Forest fire analysis method and system based on satellite remote sensing

Abstract

The application provides a forest fire analysis method and system based on satellite remote sensing, acquires a continuous acquisition thermal infrared, near infrared and visible light dimension satellite remote sensing image sequence of a target forest area, performs time sequence correlation modeling to generate a time sequence correlation feature set; inputs the set into a pre-trained forest fire dynamic analysis model to extract a forest fire suspected area feature set; based on a pre-set forest fire space-time evolution rule library, performs multi-condition constraint verification on the suspected area features to obtain a confirmed area feature set; generates forest fire real-time state description information and spread trend prediction description information according to the confirmed area features; finally generates forest fire prevention and control instructions containing area coordinate chains and sends them to a target forest management terminal. The application realizes accurate identification and dynamic prediction of forest fires through multi-dimensional time sequence correlation, dynamic model analysis and space-time rule verification.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, in particular to a forest fire analysis method and system based on satellite remote sensing. BACKGROUND

[0002] As a global natural disaster, the monitoring and prevention of forest fires are crucial for ecological protection and human safety. Traditional forest fire monitoring mainly relies on ground patrols, observation towers, and aerial inspections, which have limitations such as limited coverage, delayed response, and geographical constraints. With the development of remote sensing technology, satellite remote sensing has gradually become the core means of forest fire monitoring due to its wide coverage and fast data updating. However, existing satellite remote sensing fire analysis methods still have significant shortcomings: on the one hand, the temporal correlation of multi-dimensional remote sensing data (such as thermal infrared, near-infrared, and visible light) is not fully exploited, resulting in insufficient accuracy and sensitivity of fire feature extraction; on the other hand, the verification of suspected fire areas relies too much on single feature thresholds, lacks dynamic constraints on vegetation types, terrain slope, and other spatiotemporal evolution rules, and is easily affected by factors such as solar reflection and terrain shadow, resulting in high false positive rates. In addition, existing methods for predicting fire spread trends are mostly based on static models and do not consider the synergistic effect of real-time state and historical rules, making it difficult to meet the needs of dynamic prevention and control. Therefore, there is an urgent need for a forest fire analysis method that integrates multi-dimensional temporal correlation modeling, dynamic background suppression, and multi-condition constraint verification to improve the real-time, accuracy, and prediction reliability of fire monitoring. SUMMARY

[0003] In view of the above-mentioned problems, in combination with the first aspect of the present application, the embodiments of the present application provide a forest fire analysis method based on satellite remote sensing, which comprises:

[0004] acquiring a satellite remote sensing image sequence of a target forest area, the satellite remote sensing image sequence containing continuously collected thermal infrared, near-infrared, and visible light images covering the target forest area and corresponding carrying acquisition time period identifiers;

[0005] performing temporal correlation modeling processing on the satellite remote sensing image sequence to generate a set of temporal correlation features;

[0006] inputting the set of temporal correlation features into a pre-trained forest fire dynamic analysis model to obtain a set of forest fire suspected area features;

[0007] based on a pre-set forest fire spatiotemporal evolution rule library, performing multi-condition constraint verification processing on the set of forest fire suspected area features to obtain a set of forest fire confirmed area features;

[0008] generate real-time state description information and spread trend prediction description information of the forest fire according to the confirmed region feature set of the forest fire, the real-time state description information including each dimension feature response intensity of the fire region and vegetation burning associated features, and the spread trend prediction description information including fire expansion direction features and spread speed associated features;

[0009] generate a forest fire prevention and control instruction including a region coordinate chain based on the real-time state description information and the spread trend prediction description information, and send the forest fire prevention and control instruction to a target forest management and protection terminal.

[0010] In still another aspect, the embodiment of the present application also provides a forest fire analysis system based on satellite remote sensing, which comprises a processor and a machine readable storage medium, the machine readable storage medium is connected with the processor, the machine readable storage medium is used for storing programs, instructions or codes, and the processor is used for executing the programs, instructions or codes in the machine readable storage medium to realize the above method.

[0011] Based on the above aspects, the embodiment of the present application can comprehensively capture the dynamic feature changes of the fire in different dimensions and different time periods by acquiring a multi-dimensional satellite remote sensing image sequence (thermal infrared, near infrared, visible light) of a target forest region and performing time sequence correlation modeling processing, generate a time sequence correlation feature set including thermal radiation, vegetation reflection and color response changes, further, by using a pre-trained forest fire dynamic analysis model, combined with dimension collaborative calibration and signal enhancement technology, the background interference can be effectively suppressed, the weak fire signal can be highlighted, and a high-precision forest fire suspected region feature set can be generated. On this basis, based on a pre-set forest fire space-time evolution rule library, multi-condition constraint verification processing is performed, the authenticity of the fire suspected region can be verified from the spread speed, direction and thermal radiation response intensity, and an accurate and reliable forest fire confirmed region feature set is generated. Finally, real-time state description information (including each dimension feature response intensity of the fire region and vegetation burning associated features) and spread trend prediction description information (including fire expansion direction and spread speed associated features) are generated according to the confirmed region feature set, and a forest fire prevention and control instruction including a region coordinate chain is generated, so that the precision and real-time of fire prevention and control are realized. Through the synergistic effect of multi-dimensional data fusion, dynamic feature modeling and multi-condition constraint verification, the accuracy, timeliness and anti-interference ability of forest fire monitoring are significantly improved. BRIEF DESCRIPTION OF DRAWINGS

[0012] Figure 1 is an execution flow schematic diagram of the forest fire analysis method based on satellite remote sensing provided by the embodiment of the present application.

[0013] Figure 2is a schematic diagram of exemplary hardware and software components of a satellite remote sensing based forest fire analysis system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The present invention will be described in detail below with reference to the accompanying drawings, Figure 1 is a flowchart of a satellite remote sensing based forest fire analysis method according to an embodiment of the present invention, which will be described in detail below.

[0015] Step S110: Obtain a satellite remote sensing image sequence of a target forest region, which includes continuously collected thermal infrared dimension images, near-infrared dimension images and visible light dimension images covering the target forest region and corresponding carrying collection time period identifiers.

[0016] In this embodiment, in order to analyze the forest fire in the target forest region, the satellite remote sensing image sequence of the region needs to be obtained first. The satellite continuously collects the target forest region through specific sensors to obtain thermal infrared dimension images, near-infrared dimension images and visible light dimension images. The thermal infrared dimension images can reflect the thermal radiation of the target forest region and play an important role in detecting high-temperature areas of the fire; the near-infrared dimension images can reflect the growth conditions and reflection characteristics of the vegetation and help analyze the impact of the fire on the vegetation; the visible light dimension images provide intuitive visual information of the target forest region, such as terrain and topography, etc.

[0017] Step S120: Perform time sequence correlation modeling processing on the satellite remote sensing image sequence to generate a time sequence correlation feature set.

[0018] In order to mine the time sequence correlation information between different dimension images in the satellite remote sensing image sequence, time sequence correlation modeling processing needs to be performed.

[0019] Step S121: Group the satellite remote sensing image sequence according to the collection time period, determine a reference image group and a comparison image group of adjacent collection time periods, the reference image group is a set of thermal infrared dimension images, near-infrared dimension images and visible light dimension images collected in the previous time period, and the comparison image group is a set of thermal infrared dimension images, near-infrared dimension images and visible light dimension images collected in the next time period.

[0020] Group the satellite remote sensing image sequence according to the collection time period, and select two adjacent collection time periods. The thermal infrared dimension images, near-infrared dimension images and visible light dimension images collected in the previous time period form the reference image group, and the corresponding images collected in the next time period form the comparison image group. By comparing the two groups of images, the feature change of the target forest region within the adjacent time periods can be analyzed.

[0021] Step S122: For the thermal infrared dimension, compare and analyze the thermal infrared dimension images in the reference image set and the thermal infrared dimension images in the comparison image set, extract the thermal radiation response changes of the corresponding image regions, and generate a thermal infrared dimension spectral response change feature map, wherein the feature value of each image region in the thermal infrared dimension spectral response change feature map reflects the thermal radiation response change of the image region.

[0022] For the thermal infrared dimension, compare the thermal infrared dimension images in the reference image set and the thermal infrared dimension images in the comparison image set pixel by pixel. Calculate the thermal radiation response change of each corresponding image region, and in this way, extract the thermal radiation response change information. Integrate these information into the thermal infrared dimension spectral response change feature map, and the feature value of each image region in the response change feature map represents the thermal radiation response change of the region.

[0023] Step S123: For the near-infrared dimension, compare and analyze the near-infrared dimension images in the reference image set and the near-infrared dimension images in the comparison image set, extract the vegetation reflection response changes of the corresponding image regions, and generate a near-infrared dimension spectral response change feature map, wherein the feature value of each image region in the near-infrared dimension spectral response change feature map reflects the vegetation reflection response change of the image region.

[0024] In the near-infrared dimension, the near-infrared dimension images of the reference image set and the comparison image set are also compared and analyzed. Through comparison, the vegetation reflection response changes of the corresponding image regions are extracted. These change information generates a near-infrared dimension spectral response change feature map, wherein the feature value of each image region reflects the vegetation reflection response change of the region.

