directly extract the pixel value of the specified band. For vegetation index features, calculate the pixel value of the corresponding band according to the index calculation formula. For texture features, use texture analysis algorithms (such as gray level co-occurrence matrix, wavelet transform, etc.) to process the image and extract the texture metric value.

[0061] Standardize the key features to obtain standardized data; use the selected key features and standardized data to apply the K-means clustering algorithm; initialize the cluster center, iteratively assign each pixel to the corresponding cluster center, and update the position of the cluster center until the preset number of iterations is reached to obtain the clustering result, which specifically includes: preprocessing the extracted key feature data, including removing outliers, filling missing values, etc.; use appropriate standardization methods (such as Z-score standardization, Min-Max standardization, etc.) to standardize the key feature data, which can eliminate the dimensional differences between different features so that they can be compared and clustered on the same scale.

[0062] Determine the number of clusters K based on the vegetation types and distribution characteristics of the target area, as well as prior knowledge or test results. Randomly select K pixels as the initial cluster centers. These centers can be any pixels in the image, but it is usually recommended to select them as dispersedly as possible to cover the entire feature space. For each pixel in the image, calculate its distance from the K cluster centers. The distance can be measured using Euclidean distance, Manhattan distance, or other appropriate distance metrics, and assign each pixel to the nearest cluster center. For each cluster center, calculate the mean of all its assigned pixels and use the mean as the new cluster center position. Repeat the above pixel assignment and center update process until the preset number of iterations is reached. After iteration, each pixel is assigned to a cluster center, thus forming K clusters, which represent different vegetation types or vegetation states.

[0063] Each clustering result is marked and matched with the actual vegetation type to obtain the distribution data of each type of vegetation, including: analyzing the clustering results, observing the characteristic distribution and number of pixels of each cluster, and preliminarily judging the vegetation types they may correspond to; verifying and adjusting the clustering results in combination with field surveys or existing vegetation distribution maps, and determining the vegetation type corresponding to each cluster by comparing the clustering results with the actual vegetation type; generating the distribution data of each type of vegetation based on the clustering results and the corresponding vegetation type.

[0064] In an embodiment of the present invention, the fused image combines the high spatial resolution of the panchromatic image and the rich spectral information of the multispectral image, so that the detailed features and spectral characteristics of the vegetation are fully displayed. The use of this comprehensive information greatly improves the accuracy of vegetation classification. By selecting key features that reflect the characteristics of vegetation, such as spectral reflectance and vegetation index, the growth state and type differences of vegetation can be captured more accurately, providing a strong basis for classification. Standardizing the key features eliminates the dimensional differences between different features, allowing the clustering algorithm to treat each feature more fairly and accurately, further improving the accuracy of classification. The application of automated classification methods such as the K-means clustering algorithm can quickly process a large amount of image data, greatly improving the efficiency of vegetation classification and reducing labor costs.

[0065] In a preferred embodiment of the present invention, the above step 14, based on the distribution data of various types of vegetation and in combination with meteorological factors, constructs a combustible moisture content model for various types of vegetation, and calculates the moisture content of vegetation based on the combustible moisture content model, including: Step 141, obtain the distribution data of various types of vegetation and meteorological data related to the moisture content of vegetation, specifically including: obtaining the spatial distribution information of various types of vegetation through remote sensing technology (such as satellite images, drone aerial photography, etc.) or ground surveys, including vegetation type, coverage, height, etc., which are usually stored in the form of raster data or vector data. Obtain meteorological factor data related to vegetation moisture content, such as temperature, humidity, precipitation, wind speed, solar radiation, etc. from meteorological stations, meteorological satellites or other meteorological data sources. These data are usually time series data and need to be sorted and organized according to timestamps.

[0066] Step 142, obtaining the correlation between the moisture content of combustibles of different vegetation types and meteorological factors, and constructing a neural network model based on the correlation and data characteristics, specifically including: Collect fuel moisture data for different vegetation types. These data can come from field measurements, remote sensing monitoring, or historical records. Also collect meteorological data that may be related to fuel moisture content, such as temperature, humidity, precipitation, wind speed, solar radiation, evaporation, etc. These data match the moisture content data in time and space. Check whether there are missing values, outliers, or duplicate values ​​in the dataset, and perform corresponding processing, such as filling missing values, removing outliers, or merging duplicate values. Ensure the time series integrity of the data, and perform interpolation or other processing for incomplete data.

