Forest fire time sequence sample construction method and product

Abstract

The invention discloses a construction method of a forest fire time sequence sample and a product. Comprising the steps of detecting a fire range area from a target forest area based on interference coherence change of SAR images of two adjacent time sequence nodes in N time sequence nodes; respectively extracting multi-dimensional SAR characteristic parameters of the fire range area at two adjacent time sequence nodes, and determining multi-dimensional characteristic change information between the two adjacent time sequence nodes based on the multi-dimensional SAR characteristic parameters of the two adjacent time sequence nodes; on the basis of the multi-dimensional feature change information, the combustion state type of the fire range area is judged, and a combustion state type label associated with the second time sequence node is obtained; and in combination with the multi-temporal SAR images corresponding to the N time sequence nodes, the multi-dimensional feature change information and the combustion state type labels thereof, constructing a forest fire time sequence sample, so as to train a forest fire identification model based on the forest fire time sequence sample. According to the invention, the fire risk can be effectively and accurately identified.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of environmental change monitoring, in particular to the technical field of fire risk identification, and more particularly to a method for constructing a forest fire time sequence sample, a method for training a forest fire identification model, a device, an apparatus, a medium and a product. BACKGROUND

[0002] At present, the monitoring of forest fires mainly relies on integrated monitoring means of "sky-ground", which specifically includes ground patrol, lookout tower monitoring, aerial patrol (including manned and unmanned aircraft) and satellite remote sensing technology.

[0003] However, the traditional forest fire monitoring methods, such as manual inspection, aerial aircraft and satellite remote sensing, have obvious limitations in practical application. Specifically, manual inspection is low in efficiency, limited in coverage and poor in safety; aerial aircraft monitoring is high in cost and is strictly restricted by weather and visibility; satellite remote sensing is difficult to achieve rapid real-time fire capture and accurate assessment due to long revisit period and insufficient spatial resolution. Based on this, the inherent defects of these traditional methods make it difficult to meet the urgent needs of modern forest fire prevention when dealing with sudden and rapidly spreading forest fires, and it is difficult to effectively and accurately identify fire risks. SUMMARY

[0004] Therefore, the embodiments of the present application provide a method for constructing a forest fire time sequence sample, a device, a system, a medium and a product, which can effectively and accurately identify fire risks.

[0005] In a first aspect, the embodiments of the present application provide a method for constructing a forest fire time sequence sample, which comprises: obtaining synthetic aperture radar (SAR) images of a target forest area at N time sequence nodes in a fire process to obtain multi-temporal SAR images; detecting a fire range area from the target forest area based on interference coherence changes of SAR images of two adjacent time sequence nodes among the N time sequence nodes, wherein the two adjacent time sequence nodes include a first time sequence node and a second time sequence node, and the first time sequence node is a previous time sequence node of the second time sequence node; extracting multi-dimensional SAR feature parameters of the fire range area at the two adjacent time sequence nodes based on the multi-temporal SAR images, and determining multi-dimensional feature change information between the two adjacent time sequence nodes based on the multi-dimensional SAR feature parameters of the two adjacent time sequence nodes, wherein the multi-dimensional SAR feature parameters include full polarization scattering features and vegetation coverage; determining a burning state type of the fire range area based on the multi-dimensional feature change information between the two adjacent time sequence nodes to obtain a burning state type label associated with the second time sequence node; and constructing a forest fire time sequence sample by combining the multi-temporal SAR images corresponding to the N time sequence nodes, the multi-dimensional feature change information, and the burning state type label, so as to train a forest fire identification model based on the forest fire time sequence sample, wherein the forest fire identification model is used to identify the burning state type corresponding to the fire range area.

[0006] In a second aspect, the embodiments of the present application provide a method for training a forest fire identification model, which comprises: training a large model based on the forest fire time sequence sample of the first aspect to obtain the forest fire identification model.

[0007] In a third aspect, an embodiment of the present application provides a device for constructing a forest fire time sequence sample, the device comprising: an acquisition module configured to acquire synthetic aperture radar (SAR) images of a target forest area at N time sequence nodes in a fire process to obtain multi-temporal SAR images; a detection module configured to detect a fire range area from the target forest area based on interference coherence changes of SAR images of two adjacent time sequence nodes among the N time sequence nodes, wherein the two adjacent time sequence nodes include a first time sequence node and a second time sequence node, and the first time sequence node is a previous time sequence node of the second time sequence node; a determination module configured to extract multi-dimensional SAR feature parameters of the fire range area at the two adjacent time sequence nodes based on the multi-temporal SAR images, and determine multi-dimensional feature change information between the two adjacent time sequence nodes based on the multi-dimensional SAR feature parameters of the two adjacent time sequence nodes, wherein the multi-dimensional SAR feature parameters include full polarization scattering features and vegetation coverage; a discrimination module configured to discriminate a burning state type of the fire range area based on the multi-dimensional feature change information between the two adjacent time sequence nodes to obtain a burning state type label associated with the second time sequence node; and a sample construction module configured to construct a forest fire time sequence sample by combining the multi-temporal SAR images corresponding to the N time sequence nodes, the multi-dimensional feature change information, and the burning state type label, and train a forest fire identification model based on the forest fire time sequence sample, wherein the forest fire identification model is used to identify the burning state type corresponding to the fire range area.

[0008] In a fourth aspect, an embodiment of the present application provides a device for training a forest fire identification model, comprising: a training module configured to train a large model based on the forest fire time sequence sample of the first aspect to obtain the forest fire identification model.

[0009] In a fifth aspect, an embodiment of the present application provides an electronic device, comprising: a processor and a memory storing computer program instructions; and the processor implements the steps of the method for constructing the forest fire time sequence sample of the first aspect when executing the computer program instructions.

[0010] In a sixth aspect, an embodiment of the present application provides a computer readable storage medium, the computer readable storage medium storing computer program instructions, and the computer program instructions are executed by a processor to implement the steps of the method for constructing the forest fire time sequence sample of the first aspect.

[0011] In a seventh aspect, an embodiment of the present application provides a computer program product stored in a non-volatile storage medium, and the computer program product is executed by a processor to implement the steps of the method for constructing the forest fire time sequence sample of the first aspect.

[0012] In an eighth aspect, an embodiment of the present application provides a chip, which comprises a processor and a communication interface, the communication interface and the processor are coupled, and the processor is configured to run programs or instructions to implement the steps of the method for constructing a forest fire time sequence sample according to the first aspect.

[0013] The present application provides a method, device, system, medium and product for constructing a forest fire time sequence sample, which can effectively overcome the limitations of traditional monitoring methods by using multi-temporal synthetic aperture radar (SAR) images as a data source. Specifically, SAR has the ability to acquire data at any time and in any weather, and is not restricted by light conditions and weather factors such as clouds, fog and smoke. Therefore, the problem of optical remote sensing being unable to see in bad weather and aviation monitoring being limited by visibility can be solved. Further, by extracting and analyzing the time sequence change information of multi-dimensional SAR feature parameters in the fire range area, the differences in microwave scattering mechanism of different burning states can be mined based on the collaborative analysis and cross verification of full polarization scattering characteristics and vegetation coverage, the uncertainty of single parameter discrimination is overcome, the fine discrimination of the burning state type is realized, the identification accuracy and reliability of different burning state types in the fire process are improved, accurate input is provided for subsequent dynamic assessment of fire risk level, and the problem of rough fire identification type and inaccurate assessment by traditional methods is solved. Finally, by integrating the images, feature change information and determined burning state labels of multiple time sequence nodes, the structured time sequence sample constructed can provide reliable data support for training a machine learning model for automatically and accurately identifying fire types, and further promote the realization of rapid and accurate forest fire dynamic monitoring and risk assessment. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings in the embodiments of the present application.

