A fire monitoring method and device based on high-orbit staring satellite thermal anomaly evolution and smoke diffusion coupling

Abstract

The application discloses a fire monitoring method and device based on thermal anomaly evolution and smoke diffusion coupling of a high-orbit staring satellite, and belongs to the technical field of fire monitoring. The method solves the problem that the dynamic evolution law of smoke diffusion cannot be effectively captured due to the interference of smoke and non-smoke in a complex scene. The method comprises the following steps: acquiring an observation image sequence, calculating a thermal anomaly continuous growth index of each pixel based on the observation image sequence, performing smoke anomaly feature extraction on the observation image sequence to obtain smoke anomaly features, performing smoke diffusion analysis on the smoke anomaly features by using a preset smoke diffusion dynamics model to obtain smoke diffusion analysis results, inputting the thermal anomaly continuous growth index and the smoke diffusion analysis results into a preset space-time coupling model to obtain a fire point-smoke space-time coupling index, and performing fire monitoring on a target region based on the fire point-smoke space-time coupling index to obtain a monitoring result. The application realizes accurate monitoring of the spreading situation of a fire.

Description

Technical Field

[0001] This invention belongs to the technical field of fire monitoring, and specifically relates to a fire monitoring method and device based on the coupling of thermal anomaly evolution and smoke diffusion by a high-orbit staring satellite. Background Technology

[0002] Fires are characterized by their sudden onset, rapid spread, and wide-ranging impact, posing a serious threat to the ecological environment and the safety of people's lives and property. Satellite remote sensing technology, due to its wide coverage and high observation frequency, has become an important means of fire monitoring. In the early stages of a fire, open flames are often not yet fully visible, but the smoke produced by combustion can be detected by satellite sensors. Therefore, smoke identification based on satellite remote sensing is a key technological approach for achieving early fire warning.

[0003] For smoke recognition in satellite remote sensing images, existing technologies mostly adopt deep learning methods based on single-temporal remote sensing images. These methods use convolutional neural networks to extract the spatial texture and spectral features of smoke from single-temporal remote sensing images, extract smoke-related features from the spatial texture and spectral features, and thus determine whether a fire has occurred based on the smoke. After confirming that a fire has occurred, the smoke area is segmented and extracted, thereby achieving accurate judgment and extraction of the smoke area.

[0004] However, in the aforementioned related technologies, due to insufficient utilization of long-term remote sensing observation information, low recognition accuracy in complex scenes, and susceptibility to interference from high-temperature background targets or clouds and fog when extracting fire smoke from satellite remote sensing images, the smoke extraction method based on single-temporal remote sensing influences cannot capture the dynamic evolution of smoke generation, diffusion, and dissipation, and thus cannot achieve accurate fire monitoring. Summary of the Invention

[0005] This application provides a fire monitoring method and apparatus based on the coupling of thermal anomaly evolution and smoke diffusion using a high-orbit staring satellite. This method addresses the problem of difficulty in accurately distinguishing between smoke and non-smoke interference in complex scenarios, thus hindering the effective capture of the dynamic evolution of smoke diffusion. This application achieves accurate monitoring of the fire spread situation.

[0006] On one hand, embodiments of this application provide a fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, the method comprising: Acquire a sequence of observation images of the target area, wherein the sequence of observation images includes multiple consecutive observation images; Based on the observed image sequence, the thermal anomaly growth index of each pixel in each observed image is calculated. The thermal anomaly growth index is used to characterize the thermal anomaly growth characteristics of the target area. Smoke anomaly features are extracted from the observed image sequence to obtain smoke anomaly features; The smoke diffusion dynamics model is used to analyze the abnormal characteristics of the smoke and obtain the smoke diffusion analysis results. The smoke diffusion analysis results include the smoke centroid trajectory, smoke diffusion velocity, smoke diffusion direction, smoke area and smoke area growth rate. The preset smoke diffusion dynamics model is used to analyze the dynamic change law in the smoke diffusion process. The thermal anomaly growth index and the smoke diffusion analysis results are input into a preset spatiotemporal coupling model. The preset spatiotemporal coupling model performs spatiotemporal coupling calculation on the thermal anomaly growth index and the smoke diffusion analysis results to obtain the fire point-smoke spatiotemporal coupling index. Fire monitoring is performed on the target area based on the fire point-smoke spatiotemporal coupling index, and monitoring results are obtained.

[0007] Optionally, calculating the thermal anomaly persistence growth index for each pixel in each of the observed image sequences includes: Perform thermal infrared radiation inversion on the observed images to calculate the brightness temperature value of each pixel in each observed image; For any pixel, a brightness temperature time series corresponding to each pixel is constructed based on the brightness temperature value of each pixel; Neighborhood enhancement analysis is performed on multiple pixels within a preset neighborhood window to obtain a local thermal anomaly coefficient, which is used to characterize the degree of thermal anomaly between the pixel and the background image. When the local thermal anomaly coefficient meets the preset thermal anomaly conditions, the brightness temperature change rate at adjacent moments is calculated based on the brightness temperature time series. The thermal anomaly growth index is obtained by summing the products of the brightness temperature change rate at adjacent time points and the preset time decay weight.

[0008] Optionally, neighborhood enhancement analysis is performed on multiple pixels within a preset neighborhood window to obtain local thermal anomaly coefficients, including: Each pixel is subjected to neighborhood thermal anomaly enhancement processing to obtain multiple target enhanced pixels; Based on each of the target augmented pixels, a neighborhood window is constructed, the neighborhood window including multiple target augmented pixels; Based on the multiple target enhancement pixels within the neighborhood window, calculate the mean and standard deviation of the neighborhood background brightness temperature; A local thermal anomaly coefficient is constructed based on the mean and standard deviation of the background brightness temperature in the neighborhood.

[0009] Optionally, after obtaining the thermal anomaly growth index by summing the products of the brightness temperature change rate at adjacent time points and a preset time decay weight, the method further includes: Obtain the number of brightness temperatures and the time interval based on the brightness temperature data in the brightness temperature time series; Based on the ratio of the brightness temperature quantity to the time interval, a thermal anomaly persistence index is determined, which is used to constrain the continuous growth characteristics of thermal anomalies. If both the thermal anomaly persistence index and the thermal anomaly growth index meet the fire point identification criteria, the target pixel is identified as the target candidate region.

[0010] Optionally, before calculating the thermal anomaly persistence growth index for each pixel in each of the observed images based on the observed image sequence, the method further includes: The observed image sequence is subjected to data correction to obtain a corrected observed image sequence; the data correction includes: radiometric correction, atmospheric correction and geometric correction. The corrected observation image sequence is divided to obtain the first phase image and the image to be registered. The image to be registered includes all other observation images in the corrected observation influence sequence except for the first phase image. Feature points are extracted from the first temporal image and the image to be registered to obtain a feature point set, which includes a multidimensional feature descriptor for each feature point. The multidimensional feature descriptors of the first temporal image and the multidimensional feature descriptors of the image to be registered are matched to obtain an initial matching set, which includes multiple point pairs with the highest matching degree. Based on multiple point pairs in the initial matching set, an affine transformation model is constructed; The image to be registered is mapped to a preset coordinate system according to the affine transformation model to obtain a spatially registered image; When the spatially registered images meet the temporal analysis conditions, a spatiotemporal registration observation image sequence is constructed based on the spatially registered images under a preset temporal sequence. The temporal analysis conditions include that the registration accuracy is within the preset registration accuracy range. Based on the observed image sequence, the thermal anomaly persistence growth index for each pixel in each observed image is calculated, including: Based on the spatiotemporal registration observation image sequence, the thermal anomaly growth index of each pixel in each observation image is calculated.

[0011] Optionally, the step of extracting smoke anomaly features from the observed image sequence to obtain smoke anomaly features includes: Based on the observed image sequence, the smoke spatial texture features and thermal anomaly correlation features of the smoke in each observed image are extracted; The smoke spatial texture features and the thermal anomaly correlation features are fused using dual-flow features to obtain dual-flow features; The dual-flow features are input into a preset temporal modeling model. The preset temporal modeling model performs temporal feature analysis on the dual-flow features to obtain the temporal feature sequence corresponding to the smoke. The temporal feature sequence includes historical and future information in the smoke diffusion process, which is used to characterize the diffusion trend, morphological changes and motion patterns of the smoke target in the observed image sequence. The temporal feature sequence is decoded at resolution using a preset decoding network to obtain the probability of each pixel in the observed image belonging to a smoke region; Based on the aforementioned anomaly probability, a smoke segmentation mask is determined; The temporal feature sequence and the smoke segmentation mask are identified as smoke anomaly features.

[0012] Optionally, the step of using a preset smoke diffusion dynamics model to analyze the abnormal characteristics of smoke and obtain the smoke diffusion analysis results includes: Anomaly detection is performed on the smoke segmentation mask to obtain anomaly detection results; When the anomaly detection result is determined to be without anomaly, the smoke centroid is extracted based on the smoke segmentation mask to obtain the smoke centroid trajectory, which includes the smoke centroid coordinates at any time. Based on the centroid coordinates of the smoke at any two adjacent moments, determine the smoke displacement distance, smoke diffusion velocity, and smoke diffusion direction; Obtain the total number of smoke pixels and the actual ground surface area at the target time; The first smoke area at the target time and the second smoke area at the previous time are determined by multiplying the total number of smoke pixels at the target time with the actual ground surface area. Based on the first smoke area and the second smoke area, determine the smoke area growth rate at the target time. The trajectory of the smoke centroid, the smoke diffusion velocity, the smoke diffusion direction, the smoke area, and the smoke area growth rate are determined as the results of the smoke diffusion analysis.

[0013] Optionally, before inputting the thermal anomaly growth index and the smoke diffusion analysis results into the preset spatiotemporal coupling model, the method further includes: Obtain concurrent meteorological wind field data for the target area; Spatial registration and temporal matching are performed on the meteorological wind field data to obtain a wind field dataset that is spatiotemporally aligned with the observed image sequence. The wind field dataset includes regional wind speed and regional wind direction. Based on the wind direction and smoke diffusion direction in the area, directional consistency detection is performed to obtain the directional detection result; Based on the wind speed and smoke diffusion rate in the area, a speed consistency detection is performed to obtain the speed detection result; Based on the direction detection results and the velocity detection results, a wind field consistency index is constructed; The wind field consistency index is continuously averaged over time to obtain the average wind field consistency index. Based on the average wind field consistency index and the preset wind field verification threshold, the wind field constraint results are determined; When the wind field constraint result is wind-constrained, the thermal anomaly growth index and smoke diffusion analysis results are input into a preset spatiotemporal coupling model, including: The wind field consistency index, the thermal anomaly growth index, and the smoke diffusion analysis results are input into a preset spatiotemporal coupling model.

