Forest fire dynamic monitoring method and system based on multi-source data fusion

Abstract

The invention discloses a forest fire dynamic monitoring method and system based on multi-source data fusion, and relates to the technical field of fire dynamic monitoring, and the method comprises the steps: carrying out the calculation of the fire occurrence probability of a forest region, and outputting a dynamic risk thermodynamic diagram; non-uniform deployment of integrated sensing units is carried out to form a ground fire sensing network; large-range abnormal hot spot distribution is obtained; performing abnormal hot spot cooperative verification, and outputting a real-time fire risk alarm; matching a high-credibility unmanned aerial vehicle according to the fire risk alarm to execute real-time fire image acquisition; fire spreading modeling is carried out in combination with real-time meteorological data, and a fire spreading trend prediction result is obtained; and executing high-response fire extinguishing management and control of the forest area. According to the invention, the technical problem of low fire extinguishing management and control efficiency caused by inaccurate forest fire monitoring risk in the prior art is solved, and the technical effects of realizing dynamic monitoring and efficient management and control of forest fire and improving the efficiency and timeliness of fire extinguishing management and control are achieved.

Description

Technical Field

[0001] This invention relates to the field of dynamic fire monitoring technology, specifically to a method and system for dynamic forest fire monitoring based on multi-source data fusion. Background Technology

[0002] In the field of forest resource protection, the timeliness and accuracy of fire monitoring are crucial to reducing fire losses. Traditional forest fire monitoring methods mostly rely on a single data source (such as satellite remote sensing or ground sensors), which suffers from problems such as limited monitoring range, insufficient data accuracy, and delayed response. For example, although satellite remote sensing can achieve large-scale scanning, its resolution is low and it is greatly affected by weather, making it difficult to accurately identify early fires; ground sensors are costly to deploy and prone to monitoring blind spots in complex terrain, resulting in untimely fire warnings and failing to meet the high-response requirements of forest fire prevention and control.

[0003] Existing technologies suffer from inaccurate forest fire monitoring risks, leading to low efficiency in fire fighting and control. Summary of the Invention

[0004] This application provides a method and system for dynamic monitoring of forest fires based on multi-source data fusion, which is used to address the technical problem of inaccurate forest fire monitoring risks in existing technologies, leading to low efficiency in fire fighting and control.

[0005] In view of the above problems, this application provides a method and system for dynamic monitoring of forest fires based on multi-source data fusion.

[0006] The first aspect of this application provides a method for dynamic monitoring of forest fires based on multi-source data fusion, the method comprising: The probability of forest fire occurrence is calculated based on multi-source fire auxiliary data, and a dynamic risk heat map is output. Based on the risk gradient characteristics of the dynamic risk heat map, non-uniform deployment of integrated sensing units is implemented to form a ground-based fire sensing network. Short-cycle scanning of the forest area is performed using medium- and low-resolution satellites to obtain a large-scale distribution of abnormal hotspots. Based on this large-scale distribution of abnormal hotspots, the ground-based fire sensing network is locally triggered to perform collaborative verification of abnormal hotspots, outputting real-time fire risk alarms. After receiving the real-time fire risk alarms transmitted from the ground-based fire sensing network via a low-power wide-area network, the fire monitoring platform matches high-reliability UAVs to perform real-time fire image acquisition based on the fire risk alarms. After estimating fire boundary and fire intensity parameters based on the real-time fire images, fire spread modeling is performed in conjunction with real-time meteorological data to obtain fire spread trend prediction results. Resource scheduling optimization is performed on the fire spread trend prediction results to obtain real-time fire suppression strategies, and high-response fire suppression and control of the forest area is implemented.

[0007] A second aspect of this application provides a forest fire dynamic monitoring system based on multi-source data fusion, the system comprising: The heatmap output module is used to calculate the probability of forest fire occurrence based on multi-source fire auxiliary data and output a dynamic risk heatmap. The sensor network acquisition module is used to perform non-uniform deployment of integrated sensor units based on the risk gradient characteristics of the dynamic risk heatmap, forming a ground-based fire sensor network. The abnormal hotspot distribution acquisition module is used to obtain a large-scale abnormal hotspot distribution by performing short-period scanning of the forest area using medium- and low-resolution satellites. The fire risk alarm module is used to locally trigger the ground-based fire sensor network to perform abnormal hotspot collaborative verification based on the large-scale abnormal hotspot distribution, and output a real-time fire risk alarm. The system includes: a fire risk alarm module; a fire image acquisition module, used by the fire monitoring center to receive real-time fire risk alarms transmitted by the ground fire sensor network through a low-power wide area network, and then matching a high-reliability UAV to perform real-time fire image acquisition based on the fire risk alarms; a fire spread trend prediction module, used to estimate the fire boundary and fire intensity parameters based on the real-time fire images, and then combine real-time meteorological data to model the fire spread and obtain a fire spread trend prediction result; and a fire suppression and control module, used to optimize resource scheduling based on the fire spread trend prediction result to obtain a real-time fire suppression strategy and execute high-response fire suppression and control in the forest area.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: The probability of forest fires is calculated based on multi-source fire-related auxiliary data, and a dynamic risk heat map is output. A non-uniform deployment of integrated sensor units forms a ground-based fire sensor network. Short-cycle scanning of forest areas is performed using medium- and low-resolution satellites to obtain a wide range of abnormal hotspot distributions. The ground-based fire sensor network is locally triggered to perform collaborative verification of abnormal hotspots, outputting real-time fire risk alarms. Based on the fire risk alarms, high-reliability UAVs are matched to perform real-time fire image acquisition. After estimating fire boundary and fire intensity parameters based on the real-time fire images, fire spread modeling is performed in conjunction with real-time meteorological data to obtain fire spread trend prediction results. High-response fire suppression and control are implemented in the forest area. This achieves the technical effect of dynamic monitoring and efficient management of forest fires, improving the efficiency and timeliness of fire suppression and control. Attached Figure Description

[0009] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0010] Figure 1 A flowchart illustrating the forest fire dynamic monitoring method based on multi-source data fusion provided in this application embodiment; Figure 2 A schematic diagram of the structure of a forest fire dynamic monitoring system based on multi-source data fusion provided in an embodiment of this application.