[0015] Figure 1 is a flowchart of a method for constructing a forest fire time sequence sample provided by an embodiment of the present application; Figure 2 is an exemplary flowchart of a method for constructing a forest fire time sequence sample provided by an embodiment of the present application; Figure 3 is an exemplary flowchart of a method for constructing a forest fire time sequence sample provided by another embodiment of the present application; Figure 4 is an exemplary flowchart of a method for constructing a forest fire time sequence sample provided by another embodiment of the present application; Figure 5 is a structural diagram of a device for constructing a forest fire time sequence sample provided by an embodiment of the present application; Figure 6is a hardware structure schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0016] The principles and spirits of the present application will be described below with reference to a number of exemplary embodiments. It should be understood that the purpose of providing these embodiments is to make the principles and spirits of the present application clearer and more thorough, so that those skilled in the art can better understand and implement the principles and spirits of the present application. The exemplary embodiments provided herein are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments herein, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0017] In this article, terms such as first, second, third, etc. are only used to distinguish one entity (or operation) from another entity (or operation), and do not require or imply any order or association between the entities (or operations).

[0018] Before describing the technical solutions provided by the embodiments of the present application, in order to facilitate the understanding of the embodiments of the present application, the present application first specifically describes the problems existing in the related art: Forest fire is a sudden and destructive natural disaster, and it is crucial to achieve early and accurate monitoring and dynamic assessment. Traditional monitoring methods such as optical remote sensing are easily affected by weather conditions such as clouds, smoke and fog, and there are often observation blind spots when a fire occurs. Synthetic aperture radar (SAR) as an active microwave remote sensing technology has all-weather imaging capability to penetrate clouds and smoke, providing a new solution for forest fire monitoring.

[0019] However, how to convert SAR data into structured information that can accurately reflect the dynamic evolution process of the fire, especially the different burning states (such as undergrowth fire and crown fire), is still a challenge. The related art focuses on the simple identification of the fire area or the statistics of the burned area, and lacks the fine description and time sequence tracking of the heterogeneous burning states in the fire field, which is difficult to support fire spread prediction and accurate rescue command. Therefore, there is an urgent need for a method that can automatically construct a time sequence sample containing a burning state type label to provide a high-quality data basis for training an intelligent fire identification and risk assessment model.

[0020] In view of this, the embodiments of the present application provide a forest fire time sequence sample construction method, a forest fire identification model training method, device, equipment, medium and product, which can solve at least one of the above technical problems.

[0021] The forest fire time sequence sample construction method provided by the embodiments of the present application will be described in detail below in conjunction with the drawings, specific embodiments and application scenarios.

[0022] Figure 1 This is a flowchart illustrating a method for constructing forest fire time series samples according to an embodiment of this application. The execution entity of this method can be a forest fire time series sample construction device.

[0023] The following example illustrates the forest fire time series sample construction method of this application, using the forest fire time series sample construction device as the executing entity. It should be noted that the aforementioned executing entity and application scenario do not constitute a limitation on this application.

[0024] like Figure 1 As shown, the method for constructing forest fire time series samples provided in this application embodiment may include steps 110-150.

[0025] Step 110: Obtain synthetic aperture radar (SAR) images of the target forest area at N time-series nodes during the fire process to obtain multi-temporal SAR images; Step 120: Based on the interferometric coherence changes of SAR images of two adjacent time-series nodes in N time-series nodes, the fire range area is detected from the target forest area; Step 130: Extract the multi-dimensional SAR feature parameters of the fire range area at two adjacent time nodes respectively, and determine the multi-dimensional feature change information between the two adjacent time nodes based on the multi-dimensional SAR feature parameters of the two adjacent time nodes. Step 140: Based on the multi-dimensional feature change information between two adjacent time nodes, determine the combustion state type of the fire range area and obtain the combustion state type label associated with the second time node; Step 150: Combine the multi-temporal SAR images, multi-dimensional feature change information and their combustion state type labels corresponding to N time-series nodes to construct forest fire time-series samples, and train the forest fire identification model based on the forest fire time-series samples.

[0026] The method for constructing time-series samples of forest fires provided in this application, by using multi-temporal synthetic aperture radar (SAR) images as the data source, can effectively overcome the limitations of traditional monitoring methods. Specifically, SAR has the ability to acquire data around the clock and in all weather conditions, and is not restricted by lighting conditions or meteorological factors such as clouds, fog, and smoke. Therefore, it can solve the problems of optical remote sensing being invisible in severe weather and aerial monitoring being limited by visibility. Furthermore, by extracting and analyzing the temporal variation information of multi-dimensional SAR feature parameters within the fire area, collaborative analysis and cross-validation can be performed based on fully polarimetric scattering characteristics and vegetation coverage. This allows for the discovery of differentiated characteristics of different combustion states in microwave scattering mechanisms, overcoming the uncertainty of single-parameter discrimination, achieving refined discrimination of combustion state types, improving the accuracy and reliability of identifying different combustion state types during the fire process, providing accurate input for subsequent dynamic assessment of fire risk levels, and solving the problems of coarse fire type identification and inaccurate assessment by traditional methods. Ultimately, by integrating images, feature change information, and combustion status labels from multiple time-series nodes, the constructed structured time-series samples can provide reliable data support for training automated, high-precision machine learning models for identifying fire types, thereby promoting rapid and accurate dynamic monitoring and risk assessment of forest fires.

[0027] The specific implementation of the above steps will be described in detail below with reference to specific embodiments.

[0028] Step 110 involves acquiring synthetic aperture radar (SAR) images of the target forest area at N time-series nodes during the fire process, resulting in multi-temporal SAR images.

[0029] In step 110, N is a positive integer, and a time sequence node refers to a specific time point in the data collection. A series of time sequence nodes arranged in chronological order constitute the observation time series. The N time sequence nodes involve the period before and after the fire. In this application, the data collection time point before the fire can be taken as the first time sequence node, and the subsequent N-1 time sequence nodes are all data collection time points after the fire.

[0030] Specifically, this application utilizes a drone equipped with a fully polarimetric SAR sensor to automatically cruise the target forest area along a predetermined route and collect data. By repeatedly flying the same route, multi-temporal SAR images of the target forest area at N time points during the fire process can be obtained.

[0031] Among them, the preferred method is to use the full polarimetric SAR imaging mode. Full polarimetric data can record the backscattering information of ground objects under different polarization combinations, providing a richer data foundation for subsequent fine feature extraction.

[0032] According to an embodiment of this application, optionally, after obtaining the multi-temporal SAR image and before detecting the fire range area, the method may further include: preprocessing the multi-temporal SAR images of N time-series nodes, the preprocessing including: filtering the multi-temporal SAR images to suppress image noise; performing geometric correction on the filtered multi-temporal SAR images to match the SAR images with geographic coordinates; and performing registration processing on the geometrically corrected multi-temporal SAR images to align the SAR images of different time-series nodes in space.

[0033] Specifically, SAR images often contain noise, which can be removed using median filtering. By replacing each pixel with the median of its neighborhood window, random noise is effectively suppressed while preserving edge structures and improving image clarity. The directly observed backscattering coefficients are extracted to provide input for subsequent vegetation cover extraction. Combining the UAV's GPS position, IMU attitude parameters, and imaging parameters, geometric corrections are performed on the image, including slant range correction, azimuth correction, and map projection transformation. This achieves accurate matching between the image and map locations, providing fundamental support for fire area localization and multi-temporal image alignment.

[0034] For multi-temporal SAR images, cross-correlation can be used to calculate the correlation between corresponding regions in the images, obtain the correlation matrix between the two images, and adjust the transformation parameters of the images to maximize the correlation coefficient between the two images, so as to ensure the spatial consistency of the image data.

[0035] For example, correlation can be calculated based on the pixel values ​​of the images, for two SAR images. and The correlation can be calculated using the cross-correlation function of the following formula (1). : (1) Where μ1 and μ2 are respectively and The average pixel value, These are pixel coordinates.

[0036] In step 120, the fire range area is detected from the target forest area based on the interferometric coherence change of the SAR images of two adjacent time-series nodes in N time-series nodes.

[0037] In step 120, two adjacent time-series nodes include a first time-series node and a second time-series node, with the first time-series node being the previous time-series node of the second time-series node. For each pair of adjacent time-series nodes, a fire range area can be detected based on the two SAR images. The area where the change in interferometric coherence is less than a preset coherence threshold can be identified as the fire range area. The preset coherence threshold can be set according to specific needs, and this application does not impose any specific limitations on it.