[0014] Optionally, after determining the target pixel as a target candidate region, the method further includes: The brightness temperature time series of each pixel in the target candidate region is normalized and smoothed to generate the thermal waveform of each pixel; Extract the waveform morphology features of the thermal waveform and construct a thermal waveform feature vector; the waveform morphology features include at least one of the following: rising phase features, peak phase features, falling phase features, overall waveform features, and periodic and repetitive features; The thermal waveform feature vector is matched with a preset multi-type heat source waveform template library to identify the heat source type of the thermal anomaly candidate region; the waveform template library contains standard waveforms of at least two heat source types and corresponding waveform feature parameter ranges. Based on the identified heat source type, the target candidate areas are screened, and non-fire type thermal anomaly areas are excluded from subsequent calculations.

[0015] On the other hand, embodiments of this application provide a fire monitoring device based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, the device comprising: An acquisition module is used to acquire an observation image sequence of a target area, wherein the observation image sequence includes multiple consecutive observation images; The calculation module is used to calculate the thermal anomaly growth index of each pixel in each of the observed images based on the observed image sequence. The thermal anomaly growth index is used to characterize the thermal anomaly growth characteristics of the target area. The extraction module is used to extract smoke anomaly features from the observed image sequence to obtain smoke anomaly features; The analysis module is used to perform smoke diffusion analysis on the abnormal characteristics of the smoke using a preset smoke diffusion dynamics model, and obtain the smoke diffusion analysis results. The smoke diffusion analysis results include the smoke centroid trajectory, smoke diffusion velocity, smoke diffusion direction, smoke area, and smoke area growth rate. The preset smoke diffusion dynamics model is used to analyze the dynamic change law in the smoke diffusion process. The input module is used to input the thermal anomaly growth index and the smoke diffusion analysis results into a preset spatiotemporal coupling model. The preset spatiotemporal coupling model performs spatiotemporal coupling calculation on the thermal anomaly growth index and the smoke diffusion analysis results to obtain the fire point-smoke spatiotemporal coupling index. The monitoring module is used to monitor the target area for fire based on the fire point-smoke spatiotemporal coupling index and obtain monitoring results.

[0016] Compared with existing technologies, the fire monitoring method and apparatus based on the coupling of thermal anomaly evolution and smoke diffusion using high-orbit staring satellites in this application acquires an observation image sequence of the target area and determines the thermal anomaly growth index of each pixel in each observation image. Simultaneously, it extracts smoke anomaly features from the observation image sequence and uses a preset smoke diffusion dynamics model to perform diffusion analysis on the smoke anomaly features, obtaining multi-dimensional smoke diffusion analysis results including smoke centroid trajectory, diffusion velocity, diffusion direction, smoke area, and area growth rate. The thermal anomaly growth index and the smoke diffusion analysis results are then input into a preset spatiotemporal coupling model. Through joint discrimination of fire points and smoke, a fire point-smoke spatiotemporal coupling index is obtained, and fire monitoring of the target area is performed based on this index, outputting monitoring results. This enables rapid response in the early stages of a fire, effectively suppressing false alarms and missed alarms due to complex weather conditions such as cloudy skies and heavy fog, improving the accuracy of fire location and the reliability of monitoring results, and providing scientific and real-time data support for early fire warning, fire development trend assessment, and emergency dispatch decisions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the embodiments of this specification. For those skilled in the art, other drawings can be obtained based on these drawings. Figure 1 This is a schematic flowchart of a fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, according to one embodiment of this application. Figure 2 This is a structural block diagram of a fire monitoring device based on the coupling of thermal anomaly evolution and smoke diffusion of a high-orbit staring satellite, provided in another embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] To facilitate understanding of the embodiments of this application, further explanation and description will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of this application. In the drawings, the dimensions and relative dimensions of components may be exaggerated for clarity and / or descriptive purposes. When exemplary embodiments can be implemented differently, a specific process sequence may be performed in a different order than that described. For example, two consecutively described processes may be performed substantially simultaneously or in the reverse order of their description. Furthermore, the same reference numerals denote the same components.

[0020] The terminology used herein is for the purpose of describing particular embodiments and is not intended to be limiting. As used herein, unless the context clearly indicates otherwise, the singular forms “a” and “the” are intended to include the plural forms as well. Furthermore, when the terms “comprising” and / or “including” and variations thereof are used in this specification, it indicates the presence of the stated features, integrals, steps, operations, parts, components, and / or groups thereof, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, parts, components, and / or groups thereof. It should also be noted that, as used herein, the terms “substantially,” “about,” and other similar terms are used as approximate terms rather than as terms of degree, thus explaining the inherent biases in measurements, calculated values, and / or provided values ​​that would be recognized by one of ordinary skill in the art.

[0021] To address the problems of existing technologies, this application provides a fire monitoring method and apparatus based on the coupling of thermal anomaly evolution and smoke diffusion using a high-orbit staring satellite. Compared with existing technologies, this application acquires an observation image sequence of the target area and determines the thermal anomaly growth index of each pixel in each observation image. This thermal anomaly growth index enables accurate identification of fire points. Simultaneously, smoke anomaly features are extracted from the observation image sequence, and a preset smoke diffusion dynamics model is used to analyze the diffusion of these features, obtaining multi-dimensional smoke diffusion analysis results including smoke centroid trajectory, diffusion velocity, diffusion direction, smoke area, and area growth rate. The thermal anomaly growth index and the smoke diffusion analysis results are then input into a preset spatiotemporal coupling model. By jointly identifying fire points and smoke, accurate fire monitoring of the target area is achieved. This allows for rapid response in the early stages of a fire, effectively suppressing false alarms and missed alarms caused by complex weather conditions such as cloudy skies and heavy fog, improving the accuracy of fire location and the reliability of monitoring results, and providing scientific and real-time data support for early fire warning, fire development trend assessment, and emergency dispatch decisions.

[0022] The following section first introduces a fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, as provided in the embodiments of this application.

[0023] Figure 1 This illustration shows a flowchart of a fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion using a high-orbit staring satellite, according to an embodiment of this application. Figure 1 As shown, a fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite may include steps S101-S106: S101, acquire the observation image sequence of the target area.

[0024] Specifically, when acquiring observation image sequences, the Gaofen-4 satellite can be used to acquire observation image sequences for the target area. The target area is the monitoring coverage area of ​​the Gaofen-4 satellite, and the observation image sequence can include multiple consecutive observation images. For example, the observation images can include visible and near-infrared (VNIR) images (wavelength range covering (0.45~0.90μm), specifically including blue (0.45~0.52μm), green (0.52~0.60μm), red (0.63~0.69μm), and near-infrared (0.76~0.90μm)) with a total of four multispectral bands and one panchromatic band (0.45~0.90μm), with a spatial resolution of 50 meters, and IRS / mid-wave infrared (MWIR) images (wavelength range covering (3.50~4.10μm), with a spatial resolution of 400 meters).

[0025] S102, based on the observed image sequence, calculate the thermal anomaly growth index for each pixel in each observed image.

[0026] In this embodiment, the thermal anomaly growth index can be used to characterize the thermal anomaly growth characteristics of the target area.

[0027] When identifying fire points in target areas, the temporal growth characteristics of thermal anomalies can be characterized by analyzing the continuous temporal brightness and temperature variation patterns in the observed impact sequence. This allows for effective differentiation between real and false fire points, improving the accuracy of fire point identification. For example, the minute-level continuous observation capability of the Gaofen-4 satellite can be used to model the continuous growth process of thermal anomaly targets. Forest fires typically exhibit a continuous and intensifying thermal anomaly evolution in their early stages, while non-fire targets such as industrial heat sources and bare land warming show relatively stable or periodic fluctuations. HAGI (High-Intensity Gravity Anomaly Analysis) uses time-weighted cumulative calculation of continuous temporal brightness and temperature variation rates to quantitatively characterize the continuous growth characteristics of thermal anomalies, thereby improving early fire identification capabilities.

[0028] S103, extract smoke anomaly features from the observed image sequence to obtain smoke anomaly features.

[0029] In this embodiment, when extracting smoke, different smoke anomaly features can be extracted for different features. For example, for visible and near-infrared features in the observation image sequence, smoke spatial texture features can be extracted to extract the physical morphology of smoke. For mid-wave infrared features in the observation image sequence, thermal anomaly correlation features can be extracted to extract smoke attribute information.

[0030] Specifically, a neural network model can be used to extract abnormal features of smoke. For example, this neural network model can use a dual-stream Vision-LSTM smoke recognition network, which can include a visible infrared feature extraction branch, a mid-wave infrared feature extraction branch, a cross-modal feature fusion module, a Vision-LSTM temporal modeling module, and a smoke segmentation output module. Among them, the visible and near-infrared feature extraction branch extracts the spatial texture, edge, and morphological features of smoke; the mid-wave infrared feature extraction branch extracts thermal anomalies, thermal diffusion, and combustion zone radiation features; the cross-modal feature fusion module fuses dual-stream features using a cross-attention mechanism; the Vision-LSTM temporal modeling module learns the dynamic laws of smoke generation, diffusion, and dissipation; and the smoke segmentation output module decodes and generates a smoke probability map and a binary mask.

[0031] S104. Using a preset smoke diffusion dynamics model, smoke diffusion analysis is performed on the abnormal characteristics of smoke to obtain the smoke diffusion analysis results.

[0032] In this embodiment, the preset smoke diffusion dynamics model can be pre-trained. By learning the dynamic laws of smoke diffusion during existing fires, the model can be trained, enabling it to analyze abnormal smoke characteristics and the changes in the location and area of ​​the smoke region to construct dynamic parameters such as diffusion speed, direction, and intensity, thereby achieving dynamic analysis of the fire development trend.

[0033] S105, input the thermal anomaly growth index and smoke diffusion analysis results into the preset spatiotemporal coupling model. The preset spatiotemporal coupling model performs spatiotemporal coupling calculation on the thermal anomaly growth index and smoke diffusion analysis results to obtain the fire point-smoke spatiotemporal coupling index.