[0011] Explanation of reference numerals in the attached diagram: Heat map output module 10, sensor network acquisition module 20, abnormal hotspot distribution acquisition module 30, fire risk alarm module 40, fire image acquisition module 50, fire spread trend prediction module 60, and fire extinguishing control module 70. Detailed Implementation

[0012] This application provides a method and system for dynamic monitoring of forest fires based on multi-source data fusion, which addresses the technical problem of inaccurate forest fire monitoring in existing technologies, leading to low efficiency in fire fighting and control.

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0014] Example 1, as Figure 1 As shown, this application provides a method for dynamic monitoring of forest fires based on multi-source data fusion, the method comprising: Step S100: Calculate the probability of forest fire occurrence based on multi-source fire auxiliary data and output a dynamic risk heat map.

[0015] Specifically, multi-source fire auxiliary data is decomposed to obtain fire precursor feature data (including air pollution data, land use data, and historical fire records) and satellite remote sensing data (including low- and medium-resolution thermal infrared data and high-resolution multispectral data). CO concentration and aerosol index are extracted from the air pollution data as air pollution features. Vegetation type encoding is performed on the land use data to generate land use classification features. Spatial recurrence frequency features are extracted from historical fire records. Surface temperature anomalies in the low- and medium-resolution thermal infrared data and the NDVI vegetation index calculated based on high-resolution multispectral data are used as remote sensing derived features. The forest area is divided into planar grids to obtain a unified geographic grid. Based on this grid, the above features are aligned and segmented to obtain a multidimensional feature vector array. The random forest algorithm (by adjusting the ratio of positive to negative samples in the gridded sample set to 1:3 through SMOTE oversampling, and compressing the decision tree depth to ≤10 layers after convergence training and pruning) is used to perform multi-threaded fire occurrence probability calculation on the multidimensional feature vector array, outputting a fire occurrence probability array. Then, the array is spatially clustered based on the watershed algorithm to generate a dynamic risk heat map.

[0016] Step S200: Based on the risk gradient characteristics of the dynamic risk heat map, perform non-uniform deployment of integrated sensing units to form a ground fire sensing network.

[0017] Specifically, based on the risk gradient characteristics presented by the dynamic risk heat map (i.e., the distribution of the probability of fire occurrence in different areas), the integrated sensing units are deployed non-uniformly. The integrated sensing units are densely deployed in areas with higher risk levels and sparsely deployed in areas with lower risk levels. The integrated sensing units specifically integrate infrared thermal imaging sensors, smoke particulate sensors, and meteorological micro-stations (used to collect local wind speed and local wind direction data). This differentiated deployment method forms a ground fire sensing network covering the entire forest area, ensuring the monitoring density and sensitivity of high-risk areas, while taking into account the overall monitoring range and resource allocation efficiency.

[0018] Step S300: Conduct short-period scans of the forest area using medium- and low-resolution satellites to obtain the distribution of large-scale anomalous hotspots.

[0019] Specifically, the fire monitoring center receives short-cycle raw data streams of forest areas transmitted back by medium- and low-resolution satellites. After extracting infrared bands from the data stream, it performs radiometric calibration and outputs a brightness-temperature raster map. It retrieves historical fire point temperature characteristics from historical fire records and sets a dynamic threshold distribution based on the vegetation distribution attributes of the forest area. It then uses this dynamic threshold distribution to slide through the brightness-temperature raster map, identifies and marks areas that meet the fire point characteristics, and finally outputs a large-scale distribution of abnormal hotspots in vector format.

[0020] Step S400: Based on the large-scale abnormal hotspot distribution, the local area triggers the ground fire sensor network to perform abnormal hotspot collaborative verification and outputs a real-time fire risk alarm.

[0021] Specifically, the distribution of large-scale abnormal hotspots is analyzed, and the center coordinates of multiple hotspot clusters are extracted (with the influence radius and average brightness and temperature values ​​of the hotspot clusters as identifiers). This distribution is projected onto a ground-based fire sensing network, and the fire sensing network covering the hotspot area is selected. Using the influence radius and center coordinates of multiple hotspot clusters as segmentation constraints, the covering fire sensing network is divided into multiple local fire sensing networks. These local networks are activated to transmit multiple sets of multi-source resampling data. The baseline smoke concentration characteristics are first applied to initialize and verify the multiple sets of data, and then multiple average brightness and temperature values ​​are combined for mapping verification. Finally, a real-time fire risk alarm is output.

[0022] Step S500: After receiving the real-time fire risk alarm transmitted by the ground fire sensor network through the low-power wide area network, the fire monitoring center matches a high-reliability UAV to perform real-time fire image acquisition based on the fire risk alarm.

[0023] Specifically, the fire monitoring platform receives real-time fire risk alarms transmitted from the ground fire sensor network via a low-power wide-area network. These alarms include the location coordinates, risk level, and relevant monitoring parameters of abnormal hotspots. Based on the location information, area range, and urgency of the abnormal hotspots in the alarms, the platform selects highly reliable drones (such as those with good equipment condition, sufficient battery life, and suitability for the terrain environment) from the drone swarm. The successfully matched drones are then dispatched to the target area to perform real-time fire image acquisition according to a preset route, obtaining image data containing real-time fire conditions and transmitting it back to the fire monitoring platform.

[0024] Step S600: After estimating the fire boundary and fire intensity parameters based on the real-time fire images, fire spread modeling is performed in conjunction with real-time meteorological data to obtain the fire spread trend prediction results.

[0025] Specifically, based on the regional coverage of real-time fire risk warnings, real-time meteorological data (including local wind speed and direction) is retrieved from multiple sets of multi-source recollection data transmitted from the ground fire sensor network; real-time fire images are stitched together based on the flight trajectory of drones, fire isotherms are extracted and converted into continuous fire lines to determine the fire boundary; the fire intensity parameters are quantitatively output by solving the thermal imaging temperature distribution of real-time fire images; starting from the fire boundary, the fire intensity parameters are simulated over time based on real-time meteorological data and topographic slope data of forest areas, and finally the predicted results of the fire spread trend within the preset rescue blocking window are obtained.

[0026] Step S700: Optimize resource allocation based on the fire spread trend prediction results to obtain a real-time fire suppression strategy, and execute high-response fire suppression control in the forest area.