[0067] Standardize or normalize the data to eliminate the influence of different dimensions and orders of magnitude on the correlation analysis. You can choose to use Z-score standardization, Min-Max normalization and other methods. Use statistical methods (such as Pearson correlation coefficient) to calculate the correlation coefficient between the moisture content of combustibles and various meteorological factors. According to the absolute value of the correlation coefficient, preliminarily determine which meteorological factors may have significant correlation with the moisture content; perform a significance test on the correlation coefficient to determine the statistical significance level of the correlation. You can use t-test, F-test or other appropriate statistical test methods; according to the correlation coefficient and significance test results, determine which meteorological factors have a significant impact on the moisture content. Analyze the possible relationship between these meteorological factors and the moisture content, such as positive correlation, negative correlation, nonlinear relationship, etc.

[0068] According to the results of correlation analysis and data characteristics (such as data scale, feature dimension, degree of nonlinearity, time series characteristics, etc.), evaluate the applicability of different neural network architectures; consider whether it is necessary to process time series data (such as using recurrent neural networks) or spatial data (such as using convolutional neural networks); based on data characteristics and analysis results, select the appropriate neural network architecture. For example, if the data has significant time series characteristics, you can choose a long short-term memory network (LSTM) or a gated recurrent unit (GRU); if the data has a spatial structure, you can choose a convolutional neural network (CNN); if the data scale is large and the feature dimension is high, you can consider using a deep neural network (DNN) or a residual network (ResNet), etc. According to the data scale and complexity, set the parameters of the neural network such as the number of layers, number of neurons, and activation function. The number of layers and the number of neurons should be determined according to the complexity and computing power of the model to avoid overfitting or underfitting. The activation function can be selected from ReLU, Sigmoid, Tanh, etc., and the most appropriate activation function should be selected according to the specific problem and network architecture. According to the actual situation, you can also consider whether to use regularization techniques (such as Dropout, L2 regularization, etc.) to prevent overfitting, whether to use batch normalization to speed up the training process and improve the stability of the model, and whether to use optimization algorithms (such as Adam, SGD, etc.) to optimize the model training process. Use the selected neural network architecture and set parameters to build the initial neural network model, use part of the training data to preliminarily train the model, and use the verification data to preliminarily verify the model. According to the verification results, adjust the network parameters or architecture to optimize the performance of the model.

[0069] Step 143, selecting features related to the moisture content of combustibles from the meteorological data as input data for the neural network model, specifically includes: Review the correlation analysis results obtained in step 142, especially those meteorological factors that are significantly correlated with the moisture content of combustibles. Determine which meteorological factors (such as temperature, humidity, precipitation, etc.) have a high correlation coefficient with the moisture content. Based on the correlation analysis results, select those meteorological factors that are closely related to the moisture content of combustibles and are easy to obtain and interpret in practical applications as input features of the model. Record the name, unit, and data source of the selected feature. Extract the data of the selected feature from the original meteorological data source, ensure the integrity and accuracy of the data, and organize the data into a format suitable for model input, such as time series data or tabular data. Check whether there are missing values ​​in the data set, record the number and location of missing values, and select an appropriate filling method based on the characteristics of the data and the distribution of missing values.

[0070] For example, for time series data, you can use forward filling, backward filling, linear interpolation, or a time series-based forecasting model to fill missing values. For non-time series data, you can consider using methods such as mean filling, median filling, or mode filling. Use statistical methods (such as box plots, Z-score, etc.) to identify outliers in the data; select appropriate processing methods based on the nature and source of outliers. If the outliers are caused by data entry errors or measurement errors, they can be corrected or deleted. If the outliers are caused by special events or extreme conditions, and these events or conditions are meaningful in model predictions, these outliers can be retained. Otherwise, consider using replacement (such as replacement with the mean, median, or interpolation) or deletion methods to handle outliers. Select an appropriate normalization or standardization method based on the distribution characteristics of the data and the requirements of the model. For example, if the data distribution is relatively uniform, you can use Min-Max normalization; if the data distribution has extreme values ​​or skewed distribution, you can use Z-score standardization.