[0034] In this embodiment, the preset spatiotemporal coupling model can analyze the spatial distance and temporal differences between fire points and smoke to output a fire-smoke spatiotemporal coupling index. This index, through probability mapping, enables refined judgment of fire events. The fire-smoke spatiotemporal coupling index characterizes the correlation between fire points and smoke, allowing for accurate fire identification by considering both fire points and smoke. S106, Fire monitoring is performed on the target area based on the fire-smoke spatiotemporal coupling index to obtain monitoring results.

[0035] In this embodiment, the fire point-smoke spatiotemporal coupling index is mapped to a fire probability value, which represents the probability that the fire point and the smoke belong to the same fire event.

[0036] Specifically, a probability mapping function is used to map the fire-smoke spatiotemporal coupling index to a probability range of 0 to 1, realizing the conversion from a quantitative index to a probabilistic output. First, a decision threshold is set. When the fire probability value is greater than the decision threshold, it is determined that there is a coupling relationship between the fire point and the smoke, and a fire event is confirmed. When the fire probability value is less than the decision threshold, it is determined to be a non-fire target. Finally, fire event information is output, including the fire point location coordinates, the fire point occurrence time, the smoke coverage area, the smoke diffusion direction, and the fire probability value, realizing the fire identification result.

[0037] In this embodiment, by acquiring an observation image sequence of the target area, the thermal anomaly growth index of each pixel in each observation image is determined. This thermal anomaly growth index enables accurate identification of fire points. Simultaneously, smoke anomaly features are extracted from the observation image sequence, and a preset smoke diffusion dynamics model is used to analyze the diffusion of these features, yielding multi-dimensional smoke diffusion analysis results including smoke centroid trajectory, diffusion velocity, diffusion direction, smoke area, and area growth rate. The thermal anomaly growth index and the smoke diffusion analysis results are then input into a preset spatiotemporal coupling model. Through joint discrimination of fire points and smoke, accurate fire monitoring of the target area is achieved. This approach enables rapid response in the early stages of a fire, effectively suppressing false alarms and missed alarms caused by complex weather conditions such as cloudy skies and heavy fog, improving the accuracy of fire location and the reliability of monitoring results, and providing scientific and real-time data support for early fire warning, fire development trend assessment, and emergency dispatch decisions.

[0038] In some embodiments, in order to accurately monitor fires, it is also necessary to preprocess the observed image sequence; therefore, before S102, the method may further include: Data correction is performed on the observed image sequence to obtain the corrected observed image sequence; data correction includes: radiometric correction, atmospheric correction and geometric correction. The corrected observation image sequence is divided into the first phase image and the image to be registered. The image to be registered includes all other observation images in the corrected observation influence sequence except for the first phase image. Feature points are extracted from the first temporal image and the image to be registered to obtain a set of feature points, which includes a multi-dimensional feature descriptor for each feature point. The multidimensional feature descriptors of the first temporal image and the multidimensional feature descriptors of the image to be registered are matched to obtain an initial matching set, which includes multiple point pairs with the highest matching degree. Based on multiple point pairs in the initial matching set, an affine transformation model is constructed. The image to be registered is mapped to a preset coordinate system according to the affine transformation model to obtain a spatially registered image; Under the condition that the spatially registered images meet the temporal analysis conditions, a spatiotemporal registration observation image sequence is constructed based on the spatially registered images under a preset temporal sequence. The temporal analysis conditions include that the registration accuracy is within the preset registration accuracy range. S102 may include: Based on the spatiotemporal registration observation image sequence, the thermal anomaly continuous growth index of each pixel in each observation image is calculated.

[0039] In this embodiment, to eliminate sensor response differences, atmospheric influences, and geometric distortions, and to provide a unified data foundation for subsequent thermal anomaly analysis, smoke extraction, and fire identification, the acquired observation image sequence is sequentially subjected to radiometric correction, atmospheric correction, and geometric correction.

[0040] In radiometric correction, the digital quantization (DN) values ​​of the observed image sequence can be converted into apparent radiance to eliminate response differences between different sensors and different time phases. For example, the apparent radiance formula is calculated as follows:

[0041] in, This refers to apparent radiance, that is, the radiance received at the entrance pupil of the sensor. This represents the original grayscale value of the i-th band recorded by the sensor; This represents the gain of the i-th band, with units consistent with radiance. This represents the offset of the i-th band, with units consistent with radiance.

[0042] When performing atmospheric correction, the 6S model is used to perform atmospheric correction on the radiometrically corrected image to eliminate the influence of Rayleigh scattering, aerosol scattering, and water vapor absorption on surface information, and to obtain the surface reflectance. The formula for calculating surface reflectance is as follows:

[0043] in, Represents surface reflectance. This represents the functional relationship, i.e., the physical calculation algorithm of the 6S atmospheric radiative transfer model, where L represents apparent radiance. Indicates the zenith angle of the sun. Indicates the observed zenith angle.

[0044] It is worth noting that the 6S model is a conventional atmospheric correction model in this field, and will not be elaborated further here.

[0045] Because Gaofen-4 imagery suffers from geometric distortion due to attitude, orbit, and terrain, geometric correction can be performed by reading Rational Polynomial Coefficients (RPC) parameters and combining them with a Digital Elevation Model (DEM) to improve the positional consistency among multiple consecutive observation images. The RPC parameters include a normalization parameter (OFF / SCALE) and a mathematical mapping parameter set of 80 rational polynomial coefficients (40 in each row and column). This set is used to establish a high-precision geometric correspondence between the ground's three-dimensional coordinates (longitude, latitude, and elevation) and the image's two-dimensional pixel coordinates (rows and columns). This process involves establishing a pixel-to-geographic coordinate mapping, introducing a DEM to compensate for terrain elevation differences, interpolation and resampling, and generating an orthorectified image in a unified coordinate system.

[0046] As an example, when performing geometric correction, the RPC parameter file is read first. The parameters in Tables 1 and 2 below are read from the RPC parameter file attached to the Gaofen-4 satellite image: Table 1 Normalized parameters (OFF / SCALE)

[0047] Table 2 Rational polynomial coefficients

[0048] After reading the RPC parameters, an RPC mapping model between ground coordinates and image coordinates can be established, specifically: First, for any ground point coordinates (longitude X, latitude Y, elevation Z), normalize them to the interval [-1, 1]. The normalization calculation formula is as follows:

[0049]

[0050]

[0051] in, Indicates normalized longitude. Indicates normalized latitude; X represents the normalized elevation; Y represents the longitude coordinates of the ground point; Z represents the latitude coordinates of the ground point; Indicates the longitude offset; Indicates latitude offset; Indicates the elevation offset; Indicates the longitude scaling factor; Indicates the latitude scaling factor; This represents the normalized longitude based on the elevation scaling factor.

[0052] For rational polynomials with coefficients, each rational polynomial contains 20 terms, and the degree of all terms (the sum of their exponents) does not exceed 3. The specific expansion is as follows:

[0053] in, This represents a general cubic polynomial function used to characterize the mapping relationship between normalized ground coordinates and image coordinates. It is not a single function, but rather a common template for four independent polynomials (row numerator, row denominator, column numerator, and column denominator). This represents the normalized ground coordinates (longitude, latitude, and elevation), with a numerical range typically in the [-1, 1] interval, and serves as the input variable for the RPC polynomial. The first polynomial represents the second polynomial. The term coefficients are used to control the contribution of the corresponding term to the final mapping result; there are 20 coefficients in total, and the specific values ​​are determined by the satellite manufacturer through geometric calibration and read from the RPC file attached to the image. Corresponding constant term 1, It indicates the reference offset of the image center, reflecting the basic position of the image coordinates when the ground coordinates are zero; correspond , This indicates linear scaling along the longitude direction, reflecting the degree to which changes in ground longitude affect the row coordinates of the image. correspond , This indicates linear scaling in the latitudinal direction, reflecting the degree to which changes in ground latitude affect the row coordinates of the image; correspond , It represents the linear effect of elevation, reflecting the projection offset of the image row coordinates due to topographic relief; correspond , This indicates longitude-latitude coupling, reflecting the nonlinear effect of the joint changes in ground plane coordinates on row coordinates; correspond , This indicates longitude-elevation coupling, reflecting the differences in projection deformation caused by topographic relief at different longitude locations; correspond , This indicates latitude-elevation coupling, reflecting the differences in projection deformation caused by topographic relief at different latitude locations; correspond , This indicates a secondary curvature along the longitude direction, reflecting the optical distortion of the imaging system along the longitude direction; correspond , This indicates a secondary curvature in the latitudinal direction, reflecting the optical distortion of the imaging system along the latitudinal direction; correspond , It represents the secondary influence in the elevation direction, reflecting the nonlinear projection effect of topographic relief; correspond , It represents the three-order coupling of latitude, longitude, and altitude, reflecting the complex high-order deformation under the combined action of the three; correspond , This indicates third-order distortion in the longitude direction, reflecting more complex higher-order curvature in the longitude direction; correspond , This represents the asymmetric higher-order coupling of longitude and latitude; correspond , This indicates an asymmetric higher-order coupling between longitude and elevation; correspond , This represents the asymmetric higher-order coupling of latitude and longitude; correspond , This indicates third-order distortion in the latitudinal direction; correspond , This represents the asymmetric higher-order coupling of latitude and elevation; correspond , This indicates the coupling between longitude square and elevation. correspond , This indicates the coupling between latitude square and elevation. correspond , This indicates the third-order effect in the elevation direction.

[0054] In this context, each rational polynomial's coefficient polynomial (row numerator, row denominator, column numerator, column denominator) adopts the aforementioned unified form, differing only in the value of the coefficient multiplied by each term (i.e., ...). to (corresponding to the coefficients of each group respectively).

[0055] After constructing the cubic polynomial, normalized image coordinates can be calculated for any ground point coordinates, as shown in the following formula:

[0056]

[0057] in, Represents normalized row coordinates; Represents normalized column coordinates; Represents the row coordinate numerator polynomial; Represents the row coordinate denominator polynomial; Represents the column coordinate numerator polynomial; Represents the column coordinate denominator polynomial; Then, inverse normalization is performed to obtain the image pixel coordinates: Line= × +

[0058] Sample= × +

[0059] Where Line represents the image row coordinates; Sample represents the image column coordinates; Indicates the row coordinate scaling factor; Indicates the row coordinate offset; Indicates the column coordinate scaling factor; This indicates the column coordinate offset.