[0025] Step S124: For the visible light dimension, compare and analyze the visible light dimension images in the reference image set and the visible light dimension images in the comparison image set, extract the color response changes of the corresponding image regions, and generate a visible light dimension spectral response change feature map, wherein the feature value of each image region in the visible light dimension spectral response change feature map reflects the color response change of the image region.

[0026] For the visible light dimension, compare the visible light dimension images of the reference image set and the comparison image set. Extract the color response changes of the corresponding image regions, and generate a visible light dimension spectral response change feature map. The feature value of each image region in the response change feature map reflects the color response change of the region.

[0027] Step S125: Construct a cross-dimension time correlation matrix, and take the feature values of the corresponding image regions in the thermal infrared dimension spectrum response change feature map, the near-infrared dimension spectrum response change feature map and the visible light dimension spectrum response change feature map after standardization processing as matrix elements, and calculate the time correlation coefficient between the feature values of the corresponding image regions in different dimensions. The time correlation coefficient represents the synchronism of the feature response changes in different dimensions.

[0028] A cross-dimension time correlation matrix is constructed, and the feature values of the corresponding image regions in the thermal infrared dimension spectrum response change feature map, the near-infrared dimension spectrum response change feature map and the visible light dimension spectrum response change feature map are standardized. Standardization processing can make the feature values in different dimensions comparable. The standardized feature values are taken as matrix elements, and the time correlation coefficient between the feature values of the corresponding image regions in different dimensions is calculated. This time correlation coefficient can measure the synchronism of the feature response changes in different dimensions, that is, whether the feature changes in different dimensions have consistency in time.

[0029] Step S126: According to the time correlation coefficient, the weighted fusion processing is performed on the standardized spectrum response change feature maps of each dimension to generate a cross-dimension collaborative correlation feature map.

[0030] According to the calculated time correlation coefficient, the weighted fusion processing is performed on the standardized spectrum response change feature maps of each dimension. The dimension with a larger time correlation coefficient is given a higher weight in the fusion process, and vice versa. Through this weighted fusion, a cross-dimension collaborative correlation feature map is generated, which integrates the feature response change information in different dimensions.

[0031] Step S127: Image region segmentation is performed on the cross-dimension collaborative correlation feature map to obtain a plurality of collaborative correlation feature sub-regions, and statistical features of each collaborative correlation feature sub-region are extracted, the statistical features including feature response mean, feature response variance and feature response extreme value.

[0032] The cross-dimension collaborative correlation feature map is segmented into a plurality of collaborative correlation feature sub-regions. Each sub-region has similar feature response changes. Then, the statistical features of each collaborative correlation feature sub-region are extracted, including the feature response mean, the feature response variance and the feature response extreme value. The feature response mean reflects the average level of the feature response in the sub-region, the feature response variance reflects the dispersion degree of the feature response, and the feature response extreme value represents the maximum and minimum values of the feature response in the sub-region.

[0033] Step S128: The statistical features are associated and integrated with the regional level features of the thermal infrared dimension spectral response change feature map, the near-infrared dimension spectral response change feature map, and the visible light dimension spectral response change feature map respectively, to generate a time sequence associated feature set.

[0034] The extracted statistical features are associated and integrated with the regional level features of the thermal infrared dimension spectral response change feature map, the near-infrared dimension spectral response change feature map, and the visible light dimension spectral response change feature map respectively. The regional level features include the feature change information of different regions in each dimension image. Through this association and integration, a time sequence associated feature set is generated, which contains the associated feature information of different dimension images in the time sequence.

[0035] Step S130: The time sequence associated feature set is input into the pre-trained forest fire dynamic analysis model to obtain a forest fire suspected area feature set.

[0036] The pre-trained forest fire dynamic analysis model is used to process the time sequence associated feature set to identify the forest fire suspected area.

[0037] Step S131: The thermal infrared dimension related features and the near-infrared dimension related features are extracted from the time sequence associated feature set, and the thermal infrared dimension related features and the near-infrared dimension related features are input into the dimension collaborative calibration layer of the forest fire dynamic analysis model to identify a set of thermal radiation abnormal regions in the thermal infrared dimension related features and a set of vegetation reflection abnormal regions in the near-infrared dimension related features.

[0038] The thermal infrared dimension related features and the near-infrared dimension related features are extracted from the time sequence associated feature set. These two types of features are input into the dimension collaborative calibration layer of the forest fire dynamic analysis model. In this layer, through specific algorithms and rules, a set of thermal radiation abnormal regions in the thermal infrared dimension related features is identified, and these regions have significantly higher thermal radiation response than normal levels; at the same time, a set of vegetation reflection abnormal regions in the near-infrared dimension related features is identified, and these regions have a large difference in vegetation reflection response from normal conditions.

[0039] Step S132: The set of thermal radiation abnormal regions and the set of vegetation reflection abnormal regions are respectively subjected to binary processing to generate abnormal region identification maps with the same spatial resolution, and then the spatial overlap ratio of the image regions corresponding to the two abnormal region identification maps is calculated to screen out a set of common abnormal regions with a spatial overlap ratio satisfying a preset condition.

[0040] The two sets of abnormal regions are binarized respectively, and the abnormal regions are marked with a specific value and the normal regions are marked with another value, to generate abnormal region identification maps with the same spatial resolution. Then, the spatial overlap ratio of the image regions corresponding to the two abnormal region identification maps is calculated. The regions whose spatial overlap ratio meets the preset condition are selected as the common abnormal region set. These common abnormal regions are more likely to be the regions where the forest fire occurs.

[0041] Step S133: constructing a dimension calibration mapping relationship based on the common abnormal region set, adjusting the response scale of the normalized near-infrared dimension related feature so that the change trend of the normalized near-infrared dimension related feature at the common abnormal region set is consistent with that of the normalized thermal infrared dimension related feature, to obtain the calibrated near-infrared dimension feature.

[0042] A dimension calibration mapping relationship is constructed based on the common abnormal region set. First, the near-infrared dimension related feature and the thermal infrared dimension related feature are normalized respectively, so that they have the same dimension and scale. Then, according to the dimension calibration mapping relationship, the response scale of the normalized near-infrared dimension related feature is adjusted so that the change trend of the normalized near-infrared dimension related feature at the common abnormal region set is consistent with that of the normalized thermal infrared dimension related feature. In this way, the scale difference between different dimension features can be eliminated, and the calibrated near-infrared dimension feature is obtained.

[0043] Step S1331: for each abnormal region in the common abnormal region set, the thermal radiation response change amount of the abnormal region in the thermal infrared dimension related feature and the vegetation reflection response change amount of the abnormal region in the near-infrared dimension related feature are extracted respectively.

[0044] For each abnormal region in the common abnormal region set, the thermal radiation response change amount and the vegetation reflection response change amount are extracted from the thermal infrared dimension related feature and the near-infrared dimension related feature respectively. The thermal radiation response change amount reflects the change degree of the thermal radiation of the region, and the vegetation reflection response change amount reflects the change of the vegetation reflection characteristics of the region.

[0045] Step S1332: the thermal radiation response change amount and the vegetation reflection response change amount are normalized respectively, a correlation regression model is constructed with the normalized thermal radiation response change amount as the independent variable and the normalized vegetation reflection response change amount as the dependent variable, the regression correlation parameter of the correlation regression model is calculated, and the correlation regression model describes the correlation between the thermal radiation response change amount and the vegetation reflection response change amount.

[0046] The extracted thermal radiation response change and the extracted vegetation reflectance response change are standardized respectively to make them comparable. A regression correlation model is constructed with the standardized thermal radiation response change as the independent variable and the standardized vegetation reflectance response change as the dependent variable. The regression correlation parameters of the regression correlation model are calculated by analyzing the data in the common abnormal region set, and the regression correlation parameters describe the correlation between the thermal radiation response change and the vegetation reflectance response change.

[0047] Step S1333: generating a dimension calibration function based on the regression correlation model, the input of the dimension calibration function being the vegetation reflectance response change of the near-infrared dimension-related feature, and the output being the calibrated vegetation reflectance response change.

[0048] A dimension calibration function is generated according to the calculated regression correlation model. The input of the dimension calibration function is the vegetation reflectance response change of the near-infrared dimension-related feature, and the output is the calibrated vegetation reflectance response change. Through the function, the near-infrared dimension-related feature can be calibrated to make it consistent with the change trend of the thermal infrared dimension-related feature at the common abnormal region set.

[0049] Step S1334: extracting the vegetation reflectance response change of all image regions contained in the near-infrared dimension-related feature from the time-series correlation feature set, and inputting the vegetation reflectance response change of each image region into the dimension calibration function respectively to obtain the calibrated vegetation reflectance response change corresponding to each image region.

[0050] The vegetation reflectance response change of all image regions contained in the near-infrared dimension-related feature is extracted from the time-series correlation feature set. The vegetation reflectance response change of each image region is input into the dimension calibration function respectively to obtain the calibrated vegetation reflectance response change corresponding to each image region.