[0027] Specifically, based on the predicted fire spread trend (including changes in the fire spread path, range, and intensity within the pre-set rescue blocking window), and taking into account the terrain conditions of the forest area, the distribution and capabilities of existing firefighting resources (such as the location of firefighting teams, the number of equipment, and the distribution of water sources), resource allocation is optimized. By evaluating the efficiency and effectiveness of different resource allocation schemes in controlling the fire, the optimal scheme is selected to form a real-time firefighting strategy. Based on this real-time firefighting strategy, the corresponding firefighting resources are quickly dispatched to the target area to carry out targeted firefighting operations, achieving high-response firefighting control of forest fires and minimizing fire losses.

[0028] In one possible implementation, step S100 further includes: Step S110: Decompose the multi-source fire auxiliary data to obtain fire precursor feature data and satellite remote sensing data, wherein the fire precursor feature data includes air pollution data, land use data and historical fire records, and the satellite remote sensing data includes medium and low resolution thermal infrared data and high resolution multispectral data.

[0029] Step S120: Using the multi-source fire auxiliary data and satellite remote sensing data as input feature vectors, the probability of forest fire occurrence is calculated using the random forest algorithm, and the output is a dynamic risk heat map.

[0030] Specifically, the multi-source fire auxiliary data is decomposed to separate fire precursor feature data and satellite remote sensing data. The fire precursor feature data includes air pollution data (from which CO concentration, aerosol index, and other information can be extracted), land use data (which requires vegetation type coding), and historical fire records (used to extract spatial recurrence frequency characteristics). The satellite remote sensing data includes medium- and low-resolution thermal infrared data (from which surface temperature anomalies can be extracted) and high-resolution multispectral data (used to calculate the NDVI vegetation index).

[0031] The multi-source fire auxiliary data (including air pollution data, land use data, and historical fire records) and satellite remote sensing data (including medium- and low-resolution thermal infrared data and high-resolution multispectral data) obtained from the decomposition are integrated into the input feature vector. CO concentration and aerosol index are extracted from the air pollution data as air pollution features. Vegetation type encoding is performed on the land use data to generate land use classification features. Spatial recurrence frequency features are extracted from historical fire records. Surface temperature anomalies from medium- and low-resolution thermal infrared data and NDVI vegetation index calculated from high-resolution multispectral data are used as remote sensing derived features. The forest area is divided into a planar grid to obtain a unified geographic grid. Based on this grid, the above features are aligned and segmented to form a multi-dimensional feature vector array. The random forest algorithm is used to train the model with a gridded sample set with a positive to negative sample ratio adjusted to 1:3 by SMOTE oversampling (decision tree depth ≤ 10 layers after pruning). Multi-threaded fire occurrence probability calculation is performed on the multi-dimensional feature vector array to output the fire occurrence probability array. Then, the watershed algorithm is used to perform spatial clustering on the array to finally generate a dynamic risk heat map.

[0032] In one possible implementation, step S120 further includes: Step S121: Extract CO concentration and aerosol index from the air pollution data as air pollution characteristics.

[0033] Step S122: Encode the vegetation type of the land use data to generate land use classification features.

[0034] Step S123: Extract spatial recurrence frequency features from historical fire records.

[0035] Step S124: The surface temperature anomaly extracted from the low-to-medium resolution thermal infrared data and the NDVI vegetation index calculated based on the high-resolution multispectral data are used as remote sensing derived features.

[0036] Step S125: Divide the forest area into planar grids to obtain a unified geographic grid. Then, based on the unified geographic grid, align and segment the atmospheric pollution features, land use classification features, spatial recurrence frequency features, and remote sensing derived features to obtain a multidimensional feature vector array.

[0037] Step S126: Based on the random forest model, perform multi-threaded fire occurrence probability calculation on the multi-dimensional feature vector array and output the fire occurrence probability array.

[0038] Step S127: Spatial clustering of the fire occurrence probability array is performed based on the watershed algorithm to generate the dynamic risk heat map.

[0039] Specifically, from the air pollution data in the multi-source fire auxiliary data, key indicators that can reflect the correlation between the degree of air pollution and potential fires are extracted, namely CO concentration (carbon monoxide concentration, whose abnormal increase may be related to the combustion process) and aerosol index (reflecting the distribution and concentration of aerosol particles in the atmosphere, which can help judge the smoke diffusion situation). These two indicators are used together as air pollution characteristics to provide basic characteristic data for the subsequent calculation of the probability of forest fires.

[0040] For land use data in multi-source fire auxiliary data, systematic coding is carried out according to the different types of vegetation (such as coniferous forest, broad-leaved forest, shrub forest, etc.) and distribution characteristics in the forest area. Each type of vegetation is assigned a unique identification code through preset classification rules, thereby generating land use classification features that can clearly distinguish different vegetation types. This provides structured feature data support for subsequent analysis of fire occurrence probability in combination with vegetation attributes.

[0041] From historical fire records in multi-source fire monitoring data, key information reflecting the spatial recurrence pattern of fires within forest areas is extracted, namely, the spatial recurrence frequency characteristic. This characteristic is calculated by statistically analyzing the number of fires and the time intervals between them in a specific geographical unit (such as a rasterized forest sub-region) over a historical period. This calculates the fire recurrence frequency value for each unit, thereby quantifying the probability of fires recurring in different areas. This provides characteristic data reflecting the spatial distribution pattern of historical fires for subsequent random forest algorithms to calculate the probability of fire occurrence.

[0042] From the low-to-medium resolution thermal infrared data in satellite remote sensing data, surface temperature anomalies that reflect deviations from the normal range in forest areas are extracted (abnormally high temperatures may indicate potential fires). Simultaneously, based on the high-resolution multispectral data in satellite remote sensing data, the ratio of near-infrared to red reflectance (i.e., the NDVI vegetation index, used to characterize vegetation cover and growth status; vegetation density is related to the risk of fire spread) is calculated. The aforementioned surface temperature anomalies and the NDVI vegetation index are used together as remote sensing derived features to provide key feature data from satellite remote sensing for subsequent calculations of the probability of forest fires.

[0043] The forest area is divided into planar grids, creating a unified geographic grid (each grid has a clear spatial coordinate boundary). Then, based on the spatial coordinate system of this unified geographic grid, the extracted atmospheric pollution features (CO concentration, aerosol index), land use classification features (vegetation type coding results), spatial recurrence frequency features (historical fire spatial recurrence statistics), and remote sensing derived features (surface temperature anomalies, NDVI vegetation index) are aligned to ensure that each feature is spatially matched. These features are then segmented by grid unit, so that each grid unit corresponds to a dataset containing the above four types of features. Finally, these features are integrated to form a multidimensional feature vector array covering the entire forest area, providing spatially aligned multidimensional input data for subsequent fire probability calculations.