[0060] By first normalizing the influence coordinates and then performing inverse normalization, the mapping from ground coordinates to image coordinates is completed. .

[0061] Subsequently, in order to compensate for the geometric distortion caused by terrain undulations, it is necessary to provide accurate elevation values ​​for each ground location. For example, first load the DEM data covering the target area and obtain the elevation values ​​stored in the regular grid. Next, for the target geographic coordinates If the point is not on the DEM grid, use bilinear interpolation to obtain the accurate elevation value.

[0062] For example, set a target point If the target point falls between four grid points, then the elevation value of the target point is: Z = (1-u)(1-v) +u(1-v) +(1-u)v +uv

[0063] in, , The relative position of the target point within the grid cell (value range 0~1); , , , This represents the elevation values ​​of the four grid points.

[0064] Next, since the elevation values ​​in the RPC model affect the calculation results of image coordinates, an iterative method can be used to solve for the precise image point positions. The specific steps are as follows: For each pixel in the output orthophoto First, the elevation value of the point is obtained using the DEM. ;Will Substitute into the RPC model to calculate the corresponding image coordinates. .

[0065] It is worth noting that since the elevation provided by the DEM is the ground surface elevation, if the target point is the apex of a ground feature (such as a tree canopy or the top of a building), the ground elevation corresponding to the actual imaging location should be the ground surface elevation.

[0066] Finally, to ensure accuracy while preserving texture details, the target location can also be... Take the area around the target location Neighboring pixels (16 pixels) are used to determine the interpolation neighborhood;

[0067]

[0068] in, This represents the original image row coordinates calculated by the RPC model; This indicates rounding the Line down. This represents the column coordinates of the original image calculated by the RPC model; This indicates that Sample is rounded down.

[0069] Since the original image coordinates calculated by the RPC model are usually decimals when determining the interpolation neighborhood, the pixel value at that location cannot be directly obtained. Therefore, interpolation resampling is needed to calculate the grayscale value of the target point. For example, a cubic convolution kernel function can be used to calculate weights based on the distance from the target point to the center of the neighboring pixels: a weight of 1 when the distance is 0, a positive weight when the distance is within 1, a negative weight (for sharpening) when the distance is between 1 and 2, and a weight of zero when the distance exceeds 2. The weights in the row and column directions are calculated separately, and multiplied to obtain the final weight for each neighboring pixel. Then, the grayscale values ​​of the 16 neighboring pixels are weighted and summed to obtain the interpolation result for the target point.

[0070] After image resampling, the corrected and resampled observation images satisfy the following:

[0071] in, These are the original image pixel coordinates. These are the corresponding geographic coordinates.

[0072] Furthermore, due to the influence of factors such as slight changes in satellite attitude, orbital drift, Earth's rotation, and atmospheric refraction during long-term continuous observation, the images acquired at different times will have pixel-level positional shifts. In order to ensure the spatial consistency of the same ground target in continuous temporal images, it is necessary to perform temporal registration processing on the continuous observation images.

[0073] Specifically, the first step is to select reference images: For example, in a continuous observation sequence, the first temporal image is selected as the reference image. The formula for selecting the reference image is as follows:

[0074] The remaining temporal images are used as the images to be registered. The images to be registered are:

[0075] t=2,3,4, ,n Then, the SIFT algorithm can be used to extract feature points from the first temporal image and the image to be registered, resulting in a feature point set. This feature point set can include a multi-dimensional feature descriptor for each feature point. The specific steps are as follows: First, the scale space is established using the following formula:

[0076] in, Represents scale space; Represents the Gaussian function (smoothing kernel); This refers to the original image.

[0077] Then, based on the difference of Gaussians, key points with invariant scale are detected. The specific calculation formula is as follows:

[0078] in, This represents the difference of Gaussians (image difference result); This represents a Gaussian blurred image after scaling (i.e., the next layer is even more blurred). The table shows the spatial coordinates (two-dimensional planar positions) of pixels in the image; where... Typically corresponds to column coordinates (horizontal direction). Typically corresponds to row coordinates (vertical direction); Represents the "current scale" in scale space; It represents the constant multiplier factor between two adjacent scales.

[0079] Thus, the set of feature points is obtained:

[0080]

[0081] After feature point extraction, each feature point is represented as a 128-dimensional descriptor vector. Then, feature matching is performed to find the closest point pair as the initial match.

[0082] Specifically, the Euclidean distance between feature descriptors is calculated to find the nearest neighbor matching point. The formula for calculating the nearest neighbor matching point is as follows:

[0083] in, This represents the distance between the i-th feature point of the reference image and the j-th feature point of the image to be registered; This represents the k-th component of the feature descriptor for the i-th point of the reference image; This represents the k-th component of the feature descriptor for the j-th point of the image to be registered; This indicates that the SIFT descriptor is a 128-dimensional vector.

[0084] The multidimensional feature descriptors of the first temporal image and the multidimensional feature descriptors of the image to be registered are matched to obtain an initial matching set. The specific calculation formula for the initial matching set is as follows:

[0085] in, Indicates the initial matching set. Indicates the first in the reference image One feature point; Indicates the image to be registered with Matching the first There are 1 feature point; i represents the index of the matching point pair, with a value range of 1000. N, where N is the total number of matching pairs in the initial matching set.

[0086] Based on the initial matching set A transformation model is created by randomly selecting matching point pairs, and the reprojection error is calculated using the following formula:

[0087] in, This represents the reprojection error value of the i-th pair of matching points, in pixels; The column coordinates (horizontal direction) represent the actual location, in pixels; The row coordinates (vertical direction) representing the actual position, in pixels; The column coordinates (horizontal direction) represent the predicted location, in pixels; The row coordinates (vertical direction) of the predicted position are represented in pixels.

[0088] Based on the reprojection error, interior point determination is performed. If the reprojection error is less than the threshold ε, If the match is found to be an interior point, then the pair is considered an interior point; otherwise, it is considered an exterior point. The pair with the highest matching degree is obtained. The formula for determining an interior point is as follows:

[0089] in, This represents the pixel distance between the predicted location and the actual location. This is a preset threshold.

[0090] Subsequently, an affine transformation model is established based on the point pairs with the highest matching degree. The mathematical formula for calculating the affine transformation model is as follows:

[0091] in, , where is the original point coordinate on the image before transformation. Indicates the coordinates of the corresponding point on the transformed image; The scaling and rotation coupling coefficients representing direction; This represents the shear and rotation coupling coefficient with respect to the x-axis. This represents the shear and rotation coupling coefficient with respect to the direction y; The scaling and rotation coupling coefficients representing direction; Indicates the amount of translation in the horizontal direction; Indicates the amount of translation in the direction (vertical direction).

[0092] To smooth noise while preserving texture details and obtain a spatially registered image, sampling can be performed using an affine transformation model. Cubic convolution interpolation (16-neighborhood weighted) can be used to transform the image to be registered to the reference coordinate system. The formula for calculating the spatially registered image is as follows:

[0093] in, This represents the interpolation result (target pixel value) to be calculated. Represents the transformed non-integer coordinates; This represents an offset from -1 to 2, covering 4×4=16 neighboring pixels around the target point; Indicates the interpolation weights; The grayscale value of neighboring pixels in the original image.

[0094] Subsequently, in this embodiment, the registration accuracy can be measured by the root mean square error (RMSE) to determine the coordinate deviation between the registered image and the reference image on highly reliable matching point pairs. An RMSE < 0.5 pixels is considered acceptable, indicating that the spatial consistency requirement of the observed image sequence is met. The formula for calculating the registration accuracy evaluation is as follows:

[0095] Where RMSE represents the root mean square error, in pixels; N represents the total number of highly reliable matching point pairs (inliers). This represents the deviation of the i-th pair of matching points in the x-direction, in pixels; This represents the deviation of the i-th pair of matching points in the y-direction, in pixels.

[0096] Finally, assuming the spatially registered images meet the temporal analysis conditions, and under a preset temporal sequence, a spatiotemporal registration observation image sequence is constructed based on the spatially registered images to form a spatiotemporal registration observation image sequence under a unified spatial reference. The mathematical formula for constructing the spatiotemporal registration observation image sequence is as follows:

[0097] in, This represents a sequence of spatiotemporal registration observation images; This represents the first image in the sequence; This represents the nth image in the sequence.

[0098] Finally, after completing all temporal registrations, ensure that any pixel (x, y) corresponds to the same geographical location at different times.

[0099] It is worth noting that since radiometric correction relies on the original physical values ​​of pixels, it must be completed before resampling changes the pixel position; geometric correction is to establish a geographic reference and is a prerequisite for temporal registration. The three constitute a progressive chain of "physical → geometric → temporal", and the order cannot be reversed. Therefore, the above-mentioned radiometric correction, atmospheric correction, geometric correction and temporal registration are performed in sequence.

[0100] Based on the observation image sequence that has undergone radiometric correction, atmospheric correction, geometric correction and temporal registration in sequence, the thermal anomaly growth index of each pixel in each observation image is calculated to improve the accuracy of subsequent fire monitoring.

[0101] In some other embodiments, S102 may include: Perform thermal infrared radiation inversion on the observed images to calculate the brightness temperature value of each pixel in each observed image; For any pixel, a brightness temperature time series corresponding to each pixel is constructed based on the brightness temperature value of each pixel; Neighborhood enhancement analysis is performed on multiple pixels within a preset neighborhood window to obtain local thermal anomaly coefficients, which are used to characterize the degree of thermal anomaly between the pixel and the background image. Under the condition that the local thermal anomaly coefficient meets the preset thermal anomaly conditions, the brightness temperature change rate between adjacent time moments is calculated based on the brightness temperature time series. The thermal anomaly growth index is obtained by summing the products of the brightness temperature change rate at adjacent time points and the preset time decay weight.

[0102] In this embodiment, in order to improve the early fire identification capability, characterize the time growth characteristics of thermal anomalies, effectively distinguish between real fire points and false fire points, and improve the fire point identification accuracy, the continuous growth characteristics of thermal anomalies are quantitatively characterized by time-weighted cumulative calculation of the brightness temperature change rate over continuous time phases.

[0103] Specifically, the brightness temperature value of each pixel in each observation image is calculated using the thermal infrared radiation inversion method. The formula for calculating the brightness temperature value of each pixel in each observation image is as follows:

[0104] in, Indicates brightness temperature, measured in Kelvin (K). It is the second radiation constant; Wave number, unit: cm -1 ; The spectral radiance is obtained after radiometric calibration; is the first radiation constant.