[0071] Step 144, using historical data to train the neural network model to obtain a trained neural network model, specifically includes: collecting all available historical data, which should cover all features and labels (or target variables) that the model needs to learn. Processing missing values, outliers, duplicate values, etc. in the data to ensure data quality. Preprocessing work such as data type conversion, normalization or standardization may also be required.

[0072] Data partitioning: Training set: used to train the model, usually accounting for 60%-80% of the total data.

[0073] Validation set: used to evaluate the performance of the model during training, adjust hyperparameters (such as learning rate, batch size, etc.), and prevent overfitting. It usually accounts for 10%-20% of the total data.

[0074] Test set: used to finally evaluate the generalization ability of the model and ensure that the model performs well on unseen data. It usually accounts for 10%-20% of the total data.

[0075] Select a suitable neural network architecture (such as CNN, RNN, Transformer, etc.) according to the task type (such as classification, regression, sequence prediction, etc.) and data characteristics. Initialize the weights and biases of the neural network randomly or use the parameters of a pre-trained model for initialization. Select a suitable loss function (such as cross entropy loss, mean square error, etc.) according to the task type to measure the difference between the model prediction and the true value. Select an optimization algorithm (such as SGD, Adam, RMSprop, etc.) to update the model parameters according to the gradient of the loss function. Input the training data into the neural network and calculate the output value (prediction value) layer by layer. Use the loss function to calculate the difference between the predicted value and the true value (loss value). Calculate the gradient based on the loss value and propagate the gradient backward layer by layer using the chain rule. Use the optimizer to update the weights and biases of the neural network according to the gradient to reduce the loss value. Repeat the process of forward propagation, loss calculation, backpropagation, and parameter update until the preset number of training rounds is reached, or the loss value converges to a satisfactory level, or the model performance (such as accuracy, F1 score, etc.) is no longer significantly improved. After each training cycle (or a certain number of cycles), use the validation set to evaluate the performance of the model, such as calculating accuracy, recall, F1 score, area under the ROC curve (AUC) and other indicators. According to the performance of the validation set, adjust the learning rate, batch size, number of network layers, number of neurons and other hyperparameters to optimize the model performance. Use regularization techniques (such as L1 / L2 regularization, dropout, etc.) and early stopping strategies to prevent the model from overfitting. After the model training is completed, use the test set to perform a final evaluation of the model to ensure that the model has good generalization ability.

[0076] Step 145, verifying and optimizing the trained neural network model through cross-validation and grid search methods to obtain the final neural network model, specifically including: using a cross-validation method (such as K-fold cross-validation) to verify the model. Divide the training set into K subsets, use K-1 subsets for training each time, and the remaining subset is used for verification. Repeat this process K times, and select a different subset as the verification set each time. Evaluate the stability of the model by calculating the average performance of K verifications; use the grid search method to optimize the hyperparameters of the model (such as learning rate, number of network layers, number of neurons, etc.). Set a set of hyperparameter value ranges, and find the optimal parameter combination by traversing all possible combinations. In each traversal, use the cross-validation method to evaluate the performance of the model, and select the parameter combination with the best performance as the parameters of the final model.

[0077] Step 146, preprocessing the real-time meteorological data to obtain preprocessed meteorological data, specifically including: obtaining the latest meteorological data from real-time meteorological data sources (such as meteorological stations, meteorological satellites, etc.); checking whether the data has missing values, abnormal values ​​or error values, and performing corresponding processing (such as filling missing values, removing abnormal values, etc.); normalizing or standardizing the real-time meteorological data according to the statistical characteristics of historical data (such as mean, standard deviation, etc.) to ensure the consistency and comparability of the data; according to the features selected in step 143, extracting corresponding feature values ​​from the real-time meteorological data as input data of the model.

