Forest fire monitoring method based on low earth orbit satellite

Abstract

The invention discloses a forest fire monitoring method based on a low earth orbit satellite, and the method comprises the steps: obtaining data through multiple sensors, and carrying out the denoising and smoothing processing, and generating a fusion data set; a convolutional neural network is combined with multi-scale analysis to enhance fire point features, and a feature set is formed. And matching the multi-satellite collaborative observation data with the feature set, and generating a high-frequency observation enhanced data set through time sequence alignment. A deep learning and transfer learning optimization model is adopted, key area features are extracted, multi-satellite data are shared through distributed calculation, and a coverage enhancement data set is obtained. And dynamically adjusting a multi-satellite scheduling strategy based on the timeliness evaluation model, and outputting a scheduling result. The robustness of an algorithm is optimized through reinforcement learning, a robustness enhancement model is obtained in combination with complex meteorological condition simulation, the data fusion precision is verified by using a multi-source data verification technology, and a final monitoring result is output. According to the method, the accuracy, timeliness and coverage integrity of fire point detection are remarkably improved.

Description

Technical Field

[0001] The present invention belongs to the field of forest fire monitoring, and particularly relates to a forest fire monitoring method based on low-earth orbit satellites. Background Art

[0002] Forest fire monitoring is a key topic in the fields of ecological protection and disaster prevention and control. Its importance is self-evident and is directly related to the stability of the ecosystem, the safety of human life and property, and the sustainable development of social economy. The occurrence frequency and destructiveness of forest fires have increased significantly, and there is an urgent need for efficient and accurate monitoring technologies to address this severe challenge. However, the current monitoring methods based on ground sensors or traditional remote sensing technologies have obvious limitations. Ground monitoring is limited by a narrow coverage range, high deployment costs, and difficulty in implementation in remote areas; traditional geostationary satellite remote sensing has insufficient resolution and low observation frequencies, making it difficult to meet the requirements of real-time and refined monitoring. These defects lead to difficulties in identifying fire spots in the initial stage of the fire, lagging early warnings, inaccurate fire situation assessments, and it is difficult to effectively guide fire extinguishing decisions.

[0003] In this context, the core technical challenges faced in the field of forest fire monitoring have gradually emerged. Among them, the accuracy of data fusion, the real-time performance of intelligent analysis, and the coverage ability of multi-source collaboration have become key bottlenecks. First, heterogeneous data obtained by multiple sensors (such as different bands and resolutions) is vulnerable to noise interference during the fusion process, affecting the accuracy of fire spot detection. Second, the real-time analysis of image data is limited by the adaptability of the algorithm to complex scenes. Especially in areas with variable meteorological conditions or complex terrains, false alarms and missed detections are prominent. Finally, the observations of a single satellite are restricted by orbits and time, and it is impossible to achieve high-frequency and all-weather coverage. Moreover, the data sharing and complementary mechanism of multi-satellite collaboration is not yet mature, resulting in insufficient timeliness and integrity of monitoring. These technical problems jointly restrict the transformation of forest fire monitoring from passive response to active early warning. Therefore, how to achieve precise fusion of multi-sensor data on a low-earth orbit satellite platform, improve the real-time performance and robustness of intelligent fire spot detection algorithms, and optimize the high-frequency coverage ability through multi-satellite collaboration has become a key issue in enhancing the dynamic monitoring and early warning effect of forest fires. The solution to this problem will directly promote the monitoring technology towards intelligence and systematization, providing more reliable technical support for forest fire prevention and control. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a forest fire monitoring method based on low-earth orbit satellites, including:

[0005] Obtain multi-sensor data from a low-earth orbit satellite platform, process the heterogeneity of the sensor data through a preset denoising model, and smooth the data of different bands using adaptive filtering technology to obtain a fused data set;

[0006] For the fused dataset, when performing complex scene recognition, if the influence of noise interference exceeds a preset threshold, features are extracted through a convolutional neural network, and the fire point features are enhanced by combining multi-scale analysis techniques to obtain a feature set;

[0007] Obtain multi-satellite collaborative observation data, and match the multi-source collaborative coverage data with the feature set through time series alignment technology to obtain an enhanced dataset with enhanced high-frequency observation capabilities;

[0008] For the enhanced dataset, if the complex scene recognition requirements increase, the adaptability of image analysis is optimized through a deep learning algorithm, and the model parameters are adjusted by combining transfer learning techniques to output an optimization result;

[0009] Extract key region features from the optimization result, and use a distributed computing framework to perform data sharing processing on multi-satellite collaborative complementarity to obtain a coverage-enhanced dataset;

[0010] According to the coverage-enhanced dataset, analyze the observation frequency and integrity through a preset timeliness evaluation model. If the timeliness integrity is lower than expected, dynamically adjust the multi-satellite collaborative scheduling strategy to output a scheduling result;

[0011] For the scheduling result, use reinforcement learning technology to iteratively optimize the robustness of the algorithm, and obtain a robustness-enhanced model by simulating the fire point detection task under complex meteorological conditions;

[0012] Obtain the final fire point detection data from the robustness-enhanced model, and verify the data fusion accuracy through multi-source data verification technology to output the final monitoring result.

[0013] Preferably, the process of obtaining the fused dataset includes:

[0014] Obtain sensor data generated by multiple sensors from a low-earth orbit satellite, process the heterogeneous characteristics through a preset denoising model, and use a low-earth orbit satellite platform to carry multiple sensors to obtain multi-source data of the forest area;

[0015] Preprocess the obtained sensor data to eliminate data heterogeneity between different sensors to obtain the fused dataset; wherein, the preprocessing includes data format unification, time synchronization, and spatial registration.

[0016] Preferably, the process of obtaining the feature set includes:

[0017] Extract features from the fused dataset through a convolutional neural network, and use a spatial pyramid transform or a feature pyramid network structure to fuse feature maps of different scales;

[0018] Meanwhile, during the feature extraction process, an attention mechanism is introduced to enhance the basic features, and weights are assigned to different features to highlight the fire point features and suppress noise interference;

[0019] The enhanced features are fused with the multi-scale features to obtain the feature set.

[0020] Preferably, the process of obtaining the enhanced data set includes:

[0021] Multi-source data is obtained through multi-satellite observations, and the time series alignment technology is used to process data alignment to obtain the first aligned data set;

[0022] Cooperative coverage information is extracted from the first aligned data set, and is fused with the features of the feature set through matching processing to obtain the second fused data set;

[0023] For the second fused data set, coverage range analysis is adopted to determine the spatial distribution characteristics of the multi-source data, and a distribution characteristic set is obtained;

[0024] If the distribution characteristic set meets the preset threshold conditions, the distribution characteristic set is classified by the random forest algorithm to judge the high-frequency observation area, and a region set is obtained;

[0025] According to the corresponding relationship between the region set and the time series, the time distribution characteristics of high-frequency observations are obtained to obtain a time feature set;

[0026] Through the mapping relationship between the time feature set and the enhanced data set, the enhanced area of the observation ability is determined to obtain an enhanced area set;

[0027] The enhanced area set is optimized by data enhancement technology to obtain an optimized enhanced data set.