[0044] A multi-threaded fire probability calculation is performed on a multi-dimensional feature vector array based on a random forest model, outputting a fire probability array. First, a gridded sample set labeled with fire probability is prepared using historical monitoring data, and the ratio of positive to negative samples is adjusted to 1:3 using SMOTE oversampling technology. Multi-source feature engineering is performed on historical precursor feature data to obtain a standardized feature matrix. Random forest parameters are configured (500 decision trees, maximum tree depth of 15 layers, minimum number of samples per node of 5, and Gini coefficient used as the node splitting criterion). The model is then trained using the standardized feature matrix and the gridded sample set to convergence based on out-of-bag error monitoring, outputting an initial forest model. The decision tree depth is then compressed to ≤10 layers using a pruning algorithm, persisting as a lightweight fire probability prediction model. Finally, multi-threaded parallel computation is performed on the multi-dimensional feature vector array based on this model to obtain the fire probability of each geographic grid unit, which is then integrated to form the fire probability array.

[0045] A dynamic risk heat map is generated by spatially clustering a fire occurrence probability array based on the watershed algorithm. The fire occurrence probability array is treated as a "terrain surface" containing different probability values, with the fire occurrence probability of each grid cell used as "altitude." The watershed algorithm identifies continuous areas with similar probability values ​​(i.e., risk units formed by clustering), and assigns corresponding visual distinguishing markers (such as color gradients) based on the probability value range of each area. The result is a dynamic risk heat map that intuitively displays the spatial distribution of different fire risk levels within a forest area, clearly showing the boundaries and ranges of high-risk, medium-risk, and low-risk areas.

[0046] In one possible implementation, step S126 further includes: Step S1261: Prepare a gridded sample set with fire probability labels based on historical monitoring data, wherein the ratio of positive to negative samples in the gridded sample set is adjusted to 1:3 using SMOTE oversampling technology.

[0047] Step S1262: Perform multi-source feature engineering on the historical precursor feature data to obtain a standardized feature matrix.

[0048] Step S1263: After configuring the random forest parameters of the random forest model, use the standardized feature matrix and the gridded sample set as training data to perform convergence training on the random forest model based on out-of-bag error monitoring, and output the initial forest model.

[0049] Step S1264: After compressing the decision tree depth of the initial forest model to ≤10 layers using a pruning algorithm, the initial forest model is persisted as a lightweight fire probability prediction model.

[0050] Step S1265: Based on the fire occurrence probability prediction model, perform multi-threaded fire occurrence probability calculation on the multi-dimensional feature vector array and output the fire occurrence probability array.

[0051] Specifically, based on historical monitoring data, a gridded sample set with fire probability labels is prepared. Each sample corresponds to a geographic grid cell after the forest area is divided into grids, and the label is the probability of fire occurrence in that cell in history. To solve the problem of sample class imbalance, the SMOTE oversampling technique is used to process the sample set. By artificially synthesizing minority class samples (i.e., grid cell samples that have experienced fires in the past), the ratio of positive samples (grids that have experienced fires) to negative samples (grids that have not experienced fires) in the gridded sample set is adjusted to 1:3, thereby providing balanced sample data support for the subsequent training of the random forest model.

[0052] Multi-source feature engineering is performed on historical precursor feature data (covering multiple sources such as air pollution, land use, historical fire records, and satellite remote sensing data). This includes filling missing values, correcting outliers, and normalizing or standardizing features of different magnitudes. At the same time, non-numerical features are transformed into numerical features through feature transformation (such as encoding and normalization). Finally, a standardized feature matrix with a unified structure and consistent scale is formed, providing standardized and effective input data for training random forest models.

[0053] Configure the relevant parameters of the random forest model, including setting the number of decision trees to 500, the maximum tree depth to 15 layers, the minimum number of samples per node to 5, and using the Gini coefficient as the node splitting criterion. Then, input the processed standardized feature matrix and the gridded sample set as training data into the random forest model. The model is trained to convergence by using out-of-bag error monitoring, that is, using out-of-bag samples that did not participate in the decision tree training to evaluate the model performance. The model parameters are continuously adjusted until the out-of-bag error stabilizes within the preset range. After training is completed, the initial forest model is output.

[0054] The initial forest model trained using a pruning algorithm is optimized by removing redundant branch nodes from the decision trees and compressing the depth of each decision tree to ≤10 layers to reduce model complexity and computational resource consumption. After pruning and compression, the optimized initial forest model is persistently stored to form a lightweight fire probability prediction model. This model retains the necessary predictive capabilities while being easy to load and run quickly and efficiently in practical applications, providing reliable and efficient model support for subsequent fire probability calculations.

[0055] By using a fire occurrence probability prediction model, the multi-dimensional feature vector array obtained by aligning and segmenting through a unified geographic grid is processed. A multi-threaded parallel computing approach is adopted to solve the fire occurrence probability of the feature vectors of multiple grid cells simultaneously to improve computational efficiency. After calculation by the model, each grid cell will obtain a corresponding fire occurrence probability value. These probability values ​​are integrated according to the spatial location of the grid cells to form a fire occurrence probability array covering the entire forest area, providing a data foundation for the subsequent generation of dynamic risk heat maps.

[0056] In one possible implementation, step S300 further includes: Step S310: The fire monitoring center receives the short-cycle raw data stream transmitted back by the medium- and low-resolution satellite.

[0057] Step S320: After extracting the infrared band from the short-cycle raw data stream, perform radiometric calibration and output a brightness-temperature raster map.

[0058] Step S330: After retrieving the historical fire point temperature characteristics from the historical fire records, set a dynamic threshold distribution based on the historical fire point temperature characteristics and the vegetation distribution attributes of the forest area.

[0059] Step S340: Apply the dynamic threshold distribution to slide through the brightness temperature raster to identify fire points and output the large-scale abnormal hotspot distribution in vector format.

[0060] Specifically, the fire monitoring center is responsible for receiving short-cycle raw data streams transmitted back from medium- and low-resolution satellites. These data streams contain raw observation information of the forest area collected by satellite remote sensing technology in a short period of time, providing basic data support for subsequent processing and analysis of the distribution of abnormal hotspots in the forest area.