[0078] Step 147, inputting the preprocessed meteorological data into the final neural network model to obtain the moisture content prediction value of each type of vegetation, specifically including: loading the final neural network model obtained in step 145; using the preprocessed real-time meteorological data as the input data of the model; and calculating the output value of the model through forward propagation, that is, the moisture content prediction value of each type of vegetation.

[0079] In the embodiment of the present invention, step 141 ensures that the data required for model construction is fully covered, including the distribution of various types of vegetation and the meteorological factors related thereto, so as to provide a rich information basis for the model; using accurate and reliable vegetation distribution and meteorological data can improve the accuracy of subsequent model predictions; obtaining vegetation and meteorological data at the same time to ensure the consistency of the two in time and space is helpful to reveal the true relationship between them. Step 142, by analyzing the correlation between the moisture content of combustibles of different vegetation types and meteorological factors, the complex relationship between them can be understood more accurately; a neural network model is constructed according to the characteristics of the data, so that the model is more in line with the needs of practical problems and the accuracy and reliability of prediction are improved; the neural network model has strong learning ability and adaptability, can handle nonlinear relationships and complex patterns, and provides strong support for moisture content prediction. Step 143, by selecting meteorological features closely related to the moisture content of combustibles as input data, redundant information can be reduced, the operating efficiency and prediction accuracy of the model can be improved, and the optimized feature set makes the model more concise and clear; features are selected for specific vegetation types and meteorological conditions, so that model predictions are more targeted. Step 144, by training historical data, the neural network model can learn the historical laws and trends between vegetation moisture content and meteorological factors. During the training process, the model can be preliminarily verified and adjusted to ensure the performance and stability of the model. The trained model provides a basic framework and reference point for subsequent optimization and verification.

[0080] Step 145, through the cross-validation method, the stability and generalization ability of the model can be evaluated, and the risk of overfitting can be reduced. The grid search method can systematically search the model parameter space, find the optimal parameter combination, and improve the prediction performance of the model. The verified and optimized model is more reliable and can play a better role in practical applications. Step 146, through preprocessing steps (such as cleaning, denoising, normalization, etc.), the quality and availability of real-time meteorological data can be improved. The preprocessed meteorological data meets the model input requirements, ensuring that the model can run normally and accurately. High-quality data input can reduce errors and uncertainties in the model prediction process. Step 147, using the trained neural network model and preprocessed real-time meteorological data, the moisture content of various types of vegetation can be predicted in real time.

[0081] In a preferred embodiment of the present invention, the above step 15 may include: The population size is the number of individuals that evolve simultaneously in the genetic algorithm. This value needs to be determined based on the complexity of the problem and the computing resources. A larger population can increase diversity, but it also increases the amount of computation. Usually, the population size is set to tens to hundreds of individuals. Binary encoding is used to represent the weights of surface temperature, vegetation coverage, and vegetation moisture content as binary strings. For example, if each weight is represented by an 8-bit binary number, then a weight combination requires 24 bits (3 weights × 8 bits / weight). In this way, each individual can be represented by a 24-bit binary string. Determines how many proportions of individuals are generated through crossover operations when generating a new generation. It is usually set between 0.6 and 0.9. Determines the probability that individual genes mutate when generating a new generation. It is usually set between 0.01 and 0.1; the termination condition can be reaching the maximum number of iterations.

[0082] In a preferred embodiment of the present invention, the above step 16 may include: The fitness function is used to evaluate the quality of each individual (i.e., a set of weight parameters). In this problem, the fitness function may be defined based on the calculation results of the dynamic danger index of forest fire risk. For example, the danger index calculated using these weight parameters can be compared with the actual observed fire risk conditions, and the error or accuracy rate can be calculated as the fitness value. A higher fitness value indicates that this set of weight parameters is better.