[0028] Preferably, the process of optimizing the image analysis adaptability through a deep learning algorithm, adjusting the model parameters by combining transfer learning technology, and outputting the optimization result includes:

[0029] The enhanced data set is initially processed by a deep learning algorithm to obtain a preliminary analysis result;

[0030] According to the preliminary analysis result, the transfer learning technology is used to adjust the model parameters for complex scenarios to obtain an optimized parameter set;

[0031] If the optimized parameter set meets the preset threshold conditions, the complex scenario is feature-extracted by image analysis technology to obtain a feature set;

[0032] The enhanced data set is matched with the feature set to obtain a matching data set;

[0033] Determine the distribution characteristics of scene recognition through the matching data set to obtain a distribution feature set;

[0034] Obtain a mapping data set according to the mapping relationship between the distribution feature set and the optimization parameter set;

[0035] Perform optimization processing on the mapping data set using data augmentation techniques to obtain an optimization result.

[0036] Preferably, the process of extracting key region features from the optimization result and performing data sharing processing on the multi-star collaborative complementarity using a distributed computing framework to obtain a coverage enhancement data set includes:

[0037] Determine the distribution range of region features through key regions, and obtain a preliminary feature set using feature extraction techniques;

[0038] Divide the task units of multi-star collaboration according to the preliminary feature set to obtain a collaborative complementary allocation scheme;

[0039] Perform data sharing processing on the allocation scheme using a distributed computing framework to obtain a shared data group;

[0040] If the shared data group meets the preset threshold conditions, adjust the data range through coverage enhancement techniques to obtain an enhanced data group;

[0041] Perform processing flow optimization on the enhanced data group to obtain optimized data units;

[0042] Judge the complementary distribution of multi-star collaboration through the mapping relationship between the optimized data unit and the region feature to obtain a coverage enhancement data set;

[0043] Perform verification processing on the coverage enhancement data set using a computing framework to determine the final distribution characteristics.

[0044] Preferably, according to the coverage enhancement data set, analyze the observation frequency and integrity through a preset timeliness evaluation model. If the timeliness integrity is lower than expected, the process of dynamically adjusting the multi-star collaborative scheduling strategy and outputting the scheduling result includes:

[0045] Analyze the observation frequency and integrity through the coverage enhancement data set using a preset timeliness evaluation model to obtain an analysis data group;

[0046] Judge whether the integrity is lower than the preset threshold conditions according to the analysis data group to obtain a judgment data unit;

[0047] Process the collaborative scheduling scheme using dynamic adjustment techniques through the judgment data unit to obtain an adjustment data group;

[0048] For the adjusted data set, obtain the task allocation information for multi-satellite collaboration and determine the task allocation set;

[0049] Through the task allocation set, use the scheduling strategy optimization technology to process the data range and obtain the optimized scheduling unit;

[0050] Based on the optimized scheduling unit, judge the resource distribution characteristics of multi-satellite collaboration to obtain the scheduling result;

[0051] For the scheduling result, use the distributed computing framework to verify data consistency and determine the final scheduling data set.

[0052] Preferably, for the scheduling result, use the reinforcement learning technology to iteratively optimize the robustness of the algorithm. The process of obtaining the robustness enhanced model by simulating the fire point detection task under complex meteorological conditions includes:

[0053] For the data in the final scheduling data set, use the reinforcement learning technology to process the feature distribution under complex meteorological conditions to obtain the feature enhanced data set;

[0054] Through the feature enhanced data set, obtain the boundary conditions of the fire point detection task and determine the boundary feature set;

[0055] For the boundary feature set, use the task simulation technology to generate multi-scenario detection data to obtain the simulated data set;

[0056] Based on the simulated data set, if the distribution characteristics of the detection task exceed the preset threshold, adjust the algorithm robustness parameters to obtain the adjusted parameter set;

[0057] Through the adjusted parameter set, obtain the iteratively optimized model structure and determine the model framework;

[0058] For the model framework, use the distributed computing to verify the stability of the fire point detection to obtain the robustness enhanced model.

[0059] Preferably, the process of obtaining the final fire point detection data from the robustness enhanced model, verifying the data fusion accuracy through the multi-source data verification technology, and outputting the final monitoring result includes:

[0060] Extract the fire point detection data from the robustness enhanced model and fuse the data features through the multi-source data verification technology to obtain the fused data set;

[0061] Based on the fused data set, use the distributed computing technology to process the detection task distribution to obtain the task distribution data;

[0062] For the task distribution data, if the distribution characteristics exceed the preset threshold, adjust the verification technology parameters to determine the adjusted parameter set;

[0063] Obtain the detection data after precision fusion through the adjusted parameter set, and output the precision-enhanced data;

[0064] According to the precision-enhanced data, use the clustering analysis technique to divide the fire point detection area and obtain the area division result;

[0065] For the area division result, generate the monitoring result data through the spatial mapping technique to determine the final monitoring data;

[0066] Extract the key features from the final monitoring data to obtain the final monitoring result including the fire point distribution feature set.

[0067] Compared with the prior art, the present invention has the following advantages and technical effects:

[0068] Through technologies such as multi-sensor data acquisition, adaptive filtering, and multi-source data fusion, the present invention constructs a complete process from obtaining data from a low-earth orbit satellite platform to finally outputting the fire point monitoring result. Aiming at the problem of fire point recognition in complex scenarios, the present invention uses convolutional neural network and multi-scale analysis technologies to extract and enhance fire point features, and combines real-time stream processing and intelligent analysis models to improve the detection accuracy. At the same time, the present invention utilizes multi-satellite collaborative observation data, and through methods such as time series alignment, deep learning, and transfer learning, optimizes the adaptability of image analysis and enhances the coverage ability. In addition, the present invention also uses reinforcement learning technology to improve the algorithm robustness and ensures the fusion accuracy through multi-source data verification. This method significantly improves the accuracy, timeliness, and coverage integrity of fire point detection, and provides an effective solution for fire monitoring under complex meteorological conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0070] Figure 1 It is a schematic flowchart of the method of the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0071] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0072] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0073] Such asFigure 1 As shown in Figure 1 , in this embodiment, a forest fire monitoring method based on low-earth orbit satellites is provided, including:

[0074] Obtain multi-sensor data from a low-earth orbit satellite platform, process the heterogeneity of the sensor data through a preset denoising model, and smooth the data of different bands using an adaptive filtering technique to obtain a fused data set;