[0038] Specifically, this application can calculate the coherence of the entire SAR image from two adjacent time-series nodes to obtain a grayscale image, i.e., a coherence map, representing the coherence level of each pixel. By setting an empirical coherence threshold, the coherence is... Figure Two Value-based analysis is used to initially identify areas with coherence values ​​below a certain threshold as areas where changes have occurred, i.e., areas within the fire's range. Since changes caused by fire typically lead to a sharp decrease in coherence, these low-coherence areas are strongly correlated with the fire's burning extent.

[0039] In step 130, multi-dimensional SAR feature parameters of the fire range area at two adjacent time nodes are extracted respectively, and multi-dimensional feature change information between the two adjacent time nodes is determined based on the multi-dimensional SAR feature parameters of the two adjacent time nodes.

[0040] In step 130, multi-dimensional SAR feature parameters refer to various quantitative indicators extracted from SAR images that can reflect the characteristics of ground features from different physical perspectives. Multi-dimensional SAR feature parameters can include at least fully polarimetric scattering features and vegetation cover. Vegetation cover refers to the proportion of the vertically projected area of ​​the vegetation canopy within a pixel. Because SAR signals such as cross-polarization (e.g., HV) are sensitive to vegetation structure, they can be used to invert vegetation cover. Feature variation information refers to the differences in SAR feature parameters at the same location between different time-series nodes, such as differences, ratios, or more complex variations.

[0041] For example, for and These two time-series nodes respectively calculate the time of each cell. The characteristic parameter values ​​at time t are relative to The change over time. For example, calculating the change in scattering entropy. Changes in vegetation coverage .

[0042] According to an embodiment of this application, optionally, the aforementioned fully polarimetric scattering characteristics may include scattering entropy and average scattering angle. The step 130, which determines the change in fully polarimetric scattering characteristics between two adjacent time-series nodes based on their respective fully polarimetric scattering characteristics, may include: extracting fully polarimetric SAR data from the SAR image of each time-series node, and constructing corresponding coherence matrices based on the fully polarimetric SAR data of each time-series node; performing eigenvalue decomposition on each coherence matrix to obtain the corresponding scattering entropy and average scattering angle; determining the difference between the scattering entropy of the second time-series node and the scattering entropy of the first time-series node as the change in scattering entropy, and determining the difference between the average scattering angle of the second time-series node and the average scattering angle of the first time-series node as the change in average scattering angle; and combining the change in scattering entropy and the change in average scattering angle to obtain the change in fully polarimetric scattering characteristics.

[0043] Specifically, for the first time-series node ( ) and second time sequence node ( The fully polarimetric data of each pixel is extracted from the corresponding SAR image. Fully polarimetric SAR data can contain complex scattering information for each pixel under four polarization combinations (VV, VH, HV, HH), specifically represented as a four-channel complex scattering matrix: ,in, This represents the complex scattering information of the same polarization channel. This represents the complex scattering information of the cross-polarization channel.

[0044] Then, the coherence matrix T is generated according to the rules for constructing the complex scattering matrix and the polarization matrix: (2) Here, the superscript H indicates the conjugate transpose, and the superscript sign indicates averaging to reduce noise. ; Performing eigenvalue decomposition on matrix T yields three eigenvalues ​​and their corresponding eigenvectors: (3) (4) Where H is the scattering entropy, and A is the anisotropy. The average scattering angle, For normalized eigenvalues, For real eigenvalues, For the corresponding complex eigenvectors, .

[0045] For each time-series node ( and The coherence matrix T of the pixel is decomposed into eigenvalues ​​to obtain its eigenvalues ​​and corresponding eigenvectors. Based on the probability distribution of the eigenvalues, the scattering entropy (H), anisotropy (A), and average scattering angle (A) of the pixel are calculated. Average scattering angle It is then obtained by weighted averaging of the scattering angles corresponding to each eigenvector.

[0046] In this embodiment, by extracting fully polarimetric SAR data from multi-temporal SAR images and constructing a coherence matrix, and then obtaining the scattering entropy and average scattering angle through eigenvalue decomposition, the temporal changes in forest surface scattering mechanisms as fire progresses can be accurately quantified. By calculating the difference between scattering entropy and average scattering angle between adjacent time-series nodes, the changes in scattering entropy and average scattering angle are obtained. These two factors together constitute the changes in fully polarimetric scattering characteristics. These changes can directly and quantitatively reflect the process by which the vegetation structure caused by fire transforms from complex volume scattering to simpler surface or secondary scattering, providing crucial physical mechanism changes for subsequent precise identification of combustion state types.

[0047] According to an embodiment of this application, optionally, determining the change in vegetation coverage between two adjacent time-series nodes based on the vegetation coverage of two adjacent time-series nodes in step 120 may include: extracting the full polarimetric SAR data of the corresponding time-series node from the SAR image of each time-series node, and constructing a corresponding complex scattering matrix based on the full polarimetric SAR data of each time-series node; extracting the backscattering coefficients of the cross-polarization channels from the complex scattering matrices of the first time-series node and the second time-series node respectively, to obtain the first backscattering coefficient and the second backscattering coefficient; determining the first vegetation coverage of the pixel at the first time-series node based on the first backscattering coefficient, the first reference value, and the second reference value; determining the second vegetation coverage of the pixel at the second time-series node based on the second backscattering coefficient, the first reference value, and the second reference value; and determining the difference between the second vegetation coverage and the first vegetation coverage and the ratio of the first vegetation coverage to the first vegetation coverage as the change in vegetation coverage.

[0048] The first reference value is the backscattering coefficient of the cross-polarization channel corresponding to the pure vegetation SAR image, and the second reference value is the backscattering coefficient of the cross-polarization channel corresponding to the pure surface SAR image.

[0049] Specifically, in the SAR data prior to the fire, a large, undisturbed, dense forest area was selected, and the average value of its cross-polarization (HV) channel backscattering coefficient was calculated as the first reference value, representing the signal response of 100% vegetation cover. In the SAR data after the fire, a completely burned, bare area with no vegetation residue was selected, and the average value of its HV channel backscattering coefficient was calculated as the second reference value, representing the signal response of 0% vegetation cover.

[0050] For the pixels to be analyzed within the fire area, starting from the first time-series node ( ) and second time sequence node ( From the complex scattering matrix of , the observed backscattering coefficients of its cross-polarized (HV) channel are extracted. , denoted as the first backscattering coefficient ( ) and the second backscattering coefficient ( ).

[0051] For example, vegetation cover can be calculated using formula (5) for each time series node. : (5) in, and These are the first reference value and the second reference value, respectively.

[0052] To highlight relative changes, the change in vegetation cover It can be a relative rate of change, i.e. , The highest vegetation coverage rate, This is the second vegetation coverage rate. A negative value indicates a decrease in vegetation cover, and the magnitude of the absolute value reflects the severity of the decrease.

[0053] In this embodiment, by utilizing the sensitivity of the cross-polarization channel backscattering coefficient in fully polarimetric SAR data to vegetation structure, direct quantitative monitoring of vegetation cover before and after a fire is achieved. The cross-polarization backscattering coefficient extracted from the complex scattering matrix effectively reflects the volumetric scattering characteristics of vegetation. By introducing the backscattering coefficients of pure vegetation and pure surface endmembers as the first and second reference values, and inverting the vegetation cover at two time-series nodes based on a linear mixture model, the contributions of vegetation and surface background can be separated, thus obtaining more accurate vegetation cover estimation results. Furthermore, by calculating the relative rate of change of vegetation cover between the two time-series nodes as the change quantity, this indicator can intuitively characterize the degree and rate of change of vegetation loss caused by the fire. The resulting change in vegetation cover provides a crucial quantitative basis for subsequent accurate identification of the combustion state type, helping to distinguish fire types with different degrees of impact on the vegetation layer. Moreover, this method relies on microwave remote sensing and is not easily affected by meteorological conditions such as smoke and clouds at the fire site.