[0105] Subsequently, for any pixel, a brightness temperature time series is constructed based on the brightness temperature value of each pixel. The formula for calculating the brightness temperature time series is as follows:

[0106] Where T represents the brightness temperature time series; Indicates the brightness temperature at moment 1; Indicates the brightness temperature at time 2; Indicates the brightness temperature at time n; This indicates the number of consecutive observation phases.

[0107] Next, to further reduce noise interference and suppress isolated false detections, a 3×3 or 5×5 preset neighborhood window is established centered on the target pixel. To weaken the influence of background temperature fluctuations, neighborhood enhancement analysis is performed on multiple pixels within the preset neighborhood window to obtain local thermal anomaly coefficients, which may include: Each pixel is subjected to neighborhood thermal anomaly enhancement processing to obtain multiple target enhanced pixels; For each target augmented cell, a neighborhood window is constructed, which includes multiple target augmented cells; Based on multiple target enhancement pixels within the neighborhood window, the mean and standard deviation of the neighborhood background brightness temperature are calculated. A local thermal anomaly coefficient is constructed based on the mean and standard deviation of the domain background brightness temperature.

[0108] In this embodiment, by performing neighborhood enhancement analysis, the influence of background stability fluctuations is weakened, and the characteristics of local anomalous heat sources are highlighted. Neighborhood enhancement analysis is performed on multiple pixels within a preset neighborhood window to construct a local thermal anomaly coefficient. The calculation formula for the local thermal anomaly coefficient is as follows:

[0109] Where L is the local thermal anomaly coefficient, which is dimensionless and represents the degree of anomaly of the target pixel relative to the background; T is the brightness temperature of the target pixel, in K. The average brightness temperature of all background pixels within the preset neighborhood window, excluding the target pixel, is expressed in K. This is the standard deviation of brightness temperature for all background pixels within the preset neighborhood window, excluding the target pixel, expressed in K.

[0110] when When the value is large, it indicates that the brightness temperature of the target pixel is significantly higher than that of the local background, belonging to an isolated thermal anomaly; when... When the value is close to 0 or negative, it indicates that the target pixel has no significant difference from the background or is a cold target.

[0111] If the standard deviation of the brightness temperature of the neighboring background pixels σ = 0 (the background is completely uniform), then let =0 indicates no significant spatial anomalies.

[0112] To further improve the credibility of the fire point candidate region, a set threshold is introduced. ,when Greater than the set threshold When this occurs, it indicates that the candidate area for the fire point is highly likely, depending on the actual application scenario. A value of 2.0 to 3.0 is recommended.

[0113] To reflect the trend of temperature change, the rate of change of brightness temperature between adjacent moments can be calculated based on the brightness temperature time series. The mathematical formula for calculating the rate of change of brightness temperature between adjacent moments is as follows:

[0114] in, This represents the rate of increase of thermal anomaly at time i, in units of K / min; Indicates the time interval between adjacent observations, in minutes; when When >0, it indicates that the target is in a heating state; when When the value is less than 0, it indicates that the target is in a cooling state. This represents the brightness temperature value at time i. This represents the brightness temperature value at time i-1.

[0115] Considering that data closer to the current moment contributes more to fire identification, a time decay weight is introduced. The formula for calculating the time decay weight is as follows:

[0116] in, λ represents the time weight at time i, which is dimensionless; λ represents the time decay coefficient, controlling the decay rate of the weight over time. The value of λ ranges from 0.1 to 0.3. Within this range, the weight decay of historical data is moderate, which can highlight recent thermal anomaly changes while retaining sufficient historical trend information for continuous analysis; n represents the total number of consecutive observation phases; represents the phase sequence number; where, For the earliest time, This refers to the current moment.

[0117] By introducing a time decay weight, the influence of recent observation data in fire point determination can be increased, making the thermal anomaly persistence growth index more sensitive to the current combustion state. The thermal anomaly persistence growth index is constructed by weighting and accumulating the continuous temporal thermal anomaly growth rates. The formula for calculating the thermal anomaly persistence growth index is as follows:

[0118] in, This represents the thermal anomaly growth index, expressed in K / min. This represents the time weight at time i. This represents the growth rate of the thermal anomaly at time i. This represents the total number of consecutive observation phases; Indicates the phase number, starting from 2 and ending at n.

[0119] In some other embodiments, after S102, the method may further include: Obtain the number of brightness temperatures and the time interval based on brightness temperature data in a brightness temperature time series; Based on the ratio of brightness temperature quantity to time interval, a thermal anomaly persistence index is determined, which is used to constrain the continuous growth characteristics of thermal anomalies. If both the thermal anomaly persistence index and the thermal anomaly growth index meet the fire point identification criteria, the target pixel is identified as the target candidate region.

[0120] In this embodiment, to avoid misjudgment of stable heat sources such as industrial heat sources and building heat islands, a thermal anomaly persistence index P is introduced to constrain the continuous growth characteristics of thermal anomalies. Only when both the thermal anomaly persistence index HAGI and the thermal anomaly persistence index P meet the conditions is it determined to be a candidate area for fire point.

[0121] Specifically, based on the number and time interval of brightness temperatures in the brightness temperature time series, a thermal anomaly persistence index is calculated. The mathematical expression for the thermal anomaly persistence index is:

[0122] in, This indicates a persistent indicator, with a value ranging from 0 to 1; This indicates the number of time intervals during which the brightness temperature continuously increases during continuous observation. This indicates that a total of n brightness temperatures are obtained from continuous observations, with M = n-1 observation intervals between adjacent time phases.

[0123] It is worth noting that when A higher value indicates a more pronounced and sustained warming characteristic in the target area. For real fire points, the brightness temperature tends to rise over most observation intervals due to the continuous energy release during combustion, resulting in a typically high value. However, for transient thermal anomalies (such as solar reflection or brief industrial thermal radiation), the warming occurs only in a few phases. The value is low. Preferably, when When the value is greater than 0.7, the target is considered to have a continuous thermal anomaly index, which can be used as one of the auxiliary conditions for fire point determination.

[0124] To further determine candidate fire points, when the HAGI is greater than a preset threshold... And the persistence index P is greater than the preset threshold. When the target pixel is identified as a target candidate region, the target pixel is determined.

[0125] in, The threshold for the continuous growth index of thermal anomalies; This is the threshold for determining continuity.

[0126] In other embodiments, to further determine whether a fire has occurred in the target candidate area and achieve precise location of the fire event, the method may further include, after identifying the target pixel as the target candidate area: The brightness temperature time series of each pixel in the target candidate region is normalized and smoothed to generate the thermal waveform of each pixel. Extract the waveform morphology features of the thermal waveform and construct a thermal waveform feature vector; the waveform morphology features include at least one of the following: rising phase features, peak phase features, falling phase features, overall waveform features, and periodic and repetitive features; The thermal waveform feature vector is matched with a preset multi-type heat source waveform template library to identify the heat source type of the thermal anomaly candidate region; the waveform template library contains standard waveforms of at least two heat source types and the corresponding waveform feature parameter ranges. Based on the identified heat source type, the target candidate areas are screened, and non-fire type thermal anomaly areas are excluded from subsequent calculations.

[0127] In this embodiment, firstly, for each pixel in the extracted target candidate region, the brightness temperature time series of that pixel is obtained. Since the absolute values ​​of brightness temperature vary greatly among different heat source types, in order to eliminate the influence of dimensions and highlight the waveform morphology, the brightness temperature time series of each pixel is normalized. Specifically, the maximum and minimum brightness temperature values ​​of that pixel during the entire observation period are used as upper and lower boundaries, and the brightness temperature value at each moment is mapped to a unified numerical range of zero to one. After normalization, a curve smoothing method is used to process the time series to suppress high-frequency fluctuations caused by atmospheric disturbances and sensor noise, and to completely retain the main trend of brightness temperature evolution over time while removing noise interference. The brightness temperature time series after the above normalization and smoothing processing, together with the key stage nodes such as the starting point where the brightness temperature begins to rise continuously, the peak point where the highest value is reached, and the ending point where it falls back to the reference level, together constitute the thermal waveform of that pixel.

[0128] Next, multi-dimensional waveform morphology features capable of quantifying the morphological characteristics of each pixel's thermal waveform are extracted. These features include at least one of the following: rising phase features, peak phase features, falling phase features, overall waveform features, and periodicity and repetition features. The rising phase features include at least the rising slope, rising duration, and rising acceleration. The rising slope is the difference between the peak brightness temperature and the initial brightness temperature divided by the rising duration. The rising duration is the time from the initial moment to the peak moment, and the rising acceleration is the rate of change of the temperature rise during the rising phase. The peak phase features include at least the peak duration and peak flatness. The peak duration is the continuous duration for which the brightness temperature remains near the peak, and the peak flatness is the degree of fluctuation in the brightness temperature during this period. The falling phase features include at least... The features include: descent slope, descent duration, and descent symmetry. The descent slope is the difference between the termination brightness temperature and the peak brightness temperature divided by the descent duration. The descent duration is the time from the peak moment to the termination moment. The descent symmetry is the ratio of the absolute value of the rising slope to the absolute value of the falling slope. The overall waveform features include at least waveform width, waveform area, and waveform skewness. The waveform width is the total duration from the start moment to the termination moment. The waveform area is the integral area between the waveform curve and the baseline. The waveform skewness is a measure of the overall asymmetry of the waveform. The periodicity and repetition features include at least fluctuation period and fluctuation amplitude. The fluctuation period is detected by autocorrelation analysis of the waveform. The fluctuation amplitude is the peak-to-valley difference of the periodic fluctuation. The above features are combined to construct the thermal waveform feature vector of the pixel.

[0129] Subsequently, a multi-type heat source waveform template library is pre-constructed. This library contains standard waveforms corresponding to at least two or more heat source types, including forest fires, industrial heat sources, coalfield spontaneous combustion, solar reflection interference, volcanic thermal anomalies, and oil and gas field flares, along with the allowable value ranges of their waveform characteristic parameters. Each template is formed through statistical summarization of a large amount of historical heat waveform data of known types, and includes a template confidence weight reflecting its reliability. The current pixel's heat waveform feature vector is input into the template library and compared one by one with each type of template using a matching method. The specific matching process can adopt a two-level strategy from coarse to fine. The first level is coarse matching, which matches the heat waveform feature vectors... Each feature parameter in the eigenvector is compared with the allowable value range of the corresponding feature parameter in each template. The proportion of parameters falling within the allowable range is counted to quickly eliminate template types that are obviously mismatched. The second level is fine matching, which calculates the curve shape similarity between the thermal waveform to be identified and the standard waveform of the candidate templates selected by coarse matching. The similarity index, which measures the distance or difference between the two curves, is used for evaluation. The comprehensive matching score between the waveform to be identified and each candidate template is obtained by combining the matching ratio of coarse matching and the similarity index of fine matching. The heat source type corresponding to the template with the highest comprehensive matching score is determined as the heat source type of the pixel or the thermal anomaly area.