[0075] For the fused data set, when identifying complex scenes, if the influence of noise interference exceeds a preset threshold, extract features through a convolutional neural network, and enhance the fire point features by combining multi-scale analysis techniques to obtain a feature set;

[0076] Obtain multi-satellite collaborative observation data, match the multi-source collaborative coverage data with the feature set through time series alignment technology to obtain an enhanced data set with enhanced high-frequency observation capabilities;

[0077] For the enhanced data set, if the demand for complex scene recognition increases, optimize the adaptability of image analysis through a deep learning algorithm, and adjust the model parameters by combining transfer learning technology to output an optimized result;

[0078] Extract the key region features from the optimized result, and perform data sharing processing on the multi-satellite collaborative complementarity using a distributed computing framework to obtain a coverage-enhanced data set;

[0079] According to the coverage-enhanced data set, analyze the observation frequency and integrity through a preset timeliness evaluation model. If the timeliness integrity is lower than expected, dynamically adjust the multi-satellite collaborative scheduling strategy and output a scheduling result;

[0080] For the scheduling result, use reinforcement learning technology to iteratively optimize the robustness of the algorithm. By simulating the fire point detection task under complex meteorological conditions, obtain a robustness-enhanced model;

[0081] Obtain the final fire point detection data from the robustness-enhanced model, verify the data fusion accuracy through multi-source data verification technology, and output the final monitoring result.

[0082] Furthermore, the process of obtaining the fused data set includes:

[0083] Obtain the sensor data generated by multi-sensors from a low-earth orbit satellite, process the heterogeneous characteristics through a preset denoising model, and use a low-earth orbit satellite platform equipped with multiple sensors to obtain multi-source data of the forest area;

[0084] Preprocess the obtained sensor data to eliminate the data heterogeneity between different sensors and obtain a fused data set; where the preprocessing includes data format unification, time synchronization, and spatial registration.

[0085] Specifically, obtaining sensor data generated by multiple sensors from low-earth orbit satellites is an important technical means in the field of forest monitoring.

[0086] Exemplarily, low-earth orbit satellites can carry optical sensors, infrared sensors, and radar sensors to collect visible light images, thermal imaging data, and terrain information of the forest respectively. Taking a forest area of 500 square kilometers as an example, the optical sensor can capture the canopy coverage rate, the infrared sensor can detect fire hazards, and the radar sensor can penetrate clouds to obtain terrain undulations. This multi-source data collection method can comprehensively reflect the forest ecological status.

[0087] In a possible implementation manner, it is particularly crucial to process the heterogeneous characteristics with a preset denoising model. There are problems with inconsistent resolutions and sampling frequencies in different sensor data.

[0088] For example, each pixel of the optical sensor covers 10 meters, the infrared sensor is 50 meters, and the radar data may have a resolution of 30 meters. The denoising model unifies the resolution to 30 meters through interpolation and filtering, and at the same time eliminates cloud interference or signal noise. This processing can significantly improve the usability of the data and lay a foundation for subsequent analysis.

[0089] It should be noted that the preliminary data preprocessing includes data format unification, time synchronization, and spatial registration.

[0090] Specifically, format unification can convert the TIFF format of optical images, the DAT format of infrared, and the HDF format of radar into a unified NetCDF format for easy integrated analysis. Time synchronization ensures that the data corresponds to the same moment.

[0091] For example, adjust the 5-minute acquisition time difference caused by satellite orbit differences. Spatial registration is calibrated through a geographic coordinate system, and the pixel deviation is controlled within 1 meter. This preprocessing eliminates heterogeneity and ensures the spatial consistency of the data.

[0092] In one embodiment, when a low-earth orbit satellite platform carries multiple sensors, the configuration can be optimized.

[0093] For example, the satellite operates in an orbit at an altitude of 500 kilometers, carrying an optical camera with a resolution of 10 meters, an infrared instrument with a detection range of 50 kilometers, and a synthetic aperture radar covering 100 kilometers. When collecting data, the camera records the tree density, the infrared instrument monitors temperature anomalies, and the radar depicts the terrain contour. After the multi-source data fusion, the forest fire risk areas can be accurately identified.

[0094] Preferably, the first dataset after denoising provides a high-quality basis for subsequent applications.

[0095] For example, in a 200-kilometer forest belt, the first data set shows that the temperature in a certain area is 5 degrees Celsius higher than the average, and the canopy coverage rate decreases by 20%. Combining with radar data analysis, it is found that the low-lying terrain leads to water accumulation, affecting vegetation growth. This multi-dimensional analysis not only improves the monitoring accuracy but also provides a decision-making basis for fire prevention and ecological protection.

[0096] It can be understood that the technical effects of eliminating data heterogeneity are significant.

[0097] For example, traditional single sensors may miss fire hazards in rainy weather, while multi-sensor collaboration, after preprocessing, can complement the deficiencies of optical images through infrared data, reducing the misjudgment rate to less than 5%. This method can effectively reduce resource waste and improve the response speed in forest management.

[0098] In one embodiment, the preprocessed data set can also support dynamic monitoring.

[0099] For example, data is collected once every 3 days, and the vegetation change trend within 30 days is continuously analyzed. If the canopy coverage rate in a certain area continues to decline, combined with temperature and terrain data, the possibility of pests and diseases or human logging can be inferred. This long-term monitoring ability provides continuous support for ecological protection.

[0100] For example, through the above technology, forest management departments can quickly locate problem areas, reduce the cost of manual inspections, and improve the timeliness of disaster warnings. The collaboration and denoising processing of multi-source data jointly ensure the reliability and practicality of the data.

[0101] Furthermore, the process of obtaining the feature set includes:

[0102] Performing feature extraction on the fused data set through a convolutional neural network, and using spatial pyramid transformation or feature pyramid network structure to fuse feature maps of different scales;

[0103] At the same time, during the feature extraction process, an attention mechanism is introduced to enhance the basic features, and weights are assigned to different features to highlight the fire point features and suppress noise interference;

[0104] Fusing the enhanced features with the multi-scale features to obtain the feature set.

[0105] In a possible implementation, when a convolutional neural network (CNN) performs feature extraction on the fused data set, spatial pyramid transformation is used to capture multi-scale information. Spatial pyramid transformation generates feature representations of multiple scales by performing pooling operations on feature maps at different levels.

[0106] For example, in the forest monitoring scenario, the first fusion dataset includes canopy coverage rate, temperature distribution, and terrain undulation information. The spatial pyramid transformation can extract coarse-grained features in the global range (such as the average temperature of the entire forest) and fine-grained features in the local range (such as the canopy details of a small area). This method enables the model to simultaneously focus on large-scale trends and small-scale outliers. For example, the features of an area with abnormally concentrated temperature in a certain forest will be magnified.

[0107] Specifically, the Feature Pyramid Network (FPN) structure fuses feature maps of different depths through top-down and lateral connections.