[0054] In step 140, based on the multi-dimensional feature change information between two adjacent time nodes, the combustion state type of the fire range area is determined, and the combustion state type label associated with the second time node is obtained.

[0055] In step 140, the combustion state type can be a fine classification of the fire combustion status of the ground area represented by the pixel at a specific moment, such as "crown fire", "underground fire turning into crown fire", "stable underground fire", "extinguished or low-intensity fire", etc.

[0056] According to an embodiment of this application, optionally, the aforementioned multi-dimensional feature change information includes changes in scattering entropy, changes in average scattering angle, and changes in vegetation coverage. Step 140, based on the multi-dimensional feature change information between two adjacent time-series nodes, determines the combustion state type of the fire area. Specifically, this may include: if the change in scattering entropy indicates a decrease in entropy value, the change in average scattering angle indicates a decrease in average scattering angle, and the change in vegetation coverage indicates a decrease in coverage greater than a first preset threshold, then the combustion state type is determined to be a crown fire; if the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in coverage less than the first preset threshold but greater than a second preset threshold, then the combustion state type is determined to be a transition from understory fire to crown fire; if the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in coverage less than the second preset threshold, then the combustion state type is determined to be a stable understory fire.

[0057] In this embodiment, a set of clear logical discrimination rules is established by comprehensively considering three characteristics: changes in scattering entropy, changes in average scattering angle, and changes in vegetation coverage, and setting thresholds based on the direction and magnitude of these changes. These rules can distinguish between three typical combustion states: crown fire, understory fire transitioning to crown fire, and stable understory fire. A simplified scattering mechanism (entropy decrease, angle decrease) and rapid vegetation reduction (decline greater than a first preset threshold) are classified as crown fire; increased scattering randomness (entropy increase) and the beginning of vegetation reduction (decline between two thresholds) are classified as understory fire transitioning to crown fire; and increased scattering randomness (entropy increase) but minimal vegetation change (decline less than a second preset threshold) are classified as stable understory fire. This achieves preliminary, refined, and automatic identification of fire combustion states.

[0058] In step 150, forest fire time series samples are constructed by combining SAR images, multi-dimensional feature change information and their combustion state type labels corresponding to N time series nodes, so as to train the forest fire identification model based on the forest fire time series samples.

[0059] In step 150, the SAR image of each time-series node, the multidimensional feature change information calculated based on neighboring nodes, and the combustion state type label of each pixel at the corresponding time-series node are associated, typically using spatiotemporal coordinates (such as latitude and longitude, timestamp) as the association keywords. Forest fire time-series samples are used to train and validate machine learning or deep learning models capable of automatically identifying fire combustion states, i.e., forest fire identification models. These models are used to identify the combustion state type corresponding to the fire's extent.

[0060] In this way, the forest fire identification model can automatically and accurately identify the burning status in new and unknown fire events by learning the complex mapping relationship between feature change patterns in historical samples and burning status labels.

[0061] To further improve the accuracy of combustion state identification, according to an embodiment of this application, optionally, the multi-dimensional SAR feature parameters also include backscattering coefficients. Step 130, based on the backscattering coefficients of two adjacent time-series nodes, determines the change in backscattering intensity between two adjacent time-series nodes. Specifically, this may include: extracting the full polarimetric SAR data of the corresponding time-series node from the SAR image of each time-series node, and constructing corresponding complex scattering matrices based on the full polarimetric SAR data of each time-series node; extracting cross-scattering data from the complex scattering matrix of the first time-series node. The backscattering coefficients of the polarization channel and the co-polarization channel are used to obtain the first backscattering coefficient and the third backscattering coefficient. From the complex scattering matrix of the second time-series node, the backscattering coefficients of the cross-polarization channel and the co-polarization channel are extracted to obtain the second backscattering coefficient and the fourth backscattering coefficient. The ratio of the difference between the first backscattering coefficient and the second backscattering coefficient to the first backscattering coefficient, and the ratio of the difference between the second backscattering coefficient and the fourth backscattering coefficient to the second backscattering coefficient are obtained. The maximum value of the two ratios is determined as the change in backscattering intensity.

[0062] Specifically, for the first time-series node ( Extract the backscattering coefficients of the cross-polarization channel (VH as an example) from the complex scattering matrix of the pixel. (i.e., the first backscattering coefficient) and the backscattering coefficient of the same polarization channel (VV) (i.e., the third backscattering coefficient). Similarly, from the second time node ( )extract (Second backscattering coefficient) and (Fourth backscattering coefficient).

[0063] Calculate the rate of change of backscattering between time nodes in a cross-polarization channel: Rate of change of backscattering between time-series nodes in the same polarization channel: The maximum of the two changes mentioned above is taken as the change in backscattering intensity of that pixel, i.e. The aim is to ensure that the most significant changes caused by fire, as manifested in any backscattering feature, can be captured.

[0064] In this embodiment, by extracting the backscattering coefficients of the cross-polarized and co-polarized channels and calculating their relative change ratios, the maximum value is taken as the change in backscattering intensity. This allows for the keen detection of the most significant change patterns in radar echoes caused by fire. Based on this, the dramatic attenuation of both the cross-polarized and co-polarized channels effectively reflects the comprehensive change characteristics of radar backscattering intensity caused by vegetation burning and structural damage, supplementing the observation of the intensity dimension of the physical response to fire and further improving the accuracy of combustion state identification.

[0065] SAR images during a fire will show changes in interferometric coherence. The area where the fire occurs usually shows a lower coherence value. Therefore, interferometric coherence change detection (CCD) can be used to compare SAR image data from multiple time phases. By calculating the interferometric coherence of SAR images before and after the fire, the amount of interferometric coherence change that characterizes the stability of the surface scattering structure can be obtained.

[0066] Based on this, according to the embodiments of this application, optionally, the multi-dimensional SAR feature parameters may also include interferometric coherence. The above step 130, based on the interferometric coherence of two adjacent time-series nodes, determines the amount of change in interferometric coherence between two adjacent time-series nodes, which may include: obtaining corresponding complex SAR images from the SAR images of the first time-series node and the second time-series node respectively; calculating the interferometric coherence coefficients of the corresponding complex pixel values ​​of the two images within a predefined local window based on the complex SAR images of the first time-series node and the second time-series node, determining the interferometric coherence value between two adjacent time-series nodes, and obtaining the amount of change in interferometric coherence characterizing the stability of the surface scattering structure.

[0067] Specifically, a complex SAR image is a SAR image that preserves the amplitude and phase information of the backscattered signal at each pixel. It is obtained by calculating the normalized cross-correlation coefficient of corresponding pixels in two complex SAR images within a local spatial window. The coefficient ranges from 0 to 1 and is used to measure the consistency of the surface scattering phase between two observations. A fixed-size local window (e.g., 7x7 pixels) is defined centered on the pixel to be calculated. Within this window, the calculation... and Interferometric coherence of corresponding pixels in two complex SAR images .

[0068] For example, formula (6) can be used to compare the complex pixel values ​​of two SAR images at the same location at different times. and Calculate coherence : (6) in, This indicates that the statistical average is taken within a local window. Indicates complex conjugation. It represents the amplitude of a complex number of pixels.

[0069] In this way, an interferometric coherence map corresponding to the fire area can be calculated, where each pixel value represents the location in the fire zone. arrive Coherence over a time period. Based on the calculated interferometric coherence map, statistical features that characterize the overall change in coherence across the entire fire area are extracted as the measure of interferometric coherence variation. For example, the variation of coherence in all pixels within the area can be calculated. Values ​​such as average, median, and the proportion of pixels below a specific threshold (e.g., 0.3). A lower statistical value, such as the average... A low value, or a high proportion of low-coherence pixels, indicates that the surface scattering structure in that region underwent drastic or significant changes during the observation period.

[0070] In this embodiment, by acquiring complex SAR images and calculating their interferometric coherence coefficients, the obtained interferometric coherence values ​​are directly used as the coherence variation, which directly characterizes the stability of the surface scattering structure between adjacent time-series nodes. Since drastic surface changes caused by fires lead to a significant decrease in this value, this coherence variation can serve as a key indicator for detecting fire-affected areas and assessing the severity of their changes, further increasing the dimension for distinguishing combustion state types and thus improving its accuracy.