[0083] In a preferred embodiment of the present invention, the above step 17 may include: According to the binary coding method, a group of individuals that meet the coding requirements are randomly generated as the initial population. Each individual represents a set of weight parameters. The fitness value of each individual is calculated using the fitness function. This value will be used for subsequent selection, crossover and mutation operations. According to the fitness value, the better individual is selected as the parent generation to generate a new generation of individuals. Common selection methods include roulette selection and tournament selection. According to the crossover rate, the parent generation individuals are selected for crossover operations to generate new individuals. The crossover operation is usually achieved by exchanging the gene fragments of the parent generation individuals. According to the mutation rate, the newly generated individuals are mutated, that is, some bits in the individual gene fragments are randomly changed, which helps to increase the diversity of the population and avoid falling into the local optimal solution. Through selection, crossover and mutation operations, a new generation of population is generated. This population will be used for the next round of iteration. Check whether the preset termination conditions (such as the maximum number of iterations, fitness threshold, etc.) are met. If the conditions are met, the iteration is stopped and the current optimal individual (that is, the weight parameter combination) is output; if the conditions are not met, it returns and the iteration process continues.

[0084] In a preferred embodiment of the present invention, the fitness function The calculation formula is: ; in, Represents the index of the sample point, which is used to traverse all sample points and calculate the difference between the predicted value and the true value; Represents the total number of samples, which is a positive integer, indicating that there are a total of samples; represents the weight parameter; Represents the total number of time steps in the time series, which is a positive integer and represents the length of the entire time series; represents the index of the time step; Indicates The predicted value of samples; Indicates The true value of samples; The exponent representing the error norm; Indicated in The actual data value observed at each time step; Indicated in The predicted value at each time step; , and They respectively represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk.

[0085] In an embodiment of the present invention, the first part of the function By calculating the predicted value With the true value The absolute error between The norm is calculated and averaged to evaluate the overall prediction accuracy of the model on all sample points. This helps ensure that the optimized weight parameters enable the model to better fit the actual observed data overall. The second part of the function The dynamic change information of time series is introduced, and the predicted value is calculated and the true value The average absolute error between the rate of change (i.e., derivative) over time is used to evaluate the model's ability to capture the dynamic changes in the time series. This helps to optimize the weight parameters so that the model can more accurately reflect the changing trend of forest fire risk over time. and Provides flexibility and adjustability. By adjusting the value of p, the sensitivity of the error assessment can be changed. For example, when p = 1, it represents the average value of the absolute error; when p = 2, it represents the mean square error. It is used to balance the contribution of the two errors in the total fitness and can be adjusted according to the needs of the actual problem. The genetic algorithm optimizes the weight parameter by maximizing (or minimizing) the fitness function. , and Since the fitness function comprehensively considers the prediction accuracy and the dynamic changes of the time series, the optimized weight parameters can more comprehensively reflect the impact of surface temperature, vegetation coverage and vegetation moisture content on the dynamic forest fire hazard index. By optimizing the weight parameters, the calculation accuracy and reliability of the dynamic forest fire hazard index can be improved.

[0086] In a preferred embodiment of the present invention, the calculation formula of the forest fire risk dynamic hazard index is: ; in, It represents the dynamic danger index of forest fire risk; represents the surface temperature; Indicates vegetation coverage; Indicates the moisture content of vegetation; and represents weight; and Respectively represent the minimum and maximum values ​​of the surface temperature; Indicates the maximum value of vegetation coverage; Indicates the minimum value of vegetation moisture content; Represents the base of natural logarithms.

[0087] In the embodiment of the present invention, the formula comprehensively considers the three factors that have an important impact on forest fire risk: surface temperature, vegetation coverage, and vegetation moisture content. Surface temperature is one of the important conditions for fire occurrence, and high temperature can easily cause fire; vegetation coverage reflects the density of the forest, and a lower vegetation coverage means a higher fire risk; and vegetation moisture content directly affects the flammability of vegetation. The lower the moisture content, the easier it is to burn. By combining these three factors, the formula can more comprehensively evaluate the degree of danger of forest fire risk. The surface temperature, vegetation coverage, and vegetation moisture content in the formula are all dynamically changing parameters, which change with time, season, and climate conditions. Therefore, the formula can dynamically reflect the changes in the degree of danger of forest fire risk over time, and provide timely and accurate information for the prediction and early warning of forest fires.