[0108] Exemplarily, when processing forest data of 500 square kilometers, the shallow feature map may retain details with a resolution of 10 meters, such as the shape of tree edges; the deep feature map extracts more abstract information, such as the terrain undulation trend. After FPN fuses these feature maps, a feature set with both details and semantics is formed.

[0109] It can be understood that this fusion method is particularly suitable for forest fire monitoring because it can simultaneously retain the local high-temperature features of the fire point and the overall distribution of the surrounding environment.

[0110] It should be noted that the attention mechanism plays a key role in feature enhancement. The global attention module highlights important information by calculating the weights of each position in the feature map.

[0111] For example, in infrared data, the temperature features of the fire point will be given higher weights, while the noise caused by clouds and fog is suppressed.

[0112] In one embodiment, assuming that the abnormal temperature value in a certain area is 38 degrees Celsius, which is much higher than the average value of 32 degrees Celsius, the global attention module will enhance the feature representation of this area while reducing the weights of irrelevant areas. This enhanced feature can more accurately reflect the potential fire risk.

[0113] In one possible implementation, during multi-scale feature fusion, the enhanced features can be first weighted and superimposed, and then the number of channels can be adjusted through a convolutional layer.

[0114] For example, the feature maps generated by the spatial pyramid transformation may have three scales, with resolutions of 10 meters, 30 meters, and 50 meters respectively. After attention enhancement, they are fused into a unified feature set with a resolution of 30 meters.

[0115] Preferably, this method ensures that the characteristics of different sensor data are retained and work together. For example, the canopy details of optical data complement the temperature anomalies of infrared data.

[0116] For example, in a 200-kilometer-long forest belt, the feature set shows that a low-lying area has both a trend of rising temperature and the feature of sparse tree canopy. Combining with the analysis of terrain data, it is possible that the vegetation growth in this area is affected by waterlogging, which in turn increases the fire hazard.

[0117] It can be understood that this multi-dimensional feature fusion can provide a more reliable basis for subsequent classification or prediction tasks, such as quickly locating areas that need to be monitored with key focus.

[0118] In one embodiment, the attention mechanism can also dynamically adjust the weights.

[0119] For example, in the data of 10 consecutive days, if the temperature in a certain area continues to be high and the canopy coverage rate decreases, the model will gradually increase the attention to the features of this area. This dynamic nature helps to capture the gradual changes in the forest ecosystem and provides support for long-term monitoring.

[0120] Exemplarily, this method can also reduce false alarms because it synthesizes data over multiple days rather than making a judgment based on a single time point.

[0121] Specifically, the generation process of the feature set emphasizes the efficient integration of multi-source information. In forest monitoring, a single feature may be distorted due to weather or sensor limitations, but after multi-scale fusion and attention enhancement, the feature set can more comprehensively reflect the actual situation.

[0122] For example, the ability of radar data to penetrate clouds combined with the temperature sensitivity of infrared data makes the identification of fire points more reliable in rainy weather. This technical means significantly improves the robustness of monitoring and provides a solid guarantee for ecological management and disaster warning.

[0123] Furthermore, the process of obtaining the enhanced data set includes:

[0124] Obtaining multi-source data through multi-satellite observations, and using time series alignment technology to process data alignment to obtain the first data set after alignment;

[0125] Extracting co-coverage information from the aligned first data set, and fusing it with the features of the feature set through matching processing to obtain the second data set after fusion;

[0126] For the second data set after fusion, using coverage analysis to determine the spatial distribution characteristics of multi-source data to obtain a distribution characteristic set;

[0127] If the distribution characteristic set meets the preset threshold conditions, then classifying the distribution characteristic set through a random forest algorithm to judge the high-frequency observation areas and obtain an area set;

[0128] According to the correspondence between the area set and the time series, obtaining the time distribution characteristics of high-frequency observations to obtain a time feature set;

[0129] Determine the enhanced area of the observation ability through the mapping relationship between the time feature set and the enhanced data set, and obtain the enhanced area set;

[0130] Use data enhancement technology to optimize the enhanced area set to obtain the optimized enhanced data set.

[0131] Specifically, obtaining multi-source data through multi-satellite observation is the basis for complex scene analysis. For example, multiple remote sensing satellites can be used to observe the same area at different time periods to obtain various types of data such as optical and infrared data. Due to the differences in satellite orbits and observation times, there may be time biases in the data. When using time series alignment technology to process data alignment, interpolation adjustment can be performed based on timestamps.

[0132] Exemplarily, assume that a certain area is observed by two satellites at 8:00 and 8:05 on March 15th respectively. The two sets of data are aligned to the same time reference, such as 8:03, through linear interpolation method to obtain the aligned first data set. This alignment method can ensure the temporal consistency of subsequent analysis. Extracting the co-coverage information from the aligned first data set involves the spatial matching of multi-source data.

[0133] Specifically, the observation ranges of different satellites can also be superimposed and analyzed through the geographic coordinate system.

[0134] For example, the coverage area of satellite A is from 120° to 121° east longitude, and the coverage area of satellite B is from 120.5° to 121.5° east longitude. The co-coverage information is the data of the overlapping part from 120.5° to 121° east longitude.

[0135] After the matching process, when fusing with the features of the feature set, the co-coverage information is combined with the existing feature set through the weighted average method to generate the fused second data set. This fusion can improve the spatial integrity of the data.

[0136] For the fused second data set, when determining the spatial distribution characteristics by using the coverage range analysis, the grid division method can be based on. The observation area is divided into grids of 1 km × 1 km, and the data density in each grid is counted to obtain the distribution characteristic set. For example, if the number of data points in a certain grid is 50, exceeding the preset threshold of 30, it is marked as a high-density area. If the distribution characteristic set meets the conditions, it is classified by the random forest algorithm.

[0137] In one embodiment, the random forest classifies the area into a high-frequency observation area and a low-frequency observation area based on features such as data density and coverage frequency to obtain the area set. This classification can quickly identify key areas. When obtaining the time distribution characteristics according to the correspondence between the area set and the time series.

[0138] It should be noted that the number of observations in the high-frequency observation area within 24 hours can be counted.

[0139] For example, a certain area is observed 10 times from 6:00 to 18:00, and the time feature set shows that its active period is during the day. After mapping this time feature set with the enhanced data set, the observation ability enhancement area is determined. In one possible implementation, if the enhanced data set records the device performance parameters, the enhanced area set may be the area where the observation frequency is increased by 50%. When optimizing using data enhancement techniques, for example, by interpolating to increase the data point density, an optimized enhanced data set is generated. This optimization can improve the usability of the data.

[0140] In one embodiment, the extraction of the collaborative coverage information of multi-star observations can also be dynamically adjusted in combination with historical data. For example, according to the observation records of the past 7 days, the characteristics of the high-frequency coverage area are preferentially matched to improve the fusion efficiency.