[0130] Finally, based on the heat source type of each candidate thermal anomaly identified by waveform matching, the regions are classified and processed. When the heat source type of a region is determined to be a fire type such as forest fire or grassland fire, the region is retained as a valid target and the subsequent calculation of the thermal anomaly growth index and smoke diffusion analysis are continued. When the heat source type of a region is determined to be a non-fire type such as industrial heat source, coalfield spontaneous combustion, solar reflection interference, volcanic thermal anomaly, or oil and gas field flare, the region is removed from the candidate thermal anomaly regions and will no longer participate in any subsequent calculation steps related to fire monitoring. For candidate regions with low waveform matching scores, inability to determine the specific heat source type, or no matching with any known template, they are neither directly determined as fire types nor directly excluded. Instead, they are marked as pending and retained for subsequent processing, and further confirmed through a subsequent multi-level collaborative judgment mechanism. At the same time, when the number of regions marked as pending exceeds a preset threshold and their waveform characteristics show a consistent pattern, the template library update process is triggered. After clustering the consistent samples, new heat source type templates are created and added to the template library, enabling the template library to have self-updating capabilities as the application scenario expands.

[0131] In some other embodiments, S103 may include: Based on the observation image sequence, the spatial texture features of smoke and the thermal anomaly correlation features of smoke in each observation image are extracted; Dual-flow features are obtained by fusing the spatial texture features of smoke and the correlation features of thermal anomalies; The dual-flow features are input into the preset temporal modeling model. The preset temporal modeling model performs temporal feature analysis on the dual-flow features to obtain the temporal feature sequence corresponding to the smoke. The temporal feature sequence includes historical and future information in the smoke diffusion process, which is used to characterize the diffusion trend, morphological changes and motion patterns of the smoke target in the observed image sequence. A preset decoding network is used to perform resolution decoding on the temporal feature sequence to obtain the probability of each pixel in the observed image belonging to the smoke region; Based on the aforementioned anomaly probability, a smoke segmentation mask is determined; The temporal feature sequence and the smoke segmentation mask are used to identify the smoke anomaly features.

[0132] In this embodiment, in order to effectively model the continuous evolution of smoke, adapt to efficient calculation with a small number of parameters, and stably extract temporal features such as area and direction, data extraction is first performed on the observed image sequence to obtain near-infrared data and mid-wave infrared data. Then, the spatial texture features of smoke are extracted from the near-infrared data, and the thermal anomaly correlation features are extracted from the mid-wave infrared data. The spatial texture features of smoke may include spatial texture features, edge features and morphological features, and the thermal anomaly correlation features may include thermal anomaly features, thermal diffusion features and combustion zone radiation features.

[0133] Let the sequence of continuously observed images be represented as:

[0134] In represents the observed image at time i (including visible, near-infrared, and mid-infrared channels); n represents the number of consecutive observation phases.

[0135] Specifically, when extracting spatial texture features of smoke, the visible and near-infrared images acquired by the Gaofen-4 PMS sensor are input into a convolutional encoder for spatial feature extraction.

[0136] Let the input image be After processing by the convolutional encoder, the following is obtained:

[0137] in, This represents the extracted visible and near-infrared features; This indicates a visible and near-infrared encoder; This branch mainly extracts smoke edge features, smoke spatial texture features, smoke morphology features, and spectral reflectance features.

[0138] When extracting the correlation features of smoke thermal anomalies, the mid-wave infrared image acquired by the Gaofen-4 IRS sensor is input into the infrared encoder to extract thermal anomaly related information.

[0139] Let the input image be After passing through the convolutional encoder After processing, we get:

[0140] in, Indicates mid-infrared characteristics, This indicates a mid-wave infrared encoder.

[0141] This branch mainly extracts features of thermal anomaly regions, thermal diffusion features, temperature gradient features, and thermal radiation features of combustion regions.

[0142] To establish feature associations between the two modalities, a cross-attention mechanism was used to associate smoke spatial texture features and thermal anomalies to obtain dual-stream features. The specific fusion steps are as follows: First, using visible and near-infrared features For the query matrix, using mid-wave infrared features The key matrix and value matrix are: ; ;

[0143] in, Represents the query matrix; Represents the key matrix; Represents a value matrix; , , This represents a learnable linear projection matrix. Represents the query matrix (Query); This is the key matrix.

[0144] Next, calculate the cross-attention weights:

[0145] in, d is the normalized exponential function, which normalizes the attention weights to a probability distribution; d is the feature dimension (scaling factor, used to stabilize the gradient).

[0146] Finally, the fused features are generated: =Conv

[0147] in, Indicates the final two-stream feature; Conv 1×1 convolutions are used for dimensionality reduction and fusion; This indicates feature splicing along the channel dimension.

[0148] Through the cross-attention mechanism, the model can automatically learn the correspondence between "spatial texture features of smoke" and "thermal anomaly features", making smoke recognition pay more attention to areas related to heat sources, thereby effectively suppressing interference from cold clouds, fog and other factors.

[0149] Furthermore, in order to make full use of the long-term continuous observation image sequence of Gaofen-4, a Vision-LSTM structure was introduced to perform time-series modeling of the dual-flow characteristics of continuous time phases in order to learn the temporal evolution law of the smoke diffusion process.

[0150] Suppose the continuously acquired spatial feature sequence of smoke is:

[0151] in, Indicates the first Smoke spatial features extracted at any time; This indicates the number of phases during continuous observation.

[0152] During modeling, the spatial feature sequence of the smoke is first input into the feedforward Vision-LSTM network, and the state is updated in chronological order to obtain the feedforward hidden state:

[0153] in, Indicates the first Always in a forward-facing, hidden state; This indicates the hidden state at the previous moment.

[0154] It is worth noting that the forward Vision-LSTM is mainly used to learn the time dependencies from the past to the present in the process of smoke diffusion.

[0155] Meanwhile, the same smoke spatial feature sequence is input into the backward Vision-LSTM network, and the state is updated in reverse time order to obtain the backward hidden state:

[0156] in, Indicates the first The state is hidden backwards at any given moment; This indicates the hidden state at the next moment.

[0157] It is worth noting that the backward Vision-LSTM is used to learn the temporal information fed back from future moments to the current moment during the smoke diffusion process.

[0158] Next, feature fusion is performed on the forward hidden state and the backward hidden state to obtain the th... Bidirectional temporal characteristics at any given moment:

[0159] in, This represents the temporal characteristics after fusion; This indicates a feature splicing operation.

[0160] The final complete time-series feature sequence is obtained:

[0161] in, ; This time-series feature sequence contains both historical and future information about the smoke diffusion process, and can effectively characterize the diffusion trend, morphological changes and motion patterns of smoke targets in continuous observation phases.

[0162] Subsequently, the timing features output by the bidirectional LSTM are... Input the signal into the decoding network, gradually restore the spatial resolution, and generate the probability of each pixel belonging to the smoke region.

[0163]

[0164] in, The graph represents the probability of a pixel belonging to the smoke area; Decoder(⋅) is the decoding network, which can use transposed convolution, upsampling + convolution, or U-Net decoding structure.

[0165] Finally, a segmentation threshold is set based on the anomaly probability of the smoke area. When the probability is greater than the threshold, it is determined to be a smoke pixel:

[0166] in: For a binary smoke mask (1 represents smoke, 0 represents no smoke), τ s This is a preset threshold; the recommended value range is 0.3~0.7, which can be adjusted according to the application scenario.

[0167] In some other embodiments, S104 may include: Anomaly detection is performed on the smoke segmentation mask to obtain anomaly detection results; If the anomaly detection result indicates that there is no anomaly, the smoke centroid is extracted based on the smoke segmentation mask to obtain the smoke centroid trajectory, which includes the smoke centroid coordinates at any time. Based on the centroid coordinates of the smoke at any two adjacent moments, determine the smoke displacement distance, smoke diffusion velocity, and smoke diffusion direction; Obtain the total number of smoke pixels and the actual ground surface area at the target time; The first smoke area at the target time and the second smoke area at the previous time are determined by multiplying the total number of smoke pixels at the target time with the actual ground surface area. Based on the first smoke area and the second smoke area, determine the smoke area growth rate at the target time. The trajectory of the smoke centroid, the smoke diffusion velocity, the smoke diffusion direction, the smoke area, and the smoke area growth rate are determined as the results of the smoke diffusion analysis.

[0168] In this embodiment, by analyzing the changes in the location and area of ​​the smoke region, dynamic parameters such as diffusion speed, direction and intensity are constructed, enabling dynamic analysis of the fire development trend. The output smoke mask will be used as input data for a subsequent preset smoke diffusion dynamics model to predict fire spread trends. Misclassification (e.g., thick clouds being mistaken for smoke) can cause abnormal jumps in subsequent calculations, leading to instability in the dynamic model. Therefore, it is necessary to first perform median filtering anomaly monitoring on the smoke segmentation mask, that is, to perform median filtering on the centroid positions of continuous time series: ) in, Represents the centroid coordinates at time t. Represent the median function; This indicates the half-width of the window.

[0169] It is worth noting that if the anomaly detection result is normal, proceed to the next step; if consecutive anomalies occur, an error message needs to be generated and sent to the terminal held by the administrator to remind the administrator.

[0170] Subsequently, a continuous temporal smoke mask sequence was obtained using a dual-stream Vision-LSTM smoke recognition module:

[0171] in, This represents a continuous temporal sequence of smoke masks; Indicates the first The smoke masking results at time (binary image, 1 represents smoke, 0 represents no smoke); This represents the total number of consecutive observation phases.

[0172] For the The smoke area at any given time, let it contain The smoke resembles a 1, 2 The spatial coordinates of each smoke pixel are: The centroid coordinates of the smoke region are calculated as follows:

[0173]

[0174] in, express The x-coordinate (pixel coordinate) of the smoke centroid at time 1; Indicates the first The vertical coordinate (pixel coordinate) of the centroid of the smoke at any given moment; Indicates the first The total number of pixels contained in the smoke area at any given time.