[0088] In a preferred embodiment of the present invention, the above step 19, classifying the fire risk level according to the forest fire risk dynamic hazard index, may include: Determine reasonable fire risk classification standards based on historical data, expert experience, and the actual needs of forest fire prevention and control. For example, the fire risk level can be divided into four levels: low, medium, high, and very high, each corresponding to a specific range of danger index.

[0089] Use the weight parameters optimized in the previous step , and As well as real-time or historical surface temperature, vegetation coverage and vegetation moisture content data, substitute them into the calculation formula of the dynamic forest fire hazard index to obtain the current hazard index. Compare the calculated hazard index with the predetermined fire hazard level classification standard. Determine the current fire hazard level based on the range of the hazard index. For example, if the hazard index is in the low risk range, the fire hazard level is "low"; if the hazard index is in the high risk range, the fire hazard level is "high" or "extremely high". Since parameters such as surface temperature, vegetation coverage and vegetation moisture content change dynamically, it is necessary to continuously monitor the changes in these parameters and regularly update the dynamic forest fire hazard index and fire hazard level classification. This helps to promptly detect the rising trend of fire risk and take corresponding preventive measures to reduce the risk of fire.

[0090] For example, suppose there is a set of fire risk classification standards as follows: Low risk: H < 0.3; Medium risk: 0.3≤H<0.6; High risk: 0.6≤H<0.9; Very high risk: H ≥ 0.9; Currently, the calculated forest fire risk dynamic hazard index H = 0.75. According to the classification standard, this value falls within the high risk range, so the current fire risk level is determined to be "high risk".

[0091] like Figure 2 As shown, an embodiment of the present invention further provides a forest fire prediction system using remote sensing images, comprising: The acquisition module is used to calculate the surface temperature based on the atmospheric transmittance, surface emissivity and radiation brightness value; The classification module is used to extract key data from the preprocessed remote sensing image; calculate the normalized vegetation index using the key data; generate a new image based on the normalized vegetation index; calculate the vegetation coverage rate based on the new image and the preset threshold; reconstruct the fused image by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then performing inverse wavelet transformation; determine the number of clusters based on the vegetation characteristics of the target area, extract and standardize key vegetation features from the fused image, perform cluster analysis on the key vegetation features, and finally identify the distribution data of various types of vegetation; A construction module is used to construct a fuel moisture content model for each type of vegetation based on the distribution data of each type of vegetation and in combination with meteorological factors, and calculate the moisture content of vegetation based on the fuel moisture content model; set the population size of the genetic algorithm, where the population size represents the number of individuals; determine the binary code used to represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk; set the crossover rate, mutation rate and termination condition of the genetic algorithm; define a fitness function for evaluating the quality of each individual, where each individual represents a set of weight parameters; randomly generate an initial population that meets the encoding method, where each individual represents a set of weight parameters; calculate the fitness of each individual based on the fitness function, and generate a new generation of populations through selection, crossover and mutation operations; determine whether the preset termination condition is met, and if the termination condition is met, stop the iteration and output the current final weight parameter combination; The calculation module is used to calculate the dynamic danger index of forest fire risk based on the final combination of weight parameters, combined with surface temperature, vegetation coverage and vegetation moisture content data; and classify the fire risk level according to the dynamic danger index of forest fire risk.

[0092] It should be noted that the system is a system corresponding to the above method, and all implementation methods in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0093] The embodiment of the present invention further provides a computing device, comprising: a processor, a memory storing a computer program, wherein when the computer program is executed by the processor, the method described above is executed. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0094] The embodiment of the present invention also provides a computer-readable storage medium storing instructions, which, when executed on a computer, enable the computer to execute the method described above. All implementations in the above method embodiment are applicable to this embodiment and can achieve the same technical effect.