[0071] According to an embodiment of this application, optionally, the multi-dimensional feature change information may further include the backscattering intensity change and the interference coherence change. The above step 140, based on the multi-dimensional feature change information between two adjacent time-series nodes, determines the combustion state type of the fire range area, which may include: fusing the full polarization scattering feature change, vegetation coverage change, backscattering intensity change, and interference coherence change in the multi-dimensional feature change information to form a multi-dimensional feature change vector; and determining the multi-dimensional feature change vector based on a preset decision rule to obtain the combustion state type.

[0072] The changes in the fully polarized scattering characteristics can include the changes in scattering entropy and the changes in the average scattering angle.

[0073] Specifically, by combining and normalizing the aforementioned variable parameters in a predetermined order and manner, a structure can be constructed as follows: Figure 2The multidimensional feature change vector shown is used to characterize the change from... arrive The overall evolution of the fire area in multiple key physical dimensions at any given time can specifically include changes in interference coherence, backscattering intensity, total polarization scattering characteristics, and vegetation coverage extracted from the fire range area.

[0074] The multidimensional feature change vectors representing specific fire areas, constructed as described above, are input into a pre-defined decision rule base. These pre-defined decision rules are defined based on prior knowledge of the physical processes of different combustion states and their typical response patterns in microwave multidimensional features. Through logical judgment, the numerical patterns of the input vectors are mapped to specific combustion state type labels. For example, when the vector indicates "drastically reduced coherence, a sharp decrease in backscattering, a decrease in scattering entropy and average scattering angle, and a rapid reduction in vegetation cover," the area is determined to be in a state of combustion. The dominant combustion state at that moment was "crown fire".

[0075] In this embodiment, a multi-dimensional feature variation vector is constructed by fusing the changes in total polarization scattering characteristics, vegetation cover, backscattering intensity, and interferometric coherence. This vector integrates fire response information from multiple dimensions, including scattering mechanisms, vegetation cover, echo intensity, and surface stability, providing a more comprehensive and complementary feature set for combustion state identification. Based on this multi-dimensional vector and using preset decision rules, the uncertainty of identifying single or a few features can be overcome, further improving the accuracy and robustness of combustion state type identification.

[0076] According to embodiments of this application, optionally, the preset decision rules may include: If the change in interference coherence is within the first coherence interval, the change in backscattering intensity indicates a decrease in backscattering intensity greater than the third preset amplitude, the change in scattering entropy indicates a decrease in entropy, the change in average scattering angle indicates a decrease in average scattering angle, and the change in vegetation coverage indicates a decrease in vegetation coverage greater than the first preset amplitude threshold, then the combustion state type is determined to be crown fire. If the change in interference coherence is within the second coherence interval, the change in backscattering intensity indicates an increase in backscattering intensity, the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in vegetation coverage that is less than the first preset amplitude threshold and greater than the second preset amplitude threshold, then the combustion state type is determined to be a conversion from understory fire to crown fire. If the change in interference coherence is within the third coherence interval, the change in backscattering intensity indicates an increase in backscattering intensity, the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in coverage less than the second preset threshold, then the combustion state type is determined to be a stable forest fire.

[0077] The interval values ​​corresponding to the first, second, and third coherence intervals are arranged in ascending order of coherence; that is, the first coherence interval has the smallest value, the second coherence interval has the second largest value, and the third coherence interval has the largest value. The specific thresholds for each interval, for example, less than 0.3 for the first interval, 0.3 to 0.6 for the second interval, and greater than 0.6 for the third interval, can be determined through calibration using historical fire data or theoretical models. The thresholds for other characteristic changes, such as the first, second, and third preset amplitude thresholds, are also pre-set based on typical fire response modes.

[0078] For each pixel or region unit to be judged within the fire area, the system matches its multidimensional feature change vector with the above decision rules: Crown fire identification: The unit is identified as a crown fire if and only if the change in interference coherence of the unit falls into the first coherence interval (indicating a severe loss of coherence), the change in backscattering intensity shows a decrease exceeding the third preset amplitude (indicating a sharp attenuation of echo intensity), the change in scattering entropy is negative (indicating a decrease in scattering randomness), the change in average scattering angle is negative (indicating that the scattering mechanism tends to surface scattering), and the decrease in vegetation cover is greater than the first preset amplitude threshold (indicating a rapid reduction in vegetation).

[0079] Determining the conversion of understory fire to crown fire: When the change in interference coherence falls into the second coherence interval (indicating a moderate decrease in coherence), the change in backscattering intensity shows an upward trend (indicating a slight increase in echo), the change in scattering entropy is positive (indicating an increase in scattering randomness), and the decrease in vegetation cover is between the first and second preset threshold values ​​(indicating that vegetation has begun to decrease but has not reached a drastic level), it is determined that the understory fire has converted to crown fire.

[0080] Stable forest fire identification: When the change in interference coherence falls into the third coherence interval (indicating a small change in coherence), the change in backscattering intensity shows an upward trend, the change in scattering entropy is positive, and the decrease in vegetation cover is less than the second preset threshold (indicating that the change in vegetation cover is not significant), it is identified as a stable forest fire.

[0081] In this embodiment, by setting a preset decision rule based on a combination of multi-dimensional feature thresholds, the quantitative information of changes in interference coherence, backscattering intensity, total polarization scattering characteristics, and vegetation cover is integrated into a unified discrimination logic, effectively improving the accuracy and physical interpretability of combustion state type discrimination. Specifically, this rule divides coherence changes into different intervals from small to large and coordinates them with specific change directions and amplitude thresholds of other feature parameters. This enables the system to accurately match the differentiated physical response patterns presented by different combustion processes such as crown fire and understory fire transformation in microwave multi-dimensional features, thereby overcoming the uncertainty of single-parameter discrimination. Simultaneously, this rule automates the discrimination process with clear thresholds and logical conditions, improving processing efficiency and providing a directly deployable judgment basis for the refined identification of heterogeneous combustion states within a fire scene, enhancing the practicality and reliability of the method.

[0082] According to an embodiment of this application, optionally, it further includes: if the multi-dimensional SAR feature parameters corresponding to the second time-series node match the multi-dimensional SAR feature parameters of the target, then the combustion state type of the fire range area is determined to be extinguished or low-intensity fire.

[0083] The target multidimensional SAR feature parameters can be multidimensional SAR feature parameters corresponding to a pure surface SAR image. These can be a set of SAR feature parameters extracted from a pure surface reference area that represents a typical state of bare surface without vegetation or with vegetation completely burned. The target multidimensional SAR feature parameters can include multidimensional parameter reference values, such as at least two of the following: vegetation coverage reference value, fully polarimetric scattering feature reference value, backscattering coefficient reference value, and interferometric coherence reference value.

[0084] The multi-dimensional SAR feature parameters calculated for the fire area at the second time-series node (i.e., the current monitoring time) are quantitatively compared with the multi-dimensional SAR feature parameters of the target to determine whether they are statistically or numerically sufficiently close or consistent. If the multi-dimensional feature parameters of all or key parts of the current fire area fully match the multi-dimensional SAR feature parameters of the target, it indicates that the surface scattering characteristics of the area have recovered to a state similar to stable bare ground after a fire, i.e., the open flame has been extinguished, and combustion activity has stopped or is extremely low in intensity. Therefore, the combustion state type of the fire area at the second time-series node is determined to be "extinguished or low-intensity fire".

[0085] For example, when the multi-dimensional SAR feature parameters corresponding to the second time-series node meet the following benchmark conditions, the combustion state type is determined to be extinguished or low-intensity fire: the backscattering coefficient under the cross-polarization (HV) channel is less than the preset backscattering threshold; the interferometric coherence is greater than the preset coherence threshold (indicating that the surface is stable and there is no continuous change); the scattering entropy (H) is less than the preset scattering entropy threshold, and the average scattering angle ( The value is less than the preset scattering angle threshold, which indicates a simple scattering mechanism dominated by surface scattering; the vegetation coverage is close to zero.