[0051] Step S1335: constructing a calibration error evaluation model, and converting the calibrated vegetation reflectance response change of each image region and the corresponding thermal radiation response change into relative change rates of the same dimension or performing standardization processing, and then calculating an error measurement value.

[0052] A calibration error evaluation model is constructed to evaluate the effect of dimension calibration. The calibrated vegetation reflectance response change of each image region and the corresponding thermal radiation response change are converted into relative change rates of the same dimension or standardized to ensure their comparability. Then, an error measurement value is calculated, which reflects the difference between the calibrated vegetation reflectance response change and the corresponding thermal radiation response change.

[0053] Step S1336: Image regions with error metric values satisfying the preset error condition are screened out as valid calibration regions, and distribution density of the valid calibration regions in the entire target forest region is counted.

[0054] Image regions with error metric values satisfying the preset error condition are screened out as valid calibration regions. Distribution density of the valid calibration regions in the entire target forest region is counted, which can reflect the effectiveness and reliability of the dimensional calibration.

[0055] Step S1337: If the distribution density satisfies a preset density condition, it is determined that the dimensional calibration function is valid, and a calibrated near-infrared dimensional feature is generated based on the calibrated vegetation reflectance response changes of all image regions.

[0056] If the distribution density of the valid calibration regions satisfies the preset density condition, it means that the dimensional calibration function is valid. A calibrated near-infrared dimensional feature is generated based on the calibrated vegetation reflectance response changes of all image regions, which has a consistent change trend with the thermal infrared dimensional related features at the common anomaly region set.

[0057] Step S1338: If the distribution density does not satisfy the preset density condition, the parameters of the correlation regression model are adjusted, a non-linear correlation term is added to optimize the model, and the steps of constructing the dimensional calibration function, calculating the calibrated vegetation reflectance response changes, constructing the calibration error evaluation model to calculate the error metric values, screening the valid calibration regions, and counting the distribution density thereof are re-executed based on the optimized correlation regression model, until valid calibration regions with distribution density satisfying the preset density condition are obtained. A calibrated near-infrared dimensional feature is generated based on the calibrated vegetation reflectance response changes of all image regions at this time.

[0058] If the distribution density of the valid calibration regions does not satisfy the preset density condition, it means that the dimensional calibration effect is not good. At this time, the parameters of the correlation regression model are adjusted, a non-linear correlation term is added to optimize the model, and the steps of constructing the dimensional calibration function, calculating the calibrated vegetation reflectance response changes, constructing the calibration error evaluation model to calculate the error metric values, screening the valid calibration regions, and counting the distribution density thereof are re-executed based on the optimized correlation regression model, until valid calibration regions with distribution density satisfying the preset density condition are obtained. Then, a calibrated near-infrared dimensional feature is generated based on the calibrated vegetation reflectance response changes of all image regions at this time.

[0059] Step S134: The calibrated near-infrared dimensional feature is channel spliced with the thermal infrared dimensional related features and the visible light dimensional related features in the time series correlation feature set to generate a dimensional collaborative feature vector.

[0060] The calibrated near-infrared dimension feature is channel spliced with the thermal-infrared dimension-related feature and the visible-light dimension-related feature in the time-series correlation feature set. Channel splicing is to combine features of different dimensions in a certain order to generate a dimension-collaborative feature vector. The dimension-collaborative feature vector integrates feature information of different dimensions, which helps to more accurately identify forest fire suspected areas.

[0061] Step S135: input the dimension-collaborative feature vector into the fire signal enhancement layer of the forest fire dynamic analysis model, and perform hierarchical amplification processing on the weak fire signals in the dimension-collaborative feature vector through the hierarchical signal enhancer in the fire signal enhancement layer to generate an enhanced feature vector, and the parameters of the hierarchical signal enhancer are set based on the sample weak fire signal features in the historical fire samples.

[0062] The dimension-collaborative feature vector is input into the fire signal enhancement layer of the forest fire dynamic analysis model. The hierarchical signal enhancer in the fire signal enhancement layer sets parameters according to the sample weak fire signal features in the historical fire samples. The hierarchical signal enhancer amplifies the weak fire signals in the dimension-collaborative feature vector to enhance these weak signals, making them easier to be identified. After amplification processing, an enhanced feature vector is generated.

[0063] Step S136: perform pattern matching of the enhanced feature vector with each background feature library in the dynamic background suppression module respectively, and screen out abnormal feature components that do not match all background feature patterns, and the dynamic background suppression module includes a vegetation background feature library, a terrain background feature library and an artificial target background feature library.

[0064] The enhanced feature vector is pattern matched with each background feature library in the dynamic background suppression module. The dynamic background suppression module includes a vegetation background feature library, a terrain background feature library and an artificial target background feature library. These background feature libraries record the feature patterns of vegetation, terrain and artificial targets under normal circumstances. Through comparison, abnormal feature components that do not match all background feature patterns are screened out, and these abnormal feature components are more likely to be signals of forest fires.

[0065] Step S1361: extract a feature template of normal vegetation from the vegetation background feature library of the dynamic background suppression module, and the feature template includes a thermal radiation response range of normal vegetation in the thermal-infrared dimension, a reflection response range in the near-infrared dimension, and a color response range in the visible-light dimension.

[0066] The feature template of normal vegetation is extracted from the vegetation background feature library of the dynamic background suppression module. The feature template includes the thermal radiation response range of normal vegetation in the thermal infrared dimension, the reflection response range in the near infrared dimension, and the color response range in the visible light dimension. These ranges can be used as reference standards for determining whether it is normal vegetation.

[0067] Step S1362: Each image region feature in the enhanced feature vector is compared with the feature template of normal vegetation, respectively, to determine whether the image region feature falls within the thermal radiation response range, the reflection response range, and the color response range. If all fall within the corresponding response range, it is marked as a vegetation background feature component.

[0068] Each image region feature in the enhanced feature vector is compared with the feature template of normal vegetation. It is determined whether each image region feature falls within the corresponding range specified by the feature template in terms of thermal radiation response in the thermal infrared dimension, reflection response in the near infrared dimension, and color response in the visible light dimension. If all fall within the corresponding range, the image region feature is marked as a vegetation background feature component.

[0069] Step S1363: The feature template of terrain targets is extracted from the terrain background feature library of the dynamic background suppression module. The feature template includes the thermal radiation response distribution pattern of mountains in the thermal infrared dimension, the reflection response distribution pattern of water bodies in the near infrared dimension, and the color response distribution pattern in the visible light dimension.

[0070] The feature template of terrain targets is extracted from the terrain background feature library of the dynamic background suppression module. The feature template includes the thermal radiation response distribution pattern of mountains in the thermal infrared dimension, the reflection response distribution pattern of water bodies in the near infrared dimension, and the color response distribution pattern in the visible light dimension. These patterns can be used to identify terrain targets.

[0071] Step S1364: The image region features that are not marked as vegetation background feature components are compared with the feature template of terrain targets, respectively, to determine whether they conform to the feature distribution pattern. If so, they are marked as terrain background feature components.

[0072] The image region features that are not marked as vegetation background feature components are compared with the feature template of terrain targets, respectively. It is determined whether these image region features conform to the feature distribution pattern of terrain targets. If so, they are marked as terrain background feature components.

[0073] Step S1365: The feature template of artificial targets is extracted from the artificial target background feature library of the dynamic background suppression module. The feature template includes the thermal radiation response fluctuation pattern of buildings in the thermal infrared dimension, the reflection response uniformity pattern of roads in the near infrared dimension, and the shape feature pattern in the visible light dimension.

[0074] The feature template of the artificial target is extracted from the artificial target background feature library of the dynamic background suppression module. The feature template includes the thermal radiation response fluctuation pattern in the thermal infrared dimension, the reflection response uniformity pattern of the road in the near infrared dimension, and the shape feature pattern in the visible light dimension. These patterns can be used to identify artificial targets.

[0075] Step S1366: Compare the image region features that are not marked as vegetation or terrain background feature components with the feature templates of the artificial target respectively, and determine whether they match the feature patterns. If so, mark them as artificial target background feature components.

[0076] Compare the image region features that are not marked as vegetation or terrain background feature components with the feature templates of the artificial target respectively. Determine whether these image region features match the feature patterns of the artificial target. If so, mark them as artificial target background feature components.

[0077] Step S1367: Collect all image region features that are not marked as any of the above background feature components as abnormal feature components that do not match any background feature pattern.

[0078] Collect all image region features that are not marked as vegetation background feature components, terrain background feature components, or artificial target background feature components, and mark them as abnormal feature components that do not match any background feature pattern. These abnormal feature components may be an indication of a forest fire.

[0079] Step S137: Perform spatial clustering processing on the abnormal feature components, and group abnormal feature components that are spatially adjacent and have similar feature response change trends into the same feature cluster.

[0080] Perform spatial clustering processing on the selected abnormal feature components. Group abnormal feature components that are spatially adjacent and have similar feature response change trends into the same feature cluster. Through clustering, scattered abnormal feature components can be integrated into feature clusters with certain spatial continuity.