[0095] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A forest fire prediction method using remote sensing images, characterized in that: The method comprises: Calculate the surface temperature based on the atmospheric transmittance, surface emissivity and radiation brightness value; Extract key data from preprocessed remote sensing images; use key data to calculate the normalized vegetation index; generate a new image based on the normalized vegetation index; calculate vegetation coverage based on the new image and a preset threshold; The fused image is reconstructed by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then undergoing inverse wavelet transformation. The number of clusters is determined according to the vegetation characteristics of the target area, and key vegetation features are extracted and standardized from the fused image. Cluster analysis is performed on the key vegetation features, and finally the distribution data of various types of vegetation are identified. According to the distribution data of various types of vegetation and combined with meteorological factors, the fuel moisture content model of various types of vegetation is constructed, and the moisture content of vegetation is calculated based on the fuel moisture content model; Set the population size of the genetic algorithm, which represents the number of individuals; determine the binary code used to represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic forest fire risk index; set the crossover rate, mutation rate and termination condition of the genetic algorithm; Define a fitness function for evaluating the quality of each individual, where each individual represents a set of weight parameters; Randomly generate an initial population that meets the encoding method, where each individual represents a set of weight parameters; calculate the fitness of each individual according to the fitness function, and generate a new generation of population through selection, crossover and mutation operations; determine whether the preset termination condition is met. If the termination condition is met, stop the iteration and output the current final weight parameter combination; According to the final combination of weight parameters, combined with the surface temperature, vegetation coverage and vegetation moisture content data, the dynamic forest fire risk index is calculated; Fire risk levels are classified according to the dynamic danger index of forest fire risk.

2. A forest fire prediction method using remote sensing images according to claim 1, characterized in that: Calculate the surface temperature based on the atmospheric transmittance, surface emissivity and radiation brightness value, including: Obtain processed remote sensing image data containing thermal infrared bands, identify and select two adjacent thermal infrared bands; determine the central wavelength of the two adjacent thermal infrared bands, the difference in radiation brightness between the bands, the atmospheric transmittance and the surface emissivity; The atmospheric transmittance of two adjacent thermal infrared bands is estimated by using the atmospheric radiation transfer model and combining the atmospheric profile data of the target area. According to the land cover type, the surface emissivity of two adjacent thermal infrared bands is estimated based on the inversion method of remote sensing data; The pixel values ​​of the remote sensing image are converted into radiation brightness values; the surface temperature is calculated based on the split window algorithm, combining the atmospheric transmittance, surface emissivity and radiation brightness values.

3. A forest fire prediction method using remote sensing images according to claim 2, characterized in that: Extract key data from preprocessed remote sensing images; calculate the normalized difference vegetation index using key data; Generate a new image based on the normalized vegetation index; Calculate vegetation coverage based on the new image and preset thresholds, including: Extracting red band and near infrared band data from the processed remote sensing image data; Calculate the normalized vegetation index value based on the data of the red band and the near-infrared band; Assign the normalized vegetation index value to each pixel to generate a new normalized vegetation index image; A threshold is set, and the number of pixels in the normalized vegetation index image that are greater than the set threshold is counted, where the ratio of the number of pixels greater than the set threshold to the total number of pixels is the vegetation coverage rate.

4. The forest fire prediction method using remote sensing images according to claim 3 is characterized in that: The fused image is reconstructed by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then performing inverse wavelet transform, including: Perform wavelet decomposition on the corrected panchromatic image and multispectral image to obtain wavelet coefficients of different scales and directions corresponding to the panchromatic image and the multispectral image, respectively. The wavelet coefficients include approximate coefficients and detail coefficients. According to the preset fusion rules, the wavelet coefficients of different scales and directions corresponding to the panchromatic image and the multispectral image are fused to obtain the fused wavelet coefficients; The fused wavelet coefficients are subjected to inverse wavelet transform to reconstruct the fused image.

5. A forest fire prediction method using remote sensing images according to claim 4, characterized in that: The number of clusters is determined according to the vegetation characteristics of the target area. The key vegetation features are extracted and standardized from the fused image. The key vegetation features are clustered and analyzed. Finally, the distribution data of various types of vegetation are identified, including: According to the vegetation type and distribution characteristics of the target area, determine the corresponding cluster number K value; Select key features reflecting vegetation characteristics from the fused image, including spectral reflectance and vegetation index; Standardize the key features to obtain standardized data; use the selected key features and the standardized data to apply the K-means clustering algorithm; initialize the cluster center, iteratively assign each pixel to the corresponding cluster center, and update the position of the cluster center until a preset number of iterations is reached to obtain a clustering result; Each clustering result is marked and matched with the actual vegetation type to obtain the distribution data of each type of vegetation.