[0086] In this embodiment, by matching the multi-dimensional SAR feature parameters of the fire area at the second time-series node (current monitoring time) with the target multi-dimensional SAR feature parameters corresponding to the pure surface SAR image, it is possible to effectively determine whether the area has recovered to a stable bare ground scattering state after the fire. When the match is successful, the combustion state type is determined to be extinguished or low-intensity fire. This can accurately identify the tail end or ember area of ​​the fire, reduce the probability of false alarms, and improve the monitoring capability of the entire fire life cycle.

[0087] Optionally, if the multi-dimensional SAR feature parameters corresponding to the second time-series node are all within the reference range associated with the target multi-dimensional SAR feature parameters, then the combustion state type of the fire range area is determined to be extinguished or low-intensity fire.

[0088] The benchmark interval range can include sub-benchmark interval ranges for each dimension, and each sub-benchmark interval range is determined based on the parameter benchmark value and its error value for the corresponding dimension.

[0089] According to an embodiment of this application, optionally, after step 140 above, it may include: determining the corresponding fire risk level based on the combustion state type label associated with the second time-series node; step 150 above, which constructs the forest fire time-series sample, may specifically include: associating and storing the multi-temporal SAR images, feature change information, combustion state type labels and fire risk levels corresponding to N time-series nodes to construct the forest fire time-series sample.

[0090] Specifically, such as Figure 3 As shown, the corresponding fire risk level is determined based on the type of combustion state. This can include: if the type of combustion state is crown fire, the fire risk level of the corresponding area is determined to be high risk; if the type of combustion state is forest fire turning into crown fire, the fire risk level of the corresponding area is determined to be medium-high risk; if the type of combustion state is stable forest fire, the fire risk level of the corresponding area is determined to be medium risk; if the type of combustion state is extinguished or low-intensity fire, the fire risk level of the corresponding area is determined to be low risk.

[0091] Optionally, when a pixel exhibits obvious canopy burning characteristics and significant canopy structure damage, the area is identified as a high-risk area; when a pixel still shows understory fire but the vegetation coverage is declining significantly and there is a possibility of spreading to the canopy, it is identified as a medium-to-high-risk area; when a pixel only shows understory fire and the canopy is not significantly disturbed, it is classified as a medium-risk area; and for pixels with extinguished fires, low-intensity fires, or stable vegetation structures, they are identified as low-risk areas.

[0092] Optionally, the time series samples of forest fires can be used as a dataset for dynamic monitoring and risk assessment of forest fires. This dataset may include fields such as geographic coordinates (latitude and longitude), time index, land cover category, vegetation coverage index, burning state type label, and risk level label, and can serve as a data source for subsequent fire trend analysis, fire spread simulation, and emergency command systems.

[0093] As a concrete example, such as Figure 4 As shown, this application can construct forest fire time series samples through the following exemplary process: S1, Data Acquisition: Multi-temporal imaging of the target forest area is performed by using a UAV equipped with synthetic aperture radar (SAR) to obtain multi-temporal SAR images; S2, Data Preprocessing: Perform preprocessing such as radiometric calibration and geometric correction on multi-temporal SAR images to provide high-quality basic data for analysis; S3, Fire Area Screening: Using coherent change detection (CCD) technology, the interference coherence between adjacent time-series SAR images is analyzed to automatically and quickly screen out areas that have undergone significant changes, i.e., fire range areas. S4, Multi-dimensional SAR feature parameter extraction: Extract multi-dimensional SAR feature parameters such as interferometry, backscattering intensity, full polarization scattering characteristics, and vegetation coverage within the fire area. S5 determines multi-dimensional feature change information based on the change of multi-dimensional SAR feature parameters of adjacent time-series nodes, and intelligently identifies the specific combustion state type based on the change pattern of these multi-dimensional features and through preset discrimination rules. S6: Fire Risk Assessment: Based on the identified type of combustion and its hazard, the risk level of the fire area is assessed, and an intuitive fire risk level map is output to provide accurate intelligence support for firefighting command.

[0094] In this way, the process makes full use of SAR's all-weather observation advantages and time series analysis capabilities to achieve automated processing of forest fires from early detection and precise identification of combustion status to dynamic risk assessment.

[0095] Corresponding to the method embodiments of this application, this application also provides a training method for a forest fire identification model, comprising: training a large model based on the forest fire time series samples in the first aspect embodiments to obtain a forest fire identification model.

[0096] Corresponding to the method embodiments of this application, this application also provides an apparatus for constructing forest fire time series samples.

[0097] Figure 5 This is a schematic diagram of a device for constructing a forest fire time series sample according to an embodiment of this application. Figure 5 As shown, the forest fire time series sample construction device 500 may include: an acquisition module 510, a detection module 520, a determination module 530, a discrimination module 540, and a sample construction module 550.

[0098] The acquisition module 510 is used to acquire synthetic aperture radar (SAR) images of the target forest area at N time-series nodes during the fire process, obtaining multi-temporal SAR images. The detection module 520 is used to detect the fire range area from the target forest area based on the interferometric coherence changes of the SAR images of two adjacent time-series nodes among the N time-series nodes, wherein the two adjacent time-series nodes include a first time-series node and a second time-series node, and the first time-series node is the time-series node preceding the second time-series node. The determination module 530 is used to extract multi-dimensional SAR feature parameters of the fire range area at two adjacent time-series nodes based on the multi-temporal SAR images, and to determine the fire range area based on the multi-dimensional SAR feature parameters of the two adjacent time-series nodes. The system consists of a data collection module 540 and a sample construction module 550. The latter is used to determine the multi-dimensional feature change information between two adjacent time-series nodes, where the multi-dimensional SAR feature parameters include fully polarimetric scattering features and vegetation coverage. The former is used to determine the combustion state type of the fire range area based on the multi-dimensional feature change information between two adjacent time-series nodes, and obtain the combustion state type label associated with the second time-series node. The latter is used to construct forest fire time-series samples by combining the multi-temporal SAR images, multi-dimensional feature change information and combustion state type labels corresponding to N time-series nodes, so as to train a forest fire identification model based on the forest fire time-series samples. The forest fire identification model is used to identify the combustion state type corresponding to the fire range area.

[0099] The forest fire time-series sample construction device provided in this application, by using multi-temporal synthetic aperture radar (SAR) images as the data source, can effectively overcome the limitations of traditional monitoring methods. Specifically, SAR has all-weather, all-day data acquisition capabilities, unaffected by lighting conditions or meteorological factors such as clouds, fog, and smoke, thus solving the problems of optical remote sensing being invisible in adverse weather and aerial monitoring being limited by visibility. Furthermore, by extracting and analyzing the temporal variation information of multi-dimensional SAR feature parameters within the fire area, collaborative analysis and cross-validation based on fully polarimetric scattering characteristics and vegetation coverage can be performed to uncover the differentiated characteristics of different combustion states in microwave scattering mechanisms, overcoming the uncertainty of single-parameter discrimination, achieving refined discrimination of combustion state types, improving the accuracy and reliability of identifying different combustion state types during the fire process, providing accurate input for subsequent dynamic assessment of fire risk levels, and solving the problems of coarse fire type identification and inaccurate assessment by traditional methods. Ultimately, by integrating images, feature change information, and combustion status labels from multiple time-series nodes, the constructed structured time-series samples can provide reliable data support for training automated, high-precision machine learning models for identifying fire types, thereby promoting rapid and accurate dynamic monitoring and risk assessment of forest fires.

[0100] In some embodiments, the fully polarimetric scattering feature includes scattering entropy and average scattering angle. The determining module is specifically used to: extract fully polarimetric SAR data for the corresponding time-series SAR image from each time-series SAR image, and construct corresponding coherence matrices based on the fully polarimetric SAR data for each time-series SAR image; perform eigenvalue decomposition on each coherence matrix to obtain the corresponding scattering entropy and average scattering angle; determine the difference between the scattering entropy of the second time-series SAR and the scattering entropy of the first time-series SAR as the scattering entropy change, and determine the difference between the average scattering angle of the second time-series SAR and the average scattering angle of the first time-series SAR as the average scattering angle change; and combine the scattering entropy change and the average scattering angle change to obtain the fully polarimetric scattering feature change.