[0081] Step S138: Extract the spatial distribution features and feature intensity features of each feature cluster. The spatial distribution features include the contour shape, coverage range, and center position of the feature cluster, and the feature intensity features include the maximum feature response value and average feature response value within the feature cluster.

[0082] Extract the spatial distribution features and feature intensity features of each feature cluster. The spatial distribution features include the contour shape, coverage range, and center position of the feature cluster, which can describe the spatial distribution of the feature cluster. The feature intensity features include the maximum feature response value and average feature response value within the feature cluster, which reflect the feature intensity level of the feature cluster.

[0083] Step S139: Integrate the spatial distribution features of each feature cluster with the feature intensity features to generate a forest fire suspected area feature set containing multiple feature cluster information.

[0084] Integrate the spatial distribution features of each feature cluster with the feature intensity features. Through this integration, a forest fire suspected area feature set containing multiple feature cluster information is generated.

[0085] Step S140: Based on the preset forest fire spatiotemporal evolution rule library, perform multi-condition constraint verification processing on the forest fire suspected area feature set to obtain a forest fire confirmed area feature set.

[0086] Use the preset forest fire spatiotemporal evolution rule library to perform multi-condition constraint verification on the forest fire suspected area feature set to determine the true forest fire area.

[0087] Step S141: Extract the spatiotemporal feature patterns of forest fire spread from the forest fire spatiotemporal evolution rule library, which include the spread speed characteristics of fire in different vegetation types, the spread direction characteristics of fire under different terrain slopes, and the change characteristics of thermal radiation response intensity of fire in different collection periods.

[0088] Extract the spatiotemporal feature patterns of forest fire spread from the forest fire spatiotemporal evolution rule library. This spatiotemporal feature pattern includes the spread speed characteristics of fire in different vegetation types, the spread direction characteristics of fire under different terrain slopes, and the change characteristics of thermal radiation response intensity of fire in different collection periods.

[0089] Step S142: For each feature cluster in the forest fire suspected area feature set, extract the vegetation type information, terrain slope information, and collection period information of the location of the feature cluster.

[0090] For each feature cluster in the forest fire suspected area feature set, extract the vegetation type information, terrain slope information, and collection period information of the location of the feature cluster. Vegetation type information can be determined through near-infrared dimension information and visible light dimension information in satellite remote sensing images; terrain slope information can be obtained through digital elevation model data; and collection period information is obtained from the collection period identifier of the satellite remote sensing image.

[0091] Step S143: For each feature cluster, according to its vegetation type information, match corresponding fire spread speed features from the spatio-temporal feature pattern, and respectively determine whether the thermal radiation response intensity growth rate of the feature cluster conforms to the matched fire thermal radiation growth features, and whether the expansion rate of the feature cluster coverage conforms to the matched fire spread speed features. The feature response change rate includes the growth rate of the thermal radiation response intensity and the expansion rate of the feature cluster coverage.

[0092] For each feature cluster, according to its vegetation type information, match corresponding fire spread speed features from the spatio-temporal feature pattern. Then, respectively determine whether the thermal radiation response intensity growth rate of the feature cluster conforms to the matched fire thermal radiation growth features, and whether the expansion rate of the feature cluster coverage conforms to the matched fire spread speed features. If both the thermal radiation response intensity growth rate and the expansion rate of the feature cluster coverage conform to the matched features, it indicates that the change trend of the feature cluster is consistent with the spread law of the forest fire under the vegetation type.

[0093] Step S144: For each feature cluster, according to its terrain slope information, match corresponding fire spread direction features from the spatio-temporal feature pattern, and determine whether the expansion direction of the feature cluster conforms to the matched fire spread direction features, wherein the expansion direction of the feature cluster is determined by comparing the center position changes of the feature cluster in adjacent collection time periods.

[0094] According to the terrain slope information of each feature cluster, match corresponding fire spread direction features from the spatio-temporal feature pattern. The expansion direction of the feature cluster is determined by comparing the center position changes of the feature cluster in adjacent collection time periods, and then it is determined whether the expansion direction conforms to the matched fire spread direction features. If the expansion direction conforms, it further supports the possibility that the feature cluster is a forest fire area.

[0095] Step S145: For each feature cluster, according to its collection time period information, match corresponding fire thermal radiation response intensity change features from the spatio-temporal feature pattern, and determine whether the thermal radiation response intensity value of the feature cluster in the corresponding collection time period conforms to the matched thermal radiation response intensity change features.

[0096] According to the collection time period information of each feature cluster, match corresponding fire thermal radiation response intensity change features from the spatio-temporal feature pattern. Determine whether the thermal radiation response intensity value of the feature cluster in the corresponding collection time period conforms to the matched thermal radiation response intensity change features. If it conforms, it indicates that the thermal radiation response intensity change of the feature cluster is consistent with the law of the forest fire in the time period.

[0097] Step S146: Condition judgment is performed on each feature cluster. If the following three conditions are met simultaneously: the feature response change rate conforms to the matched fire heat radiation growth feature and fire spread speed feature; the expansion direction conforms to the matched fire spread direction feature; and the heat radiation response intensity value conforms to the matched heat radiation response intensity change feature, the feature cluster is marked as a candidate fire area feature cluster. If the three conditions are not met simultaneously, further analysis is performed on whether the feature cluster is affected by non-fire disturbance factors.

[0098] Comprehensive condition judgment is performed on each feature cluster. If the heat radiation response intensity growth rate and the expansion rate of the feature cluster coverage range conform to the matched features, the expansion direction conforms to the matched fire spread direction feature, and the heat radiation response intensity value conforms to the matched heat radiation response intensity change feature, the feature cluster is marked as a candidate fire area feature cluster.

[0099] If a feature cluster does not meet the above three conditions simultaneously, further analysis is needed to determine whether the feature cluster is affected by non-fire disturbance factors. Non-fire disturbance factors can include solar radiation, industrial heat sources, etc.

[0100] Step S147: For feature clusters that are not marked as candidate fire area feature clusters, the feature response mode is extracted, which includes the stability of the heat radiation response intensity and the concentration of the feature distribution.

[0101] For feature clusters that are not marked as candidate fire area feature clusters, the feature response mode is extracted. The feature response mode includes the stability of the heat radiation response intensity, i.e., whether the heat radiation response intensity remains relatively stable over time, and the concentration of the feature distribution, i.e., whether the feature distribution is concentrated in space.

[0102] Step S148: The feature response mode is compared with the non-fire disturbance feature mode library in the forest fire space-time evolution rule library. If a disturbance pattern is matched, the feature cluster is marked as a disturbance feature cluster and removed from the forest fire suspected area feature set.

[0103] The extracted feature response mode is compared with the non-fire disturbance feature mode library in the forest fire space-time evolution rule library. If a disturbance pattern is matched, it means that the feature cluster is caused by non-fire disturbance factors, and it is marked as a disturbance feature cluster and removed from the forest fire suspected area feature set.

[0104] Step S149: Collect the feature information of all candidate fire area feature clusters. For each candidate fire area feature cluster, the verification feature is extracted, which includes the heat radiation response sustained growth and the vegetation reflection response sustained decline.

[0105] The feature information of all candidate fire area feature clusters is collected, and the verification features of each candidate fire area feature cluster are extracted. The verification features include thermal radiation response sustained growth, i.e., whether the intensity of the thermal radiation response is continuously increasing, and vegetation reflection response sustained decline, i.e., whether the vegetation reflection response is continuously decreasing. These features can further verify whether the candidate fire area feature cluster is a true forest fire area.

[0106] Step S1410: Time series analysis is performed on the verification features of each candidate fire area feature cluster to determine whether the verification features maintain consistent change trends in consecutive collection time periods.

[0107] Time series analysis is performed on the verification features of each candidate fire area feature cluster.

[0108] Step S14101: For the verification features of each candidate fire area feature cluster, the feature response values in consecutive collection time periods are extracted to form a verification feature time series.

[0109] For the verification features of each candidate fire area feature cluster, the feature response values in consecutive collection time periods are extracted. These values are arranged in the order of the collection time periods to form a verification feature time series. The verification feature time series records the changes of the verification features over time.

[0110] Step S14102: For each verification feature time series, the change difference values of the feature response values of adjacent collection time periods are calculated to generate a respective difference value sequence, and the sign consistency of the difference value sequence is determined, which includes both positive values or both negative values.

[0111] For each verification feature time series, the change difference values of the feature response values of adjacent collection time periods are calculated. These difference values are arranged in the order of the collection time periods to generate a difference value sequence. The sign consistency of the difference value sequence is determined. If all elements in the difference value sequence are positive values or negative values, it indicates that the verification feature has a consistent change trend in adjacent collection time periods.

[0112] Step S14103: If the signs are consistent, it is preliminarily determined that the change trend is consistent; if the signs are not consistent, the moving average value of the difference value sequence is calculated, and it is determined whether the signs of the moving average values are consistent, and if they are consistent, it is still preliminarily determined that the change trend is consistent.