6. A forest fire prediction method using remote sensing images according to claim 5, characterized in that: According to the distribution data of various types of vegetation and combined with meteorological factors, the fuel moisture content model of various types of vegetation is constructed, and the moisture content of vegetation is calculated based on the fuel moisture content model, including: Obtain distribution data of various types of vegetation, as well as meteorological data related to vegetation moisture content; Obtain the correlation between the moisture content of fuels of different vegetation types and meteorological factors, and build a neural network model based on the correlation and data characteristics; Select features related to moisture content of combustibles from meteorological data as input data for the neural network model; Using historical data to train a neural network model to obtain a trained neural network model; The trained neural network model is verified and optimized through cross-validation and grid search methods to obtain the final neural network model; Preprocessing the real-time meteorological data to obtain preprocessed meteorological data; The preprocessed meteorological data are input into the final neural network model to obtain the predicted values ​​of moisture content of various types of vegetation.

7. A forest fire prediction method using remote sensing images according to claim 6, characterized in that: Fitness function The calculation formula is: ； in, Represents the index of the sample point, which is used to traverse all sample points and calculate the difference between the predicted value and the true value; Represents the total number of samples, which is a positive integer, indicating that there are a total of samples; represents the weight parameter; Represents the total number of time steps in the time series, which is a positive integer and represents the length of the entire time series; represents the index of the time step; Indicates The predicted value of samples; Indicates The true value of samples; The exponent representing the error norm; Indicated in The actual data value observed at each time step; Indicated in The predicted value at each time step; , and They respectively represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk.

8. A forest fire prediction method using remote sensing images according to claim 7, characterized in that: The calculation formula of forest fire risk dynamic hazard index is: ； in, It represents the dynamic danger index of forest fire risk; represents the surface temperature; Indicates vegetation coverage; Indicates the moisture content of vegetation; and represents weight; and Respectively represent the minimum and maximum values ​​of the surface temperature; Indicates the maximum value of vegetation coverage; Indicates the minimum value of vegetation moisture content; Represents the base of natural logarithms.

9. A forest fire prediction system using remote sensing images, characterized in that: Applied to the method according to any one of claims 1 to 8, comprising: The acquisition module is used to calculate the surface temperature based on the atmospheric transmittance, surface emissivity and radiation brightness value; The classification module is used to extract key data from the preprocessed remote sensing image; calculate the normalized vegetation index using the key data; generate a new image based on the normalized vegetation index; calculate the vegetation coverage rate based on the new image and the preset threshold; reconstruct the fused image by performing wavelet decomposition and fusion on the corrected panchromatic image and multispectral image, and then performing inverse wavelet transformation; determine the number of clusters based on the vegetation characteristics of the target area, extract and standardize key vegetation features from the fused image, perform cluster analysis on the key vegetation features, and finally identify the distribution data of various types of vegetation; A construction module is used to construct a fuel moisture content model for each type of vegetation based on the distribution data of each type of vegetation and in combination with meteorological factors, and calculate the moisture content of vegetation based on the fuel moisture content model; set the population size of the genetic algorithm, where the population size represents the number of individuals; determine the binary code used to represent the weights of surface temperature, vegetation coverage and vegetation moisture content in the calculation of the dynamic danger index of forest fire risk; set the crossover rate, mutation rate and termination condition of the genetic algorithm; define a fitness function for evaluating the quality of each individual, where each individual represents a set of weight parameters; randomly generate an initial population that meets the encoding method, where each individual represents a set of weight parameters; calculate the fitness of each individual based on the fitness function, and generate a new generation of populations through selection, crossover and mutation operations; determine whether the preset termination condition is met, and if the termination condition is met, stop the iteration and output the current final weight parameter combination; The calculation module is used to calculate the dynamic danger index of forest fire risk based on the final combination of weight parameters, combined with surface temperature, vegetation coverage and vegetation moisture content data; and classify the fire risk level according to the dynamic danger index of forest fire risk.

10. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the method as claimed in any one of claims 1 to 8.

Images