[0175] Thus, the trajectory of the smoke's center of mass is obtained:

[0176] in, Indicates the trajectory of the smoke's center of mass; express The x-coordinate (pixel coordinate) of the smoke centroid at time 1; Indicates the first The vertical coordinate (pixel coordinate) of the centroid of the smoke at any given moment.

[0177] The trajectory of the smoke's center of mass can intuitively reflect the overall migration process and movement pattern of the smoke.

[0178] Furthermore, for two consecutive observation times and The coordinates of the centroid of the smoke can be calculated as follows:

[0179]

[0180] in, Indicates time Coordinates of the smoke's centroid; Indicates time Coordinates of the centroid of the smoke.

[0181] The displacement of the smoke's center of mass can then be calculated as follows:

[0182] in, Indicates the first The displacement distance of the smoke's center of mass at any given time; the displacement is used to quantitatively describe the degree of overall movement of the smoke, and the larger the value, the more significant the change in the spatial position of the smoke per unit time.

[0183] Let the time interval between two consecutive observations be . The smoke diffusion rate is calculated as follows:

[0184] in: Indicates the first The rate at which smoke diffuses at any given moment; Indicates the distance the smoke has moved; This indicates the time interval between adjacent observations.

[0185] Calculations of smoke diffusion speed show that the greater the smoke diffusion speed, the faster the smoke spreads, and the more significant the risk of fire spread and the trend of fire spread.

[0186] Finally, the direction of smoke diffusion can be calculated based on the changes in the centroid coordinates of the smoke over continuous time.

[0187]

[0188] in, Indicates the first The direction of smoke diffusion at any given time; arctan(⋅) is the arctangent function, with a range of values ​​of ( , ).

[0189] The direction of smoke diffusion can be used to further determine the main direction of smoke diffusion, which can be used to analyze the fire's spread trend and potential areas of spread.

[0190] Furthermore, let the first... The smoke area contains Each smoke pixel corresponds to an actual area of ​​[missing information]. The smoke coverage area The calculation is as follows:

[0191] in, This indicates the area covered by smoke at time i.e., the area covered by smoke at time j. Indicates the first The total number of smoke particles at any given moment; It represents the actual surface area of ​​a single pixel.

[0192] A sequence of smoke area changes can be obtained through continuous observation:

[0193] in, This represents a sequence of changes in smoke area. This represents the total number of consecutive observation phases.

[0194] The area change sequence can intuitively reflect the trend of smoke diffusion expanding or shrinking.

[0195] Next, after obtaining the smoke area change sequence, it is necessary to calculate the smoke area growth rate. The smoke area growth rate is used to quantitatively reflect the intensity and rate of smoke diffusion. The formula for calculating the smoke area growth rate is as follows:

[0196] in, Indicates the first The rate of increase in smoke area at any given time; Indicates the current smoke area; This indicates the area of ​​the smoke at the previous moment; This indicates the time interval between adjacent observations.

[0197] It is worth noting that when The area covered by smoke continued to expand, and the fire was in the development stage. The smoke area is stable, and the fire is in a stable combustion phase. As the smoke area gradually decreases, the fire is in the decay or extinguishing stage.

[0198] Subsequently, in order to comprehensively characterize the dynamic process of smoke diffusion, a smoke diffusion feature vector is constructed. The mathematical formula for calculating the smoke diffusion feature vector is:

[0199] in, This represents the characteristic vector of smoke diffusion.

[0200] This feature vector can simultaneously characterize the motion state (velocity and direction) and diffusion intensity (area change rate) of smoke, providing multidimensional information for the comprehensive judgment of subsequent fire development trends.

[0201] Finally, to further quantify and assess the severity of smoke diffusion, the smoke centroid trajectory, smoke diffusion velocity, smoke diffusion direction, smoke area, and smoke area growth rate are defined as the results of the smoke diffusion analysis. The mathematical formula for calculating the smoke diffusion analysis results is as follows:

[0202] in, Indicates the first The smoke diffusion intensity index at any given time; This represents the normalized smoke diffusion rate; This represents the normalized growth rate of the smoke area. This is the speed weighting coefficient; Let be the area growth weighting coefficient, satisfying α + β = 1. Since... and They may have different dimensions, so Min-Max normalization is required before calculation to make them comparable.

[0203] It is worth noting that the optimal parameters are α=0.6 and β=0.4. The smoke diffusion intensity index allows for the quantitative classification of smoke diffusion intensity, providing an intuitive and quantifiable basis for fire emergency response decisions.

[0204] In other embodiments, fire smoke is continuously driven by the regional wind field during its diffusion process, and its diffusion direction and velocity are typically highly correlated with the concurrent meteorological wind field parameters. Therefore, after obtaining the smoke diffusion dynamics parameters, the smoke diffusion results can be verified in conjunction with meteorological wind field data to improve the reliability of fire smoke identification results. Therefore, before S105, the method may further include: Obtain concurrent meteorological wind field data for the target area; Spatial registration and temporal matching of meteorological wind field data are performed to obtain a wind field dataset that is spatiotemporally aligned with the observed image sequence. The wind field dataset includes regional wind speed and regional wind direction. Directional consistency detection is performed based on regional wind direction and smoke diffusion direction to obtain directional detection results; Velocity consistency detection is performed based on regional wind speed and smoke diffusion rate to obtain velocity detection results; Based on the direction detection results and velocity detection results, a wind field consistency index is constructed; The wind field consistency index is continuously averaged over time to obtain the average wind field consistency index. The wind field constraint results are determined based on the average wind field consistency index and the preset wind field verification threshold. When the wind field constraint results are wind-constrained, the thermal anomaly growth index and smoke diffusion analysis results are input into the preset spatiotemporal coupling model, including: The wind field consistency index, thermal anomaly growth index, and smoke diffusion analysis results are input into the preset spatiotemporal coupling model.

[0205] In this embodiment, to verify the rationality of the smoke diffusion process and provide reliable input for subsequent fire determination, the smoke diffusion feature vector is first obtained. Secondly, to ensure the accuracy and universality of meteorological wind field data, the meteorological wind field data uses multiple sources: Subsequent analysis of data: High Resolution Land Surface Data Assimilation System (CLDAS) fusion products; Numerical weather forecast data: GFS (Global Forecast System) data, suitable for real-time monitoring scenarios with high timeliness requirements.

[0206] To achieve spatial alignment and temporal matching with the Gaofen-4 satellite image sequence, the meteorological wind field data underwent preprocessing. This preprocessing included spatial registration using bilinear interpolation to resample the meteorological wind field data to the spatial resolution (400 meters) of the Gaofen-4 mid-wave infrared image; and temporal matching using nearest-neighbor interpolation to select the hourly meteorological wind field data closest to the satellite observation time. The result was a meteorological wind field dataset that was strictly aligned with the Gaofen-4 image sequence both temporally and spatially.

[0207] in, This represents a dataset of meteorological wind direction data. Indicates the wind speed in the area; Indicates the wind direction in the area.

[0208] Subsequently, directional consistency detection can be performed based on regional wind direction and smoke diffusion direction to obtain directional detection results. For example, the angle between the smoke diffusion direction and the wind direction can be used to reflect whether the smoke is driven by the wind field. Let:

[0209] in, This represents the minimum angle between the direction of smoke diffusion and the direction of the wind field, i.e., the directional deviation angle, and its value ranges from [value missing]. The smaller the directional deviation angle, the more consistent the smoke diffusion direction is with the wind direction; Indicates the angle of smoke diffusion direction, in degrees; This indicates the wind direction angle, expressed in degrees (°).

[0210] Because the wind direction angle has a circular periodicity, when At that time, angle correction is performed:

[0211] in: This represents the minimum angle between the direction of smoke diffusion and the direction of the wind field, with a value ranging from [value missing]. The smaller the directional deviation angle, the more consistent the smoke diffusion direction is with the wind direction.

[0212] Further construct the direction matching coefficients:

[0213] in: This represents the direction matching coefficient, with a value range of [0,1].

[0214] when When =0°, =1 indicates that the directions are completely consistent; when When = 180°, =0 indicates that the directions are completely opposite.

[0215] The closer it is to 1, the higher the degree of match between the smoke diffusion direction and the wind direction.

[0216] Subsequently, a velocity consistency test was performed based on the regional wind speed and smoke diffusion velocity to obtain the velocity test results. The degree of matching between smoke diffusion velocity and wind speed was defined as follows:

[0217] in, This represents the speed matching coefficient, with a value range of [0,1]. The larger the value, the closer the smoke speed is to the wind speed.

[0218] use Instead of a fixed threshold, the matching coefficient can be adapted to wind speeds of different magnitudes. When = When = 0, special handling is required. In this case, let =1 (no smoke diffusion under calm conditions, matching).

[0219] when The closer the value is to 1, the closer the smoke diffusion speed is to the wind speed, which conforms to the laws of wind-driven mechanics.

[0220] Next, by combining the direction detection results and velocity detection results, a wind field consistency index is constructed. The mathematical formula for calculating the wind field consistency index is:

[0221] WCI stands for Wind Field Consistency Index. The directional consistency index reflects the degree of agreement between the direction of smoke diffusion and the wind direction; The velocity consistency index reflects the degree of agreement between the smoke diffusion rate (or the movement speed of the smoke front) and the wind speed. and Represents the weighting coefficients, satisfying + =1.

[0222] Preferred parameters: =0.7, =0.3.

[0223] Parameter selection criteria: Since wind direction has a more significant impact on the smoke diffusion trajectory than wind speed, the weight of the direction matching coefficient is higher than that of the speed matching coefficient.

[0224] Finally, to reduce the impact of single-phase outliers, continuous time-series averaging is used:

[0225] in: The average wind field consistency index; This represents the number of consecutive observation phases involved in the calculation.

[0226] By using time-series averaging, random noise and transient disturbances can be effectively suppressed, improving the stability and reliability of wind field verification results.

[0227] Subsequently, a wind field verification threshold was set. The judgment rules are as follows:

[0228] in, This indicates that the smoke diffusion conforms to the wind field constraints; This indicates that the smoke diffusion does not conform to the wind field constraints.

[0229] Preferred threshold: ; Finally, the wind field consistency index, thermal anomaly growth index, and smoke diffusion analysis results are input into the preset spatiotemporal coupling model. Combined with threshold judgment of fire events, the probabilistic output of fire can be realized, which can output the fire point location, smoke range, diffusion direction, and coupling probability, thus realizing quantitative fire judgment.