[0141] It can be understood that the combination of time series alignment and feature fusion can provide a more reliable data basis for subsequent analysis. And random forest classification ensures the accuracy of area division through comprehensive judgment of multi-dimensional features. The synergistic effect of these methods ultimately provides strong support for data processing in complex scenarios.

[0142] Furthermore, the process of optimizing the adaptability of image analysis through deep learning algorithms and adjusting the model parameters in combination with transfer learning techniques includes:

[0143] Performing initial processing on the enhanced data set through deep learning algorithms to obtain a preliminary analysis result;

[0144] According to the preliminary analysis result, using transfer learning techniques to adjust the model parameters for complex scenarios to obtain an optimized parameter set;

[0145] If the optimized parameter set meets the preset threshold conditions, then feature extraction is performed on the complex scenario through image analysis techniques to obtain a feature set;

[0146] Performing matching processing on the enhanced data set using the feature set to obtain a matching data set;

[0147] Determining the distribution characteristics of scene recognition through the matching data set to obtain a distribution feature set;

[0148] Obtaining a mapping data set according to the mapping relationship between the distribution feature set and the optimized parameter set;

[0149] Performing optimization processing on the mapping data set using data enhancement techniques to obtain an optimized result.

[0150] Specifically, when initially processing the augmented dataset through deep learning algorithms, for example, in the field of satellite collaborative observation, the augmented dataset can be regarded as remotely sensed data collected by multiple satellites, containing time series and spatial information. In one possible implementation, a convolutional neural network is used to extract features from the data. The input data may be 1000 images with a resolution of 256×256, and the output of the preliminary analysis result is, for example, the probability distribution of land cover types. This method can quickly process high-dimensional data and lay a foundation for subsequent analysis.

[0151] When using transfer learning technology to adjust model parameters for complex scenarios, for example, in a complex terrain scenario with a low forest coverage rate, a model pre-trained on a large-scale general dataset, such as a network based on ResNet, can be selected, and then the parameters are fine-tuned using 200 samples in the augmented dataset. The optimized parameter set may be the adjusted weight set, which meets the threshold condition, such as an accuracy rate of 85%, so as to adapt to a specific scenario.

[0152] When extracting features from complex scenarios through image analysis technology, specifically, an edge detection algorithm can be used to extract the boundary features of ground objects. In one embodiment, a feature set is extracted from the images of the complex scenario, such as numerical values like a river width of 50 meters and a vegetation density of 0.7, to generate a feature vector. This method can highlight the key attributes of the scenario and facilitate subsequent matching.

[0153] When using the feature set to perform matching processing on the augmented dataset, the feature set is corresponding to the samples in the augmented dataset through similarity calculation. For example, the feature vector is compared with 500 samples in the dataset one by one, and the matching dataset with a similarity higher than 0.9 is selected. This matching can effectively integrate multi-source information. When determining the distribution characteristics of scene recognition through the matching dataset, in one embodiment, if the matching dataset covers an area of 1000 square kilometers, the distribution feature set may show that 70% is plain and 20% is mountain. This distribution characteristic reflects the spatial heterogeneity of the data. When obtaining the mapping dataset according to the mapping relationship between the distribution feature set and the optimized parameter set, the data weights are adjusted through the parameter set to generate a mapping dataset that more conforms to the distribution characteristics. Assuming that the weight of the plain area is increased by 10% and the mountain area is decreased by 5%, the new dataset can more accurately reflect the characteristics of the target area.

[0154] When using data augmentation technology to optimize the mapping dataset, 300 images in the dataset can be rotated, flipped, or noise can be added to generate an augmented dataset. This method can enrich data diversity and improve the reliability of subsequent applications.

[0155] Therefore, the transfer learning adjustment parameters in this embodiment can reduce the training cost, the image analysis for feature extraction can enhance the detail expression, and the data augmentation optimizes the data quality. These techniques support each other and jointly improve the efficiency and accuracy of multi-satellite collaborative observation data processing. The optimized augmented dataset can be used for disaster monitoring to quickly locate high-risk areas, demonstrating significant practical value.

[0156] Further, the process of obtaining the coverage augmented dataset by extracting the key region features from the optimization results and performing data sharing processing on the multi-satellite collaboration complementarity using a distributed computing framework includes:

[0157] Determine the distribution range of the regional features through the key regions, and obtain the preliminary feature set using feature extraction techniques;

[0158] Divide the task units of multi-satellite collaboration according to the preliminary feature set, and obtain the collaborative complementary allocation scheme;

[0159] Perform data sharing processing on the allocation scheme using a distributed computing framework to obtain a shared data group;

[0160] If the shared data group meets the preset threshold conditions, adjust the data range through coverage enhancement technology to obtain an enhanced data group;

[0161] Execute the processing flow optimization on the enhanced data group to obtain optimized data units;

[0162] Judge the complementary distribution of multi-satellite collaboration through the mapping relationship between the optimized data units and the regional features to obtain the coverage augmented dataset;

[0163] Perform verification processing on the coverage augmented dataset using a computing framework to determine the final distribution characteristics.

[0164] Specifically, multi-satellite collaboration may involve multiple satellites respectively responsible for different regions or different types of feature extraction tasks.

[0165] Exemplarily, one satellite focuses on high-resolution image acquisition, and another satellite is responsible for wide-range spectral analysis. The two form a complementarity through task unit division, thereby improving the comprehensiveness of data acquisition.

[0166] Perform data sharing processing on the allocation scheme using a distributed computing framework to obtain a shared data group. The distributed computing framework can be deployed in the cloud, and the data collected by the satellites is uploaded and shared in real time through the network.

[0167] For example, each satellite slices the 5GB of data it collects and uploads it to the computing nodes. After parallel processing, a shared data group of boundary information is generated. This method can effectively improve the efficiency of data processing. If the shared data group meets the preset threshold conditions, the data range is adjusted through overlay enhancement technology to obtain an enhanced data group.

[0168] It should be noted that the threshold conditions may be set such that the data coverage rate reaches over 90%.

[0169] In one embodiment, if the coverage rate in the shared data group is only 85%, the missing part can be completed through interpolation or data fusion technology, and finally an enhanced data group with a coverage rate of 95% is formed. This can ensure the integrity of subsequent analysis.

[0170] Perform process optimization on the enhanced data group to obtain optimized data units. The optimization process may include data cleaning and denoising processing. For example, for the boundary data with noise in the enhanced data group, abnormal points are removed through smoothing filtering technology to obtain clearer optimized data units. This optimization helps improve the accuracy of subsequent analysis. By optimizing the mapping relationship between the optimized data units and regional features, the complementary distribution of multi-satellite collaboration is judged to obtain an overlay enhancement data set.