[0113] If the signs of the difference value sequence are consistent, it is preliminarily determined that the verification feature maintains a consistent change trend in consecutive collection time periods. If the signs are not consistent, the moving average value of the difference value sequence is calculated. By calculating the average value within a certain window, the fluctuations of the difference value sequence are smoothed. It is determined whether the signs of the moving average values are consistent, and if they are consistent, it is still preliminarily determined that the change trend is consistent.

[0114] Step S14104: Trend fitting processing is performed on the verification feature time sequence, a trend fitting line is generated through a fitting algorithm, a slope attribute of the trend fitting line is determined, and it is judged whether a variation amplitude of the slope attribute satisfies a preset slope condition.

[0115] Trend fitting processing is performed on the verification feature time sequence. A trend fitting line is generated through a fitting algorithm, such as the least square method. The trend fitting line can approximately represent the variation trend of the verification feature over time. The slope attribute of the trend fitting line is determined, which reflects the variation rate of the verification feature. It is judged whether the variation amplitude of the slope attribute satisfies the preset slope condition. If it does, it means that the variation trend of the verification feature is relatively stable.

[0116] Step S14105: If it is preliminarily determined that the variation trends are consistent and the variation amplitude of the slope attribute satisfies the preset slope condition, it is determined that the verification feature maintains a consistent variation trend in the continuous collection period. If it is preliminarily determined that the variation trends are consistent but the variation amplitude of the slope attribute does not satisfy the preset slope condition, it is checked whether there are other forest fire confirmation area feature clusters around the candidate fire area feature cluster. If there are and the distance satisfies the preset distance condition, it is determined that the variation trends are consistent.

[0117] If it is preliminarily determined that the variation trends are consistent and the variation amplitude of the slope attribute satisfies the preset slope condition, it is determined that the verification feature maintains a consistent variation trend in the continuous collection period. If it is preliminarily determined that the variation trends are consistent but the variation amplitude of the slope attribute does not satisfy the preset slope condition, it is checked whether there are other forest fire confirmation area feature clusters around the candidate fire area feature cluster. If there are and the distance satisfies the preset distance condition, it means that the candidate fire area feature cluster may be affected by the surrounding fires, and it is determined that the variation trends are consistent.

[0118] Step S14106: If there are no other forest fire confirmation area feature clusters around, the candidate fire area feature cluster is marked as a to-be-observed feature cluster and is reanalyzed after the next period of satellite remote sensing images is acquired.

[0119] If there are no other forest fire confirmation area feature clusters around, the candidate fire area feature cluster is marked as a to-be-observed feature cluster. After the next period of satellite remote sensing images is acquired, the verification feature of the feature cluster is reanalyzed.

[0120] Step S14107: If it is preliminarily determined that the variation trends are inconsistent, it is directly determined that the verification feature does not maintain a consistent variation trend.

[0121] If it is preliminarily determined that the variation trends are inconsistent, it is directly determined that the verification feature does not maintain a consistent variation trend in the continuous collection period.

[0122] Step S1411: If the verification feature keeps consistent change trend in the continuous acquisition period, the candidate fire area feature cluster is determined as a forest fire confirmed area feature cluster.

[0123] If the verification feature of a candidate fire area feature cluster keeps consistent change trend in the continuous acquisition period, the candidate fire area feature cluster is determined as a forest fire confirmed area feature cluster.

[0124] Step S1412: Collect feature information of all forest fire confirmed area feature clusters, and integrate to generate a forest fire confirmed area feature set, wherein the feature information includes spatial distribution features, feature intensity features, and time change features.

[0125] Collect feature information of all forest fire confirmed area feature clusters, including spatial distribution features, feature intensity features, and time change features. Integrate the feature information to generate a forest fire confirmed area feature set. The confirmed area feature set records detailed information of all confirmed forest fire areas.

[0126] Step S150: Generate real-time state description information and spread trend prediction description information of the forest fire according to the forest fire confirmed area feature set, wherein the real-time state description information includes feature response intensity of each dimension of the fire area and vegetation burning associated features, and the spread trend prediction description information includes fire expansion direction features and spread speed associated features.

[0127] Generate real-time state description information and spread trend prediction description information of the forest fire according to the forest fire confirmed area feature set.

[0128] Step S151: For each forest fire confirmed area feature cluster in the forest fire confirmed area feature set, extract the maximum thermal radiation response value in the thermal infrared dimension, compare the maximum thermal radiation response value with a preset thermal radiation intensity level library, and determine the fire burning intensity associated features corresponding to the feature cluster based on the comparison result.

[0129] For each forest fire confirmed area feature cluster in the forest fire confirmed area feature set, extract the maximum thermal radiation response value in the thermal infrared dimension. Compare the maximum thermal radiation response value with a preset thermal radiation intensity level library. The thermal radiation intensity level library records fire burning intensity levels corresponding to different thermal radiation intensities. According to the comparison result, determine the fire burning intensity associated features corresponding to the feature cluster.

[0130] Step S152: Extract the minimum reflection response value of each forest fire confirmed area feature cluster in the near-infrared dimension, compare it with a preset vegetation burning degree level library, and determine the corresponding vegetation burning degree associated features.

[0131] The minimum reflection response value of each forest fire confirmation area feature cluster in the near-infrared dimension is extracted. The minimum reflection response value is compared with a preset vegetation burning degree level library. The vegetation burning degree level library records the vegetation burning degree level corresponding to different reflection response values. According to the comparison result, the corresponding vegetation burning degree associated feature is determined.

[0132] Step S153: The spatial contour coordinates of each forest fire confirmation area feature cluster are extracted, the area range surrounded by the contour coordinates is calculated, and the corresponding fire area range feature is determined.

[0133] The spatial contour coordinates of each forest fire confirmation area feature cluster are extracted. The area range surrounded by the contour coordinates is calculated, and the corresponding fire area range feature is determined. The fire area range feature describes the size of the fire area.

[0134] Step S154: The fire burning intensity associated feature, the vegetation burning degree associated feature, and the fire area range feature of each forest fire confirmation area feature cluster are associated and integrated to generate real-time state sub-information. All real-time state sub-information is collected and integrated to form forest fire real-time state description information covering the target forest area.

[0135] The fire burning intensity associated feature, the vegetation burning degree associated feature, and the fire area range feature of each forest fire confirmation area feature cluster are associated and integrated to generate real-time state sub-information. All real-time state sub-information is collected and integrated to form forest fire real-time state description information covering the target forest area. The real-time state description information describes the real-time state of the current forest fire.

[0136] Step S155: For each forest fire confirmation area feature cluster, according to the vegetation type information and the terrain slope information of its location, a matching fire spread prediction model is extracted from the forest fire spatio-temporal evolution rule library. The fire burning intensity associated feature and the fire area range feature of each forest fire confirmation area feature cluster are standardized and then input into the matching fire spread prediction model. The real-time state sub-information includes the fire burning intensity associated feature and the fire area range feature.

[0137] For each forest fire confirmation area feature cluster, according to the vegetation type information and the terrain slope information of its location, a matching fire spread prediction model is extracted from the forest fire spatio-temporal evolution rule library. The fire spread prediction model is established according to different vegetation types and terrain slopes and is used to predict the spread of the fire. The fire burning intensity associated feature and the fire area range feature of each forest fire confirmation area feature cluster are standardized to make them comparable. Then, the standardized features are input into the matching fire spread prediction model.

[0138] Step S156: calculating, by the fire spread prediction model, an expansion range prediction result and an expansion direction prediction vector of the feature cluster in future multiple collection time periods, extracting a direction angle attribute in the expansion direction prediction vector, and combining a terrain slope direction adjustment vector of a location where the feature cluster is located to obtain a corrected expansion direction prediction vector.

[0139] The fire spread prediction model is used to calculate an expansion range prediction result and an expansion direction prediction vector of the feature cluster in future multiple collection time periods. The expansion range prediction result describes a range in which the fire can spread in future time periods, and the expansion direction prediction vector indicates the expansion direction of the fire. The direction angle attribute in the expansion direction prediction vector is extracted, and a terrain slope direction adjustment vector of a location where the feature cluster is located is combined. The terrain slope affects the spread direction of the fire, and the adjustment vector is used to correct the expansion direction to obtain a corrected expansion direction prediction vector.

[0140] Step S157: generating, according to the expansion range prediction result and the corrected expansion direction prediction vector, spread trend sub-information corresponding to each feature cluster.

[0141] According to the expansion range prediction result and the corrected expansion direction prediction vector, spread trend sub-information corresponding to each feature cluster is generated.

[0142] Step S1571: taking a current center coordinate of each feature cluster of the forest fire confirmation area as a starting point, and determining a spread main axis direction based on the corrected expansion direction prediction vector corresponding to the feature cluster.

[0143] The current center coordinate of each feature cluster of the forest fire confirmation area is taken as a starting point, and the spread main axis direction is determined according to the corrected expansion direction prediction vector. The spread main axis direction indicates the main expansion direction of the fire.

[0144] Step S1572: in the spread main axis direction, a predicted boundary coordinate of the feature cluster in each collection time period in the future is calculated according to an area change amount indicated by the expansion range prediction result of the feature cluster, in combination with an equivalent radius or a main axis length of a current area range of the feature cluster, and a distance between the predicted boundary coordinate and a current boundary coordinate of the feature cluster is determined based on a relationship between the area change amount and a current area spatial scale.