[0101] In some embodiments, the determining module is specifically used for: extracting the full polarimetric SAR data of the corresponding time-series SAR image from the SAR image of each time-series SAR image, and constructing the corresponding complex scattering matrix based on the full polarimetric SAR data of each time-series SAR image; extracting the backscattering coefficients of the cross-polarization channel from the complex scattering matrices of the first time-series SAR image and the second time-series SAR image respectively, to obtain the first backscattering coefficient and the second backscattering coefficient; determining the first vegetation coverage of the pixel at the first time-series SAR image based on the first backscattering coefficient, the first reference value, and the second reference value; determining the second vegetation coverage of the pixel at the second time-series SAR image based on the second backscattering coefficient, the first reference value, and the second reference value; and determining the vegetation coverage change as the ratio of the difference between the second vegetation coverage and the first vegetation coverage to the first vegetation coverage; wherein the first reference value is the backscattering coefficient of the cross-polarization channel corresponding to the pure vegetation SAR image, and the second reference value is the backscattering coefficient of the cross-polarization channel corresponding to the pure surface SAR image.

[0102] In some embodiments, the multi-dimensional feature change information includes changes in scattering entropy, changes in average scattering angle, and changes in vegetation coverage. The discrimination module is specifically used to: if the change in scattering entropy indicates a decrease in entropy value, the change in average scattering angle indicates a decrease in average scattering angle, and the change in vegetation coverage indicates a decrease in vegetation coverage greater than a first preset threshold, then the combustion state type is determined to be crown fire; if the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in vegetation coverage less than the first preset threshold but greater than a second preset threshold, then the combustion state type is determined to be a transition from understory fire to crown fire; if the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in coverage less than the second preset threshold, then the combustion state type is determined to be stable understory fire.

[0103] In some embodiments, the multi-dimensional SAR feature parameters further include backscattering coefficients. The determining module is specifically used to: extract the full polarization SAR data of the corresponding time-series node from the SAR image of each time-series node, and construct the corresponding complex scattering matrix based on the full polarization SAR data of each time-series node; extract the backscattering coefficients of the cross-polarization channel and the co-polarization channel from the complex scattering matrix of the first time-series node to obtain the first backscattering coefficient and the third backscattering coefficient; extract the backscattering coefficients of the cross-polarization channel and the co-polarization channel from the complex scattering matrix of the second time-series node to obtain the second backscattering coefficient and the fourth backscattering coefficient; obtain the ratio of the difference between the first backscattering coefficient and the second backscattering coefficient to the first backscattering coefficient, and the ratio of the difference between the second backscattering coefficient and the fourth backscattering coefficient to the second backscattering coefficient, and determine the maximum value of the two ratios as the change in backscattering intensity.

[0104] In some embodiments, the multidimensional SAR feature parameters further include interferometric coherence. The determination module is specifically used to: obtain corresponding complex SAR images from the SAR images of the first time-series node and the second time-series node, respectively; and, based on the complex SAR images of the first time-series node and the second time-series node, determine the interferometric coherence value between two adjacent time-series nodes by calculating the interferometric coherence coefficients of the corresponding complex pixel values ​​of the two images within a predefined local window, thereby obtaining the interferometric coherence change that characterizes the stability of the surface scattering structure.

[0105] In some embodiments, the multi-dimensional feature change information further includes the backscattering intensity change and the interference coherence change. The discrimination module is specifically used to: fuse the total polarization scattering feature change, vegetation coverage change, backscattering intensity change and interference coherence change in the multi-dimensional feature change information to form a multi-dimensional feature change vector, wherein the total polarization scattering feature change includes the scattering entropy change and the average scattering angle change; and to discriminate the multi-dimensional feature change vector based on a preset decision rule to obtain the combustion state type.

[0106] In some embodiments, the preset decision rules include: if the change in interference coherence is within a first coherence interval, the change in backscattering intensity indicates a decrease in backscattering intensity greater than a third preset amplitude, the change in scattering entropy indicates a decrease in entropy, the change in average scattering angle indicates a decrease in average scattering angle, and the change in vegetation cover indicates a decrease in vegetation cover greater than a first preset amplitude threshold, then the combustion state type is determined to be crown fire; if the change in interference coherence is within a second coherence interval, the change in backscattering intensity indicates an increase in backscattering intensity, and the change in scattering entropy indicates an increase in entropy. If the change in vegetation cover indicates that the decrease in vegetation cover is less than the first preset threshold and greater than the second preset threshold, then the combustion state type is determined to be a transition from understory fire to crown fire. If the change in interference coherence is located in the third coherence interval, the change in backscattering intensity indicates an increase in backscattering intensity, the change in scattering entropy indicates an increase in entropy value, and the change in vegetation cover indicates that the decrease in coverage is less than the second preset threshold, then the combustion state type is determined to be a stable understory fire. The interval values ​​corresponding to the first coherence interval, the second coherence interval, and the third coherence interval are in ascending order.

[0107] In some embodiments, the discrimination module is further configured to: if the multi-dimensional SAR feature parameters corresponding to the second time-series node match the target multi-dimensional SAR feature parameters, then determine that the combustion state type of the fire range area is extinguished or low-intensity fire; wherein, the target multi-dimensional SAR feature parameters are the multi-dimensional SAR feature parameters corresponding to the pure surface SAR image.

[0108] The forest fire time series sample construction apparatus provided in this application embodiment can achieveFigures 1-4 The various processes implemented in the method embodiments can achieve the same technical effect, and will not be described again here to avoid repetition.

[0109] Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application.

[0110] like Figure 6 As shown, the electronic device 600 includes a memory 601, a processor 602, and a computer program stored in the memory 601 and executable on the processor 602.

[0111] In one example, the processor 602 described above may include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0112] Memory 601 may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the method of constructing forest fire timing samples according to the embodiments of the first aspect of this application.

[0113] The processor 602 runs a computer program corresponding to the executable program code by reading the executable program code stored in the memory 601, in order to implement the method for constructing forest fire time series samples in the embodiments of the first aspect described above.

[0114] In some examples, electronic device 600 may also include communication interface 603 and bus 610. For example, Figure 6 As shown, the memory 601, processor 602, and communication interface 603 are connected through bus 610 and complete communication with each other.

[0115] The communication interface 603 is mainly used to enable communication between various modules, systems, units, and / or devices in the embodiments of this application. Input devices and / or output devices can also be connected through the communication interface 603.

[0116] Bus 610 includes hardware, software, or both, that couples components of electronic device 600 together. For example, and not limitingly, bus 610 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-E) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 610 may include one or more buses. Although specific buses are described and illustrated in the embodiments of this application, this application considers any suitable bus or interconnection.

[0117] The electronic device provided in this application embodiment is capable of achieving Figures 1-4 The various processes implemented in the method embodiments can achieve the same technical effect, and will not be described again here to avoid repetition.

[0118] Based on the method for constructing forest fire time series samples in the above embodiments, this application embodiment can provide a computer storage medium for implementation. This computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any step in the above method embodiments.

[0119] Based on the method for constructing forest fire time series samples in the above embodiments, this application embodiment can provide a computer program product for implementation. This (computer) program product is stored in a non-volatile storage medium, and when executed by at least one processor, it implements any step in the above method embodiments.

[0120] This application also provides a chip, which includes a processor and a communication interface. The communication interface and the processor are coupled. The processor is used to run programs or instructions to implement the various processes of the above method embodiments and achieve the same technical effect. To avoid repetition, it will not be described again here.

[0121] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0122] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0123] The functional blocks shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0124] It should also be noted that the exemplary embodiments mentioned in this application describe some methods or systems based on a series of steps or systems. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0125] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, systems (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing system to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing system, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by special-purpose hardware performing the specified functions or actions, or can be implemented by a combination of special-purpose hardware and computer instructions.