[0171] Furthermore, based on the overlay enhancement data set, the observation frequency and integrity are analyzed through a preset timeliness evaluation model. If the timeliness integrity is lower than expected, the process of dynamically adjusting the multi-satellite collaboration scheduling strategy and outputting the scheduling result includes:

[0172] Through the overlay enhancement data set, analyze the observation frequency and integrity using a preset timeliness evaluation model to obtain an analysis data group;

[0173] Based on the analysis data group, judge whether the integrity is lower than the preset threshold conditions to obtain a judgment data unit;

[0174] Through the judgment data unit, process the collaborative scheduling plan using dynamic adjustment technology to obtain an adjustment data group;

[0175] For the adjustment data group, obtain the task assignment information of multi-satellite collaboration to determine the task assignment set;

[0176] Through the task assignment set, process the data range using scheduling strategy optimization technology to obtain an optimized scheduling unit;

[0177] Based on the optimized scheduling unit, judge the resource distribution characteristics of multi-satellite collaboration to obtain the scheduling result;

[0178] For the scheduling result, use a distributed computing framework to verify data consistency and determine the final scheduled data group.

[0179] Specifically, by covering and enhancing the dataset, a preset timeliness evaluation model is used to analyze the observation frequency and integrity, obtaining an analysis data group. This process can be understood as quantifying the time sensitivity of the data. For example, in the multi-satellite collaboration business, assuming that a certain area needs to be observed 10 times a day, but actually only 8 times are completed, the timeliness evaluation model will give a score based on the observation frequency and data integrity, such as 80 points, which is lower than the preset threshold of 90 points. The timeliness evaluation model will analyze whether the missing observations affect the overall coverage by combining historical data and real-time feedback.

[0180] Based on the analysis data group, judge whether the integrity is lower than the preset threshold condition, and obtain a judgment data unit. Specifically, if the integrity score of a certain area is only 75 points, which is lower than the threshold, the judgment data unit will mark that this area needs to be adjusted. Exemplarily, this kind of judgment can be based on the data fluctuation situation in the past 3 days to determine whether it is an accidental omission or a systematic problem, so as to provide a basis for subsequent scheduling.

[0181] Through the judgment data unit, use dynamic adjustment technology to process the collaborative scheduling plan, obtaining an adjusted data group. In one possible implementation, if a certain satellite cannot cover the target area due to orbital restrictions, the task time of another satellite can be dynamically adjusted. For example, originally satellite A was scheduled to perform observations at 9:00, and now it is adjusted to satellite B to cover at 9:30. The adjusted data group records this change. This method can effectively improve resource utilization.

[0182] For the adjusted data group, obtain the task assignment information for multi-satellite collaboration, and determine the task assignment set. For example, for a certain area, the task assignment set may include satellite A being responsible for morning observations, satellite B being responsible for afternoon observations, and satellite C providing backup support. Specifically, this kind of assignment will be generated according to the performance parameters and geographical locations of each satellite to ensure efficient task assignment.

[0183] Through the task assignment set, use scheduling strategy optimization technology to process the data range, obtaining an optimized scheduling unit. In one embodiment, if the coverage range of a certain area is too large, the task is split into multiple sub-areas through optimization technology. For example, the original coverage range was 1000 square kilometers, and after optimization, it was divided into 3 sub-areas, each about 330 square kilometers, and each is executed by a different satellite. This kind of splitting can significantly improve the observation accuracy. According to the optimized scheduling unit, judge the resource distribution characteristics of multi-satellite collaboration, and obtain the scheduling result.

[0184] It should be noted that the resource distribution characteristics may be reflected in the fact that a certain satellite has too high a load.

[0185] For example, satellite A undertakes 60% of the tasks, while satellite B only undertakes 20%. The sixth scheduling result will reflect this distribution and provide data support for subsequent balancing. For the sixth scheduling result, a distributed computing framework is used to verify data consistency and determine the final scheduling data set.

[0186] Exemplarily, the distributed framework will distribute the scheduling result to multiple computing nodes for verification. For example, each node separately checks whether the timestamps are synchronized and whether the coverage areas overlap, and finally confirms that the data set has no conflicts. This verification can ensure the reliability of the scheduling scheme in actual execution.

[0187] Furthermore, for the scheduling result, reinforcement learning technology is used to iteratively optimize the robustness of the algorithm. The process of obtaining a robustness-enhanced model by simulating the fire point detection task under complex meteorological conditions includes:

[0188] For the data in the final scheduling data set, reinforcement learning technology is used to process the feature distribution under complex meteorological conditions to obtain a feature-enhanced data set;

[0189] Through the feature-enhanced data set, the boundary conditions of the fire point detection task are obtained to determine the boundary feature set;

[0190] For the boundary feature set, task simulation technology is used to generate multi-scenario detection data to obtain a simulation data set;

[0191] According to the simulation data set, if the distribution characteristics of the detection task exceed the preset threshold, the algorithm robustness parameters are adjusted to obtain an adjusted parameter set;

[0192] Through the adjusted parameter set, the iteratively optimized model structure is obtained to determine the model framework;

[0193] For the model framework, distributed computing is used to verify the stability of fire point detection to obtain a robustness-enhanced model.

[0194] Specifically, when using reinforcement learning technology to process the feature distribution under complex meteorological conditions for the data in the final scheduling data set, reinforcement learning gradually optimizes the feature extraction strategy through continuous trial and error and reward mechanisms. For example, in cloudy weather or strong wind conditions, the original data may be distorted due to occlusion or noise. In one embodiment, the reward function can be set to focus on high-confidence features, and the initial learning rate is set to 0.01. After 100 training times, the clarity of the feature distribution is significantly improved, and a feature-enhanced data set is obtained. This provides a more reliable basis for subsequent tasks. When obtaining the boundary conditions of the fire point detection task through the feature-enhanced data set.

[0195] Specifically, by enhancing the feature dataset, the boundary conditions for the fire point detection task are obtained, and the boundary feature set is determined. Exemplarily, in a forest area of 500 square kilometers, the boundary conditions can be set as the temperature anomaly point being higher than 50 degrees Celsius and lasting for more than 10 minutes, and the boundary feature set is determined. This method ensures that the detection task focuses on the key area and improves the efficiency.

[0196] When generating multi-scenario detection data by using task simulation technology for the boundary feature set, in a possible implementation, three scenarios of sunny, cloudy, and night can be simulated. Preferably, the fire point brightness value is set to 80 in the sunny scenario, reduced to 50 in the cloudy scenario due to cloud interference, and increased to 90 at night due to enhanced background contrast, obtaining a set of simulated data. This multi-scenario design helps to comprehensively evaluate the detection ability.

[0197] When judging whether the distribution characteristics of the detection task exceed the preset threshold and adjusting the algorithm robustness parameters according to the set of simulated data, for example, the preset threshold is a detection rate of 90%. If the detection rate in the cloudy scenario is only 85%, then the parameter can be increased from the default value of 0.5 to 0.7 by increasing the depth of the convolutional layer or adjusting the loss weight, obtaining an adjusted parameter set. This adjustment can effectively cope with the abnormal distribution under complex conditions. When obtaining the iteratively optimized model structure through the adjusted parameter set.