[0145] In the spread main axis direction, a predicted boundary coordinate of the feature cluster in each collection time period in the future is calculated according to an area change amount indicated by the expansion range prediction result, in combination with an equivalent radius or a main axis length of a current area range of the feature cluster. The distance between the predicted boundary coordinate and a current boundary coordinate of the feature cluster is determined according to the relationship between the area change amount and the current area spatial scale. The greater the area change amount, the farther the distance between the predicted boundary coordinate and the current boundary coordinate.

[0146] Step S1573: Calculate the predicted boundary extension width of the feature cluster at each future acquisition period on both sides perpendicular to the direction of the main axis of spread, which is determined based on the vegetation density parameter at the location of the feature cluster, which is extracted from the satellite remote sensing image sequence.

[0147] Calculate the predicted boundary extension width of the feature cluster at each future acquisition period on both sides perpendicular to the direction of the main axis of spread. The predicted boundary extension width is determined based on the vegetation density parameter at the location of the feature cluster. The vegetation density parameter can be extracted from the near-infrared dimension image in the satellite remote sensing image sequence. The greater the vegetation density, the greater the predicted boundary extension width.

[0148] Step S1574: Generate the predicted area profile coordinates of the feature cluster at each future acquisition period according to the predicted boundary coordinates in the direction of the main axis of spread and the predicted boundary extension width perpendicular to the direction of the main axis of spread.

[0149] Generate the predicted area profile coordinates of the feature cluster at each future acquisition period according to the predicted boundary coordinates in the direction of the main axis of spread and the predicted boundary extension width perpendicular to the direction of the main axis of spread. These coordinates describe the area profile that the fire may expand in the future period.

[0150] Step S1575: Calculate the overlap range proportion of the predicted area profile coordinates of the feature cluster at each future acquisition period and the current area profile coordinates, and verify the rationality of the predicted area.

[0151] Calculate the overlap range proportion of the predicted area profile coordinates of the feature cluster at each future acquisition period and the current area profile coordinates. Verify the rationality of the predicted area through the proportion. If the overlap range proportion is too small or too large, it may indicate that there is a problem with the prediction result.

[0152] Step S1576: If the overlap range proportion meets the preset overlap condition, determine that the predicted area is valid.

[0153] If the overlap range proportion meets the preset overlap condition, determine that the predicted area is valid. The preset overlap condition can be set according to historical data and experience.

[0154] Step S1577: Extract the range change rate and center position offset of the valid predicted area of the feature cluster at each future acquisition period, and integrate the predicted area profile coordinates, range change rate, and center position offset of the feature cluster with the corrected spread direction prediction vector to generate the spread trend sub-information corresponding to the feature cluster.

[0155] The feature cluster is used to predict the range change rate and center position offset of the fire area in each collection period. The range change rate describes the change of the fire area range over time, and the center position offset indicates the movement of the fire center position. The predicted area contour coordinates, range change rate, and center position offset of the feature cluster are associated and integrated with the corrected expansion direction prediction vector to generate the corresponding spread trend sub-information of the feature cluster.

[0156] Step S158: Collect all the spread trend sub-information, and analyze the spread correlation between different feature clusters, which includes whether there is a spread intersection possibility.

[0157] All the spread trend sub-information is collected, and the spread correlation between different feature clusters is analyzed. The spread correlation mainly considers whether there is a spread intersection possibility, i.e., whether different fire areas can intersect in a future period.

[0158] Step S159: If there is a spread intersection possibility, calculate the intersection period prediction result, add the corresponding spread trend sub-information, integrate all the spread trend sub-information, and form the forest fire spread trend prediction description information.

[0159] If there is a spread intersection possibility, the intersection period prediction result is calculated. The intersection period prediction result is added to the corresponding spread trend sub-information. Then, all the spread trend sub-information is integrated to form the forest fire spread trend prediction description information. The spread trend prediction description information describes the future spread trend of the forest fire.

[0160] Step S160: Based on the real-time state description information and the spread trend prediction description information, generate a forest fire prevention and control instruction containing a region coordinate chain, and send the forest fire prevention and control instruction to a target forest management terminal.

[0161] Based on the real-time state description information and the spread trend prediction description information of the forest fire, a forest fire prevention and control instruction containing a region coordinate chain is generated. The region coordinate chain records the position information of the fire area, which is convenient for forest management personnel to determine the specific range of the fire. The generated forest fire prevention and control instruction is sent to the target forest management terminal, providing a decision basis for forest management personnel to take corresponding prevention and control measures.

[0162] Figure 2 A schematic diagram of exemplary hardware and software components of a satellite remote sensing-based forest fire analysis system 100 that can implement the idea of the present application is shown. For example, a processor 120 can be used in the satellite remote sensing-based forest fire analysis system 100 and used to perform the functions in the present application.

[0163] The satellite remote sensing based forest fire analysis system 100 can be a general server or a special purpose server, both of which can be used to implement the satellite remote sensing based forest fire analysis method of the present application. The present application illustrates only one server, but for the sake of convenience, the functions described in the present application can be implemented in a distributed manner on multiple similar platforms to balance the processing load.

[0164] For example, the satellite remote sensing based forest fire analysis system 100 can include a network port 110 connected to a network, one or more processors 120 for executing program instructions, a communication bus 130, and different forms of storage media 140, such as a disk, a ROM, or a RAM, or any combination thereof. The satellite remote sensing based forest fire analysis system 100 can also include program instructions stored in a ROM, a RAM, or other types of non-transitory storage media, or any combination thereof, for example. The method of the present application can be implemented according to these program instructions. The satellite remote sensing based forest fire analysis system 100 also includes an I / O interface 150 between the computer and other input and output devices.

[0165] For the sake of convenience, only one processor is described in the satellite remote sensing based forest fire analysis system 100. However, it should be noted that the satellite remote sensing based forest fire analysis system 100 in the present application can also include multiple processors, so the steps performed by one processor described in the present application can also be jointly performed or separately performed by multiple processors. For example, if the processor of the satellite remote sensing based forest fire analysis system 100 performs steps A and B, it should be understood that steps A and B can also be jointly performed by two different processors or separately performed in one processor. For example, a first processor performs step A and a second processor performs step B, or the first processor and the second processor jointly perform steps A and B.

[0166] In addition, the present application also provides a readable storage medium, in which computer executable instructions are pre-set, and when the processor executes the computer executable instructions, the satellite remote sensing based forest fire analysis method described above is implemented.

[0167] It should be noted that, in order to simplify the description of the present application and to help understand one or more embodiments of the present application, in the foregoing description of the embodiments of the present application, various features are sometimes combined into one embodiment, drawing or description thereof.

Claims

1. A forest fire analysis method based on satellite remote sensing, characterized in that, The method includes: A sequence of satellite remote sensing images of the target forest area is acquired. The satellite remote sensing image sequence includes continuously acquired thermal infrared, near-infrared and visible light images covering the target forest area, as well as corresponding acquisition time period identifiers. Perform temporal correlation modeling processing on the satellite remote sensing image sequence to generate a temporal correlation feature set; The time-series associated feature set is input into a pre-trained forest fire dynamic analysis model to obtain a feature set of suspected forest fire areas. Based on a pre-defined forest fire spatiotemporal evolution pattern database, a multi-condition constraint verification process is performed on the suspected forest fire area feature set to obtain a confirmed forest fire area feature set. The forest fire spatiotemporal evolution pattern database includes spatiotemporal feature patterns of forest fire spread. These patterns include fire spread speed characteristics in different vegetation types, fire spread direction characteristics under different terrain slopes, and fire thermal radiation response intensity variation characteristics at different data collection times. The multi-condition constraint verification process involves, based on the spatiotemporal feature patterns, determining whether the feature response change rate of each feature cluster in the suspected forest fire area feature set conforms to the corresponding vegetation type. The spread rate characteristics of the type, whether the expansion direction of the feature clusters conforms to the spread direction characteristics of the corresponding terrain slope, and whether the thermal radiation response intensity values ​​of the feature clusters conform to the thermal radiation response intensity change characteristics of the corresponding collection period are all considered. After removing interfering feature clusters, the consistency of the time series change trend of the remaining feature clusters is verified. The feature response change rate includes the growth rate of thermal radiation response intensity and the expansion rate of the feature cluster coverage area. The forest fire confirmed area feature set is a set of candidate fire area feature clusters that meet all conditions and have consistent time series change trends after the multi-condition constraint verification process. It includes the spatial distribution characteristics, feature intensity characteristics, and time change characteristics of each feature cluster. Based on the feature set of the confirmed forest fire area, real-time status description information and spread trend prediction description information of the forest fire are generated. The real-time status description information includes the response intensity of various dimensions of the fire area and the vegetation burning correlation characteristics. The spread trend prediction description information includes the fire expansion direction characteristics and the spread speed correlation characteristics. Based on the real-time status description information and the spread trend prediction description information, a forest fire prevention and control instruction containing a regional coordinate chain is generated, and the forest fire prevention and control instruction is sent to the target forest management terminal.