[0126] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for constructing a time series sample of forest fires, characterized in that, include: Synthetic Aperture Radar (SAR) images of the target forest area at N time points during the fire process are obtained to obtain multi-temporal SAR images; Based on the interferometric coherence change of SAR images of two adjacent time-series nodes in N time-series nodes, the fire range area is detected from the target forest area. The two adjacent time-series nodes include a first time-series node and a second time-series node, and the first time-series node is the time-series node preceding the second time-series node. Based on multi-temporal SAR images, multi-dimensional SAR feature parameters of the fire range area at two adjacent time nodes are extracted respectively. Based on the multi-dimensional SAR feature parameters of the two adjacent time nodes, the multi-dimensional feature change information between the two adjacent time nodes is determined. The multi-dimensional SAR feature parameters include fully polarimetric scattering features and vegetation coverage. Based on the multi-dimensional feature change information between two adjacent time-series nodes, the combustion state type of the fire range area is determined, and the combustion state type label associated with the second time-series node is obtained; By combining multi-temporal SAR images, multi-dimensional feature change information, and combustion state type labels corresponding to N time-series nodes, a forest fire time-series sample is constructed to train a forest fire identification model. The forest fire identification model is used to identify the combustion state type corresponding to the fire range area.

2. The method according to claim 1, characterized in that, Fully polarimetric scattering characteristics include scattering entropy and average scattering angle. Based on the fully polarimetric scattering characteristics of two adjacent time nodes, the changes in fully polarimetric scattering characteristics between two adjacent time nodes are determined, including: Extract the full polarimetric SAR data of the corresponding time-series SAR image from the SAR image of each time-series SAR image, and construct the corresponding coherence matrix based on the full polarimetric SAR data of each time-series SAR image. Eigenvalue decomposition is performed on each coherence matrix to obtain the corresponding scattering entropy and average scattering angle; The difference between the scattering entropy of the second time-series node and the scattering entropy of the first time-series node is determined as the change in scattering entropy, and the difference between the average scattering angle of the second time-series node and the average scattering angle of the first time-series node is determined as the change in average scattering angle. The change in the total polarization scattering characteristic is obtained by combining the change in scattering entropy and the change in average scattering angle.

3. The method according to claim 1, characterized in that, Based on the vegetation cover rate of two adjacent time-series nodes, determine the change in vegetation cover rate between the two adjacent time-series nodes, including: Extract the full polarimetric SAR data of the corresponding time-series SAR image from the SAR image of each time-series SAR image, and construct the corresponding complex scattering matrix based on the full polarimetric SAR data of each time-series SAR image. The backscattering coefficients of the cross-polarization channel are extracted from the complex scattering matrices of the first and second time-series nodes, respectively, to obtain the first backscattering coefficient and the second backscattering coefficient. Based on the first backscattering coefficient, the first reference value, and the second reference value, the first vegetation coverage of the pixel at the first time node is determined. Based on the second backscattering coefficient, the first reference value, and the second reference value, the second vegetation coverage of the pixel at the second time node is determined. The difference between the second vegetation coverage rate and the first vegetation coverage rate, and the ratio of the second vegetation coverage rate to the first vegetation coverage rate, is determined as the change in vegetation coverage rate. The first reference value is the backscattering coefficient of the cross-polarization channel corresponding to the pure vegetation SAR image, and the second reference value is the backscattering coefficient of the cross-polarization channel corresponding to the pure surface SAR image.

4. The method according to any one of claims 1-3, characterized in that, Multidimensional feature change information includes changes in scattering entropy, average scattering angle, and vegetation cover. Based on the multidimensional feature change information between two adjacent time-series nodes, the combustion state type of the fire range area is determined, including: If the change in scattering entropy indicates a decrease in entropy value, the change in average scattering angle indicates a decrease in average scattering angle, and the change in vegetation coverage indicates a decrease in vegetation coverage greater than a first preset threshold, then the combustion state type is determined to be a crown fire. If the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in vegetation coverage that is less than a first preset threshold and greater than a second preset threshold, then the combustion state type is determined to be a conversion from understory fire to crown fire. If the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in coverage less than a second preset threshold, then the combustion state type is determined to be a stable understory fire.

5. The method according to claim 1, characterized in that, The multi-dimensional SAR feature parameters also include backscattering coefficients. Based on the backscattering coefficients of two adjacent time nodes, the change in backscattering intensity between two adjacent time nodes is determined, including: Extract the full polarimetric SAR data of the corresponding time-series SAR image from the SAR image of each time-series SAR image, and construct the corresponding complex scattering matrix based on the full polarimetric SAR data of each time-series SAR image. The backscattering coefficients of the cross-polarization channel and the same-polarization channel are extracted from the complex scattering matrix of the first time-series node to obtain the first backscattering coefficient and the third backscattering coefficient. The backscattering coefficients of the cross-polarization channel and the co-polarization channel are extracted from the complex scattering matrix of the second time node to obtain the second backscattering coefficient and the fourth backscattering coefficient. The ratio of the difference between the first backscattering coefficient and the second backscattering coefficient to the first backscattering coefficient, and the ratio of the difference between the second backscattering coefficient and the fourth backscattering coefficient to the second backscattering coefficient, are obtained. The maximum value of the two ratios is determined as the change in backscattering intensity.

6. The method according to claim 1, characterized in that, The multi-dimensional SAR feature parameters also include interferometric coherence. Based on the interferometric coherence of two adjacent time-series nodes, the change in interferometric coherence between two adjacent time-series nodes is determined, including: Obtain the corresponding complex SAR images from the SAR images of the first time-series node and the second time-series node, respectively; Based on the complex SAR images of the first and second time-series nodes, the interference coherence coefficients of the corresponding complex pixel values ​​of the two images are calculated within a predefined local window to determine the interference coherence value between two adjacent time-series nodes, thereby obtaining the interference coherence variation that characterizes the stability of the surface scattering structure.

7. The method according to any one of claims 1, 5, and 6, characterized in that, The multi-dimensional feature change information also includes changes in backscattering intensity and interference coherence. Based on the multi-dimensional feature change information between two adjacent time nodes, the combustion state type of the fire range area is determined, including: The changes in total polarization scattering characteristics, vegetation coverage, backscattering intensity, and interference coherence in the multi-dimensional feature change information are fused to form a multi-dimensional feature change vector, wherein the total polarization scattering characteristic changes include scattering entropy changes and average scattering angle changes. The combustion state type is obtained by judging the multidimensional feature change vector based on the preset decision rules.

8. The method according to claim 7, characterized in that, The preset decision rules include: If the change in interference coherence is within the first coherence interval, the change in backscattering intensity indicates a decrease in backscattering intensity greater than a third preset amplitude, the change in scattering entropy indicates a decrease in entropy, the change in average scattering angle indicates a decrease in average scattering angle, and the change in vegetation coverage indicates a decrease in vegetation coverage greater than a first preset amplitude threshold, then the combustion state type is determined to be crown fire. If the change in interference coherence is within the second coherence interval, the change in backscattering intensity indicates an increase in backscattering intensity, the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in vegetation coverage that is less than the first preset amplitude threshold and greater than the second preset amplitude threshold, then the combustion state type is determined to be a conversion from understory fire to crown fire. If the change in interference coherence is located in the third coherence interval, the change in backscattering intensity indicates an increase in backscattering intensity, the change in scattering entropy indicates an increase in entropy value, and the change in vegetation coverage indicates a decrease in coverage less than the second preset threshold, then the combustion state type is determined to be a stable understory fire. The interval values ​​corresponding to the first coherence interval, the second coherence interval, and the third coherence interval are arranged from small to large.

9. The method according to claim 1, characterized in that, Also includes: If the multi-dimensional SAR feature parameters corresponding to the second time-series node match the multi-dimensional SAR feature parameters of the target, then the combustion state type of the fire range area is determined to be either extinguished or low-intensity fire. Among them, the target multidimensional SAR feature parameters are the multidimensional SAR feature parameters corresponding to the pure surface SAR image.

10. A computer program product, characterized in that, The computer program product is stored in a non-volatile storage medium, and when executed by a processor, the computer program product implements the method as described in any one of claims 1-9.

Images