[0061] The short-period raw data stream transmitted from medium- and low-resolution satellites is processed to extract infrared band information related to temperature monitoring, focusing on spectral data that reflects the thermal state of the Earth's surface. Subsequently, radiometric calibration is performed on the extracted infrared band data to eliminate interference factors such as sensor errors and atmospheric effects, converting the radiance values ​​into physically meaningful brightness-temperature values. Finally, the processed brightness-temperature values ​​are rasterized according to the geographic coordinates of the forest area to form a brightness-temperature raster map, which visually presents the spatial distribution of temperature in the region.

[0062] First, retrieve historical fire point temperature characteristics from historical fire records. These characteristics include key information such as the temperature range and temperature change patterns during past fires. Then, combine this with the vegetation distribution attributes of the forest area, such as the differences in ignition points and combustion characteristics of different vegetation types (coniferous forests, broad-leaved forests, shrubs, etc.), and comprehensively set a dynamic threshold distribution. This allows the threshold to be adaptively adjusted according to the vegetation conditions and historical fire point temperature characteristics of different areas, ensuring the accuracy of subsequent fire point judgments.

[0063] Using a pre-defined dynamic threshold distribution, the brightness-temperature grid map is traversed cell by cell in a sliding window manner. The brightness-temperature value of each cell is compared with the dynamic threshold at the corresponding location. If the brightness-temperature value exceeds the dynamic threshold, the cell is identified as a potential fire point and marked. After traversing and judging the entire brightness-temperature grid map, all marked potential fire points are integrated to form a large-scale abnormal hotspot distribution in vector format. This vector format data can accurately describe the spatial location and range of abnormal hotspots, providing a clear target area for subsequent collaborative verification of the ground fire sensing network.

[0064] In one possible implementation, step S200 further includes: The integrated sensing unit specifically integrates an infrared thermal imaging sensor, a smoke particulate sensor, and a meteorological micro-station, wherein the meteorological micro-station is used to collect local wind speed data and local wind direction data.

[0065] Specifically, the integrated sensing unit is a key device in the forest fire dynamic monitoring system used for ground-based monitoring. It integrates three core sensing components: first, an infrared thermal imaging sensor, which generates thermal images by capturing the infrared radiation of objects, thereby monitoring the temperature distribution within the area in real time and promptly detecting abnormally high-temperature points; second, a smoke particulate sensor, which detects the concentration of smoke particulate matter in the air, providing a basis for determining the presence of a fire; and third, a meteorological micro-station, specifically designed to collect local wind speed and direction data around the monitoring point. This meteorological data plays a crucial role in analyzing the potential direction and speed of fire spread. These three sensors work together to provide multi-dimensional monitoring data support for the ground-based fire sensing network.

[0066] In one possible implementation, step S400 further includes: Step S410: Analyze the large-scale abnormal hotspot distribution and extract the center coordinates of multiple hotspot clusters, wherein the center coordinates of the hotspot clusters are labeled with the influence radius and average brightness temperature value of the hotspot clusters.

[0067] Step S420: Project the large-scale abnormal hotspot distribution onto the ground fire sensing network and filter the fire sensing network to cover it.

[0068] Step S430: Using the influence radius of multiple hotspot clusters and the center coordinates of the multiple hotspot clusters as segmentation constraints, the fire coverage sensor network is segmented into multiple local fire sensor networks.

[0069] Step S440: Activate the multiple local fire sensor networks to transmit multiple sets of multi-source recollection data.

[0070] Step S450: After initializing and verifying the multiple sets of multi-source resampling data using the baseline smoke concentration characteristics, the multiple sets of multi-source resampling data are mapped and verified using multiple average brightness and temperature values, and the real-time fire risk alarm is output.

[0071] Specifically, the large-scale abnormal hotspot distribution obtained through satellite scanning is analyzed and processed to identify multiple hotspot clusters that are clustered together. The center coordinates of each hotspot cluster are extracted to determine its spatial location. At the same time, the corresponding hotspot cluster influence radius (used to characterize the spatial coverage of the hotspot cluster) and average brightness temperature value (used to reflect the overall temperature characteristics of the hotspot cluster) are attached to each center coordinate, thereby providing accurate target area information for the local triggering and collaborative verification of the subsequent ground fire sensing network.

[0072] The distribution of large-scale abnormal hotspots obtained by satellite is projected onto the coverage area of ​​the ground fire sensing network according to the spatial coordinate correspondence. By comparing the spatial location of the abnormal hotspots with the deployment location of each integrated sensing unit in the sensing network, those sensing units and their sub-networks that can cover these abnormal hotspot areas are selected, i.e., the fire sensing network is covered, so as to carry out accurate monitoring and verification of abnormal hotspot areas in the future.

[0073] Using the extracted influence radius of multiple hotspot clusters and the corresponding center coordinates of the hotspot clusters as the basis and constraint, the selected fire sensing network is spatially divided so that each segmented sub-network can cover one or more hotspot cluster areas, thereby forming multiple independent local fire sensing networks. This allows each local network to perform targeted and accurate monitoring and data collection on the hotspot cluster areas it covers.

[0074] Activation commands are issued to the multiple local fire sensing networks formed after segmentation, prompting the integrated sensing units (including infrared thermal imaging sensors, smoke particulate sensors, and meteorological micro-stations) in these local networks to start working and re-collect relevant data within their coverage areas, including multi-dimensional information such as temperature distribution, smoke concentration, local wind speed, and wind direction. Subsequently, these multi-set, multi-source recollected data are transmitted back to the fire monitoring center to provide detailed ground monitoring data support for subsequent abnormal hotspot verification.

[0075] The baseline smoke concentration characteristics were used to initialize and verify multiple sets of multi-source recollected data transmitted from multiple local fire sensor networks. This involved comparing the smoke concentration values ​​in the recollected data with preset baseline smoke concentration characteristics (such as typical smoke concentration ranges at different fire stages) to eliminate abnormal data that significantly deviated from the baseline, ensuring the basic validity of the data. Next, multiple average brightness and temperature values ​​extracted from a large-scale distribution of abnormal hotspots were used to map and verify the initialized data. The temperature information in the recollected data was correlated and compared with the average brightness and temperature values ​​of the corresponding hotspot clusters to further confirm the accuracy and consistency of the data. Finally, the results of the two verifications were combined to generate and output a real-time fire risk alarm that accurately reflects the fire situation in the forest area.