2. The forest fire analysis method based on satellite remote sensing according to claim 1, characterized in that, The step of performing temporal correlation modeling on the satellite remote sensing image sequence to generate a temporal correlation feature set includes: The satellite remote sensing image sequence is grouped according to the acquisition time period, and a reference image group and a comparison image group are determined for adjacent acquisition time periods. The reference image group is a set of thermal infrared dimension images, near infrared dimension images and visible light dimension images acquired in the previous time period, and the comparison image group is a set of thermal infrared dimension images, near infrared dimension images and visible light dimension images acquired in the subsequent time period. For the thermal infrared dimension, the thermal infrared dimension images in the reference image group and the thermal infrared dimension images in the comparison image group are compared and analyzed. The thermal radiation response changes of the corresponding image regions are extracted to generate a thermal infrared dimension spectral response change feature map. The feature value of each image region in the thermal infrared dimension spectral response change feature map reflects the thermal radiation response change of that image region. For the near-infrared dimension, the near-infrared dimension images in the reference image group and the near-infrared dimension images in the comparison image group are compared and analyzed. The vegetation reflectance response changes in the corresponding image regions are extracted to generate a near-infrared dimension spectral response change feature map. The feature value of each image region in the near-infrared dimension spectral response change feature map reflects the vegetation reflectance response changes in that image region. For the visible light dimension, the visible light dimension images in the reference image group and the visible light dimension images in the comparison image group are compared and analyzed. The color response changes of the corresponding image regions are extracted to generate a visible light dimension spectral response change feature map. The feature value of each image region in the visible light dimension spectral response change feature map reflects the color response change of that image region. A cross-dimensional temporal correlation matrix is ​​constructed. The feature values ​​of the corresponding image regions in the thermal infrared spectral response change feature map, the near-infrared spectral response change feature map, and the visible light spectral response change feature map are standardized and used as matrix elements. The temporal correlation coefficient between the feature values ​​of the corresponding image regions in different dimensions is calculated. The temporal correlation coefficient represents the synchronicity of the feature response changes in different dimensions. Based on the time-series correlation coefficient, the standardized spectral response change feature maps of each dimension are weighted and fused to generate a cross-dimensional collaborative correlation feature map. The cross-dimensional collaborative association feature map is segmented into multiple collaborative association feature sub-regions. Statistical features of each collaborative association feature sub-region are extracted. The statistical features include the feature response mean, feature response variance, and feature response extreme value. The statistical features are correlated and integrated with the regional features of the thermal infrared spectral response change feature map, the near-infrared spectral response change feature map, and the visible light spectral response change feature map to generate a time-series correlated feature set.

3. The forest fire analysis method based on satellite remote sensing according to claim 1, characterized in that, The step of inputting the time-series correlated feature set into a pre-trained forest fire dynamic analysis model to obtain a feature set of suspected forest fire areas includes: Extract thermal infrared dimension-related features and near-infrared dimension-related features from the time-series associated feature set, input the thermal infrared dimension-related features and near-infrared dimension-related features into the dimension co-calibration layer of the forest fire dynamic analysis model, identify the set of thermal radiation anomaly regions in the thermal infrared dimension-related features, and identify the set of vegetation reflection anomaly regions in the near-infrared dimension-related features. Binarization is performed on the sets of anomalous thermal radiation regions and anomalous vegetation reflection regions to generate anomalous region identification maps with the same spatial resolution. Then, the spatial overlap ratio of the corresponding image regions of the two anomalous region identification maps is calculated, and a set of common anomalous regions that meet the preset conditions for spatial overlap ratio is selected. Based on the set of common anomaly regions, a dimension calibration mapping relationship is constructed. After standardizing the near-infrared dimension-related features and the thermal infrared dimension-related features respectively, the response scale of the standardized near-infrared dimension-related features is adjusted so that it is consistent with the change trend of the standardized thermal infrared dimension-related features at the set of common anomaly regions, thus obtaining the calibrated near-infrared dimension features. The calibrated near-infrared dimensional features are channel-concatenated with the thermal infrared dimensional correlation features and visible light dimensional correlation features in the time-series correlated feature set to generate a dimensional collaborative feature vector. The dimensional collaborative feature vector is input into the fire signal enhancement layer of the forest fire dynamic analysis model. The weak fire signal in the dimensional collaborative feature vector is amplified in a hierarchical manner through the hierarchical signal enhancer to generate an enhanced feature vector. The parameters of the hierarchical signal enhancer are set based on the weak fire signal characteristics of the samples in historical fire samples. The enhanced feature vector is compared with each background feature library in the dynamic background suppression module to filter out abnormal feature components that do not match any background feature patterns. The dynamic background suppression module includes a vegetation background feature library, a terrain background feature library, and an artificial target background feature library. Spatial clustering is performed on the abnormal feature components to group the spatially adjacent abnormal feature components with similar feature response change trends into the same feature cluster; Extract the spatial distribution features and feature intensity features of each feature cluster. The spatial distribution features include the outline shape, coverage area, and center position of the feature cluster. The feature intensity features include the maximum feature response value and the average feature response value within the feature cluster. By associating and integrating the spatial distribution features and feature intensity features of each feature cluster, a feature set of suspected forest fire areas containing information on multiple feature clusters is generated.

4. The forest fire analysis method based on satellite remote sensing according to claim 3, characterized in that, The step involves constructing a dimensional calibration mapping relationship based on the set of common anomaly regions. After standardizing the near-infrared dimensional correlation features and the thermal infrared dimensional correlation features respectively, the response scale of the standardized near-infrared dimensional correlation features is adjusted to maintain consistency with the changing trend of the standardized thermal infrared dimensional correlation features at the set of common anomaly regions, resulting in calibrated near-infrared dimensional features, including: For each anomalous region in the set of common anomalous regions, extract the change in thermal radiation response in the thermal infrared dimension and the change in vegetation reflectance response in the near-infrared dimension. The changes in thermal radiation response and vegetation reflectance response were standardized respectively. The standardized changes in thermal radiation response were used as independent variables and the standardized changes in vegetation reflectance response were used as dependent variables to construct an association regression model. The regression parameters of the association regression model were calculated. The association regression model describes the relationship between the changes in thermal radiation response and vegetation reflectance response. A dimensional calibration function is generated based on the aforementioned correlation regression model. The input of the dimensional calibration function is the change in vegetation reflectance response of near-infrared dimensional correlation features, and the output is the calibrated change in vegetation reflectance response. Extract the vegetation reflectance response changes of all image regions contained in the near-infrared dimensional correlation features from the time-series associated feature set, and input the vegetation reflectance response changes of each image region into the dimensional calibration function to obtain the calibrated vegetation reflectance response changes of each image region. A calibration error assessment model is constructed. The changes in vegetation reflectance response and corresponding thermal radiation response after calibration for each image region are converted into relative change rates of the same dimension or standardized, and then the error metric is calculated. Image regions whose error metric values ​​meet preset error conditions are selected as effective calibration regions, and the distribution density of the effective calibration regions in the entire target forest region is statistically analyzed. If the distribution density meets the preset density condition, the dimensional calibration function is determined to be valid, and the calibrated near-infrared dimensional features are generated based on the calibrated vegetation reflectance change in all image regions. If the distribution density does not meet the preset density condition, the parameters of the correlation regression model are adjusted, a nonlinear correlation term is added to optimize the model, and the above steps are repeated until an effective calibration region whose distribution density meets the preset density condition is obtained. Based on the changes in vegetation reflectance response after calibration in all image regions at this time, the calibrated near-infrared dimensional features are generated.

5. The forest fire analysis method based on satellite remote sensing according to claim 1, characterized in that, The pre-defined forest fire spatiotemporal evolution law database is used to perform multi-condition constraint verification processing on the suspected forest fire area feature set to obtain the confirmed forest fire area feature set, including: Extract the spatiotemporal feature patterns of forest fire spread from the aforementioned forest fire spatiotemporal evolution law database; For each feature cluster in the suspected forest fire area feature set, extract the vegetation type information, terrain slope information and collection time information of the location of the feature cluster; For each feature cluster, the corresponding fire spread rate feature is matched from the spatiotemporal feature pattern based on its vegetation type information. It is then determined whether the growth rate of the thermal radiation response intensity of the feature cluster matches the matched fire thermal radiation growth feature, and whether the expansion rate of the feature cluster's coverage area matches the matched fire spread rate feature. The feature response change rate includes the growth rate of the thermal radiation response intensity and the expansion rate of the feature cluster's coverage area. For each feature cluster, the corresponding fire spread direction feature is matched from the spatiotemporal feature pattern based on its terrain slope information. It is then determined whether the expansion direction of the feature cluster conforms to the matched fire spread direction feature. The expansion direction of the feature cluster is determined by comparing the changes in its center position in adjacent acquisition periods. For each feature cluster, the corresponding fire thermal radiation response intensity change feature is matched from the spatiotemporal feature pattern based on its collection time period information, and it is determined whether the thermal radiation response intensity value of the feature cluster in the corresponding collection time period matches the matched thermal radiation response intensity change feature. For each feature cluster, a conditional judgment is made. If the feature response change rate, expansion direction, and thermal radiation response intensity value are all satisfied simultaneously, then the feature cluster is marked as a candidate fire area feature cluster. If the above three conditions are not met simultaneously, further analysis should be conducted to determine whether the feature cluster is affected by non-fire interference factors. For feature clusters that are not marked as candidate fire area feature clusters, their feature response patterns are extracted. The feature response patterns include the stability of thermal radiation response intensity and the concentration of feature distribution. The feature response pattern is compared with the non-fire interference feature pattern library in the forest fire spatiotemporal evolution law library. If an interference pattern is matched, the feature cluster is marked as an interference feature cluster and removed from the forest fire suspected area feature set. Collect feature information of all candidate fire area feature clusters, and extract verification features for each candidate fire area feature cluster. The verification features include the continuous increase of thermal radiation response and the continuous decrease of vegetation reflectance response. Time series analysis is performed on the verification features of each candidate fire area feature cluster to determine whether the verification features maintain a consistent trend during the continuous collection period. If they do, the candidate fire area feature cluster is identified as a confirmed forest fire area feature cluster. Collect feature information of all forest fire confirmed area feature clusters, integrate them to generate a forest fire confirmed area feature set, the feature information includes spatial distribution features, feature intensity features, and temporal variation features.