[0198] It should be noted that iterative optimization can gradually approach the best performance through multiple verifications. In one embodiment, after 5 iterations, the false alarm rate of the model for fire points is reduced from 10% to 3%, and the seventh model framework is determined. This optimization process significantly improves the practicality of the model. When verifying the fire point detection stability for the seventh model framework by using distributed computing.

[0199] It can be understood that distributed computing distributes the tasks to multiple nodes for parallel processing. For example, 1000 fire point detection tasks are respectively run on 10 computing nodes, and the result consistency of each node reaches 98%, obtaining a robustness enhanced model. This method ensures the reliability of the model in large-scale applications and improves the processing speed at the same time. Throughout the process, from feature enhancement to model optimization, each step closely focuses on the fire point detection task.

[0200] Exemplarily, reinforcement learning improves the feature quality, the boundary conditions focus on the task scope, multi-scenario simulation enriches the data diversity, and parameter adjustment and distributed verification ensure the stability and robustness. These links support each other, forming a complete solution, providing strong technical support for fire point detection in complex environments.

[0201] Furthermore, the process of obtaining the final fire point detection data from the robustness enhanced model and verifying the data fusion accuracy through multi-source data verification technology and outputting the final monitoring results includes:

[0202] Extract fire point detection data from the robustness-enhanced model, and fuse data features through multi-source data verification technology to obtain a fused data set;

[0203] According to the fused data set, use distributed computing technology to process the distribution of detection tasks to obtain task distribution data;

[0204] For the task distribution data, if the distribution characteristics exceed the preset threshold, adjust the verification technology parameters to determine the adjusted parameter set;

[0205] Through the adjusted parameter set, obtain the detection data after precision fusion and output the precision-enhanced data;

[0206] According to the precision-enhanced data, use clustering analysis technology to divide the fire point detection area to obtain the area division result;

[0207] For the area division result, generate monitoring result data through spatial mapping technology to determine the final monitoring data;

[0208] Extract key features from the final monitoring data to obtain the final monitoring result including the fire point distribution feature set.

[0209] Specifically, when extracting fire point detection data from the robustness-enhanced model, for example, under complex meteorological conditions, the fire point detection data may include multi-dimensional features such as temperature, humidity, and wind speed.

[0210] In one embodiment, assume that the fire point data in a certain area includes a temperature of 35 degrees, a humidity of 20%, and a wind speed of 5 m / s. After extraction by the model, an initial data set is formed to provide a basis for subsequent fusion.

[0211] By fusing data features through multi-source data verification technology, multi-source data can be obtained from satellite remote sensing, ground sensors, and meteorological stations. Exemplarily, satellite data may show the location coordinates of a certain fire point, the ground sensor detects an abnormal temperature, and the meteorological station provides wind direction data. When fusing, first align the timestamps and spatial positions of these data, and then integrate the features through a weighted average or voting mechanism to obtain a fused data set. This method can effectively reduce the deviation of a single data source.

[0212] When using distributed computing technology to process the distribution of detection tasks according to the fused data set, in one possible implementation, the data is sharded into multiple computing nodes. For example, a certain forest area is divided into 10 sub-areas, and each node processes 1000 data records to count the frequency and intensity of fire point occurrences, and finally summarizes to form task distribution data. This distributed processing can significantly improve efficiency. For the task distribution data, if the distribution characteristics exceed the preset threshold, adjust the verification technology parameters.

[0213] Preferably, assume that the threshold is set to 0.5 fire points per square kilometer. If a certain area reaches 0.7 fire points per square kilometer, the adjusted parameter set may be generated by increasing the verification frequency or adjusting the feature weights, which can ensure the reliability of the data.

[0214] When obtaining the detection data with precision fusion through the adjusted parameter set, the outliers in the fusion data group are filtered out. In one embodiment, if the temperature data of a certain fire point suddenly jumps to 100 degrees, which is significantly beyond the normal range, it is removed and the average value is recalculated to output the data with enhanced precision. This method helps to improve the accuracy of the data.

[0215] When dividing the fire point detection area using the clustering analysis technique based on the data with enhanced precision, the density-based clustering method is used. Exemplarily, assume that there are 50 fire point data points in a certain area. By setting the rules of a radius of 1 kilometer and a minimum number of points of 5, it is divided into 3 fire point dense areas. This division can clearly reflect the distribution law of the fire points.

[0216] For the area division result, the monitoring result data is generated through the spatial mapping technique. Specifically, the divided area is mapped to a two-dimensional grid. For example, a certain area is divided into three sub-areas A, B, and C. The number of fire points in area A is 20, in area B is 15, and in area C is 10. Intuitive monitoring result data is generated through grid processing. This mapping is convenient for subsequent analysis and visualization.

[0217] When extracting the key features from the final monitoring data, attention is paid to indicators such as fire point density, diffusion speed, and duration. For example, a certain fire point distribution feature set shows a density of 0.6 fire points per square kilometer, a diffusion speed of 2 kilometers per hour, and a duration of 3 hours.

[0218] The multi-source data fusion in this embodiment can improve the feature integrity, the distributed computing ensures the large-scale data processing ability, and the clustering analysis and spatial mapping optimize the regional management efficiency. These links support each other to form a complete analysis chain of the fire point distribution characteristics.

[0219] The above is only the preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A forest fire monitoring method based on low-earth orbit satellites, characterized in that, Including: Obtain multi-sensor data from a low-earth orbit satellite platform, process the heterogeneity of the sensor data through a preset denoising model, and smooth the data of different bands using adaptive filtering technology to obtain a fusion data set; For the fusion data set, when identifying complex scenarios, if the influence of noise interference exceeds the preset threshold, extract features through a convolutional neural network, and enhance the fire point features by combining multi-scale analysis technology to obtain a feature set; Obtain multi-satellite collaborative observation data, match the multi-source collaborative coverage data with the feature set through time series alignment technology to obtain an enhanced data set with enhanced high-frequency observation capabilities; For the enhanced data set, if the requirements for complex scenario identification increase, optimize the adaptability of image analysis through a deep learning algorithm, and adjust the model parameters by combining transfer learning technology to output an optimized result; Extract key region features from the optimized result, and use a distributed computing framework to perform data sharing processing on the multi-satellite collaborative complementarity to obtain a coverage-enhanced data set; According to the coverage-enhanced data set, analyze the observation frequency and integrity through a preset timeliness evaluation model. If the timeliness integrity is lower than expected, dynamically adjust the multi-satellite collaborative scheduling strategy and output a scheduling result; For the scheduling result, use reinforcement learning technology to iteratively optimize the robustness of the algorithm. By simulating the fire point detection task under complex meteorological conditions, obtain a robustness-enhanced model; Obtain the final fire point detection data from the robustness-enhanced model, verify the data fusion accuracy through multi-source data verification technology, and output the final monitoring result.