[0076] In one possible implementation, step S600 further includes: Step S610: Based on the regional coverage of the real-time fire risk alarm, retrieve the real-time meteorological data from the multiple sets of multi-source recollection data.

[0077] Step S620: After stitching the real-time fire image based on the flight trajectory, extract the fire isotherms, perform continuous fire line conversion, and output the fire boundary.

[0078] Step S630: Solve the thermal imaging temperature distribution based on the real-time fire image, and quantify and output the fire intensity parameters.

[0079] Step S640: Starting from the fire boundary, perform a time-progression simulation of the fire intensity parameters based on the real-time meteorological data and the topographic slope data of the forest area, and output the predicted fire spread trend within the preset rescue blocking window.

[0080] Specifically, the first step is to define the forest area covered by the real-time fire risk warning. Then, based on this area, real-time meteorological data for the corresponding area is accurately selected and retrieved from multiple sets of multi-source recollected data transmitted from multiple local fire sensor networks. This data mainly includes information such as local wind speed and local wind direction collected by meteorological micro-stations, providing key meteorological parameter support for subsequent fire spread modeling.

[0081] By recording the GPS flight trajectory of the drone, multiple frames of real-time fire images are aligned and stitched together according to the shooting position and angle to form a complete panoramic image of the fire scene. Then, thermal imaging analysis technology is used to extract isotherms of different temperatures from the stitched images, and key isotherms that characterize the edge of the fire scene are identified and screened. Subsequently, curve fitting and interpolation algorithms are used to smooth the discrete characteristic isotherms, eliminate discontinuities and convert them into continuous closed fire lines, thereby accurately defining the spatial range of fire spread, and finally outputting fire scene boundary data in vector format.

[0082] Infrared thermal imaging analysis technology is used to process real-time fire images. Through radiometric calibration, the grayscale values ​​of the images are converted into actual temperature values ​​to construct a thermal imaging temperature distribution matrix of the fire area. Subsequently, based on this matrix, indicators such as the average temperature, maximum temperature, and area ratio of high-temperature regions of the fire are calculated. Combined with a preset temperature-intensity mapping model (such as corresponding different temperature ranges to different fire intensity levels), these indicators are quantified and converted into specific fire intensity parameters, and finally output in numerical form.

[0083] The Long Short-Term Memory (LSTM) network algorithm is adopted, with the fire boundary as the initial spatial constraint. Real-time meteorological data (wind speed, wind direction) and terrain slope data are normalized and combined with fire intensity parameters to form a multi-dimensional temporal input vector. The input vector is iteratively calculated at time steps by the trained LSTM model. The hidden layer of the model captures the long-term dependencies (such as the fire acceleration trend under the continuous effect of wind speed) and short-term abrupt features (such as the sudden increase in intensity in steep terrain) in the fire spread process through the gating mechanism. Within the preset rescue blocking time window (such as 1 hour, 3 hours), the quantitative results such as the fire boundary expansion range and fire intensity distribution at each time node are output, forming a complete fire spread trend prediction sequence.

[0084] In one possible implementation, step S1263 further includes: The random forest parameters include 500 decision trees, a maximum tree depth of 15 layers, and a minimum number of samples per node of 5. The Gini coefficient is used as the node splitting criterion.

[0085] Specifically, the parameters of the random forest are set as follows: the number of decision trees is set to 500, which improves the generalization ability and prediction accuracy of the model through ensemble learning of multiple trees; the maximum tree depth is limited to 15 layers to avoid overfitting the training data and ensure the prediction effect of the model on new data; the minimum number of samples per node is 5, that is, when the number of samples of a node is less than 5, it will not be split, ensuring that the decision of each node has certain statistical significance; the node splitting criterion adopts the Gini coefficient, which is used to measure the impurity of the sample set. By selecting the feature that reduces the Gini coefficient the most, the optimal decision tree structure is constructed.

[0086] Example 2, based on the same inventive concept as the forest fire dynamic monitoring method based on multi-source data fusion in the previous examples, such as... Figure 2 As shown, this application provides a forest fire dynamic monitoring system based on multi-source data fusion. The system and method embodiments in this application are based on the same inventive concept. The system includes: The heat map output module 10 is used to calculate the probability of forest fire occurrence based on multi-source fire auxiliary data and output a dynamic risk heat map.

[0087] The sensor network acquisition module 20 is used to perform non-uniform deployment of integrated sensor units based on the risk gradient characteristics of the dynamic risk heat map to form a ground fire sensing network.

[0088] The abnormal hotspot distribution acquisition module 30 is used to obtain the distribution of large-scale abnormal hotspots by performing short-period scanning of forest areas using medium- and low-resolution satellites.

[0089] The fire risk alarm module 40 is used to trigger the ground fire sensor network to perform collaborative verification of abnormal hotspots based on the distribution of the large-scale abnormal hotspots, and output a real-time fire risk alarm.

[0090] The fire image acquisition module 50 is used by the fire monitoring center to receive the real-time fire risk alarm transmitted by the ground fire sensor network through a low-power wide area network, and then match a high-reliability UAV to perform real-time fire image acquisition based on the fire risk alarm.

[0091] The fire spread trend prediction module 60 is used to estimate the fire boundary and fire intensity parameters based on the real-time fire images, and then combine the real-time meteorological data to perform fire spread modeling and obtain the fire spread trend prediction results.

[0092] The fire suppression and control module 70 is used to optimize resource allocation based on the fire spread trend prediction results to obtain real-time fire suppression strategies and execute high-response fire suppression and control in the forest area.

[0093] Furthermore, the system is also used to implement the following functions: The multi-source fire auxiliary data is decomposed to obtain fire precursor feature data and satellite remote sensing data. The fire precursor feature data includes air pollution data, land use data, and historical fire records. The satellite remote sensing data includes medium- and low-resolution thermal infrared data and high-resolution multispectral data. The multi-source fire auxiliary data and satellite remote sensing data are used as input feature vectors, and the probability of forest fire occurrence is calculated using a random forest algorithm. The output is a dynamic risk heat map.