6. The forest fire analysis method based on satellite remote sensing according to claim 5, characterized in that, The step of performing time-series analysis on the verification features of each candidate fire area feature cluster to determine whether the verification features maintain a consistent trend of change over a continuous data collection period includes: For each candidate fire area feature cluster, the feature response values ​​of the feature cluster are extracted in multiple consecutive collection periods to form a time series of the verification features. For each verification feature time series, calculate the difference in the change of feature response values ​​between adjacent collection periods to generate their respective difference sequences, and determine the sign consistency of the difference sequences, wherein the sign consistency includes both positive values ​​and both negative values. If the signs are the same, it is preliminarily determined that the trend of change is the same; if the signs are not the same, the moving average value of the difference sequence is calculated, and it is determined whether the signs of the moving average values ​​are consistent. If they are consistent, it is still preliminarily determined that the trend of change is the same. The time series of the verification features is subjected to trend fitting processing. A trend fitting line is generated by the fitting algorithm, the slope attribute of the trend fitting line is determined, and it is determined whether the change magnitude of the slope attribute meets the preset slope condition. If it is initially determined that the trend of change is consistent and the change range of the slope attribute meets the preset slope condition, then it is determined that the verification feature has a consistent trend of change during the continuous collection period; if it is initially determined that the trend of change is consistent but the change range of the slope attribute does not meet the preset slope condition, then it is checked whether there are other forest fire confirmed area feature clusters around the candidate fire area feature cluster. If they exist and the distance meets the preset distance condition, then it is determined that the trend of change is consistent. If there are no other confirmed forest fire area feature clusters in the surrounding area, the candidate fire area feature cluster will be marked as a feature cluster to be observed and will be re-analyzed after the next period of satellite remote sensing images are obtained. If the initial determination is that the trends of change are inconsistent, then it is directly determined that the verification feature has not maintained a consistent trend of change.

7. The forest fire analysis method based on satellite remote sensing according to claim 1, characterized in that, The process of generating real-time status description information and spread trend prediction description information for forest fires based on the feature set of confirmed forest fire areas includes: For each forest fire confirmed area feature cluster in the forest fire confirmed area feature set, extract its maximum thermal radiation response value in the thermal infrared dimension, compare the maximum thermal radiation response value with a preset thermal radiation intensity level library, and determine the fire combustion intensity correlation feature corresponding to the feature cluster based on the comparison result. Extract the minimum reflectance response value of the feature clusters of each confirmed forest fire area in the near-infrared dimension, compare it with the preset vegetation burn level library, and determine the corresponding vegetation burn level related features. Extract the spatial contour coordinates of the feature clusters of each confirmed forest fire area, calculate the area range enclosed by the contour coordinates, and determine the corresponding fire area range features. The fire burning intensity correlation feature, vegetation burning degree correlation feature and fire area range feature of each forest fire confirmed area feature cluster are correlated and integrated to generate real-time status sub-information. All real-time status sub-information is collected and integrated to form real-time status description information of forest fires covering the target forest area. For each forest fire confirmed area feature cluster, based on the vegetation type information and terrain slope information of its location, a matching fire spread prediction model is extracted from the forest fire spatiotemporal evolution law database. After standardizing the fire burning intensity correlation features and fire area range features of each forest fire confirmed area feature cluster, the model is input into the matching fire spread prediction model. The real-time status sub-information includes fire burning intensity correlation features and fire area range features. The fire spread prediction model is used to calculate the predicted range and direction of the feature cluster in multiple future data collection periods. The direction angle attribute in the direction prediction vector is extracted and combined with the terrain slope direction adjustment vector at the location of the feature cluster to obtain the corrected direction prediction vector. Based on the predicted expansion range and the corrected predicted expansion direction vector, the spread trend sub-information corresponding to each feature cluster is generated. All spread trend sub-information is collected, and the spread correlation between different feature clusters is analyzed. The spread correlation includes whether there is a possibility of spread convergence. If there is a possibility of convergence and spread, the prediction results for the convergence period are calculated, the corresponding spread trend sub-information is added, and all spread trend sub-information is integrated to form a forest fire spread trend prediction description information.

8. The forest fire analysis method based on satellite remote sensing according to claim 3, characterized in that, The step of performing pattern comparison between the enhanced feature vector and each background feature library in the dynamic background suppression module, and filtering out abnormal feature components that do not match any background feature patterns, includes: The feature templates of normal vegetation are extracted from the vegetation background feature library of the dynamic background suppression module. The feature templates include the thermal radiation response range of normal vegetation in the thermal infrared dimension, the reflection response range in the near infrared dimension, and the color response range in the visible light dimension. Each image region feature in the enhanced feature vector is compared with the feature template of the normal vegetation to determine whether the image region feature falls within the thermal radiation response range, the reflection response range, and the color response range. If all of them fall within the corresponding response range, they are marked as vegetation background feature components. The feature templates of the terrain targets are extracted from the terrain background feature library of the dynamic background suppression module. The feature templates include the thermal radiation response distribution pattern of mountains in the thermal infrared dimension, the reflection response distribution pattern of water bodies in the near infrared dimension, and the color response distribution pattern in the visible light dimension. The image region features not marked as vegetation background feature components are compared with the feature template of the terrain target to determine whether they conform to the feature distribution pattern. If they do, they are marked as terrain background feature components. The feature templates of artificial targets are extracted from the artificial target background feature library of the dynamic background suppression module. The feature templates include the thermal radiation response fluctuation mode of buildings in the thermal infrared dimension, the reflection response uniformity mode of roads in the near infrared dimension, and the shape feature mode in the visible light dimension. The image region features that are not labeled as vegetation or terrain background feature components are compared with the feature template of the artificial target to determine whether they match the feature pattern. If they do, they are labeled as artificial target background feature components. Collect all image region features that are not labeled as any of the above background feature components, as anomalous feature components that do not match any of the background feature patterns.

9. The forest fire analysis method based on satellite remote sensing according to claim 7, characterized in that, The step of generating propagation trend sub-information for each feature cluster based on the expansion range prediction result and the corrected expansion direction prediction vector includes: Using the current center coordinates of each forest fire confirmed area feature cluster as the origin, the direction of the main spread axis is determined based on the corrected spread direction prediction vector corresponding to the feature cluster. In the direction of the main spread axis, based on the area change indicated by the expansion range prediction result of the feature cluster, and combined with the equivalent radius or main axis length of the current area of ​​the feature cluster, the predicted boundary coordinates of the feature cluster are calculated for each future acquisition period. The distance between the predicted boundary coordinates and the current boundary coordinates of the feature cluster is determined based on the relationship between the area change and the current spatial scale of the area. On both sides perpendicular to the main spread axis, the predicted boundary expansion width of the feature cluster is calculated for each future acquisition period. The predicted boundary expansion width is determined based on the vegetation density parameter at the location of the feature cluster, which is extracted from the satellite remote sensing image sequence. Based on the predicted boundary coordinates along the main propagation axis and the predicted boundary expansion width perpendicular to the main propagation axis, the predicted region contour coordinates of the feature cluster are generated for each future acquisition period. Calculate the proportion of overlap between the predicted region contour coordinates and the current region contour coordinates for each future acquisition period to verify the rationality of the predicted region. If the overlap ratio meets the preset overlap condition, the predicted region is determined to be valid; Extract the range change rate and center position offset of the effective prediction area of ​​the feature cluster in each future acquisition period. Then, associate and integrate the prediction area contour coordinates, range change rate, and center position offset of the feature cluster with the corrected expansion direction prediction vector to generate the spread trend sub-information corresponding to the feature cluster.

10. A forest fire analysis system based on satellite remote sensing, characterized in that, The satellite remote sensing-based forest fire analysis system includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the satellite remote sensing-based forest fire analysis method according to any one of claims 1-9.

Images