2. The method according to claim 1, wherein: The process of obtaining the fusion data set includes: Obtain sensor data generated by multiple sensors from a low-earth orbit satellite, process the heterogeneous characteristics through a preset denoising model, and use a low-earth orbit satellite platform equipped with multiple sensors to obtain multi-source data of a forest area; Preprocess the obtained sensor data to eliminate data heterogeneity between different sensors to obtain the fusion data set; wherein, the preprocessing includes data format unification, time synchronization, and spatial registration.

3. The method according to claim 1, wherein: The process of obtaining the feature set includes: Extract features from the fusion data set through a convolutional neural network, and use a spatial pyramid transformation or a feature pyramid network structure to fuse feature maps of different scales; At the same time, during the feature extraction process, introduce an attention mechanism to enhance the basic features, and assign weights to different features to highlight the fire point features and suppress noise interference; Fuse the enhanced features with the multi-scale features to obtain the feature set.

4. The method according to claim 1, wherein: The process of obtaining the enhanced data set includes: Obtain multi-source data through multi-satellite observations, and use time series alignment technology to process data alignment to obtain a first aligned data set; Extract collaborative coverage information from the aligned first data set, and fuse it with the features of the feature set through matching processing to obtain a second fused data set; For the fused second data set, coverage analysis is adopted to determine the spatial distribution characteristics of multi-source data, and a distribution characteristic set is obtained; If the distribution characteristic set meets the preset threshold conditions, the distribution characteristic set is classified by the random forest algorithm to judge the high-frequency observation areas, and a region set is obtained; According to the corresponding relationship between the region set and the time series, the time distribution characteristics of high-frequency observations are obtained, and a time characteristic set is obtained; Through the mapping relationship between the time characteristic set and the enhanced data set, the enhanced areas of the observation ability are determined, and an enhanced area set is obtained; The enhanced area set is optimized by using data enhancement technology to obtain an optimized enhanced data set.

5. The method according to claim 1, wherein The process of optimizing the image analysis adaptability through a deep learning algorithm, adjusting the model parameters by combining transfer learning technology, and outputting the optimization result includes: Performing initial processing on the enhanced data set through a deep learning algorithm to obtain a preliminary analysis result; According to the preliminary analysis result, the transfer learning technology is used to adjust the model parameters for complex scenarios to obtain an optimized parameter set; If the optimized parameter set meets the preset threshold conditions, feature extraction is performed on the complex scenarios through image analysis technology to obtain a feature set; The enhanced data set is matched by using the feature set to obtain a matched data set; The distribution characteristics of scene recognition are determined through the matched data set to obtain a distribution feature set; According to the mapping relationship between the distribution feature set and the optimized parameter set, a mapped data set is obtained; The mapped data set is optimized by using data enhancement technology to obtain an optimization result.

6. The method according to claim 1, wherein The process of extracting key region features from the optimization result and performing data sharing processing on the multi-satellite collaborative complementarity by using a distributed computing framework to obtain a coverage-enhanced data set includes: Determining the distribution range of region features through key regions, and obtaining a preliminary feature set by using feature extraction technology; Dividing the task units of multi-satellite collaboration according to the preliminary feature set to obtain a collaborative complementary allocation scheme; Performing data sharing processing on the allocation scheme by using a distributed computing framework to obtain a shared data group; If the shared data group meets the preset threshold conditions, the data range is adjusted by using coverage enhancement technology to obtain an enhanced data group; Performing processing flow optimization on the enhanced data group to obtain an optimized data unit; Judging the complementary distribution of multi-satellite collaboration through the mapping relationship between the optimized data unit and the region features to obtain a coverage-enhanced data set; Verifying the coverage-enhanced data set by using a computing framework to determine the final distribution characteristics.

7. The method according to claim 1, wherein According to the coverage-enhanced data set, analyzing the observation frequency and integrity through a preset timeliness evaluation model. If the timeliness integrity is lower than expected, dynamically adjusting the multi-satellite collaborative scheduling strategy and outputting the scheduling result includes: Analyzing the observation frequency and integrity through a preset timeliness evaluation model by using the coverage-enhanced data set to obtain an analysis data group; Based on the analyzed data group, determine whether the integrity is lower than a preset threshold condition, and obtain a determination data unit; Through the determination data unit, adopt dynamic adjustment technology to process the collaborative scheduling scheme to obtain an adjusted data group; For the adjusted data group, obtain the task allocation information of multi-satellite cooperation and determine the task allocation set; Through the task allocation set, adopt scheduling strategy optimization technology to process the data range to obtain an optimized scheduling unit; Based on the optimized scheduling unit, judge the resource distribution characteristics of multi-satellite cooperation to obtain a scheduling result; For the scheduling result, adopt a distributed computing framework to verify data consistency and determine the final scheduling data group.

8. The method according to claim 1, wherein For the scheduling result, the process of iteratively optimizing the robustness of the algorithm using reinforcement learning technology and obtaining a robustness-enhanced model by simulating the fire point detection task under complex meteorological conditions includes: For the data of the final scheduling data group, adopt reinforcement learning technology to process the feature distribution under complex meteorological conditions to obtain a feature-enhanced data set; Through the feature-enhanced data set, obtain the boundary conditions of the fire point detection task and determine the boundary feature set; For the boundary feature set, adopt task simulation technology to generate multi-scenario detection data to obtain a simulated data group; Based on the simulated data group, if the distribution characteristics of the detection task exceed the preset threshold, adjust the algorithm robustness parameters to obtain an adjusted parameter set; Through the adjusted parameter set, obtain the iteratively optimized model structure and determine the model framework; For the model framework, adopt distributed computing to verify the stability of fire point detection to obtain a robustness-enhanced model.

9. The method according to claim 1, wherein The process of obtaining the final fire point detection data from the robustness-enhanced model, verifying the data fusion accuracy through multi-source data verification technology, and outputting the final monitoring result includes: Extract the fire point detection data from the robustness-enhanced model, and fuse the data features through multi-source data verification technology to obtain a fused data group; Based on the fused data group, adopt distributed computing technology to process the detection task distribution to obtain task distribution data; For the task distribution data, if the distribution characteristics exceed the preset threshold, adjust the verification technology parameters to determine the adjusted parameter set; Through the adjusted parameter set, obtain the detection data after precision fusion and output precision-enhanced data; Based on the precision-enhanced data, adopt clustering analysis technology to divide the fire point detection area to obtain an area division result; For the area division result, generate monitoring result data through spatial mapping technology to determine the final monitoring data; Extract the key features from the final monitoring data to obtain the final monitoring result including the fire point distribution feature set.

Images