[0094] Furthermore, the system is also used to implement the following functions: CO concentration and aerosol index are extracted from the air pollution data as air pollution features; vegetation type is encoded from the land use data to generate land use classification features; spatial recurrence frequency features are extracted from historical fire records; surface temperature anomalies extracted from the low- and medium-resolution thermal infrared data and the NDVI vegetation index calculated based on the high-resolution multispectral data are used as remote sensing derived features; the forest area is divided into planar grids to obtain a unified geographic grid, and the air pollution features, land use classification features, spatial recurrence frequency features, and remote sensing derived features are aligned and segmented based on the unified geographic grid to obtain a multidimensional feature vector array; based on the random forest model, multi-threaded fire occurrence probability calculation is performed on the multidimensional feature vector array to output a fire occurrence probability array; spatial clustering is performed on the fire occurrence probability array based on the watershed algorithm to generate the dynamic risk heat map.

[0095] Furthermore, the system is also used to implement the following functions: A gridded sample set labeled with fire probability is prepared based on historical monitoring data, wherein the ratio of positive to negative samples in the gridded sample set is adjusted to 1:3 using SMOTE oversampling technology; multi-source feature engineering is performed on historical precursor feature data to obtain a standardized feature matrix; after configuring the random forest parameters of the random forest model, the standardized feature matrix and the gridded sample set are used as training data to perform convergence training on the random forest model based on out-of-bag error monitoring, and an initial forest model is output; after compressing the decision tree depth of the initial forest model to ≤10 layers using a pruning algorithm, the initial forest model is persisted as a lightweight fire occurrence probability prediction model; based on the fire occurrence probability prediction model, multi-threaded fire occurrence probability calculation is performed on the multi-dimensional feature vector array, and the fire occurrence probability array is output.

[0096] Furthermore, the system is also used to implement the following functions: The fire monitoring center receives short-period raw data streams transmitted back from the medium- and low-resolution satellites; after extracting infrared bands from the short-period raw data streams, it performs radiometric calibration and outputs a brightness-temperature raster map; after retrieving historical fire point temperature characteristics from historical fire records, it sets a dynamic threshold distribution based on the historical fire point temperature characteristics and the vegetation distribution attributes of the forest area; it applies the dynamic threshold distribution to slide through the brightness-temperature raster map to identify fire points and outputs the large-scale abnormal hotspot distribution in vector format.

[0097] Furthermore, the system is also used to implement the following functions: The integrated sensing unit specifically integrates an infrared thermal imaging sensor, a smoke particulate sensor, and a meteorological micro-station, wherein the meteorological micro-station is used to collect local wind speed data and local wind direction data.

[0098] Furthermore, the system is also used to implement the following functions: The large-scale abnormal hotspot distribution is analyzed, and the center coordinates of multiple hotspot clusters are extracted. These center coordinates are labeled with the influence radius and average brightness / temperature value of the hotspot cluster. The large-scale abnormal hotspot distribution is projected onto the ground-based fire sensing network, and the fire sensing network is filtered to cover it. Using the influence radius and center coordinates of the multiple hotspot clusters as segmentation constraints, the fire sensing network is divided into multiple local fire sensing networks. These local fire sensing networks are activated to transmit multiple sets of multi-source re-collected data. After initializing and verifying the multiple sets of multi-source re-collected data using baseline smoke concentration characteristics, the multiple sets of multi-source re-collected data are then verified using multiple average brightness / temperature values, and a real-time fire risk alarm is output.

[0099] Furthermore, the system is also used to implement the following functions: Based on the area coverage of the real-time fire risk warning, real-time meteorological data is retrieved from the multiple sets of multi-source recollection data; after stitching the real-time fire image based on the flight trajectory, fire isotherms are extracted and continuous fire line conversion is performed to output the fire boundary; thermal imaging temperature distribution is solved based on the real-time fire image, and the fire intensity parameter is quantitatively output; starting from the fire boundary, the time progression simulation of the fire intensity parameter is performed based on the real-time meteorological data and the terrain slope data of the forest area, and the predicted result of the fire spread trend within the preset rescue blocking window is output.

[0100] Furthermore, the system is also used to implement the following functions: The random forest parameters include 500 decision trees, a maximum tree depth of 15 layers, and a minimum number of samples per node of 5. The Gini coefficient is used as the node splitting criterion.

[0101] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this specification. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0102] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

[0103] This specification and accompanying drawings are merely illustrative examples of this application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from its scope. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application intends to include such modifications and variations.

Claims

1. A method for dynamic monitoring of forest fires based on multi-source data fusion, characterized in that, The method includes: The probability of forest fires is calculated based on multi-source fire auxiliary data, and a dynamic risk heat map is output. Based on the risk gradient characteristics of the dynamic risk heat map, the integrated sensing units are deployed non-uniformly to form a ground fire sensing network. Short-term scans of forest areas using medium- and low-resolution satellites were conducted to obtain the distribution of large-scale anomalous hotspots. Based on the large-scale abnormal hotspot distribution, the local area triggers the ground fire sensor network to perform abnormal hotspot collaborative verification and outputs real-time fire risk alarms. After receiving the real-time fire risk alarm transmitted by the ground fire sensor network through a low-power wide area network, the fire monitoring center matches a highly reliable UAV to perform real-time fire image acquisition based on the fire risk alarm. After estimating the fire boundary and fire intensity parameters based on the real-time fire images, fire spread modeling is performed in conjunction with real-time meteorological data to obtain fire spread trend prediction results. The fire spread trend prediction results are used to optimize resource allocation to obtain a real-time fire suppression strategy, and a high-response fire suppression and control system is implemented in the forest area.

2. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 1, characterized in that, The method involves calculating the probability of forest fires based on multi-source fire auxiliary data and outputting a dynamic risk heat map. The multi-source fire auxiliary data is decomposed to obtain fire precursor feature data and satellite remote sensing data. The fire precursor feature data includes air pollution data, land use data and historical fire records. The satellite remote sensing data includes medium and low resolution thermal infrared data and high resolution multispectral data. Using the multi-source fire auxiliary data and satellite remote sensing data as input feature vectors, the probability of forest fire occurrence is calculated using the random forest algorithm, and the output is a dynamic risk heat map.

3. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 2, characterized in that, Using the multi-source fire auxiliary data and satellite remote sensing data as input feature vectors, the probability of forest fire occurrence is calculated using the random forest algorithm, and the output is a dynamic risk heat map. The method includes: CO concentration and aerosol index are extracted from the air pollution data as characteristics of air pollution. The land use data is encoded with vegetation types to generate land use classification features; Extracting spatial recurrence frequency characteristics from historical fire records; The surface temperature anomaly extracted from the low- and medium-resolution thermal infrared data and the NDVI vegetation index calculated based on the high-resolution multispectral data are used as remote sensing derived features. After dividing the forest area into planar grids to obtain a unified geographic grid, the atmospheric pollution features, land use classification features, spatial recurrence frequency features, and remote sensing derived features are aligned and segmented based on the unified geographic grid to obtain a multidimensional feature vector array. Based on the random forest model, multi-threaded fire occurrence probability calculation is performed on the multi-dimensional feature vector array, and the fire occurrence probability array is output. The fire occurrence probability array is spatially clustered based on the watershed algorithm to generate the dynamic risk heat map.

4. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 3, characterized in that, Based on a random forest model, a multi-threaded fire occurrence probability calculation is performed on the multi-dimensional feature vector array, and a fire occurrence probability array is output. The method includes: A gridded sample set with fire probability labels is prepared based on historical monitoring data, wherein the ratio of positive to negative samples in the gridded sample set is adjusted to 1:3 using SMOTE oversampling technology; Multi-source feature engineering is performed on historical precursor feature data to obtain a standardized feature matrix; After configuring the random forest parameters of the random forest model, the standardized feature matrix and the gridded sample set are used as training data to perform convergence training on the random forest model based on out-of-bag error monitoring, and the initial forest model is output. After compressing the decision tree depth of the initial forest model to ≤10 layers using a pruning algorithm, the initial forest model is persisted as a lightweight fire probability prediction model. Based on the fire occurrence probability prediction model, multi-threaded fire occurrence probability calculation is performed on the multi-dimensional feature vector array, and the fire occurrence probability array is output.

5. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 2, characterized in that, The method involves short-period scanning of forest areas using medium- and low-resolution satellites to obtain the distribution of large-scale anomalous hotspots. The fire monitoring center receives and transmits short-cycle raw data streams back from the medium- and low-resolution satellites. After extracting the infrared band from the short-cycle raw data stream, radiometric calibration is performed, and a brightness-temperature raster map is output. After retrieving the historical fire point temperature characteristics from the historical fire records, a dynamic threshold distribution is set based on the historical fire point temperature characteristics and the vegetation distribution attributes of the forest area. The dynamic threshold distribution is applied to slide through the brightness and temperature raster to identify fire points, and the large-scale abnormal hotspot distribution is output in vector format.

6. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 1, characterized in that, The integrated sensing unit specifically integrates an infrared thermal imaging sensor, a smoke particulate sensor, and a meteorological micro-station, wherein the meteorological micro-station is used to collect local wind speed data and local wind direction data.

7. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 6, characterized in that, Based on the large-scale abnormal hotspot distribution, the local area triggers the ground fire sensor network to perform abnormal hotspot collaborative verification and outputs a real-time fire risk alarm. The method includes: The large-scale abnormal hotspot distribution is analyzed, and the center coordinates of multiple hotspot clusters are extracted. The center coordinates of the hotspot clusters are labeled with the influence radius and average brightness temperature value of the hotspot clusters. The large-scale abnormal hotspot distribution is projected onto the ground fire sensing network to filter and cover the fire sensing network. Using the influence radius of multiple hotspot clusters and the center coordinates of the multiple hotspot clusters as segmentation constraints, the fire sensing network is divided into multiple local fire sensing networks. Activate the multiple local fire sensor networks to transmit multiple sets of multi-source recollected data; After initializing and verifying the multiple sets of multi-source resampling data using the baseline smoke concentration characteristics, the multiple sets of multi-source resampling data are then mapped and verified using multiple average brightness and temperature values, and the real-time fire risk alarm is output.

8. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 7, characterized in that, After estimating the fire boundary and fire intensity parameters based on the real-time fire images, fire spread modeling is performed in conjunction with real-time meteorological data to obtain fire spread trend prediction results. The method includes: Based on the regional coverage of the real-time fire risk alarm, the real-time meteorological data is retrieved from the multiple sets of multi-source recollection data. After stitching the real-time fire image based on the flight trajectory, the fire isotherms are extracted and continuous fire line conversion is performed to output the fire boundary. Based on the real-time fire image, the thermal imaging temperature distribution is solved, and the fire intensity parameter is quantified and output. Starting from the fire boundary, the fire intensity parameters are simulated over time based on real-time meteorological data and topographic slope data of the forest area, and the predicted fire spread trend within the preset rescue blocking window is output.

9. The forest fire dynamic monitoring method based on multi-source data fusion as described in claim 4, characterized in that, The random forest parameters include 500 decision trees, a maximum tree depth of 15 layers, and a minimum number of samples per node of 5. The node splitting criterion uses the Gini coefficient.

10. A forest fire dynamic monitoring system based on multi-source data fusion, characterized in that, The system is used to implement the forest fire dynamic monitoring method based on multi-source data fusion as described in any one of claims 1-9, and the system comprises: The heat map output module is used to calculate the probability of forest fire occurrence based on multi-source fire auxiliary data and output a dynamic risk heat map. The sensor network acquisition module is used to perform non-uniform deployment of integrated sensor units based on the risk gradient characteristics of the dynamic risk heat map to form a ground fire sensing network. The abnormal hotspot distribution acquisition module is used to obtain the distribution of large-scale abnormal hotspots by performing short-period scanning of forest areas using medium and low resolution satellites. The fire risk alarm module is used to trigger the ground fire sensor network to perform collaborative verification of abnormal hotspots based on the distribution of the large-scale abnormal hotspots, and output real-time fire risk alarms. The fire image acquisition module is used by the fire monitoring center to receive the real-time fire risk alarm transmitted by the ground fire sensor network through the low-power wide area network, and then match a high-reliability UAV to perform real-time fire image acquisition based on the fire risk alarm. The fire spread trend prediction module is used to estimate the fire boundary and fire intensity parameters based on the real-time fire images, and then combine real-time meteorological data to model the fire spread and obtain the fire spread trend prediction results. The fire suppression and control module is used to optimize resource allocation based on the predicted fire spread trend to obtain real-time fire suppression strategies and execute high-response fire suppression and control in the forest area.

Images