[0230] In some embodiments, in S105, when the wind field consistency index, thermal anomaly growth index and smoke diffusion analysis results are input into the preset spatiotemporal coupling model, the smoke objects participating in the coupling calculation are limited to the smoke mask area within a preset radius R (e.g., 5-10 km) centered on the target candidate area. Smoke outside this range is not encouraged to participate in the current coupling calculation.

[0231] To eliminate the influence of different dimensions, each feature is normalized:

[0232] Where X represents the results of the wind field consistency index, thermal anomaly growth index, and smoke diffusion analysis, respectively. This represents the maximum value of the aforementioned input features. The minimum value of the above input features, The input features after dimensionless elimination.

[0233] The fire point-smoke spatiotemporal coupling index is then calculated using the dimensionless input features. During the calculation, the pre-defined coupling model is constructed based on a pre-defined coupling function, which is:

[0234] Where: a+b+c=1, a is the thermal anomaly feature with the highest weight; b is the smoke diffusion feature; and c is the wind field constraint as an auxiliary constraint. The dimensionless thermal anomaly growth index. To obtain the dimensionless smoke diffusion analysis results, Wind field consistency index after dimensionless measurement In this embodiment, the thermal anomaly growth index provides reliable information (brightness temperature continues to increase) even after the initial fire point has formed. Smoke diffusion dynamics require a certain displacement to calculate velocity and direction. When the initial smoke volume is small, it may face challenges such as high velocity / direction noise and unreliable SDI values. The WCI may also be unstable, and the coupling process still uses low-reliability parameters, which may lead to initial missed detections.

[0235] Therefore, when calculating the fire point-smoke spatiotemporal coupling index CI, the weights of a, b, and c are adjusted according to the confidence level of smoke characteristics:

[0236] The confidence index is defined as follows:

[0237] Confidence level of smoke features; The smoke coverage area at time t; Smoke area reliability threshold; The cumulative displacement of the smoke centroid at time t. Fog displacement reliability threshold; minimum value function Preferably, Take 5,000 square meters (approximately 30 400-meter pixels). A distance of 100 meters was used to ensure that the smoke diffusion characteristics were statistically significant.

[0238] Dynamically adjust weights:

[0239] When the smoke area or displacement is very small The HAGI weight has been significantly increased, and initial warnings depend on hot anomalies.

[0240] Dynamic weighting of thermal anomaly features; Dynamic weighting of smoke diffusion characteristics; Dynamic weighting of wind field consistency; Maximum weight for smoke diffusion; Wind field consistency has the highest weight.

[0241] After obtaining the fire point-smoke spatiotemporal coupling index, the Sigmoid function is used to map the coupling index to a probability value:

[0242] Right now:

[0243] in: This represents the probability that the fire point and the smoke belong to the same fire event; its value range is:

[0244] When conducting fire monitoring, a decision threshold can be set:

[0245] If the detection probability is greater than the judgment threshold, it is determined that there is a coupling relationship between the fire point and the smoke, and the fire event is confirmed.

[0246] If the detection probability is less than the judgment threshold, it is determined to be a non-fire target.

[0247] Among them, the preferred option is:

[0248] Unlike existing technologies that rely on binary correlation determination based on rules such as "spatial proximity and temporal synchronization," this invention constructs a spatiotemporal coupled probability model of fire point-smoke. This model normalizes the Heat Anomaly Growth Index (HAGI), Smoke Diffusion Intensity Index (SDI), and Wind Field Consistency Index (WCI) to a unified dimension, and uses a linear coupling function CI for multi-source evidence fusion. Finally, the fire probability is output via a sigmoid function. This mechanism represents a technological upgrade from "qualitative / semi-quantitative rule-based decision-making" to "quantitative probability output," addressing the issues of existing methods such as high subjectivity, poor scenario adaptability, and inability to provide confidence assessment.

[0249] In some other embodiments, after S105, the method may further include: Obtain smoke mask sequences that reveal abnormal smoke features; Extract the smoke boundary contour for each phase of the smoke mask sequence; Boundary morphology feature parameters are calculated based on the smoke boundary contour. The boundary morphology feature parameters include at least one of fractal dimension, extended anisotropy index, and boundary convex hull ratio. The fractal dimension is used to characterize the complexity of the smoke boundary contour, the extended anisotropy index is used to characterize the degree of difference in the expansion of smoke in different directions, and the boundary convex hull ratio is used to characterize the ratio of the area of ​​the smoke region to the area of ​​its convex hull. Input the boundary morphology feature parameters into the preset fire intensity mapping model, and output the fire intensity level of the current time phase; Based on the fire intensity level in continuous time phases, a time-series dynamic analysis is performed to output the trend of fire intensity change.

[0250] In this embodiment, binary smoke masks corresponding to each consecutive time step are directly obtained from the completed smoke anomaly feature extraction results. The smoke mask at each time step is recorded in the form of a two-dimensional matrix, and the value of each element in the matrix is ​​used to indicate whether the location belongs to the smoke area. The smoke masks at all consecutive time steps are arranged in chronological order to form a smoke mask sequence. The masks at each time step in this sequence are aligned with each other in spatial position through geometric registration to ensure that the same geographical coordinates correspond to the same ground position at different times. The obtained smoke mask sequence is preprocessed morphologically. First, an erosion operation is used to eliminate isolated small noise points, and then a dilation operation is used to fill the small holes inside the smoke area, making the boundary of the smoke area more continuous and smooth. The preprocessed smoke mask sequence is used as the input data for subsequent boundary contour extraction.

[0251] Next, boundary extraction is performed for each phase in the smoke mask sequence. The specific determination rule is as follows: traverse each pixel in the smoke mask of the current phase. If the pixel belongs to the smoke region and at least one pixel in its four adjacent directions (up, down, left, right) or eight surrounding directions does not belong to the smoke region, then the pixel is marked as a boundary point. Connect all boundary points in sequence according to their adjacent relationships to form a closed outer boundary contour line of the smoke region. In order to reduce the amount of subsequent calculations while retaining the main morphological features, the extracted boundary contour line is simplified by removing redundant intermediate points located on the same straight line segment and retaining key morphological nodes such as the turning points and inflection points of the contour. Finally, the simplified smoke boundary contour point sequence is obtained.

[0252] Subsequently, box counting was used to overlay the smoke boundary profile onto a progressively enlarged square grid. The number of grids required to cover the boundary at each scale was calculated, and the slope was linearly fitted in logarithmic coordinates as the fractal dimension to quantify the degree of boundary tortuosity. When calculating the expansion anisotropy index, the plane was divided into multiple fan-shaped directional intervals centered on the smoke centroid. The maximum expansion radius in each direction was calculated, and the average expansion radius in the downwind direction and perpendicular to the wind direction was calculated based on contemporaneous wind field data. The ratio of these two values ​​was used as the index to reflect the dominant expansion degree of the smoke along the wind direction. When calculating the boundary convex hull ratio, the actual area of ​​the smoke region was divided by the area of ​​its smallest circumscribed convex polygon. The resulting ratio was used to characterize the degree to which the smoke diffusion approached unconstrained free diffusion. These three parameters together constitute the smoke boundary morphological feature vector.

[0253] Next, a fire intensity mapping model is pre-established. This model takes the smoke boundary morphology feature vector as input and the fire intensity level as output. Specifically, it can be constructed using methods such as statistical regression, numerical simulation fitting, or fuzzy comprehensive evaluation. After inputting the feature vector composed of three parameters extracted in the current time phase—fractal dimension, extended anisotropy index, and boundary convex hull ratio—into the model, the model makes a comprehensive judgment based on the positive correlation between each parameter and the fire intensity, and outputs the current fire intensity level result.

[0254] Finally, the fire intensity levels output from each continuous observation time are arranged in chronological order to construct a fire intensity time series. By calculating the difference in fire intensity levels between adjacent times, it is determined whether the fire is currently in an increasing, decreasing, or stable state. Further calculation of the temporal rate of change of the level difference identifies dynamic characteristics such as accelerated increasing, accelerated decreasing, or decelerating transitions in fire intensity. Based on a comprehensive analysis of the changing trends over multiple consecutive times, it is determined whether the fire is in a general trend of continuous increasing, continuous decreasing, fluctuating, or tending towards stability. The fire intensity level for the next time moment is predicted based on a weighted average of the changes over the most recent several times. Simultaneously, an early warning output is triggered when the fire intensity experiences a sudden jump, a sudden reversal of the trend, or high-frequency abnormal oscillations. The final output includes the current fire intensity level, changing trend, short-term prediction results, and abnormal early warning information.

[0255] Based on the fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion using a high-orbit staring satellite provided in the above embodiments, this application also provides a specific implementation of a fire monitoring device based on the coupling of thermal anomaly evolution and smoke diffusion using a high-orbit staring satellite. Please refer to the following embodiments.

[0256] First see Figure 2 The fire monitoring device 200 based on the coupling of thermal anomaly evolution and smoke diffusion of a high-orbit staring satellite provided in this application embodiment may include: The acquisition module 201 is used to acquire an observation image sequence of the target area, which includes multiple consecutive observation images. The calculation module 202 is used to calculate the thermal anomaly growth index of each pixel in each observation image based on the observation image sequence. The thermal anomaly growth index is used to characterize the thermal anomaly growth characteristics of the target area. Extraction module 203 is used to extract smoke anomaly features from the observed image sequence to obtain smoke anomaly features; Analysis module 204 is used to perform smoke diffusion analysis on abnormal smoke characteristics using a preset smoke diffusion dynamics model, and obtain smoke diffusion analysis results. The smoke diffusion analysis results include smoke centroid trajectory, smoke diffusion speed, smoke diffusion direction, smoke area and smoke area growth rate. The preset smoke diffusion dynamics model is used to analyze the dynamic change law in the smoke diffusion process. The input module 205 is used to input the thermal anomaly growth index and smoke diffusion analysis results into the preset spatiotemporal coupling model. The preset spatiotemporal coupling model performs spatiotemporal coupling calculation on the thermal anomaly growth index and smoke diffusion analysis results to obtain the fire point-smoke spatiotemporal coupling index. The monitoring module 206 is used to monitor the target area for fire based on the fire point-smoke spatiotemporal coupling index and obtain monitoring results.

[0257] Figure 3 A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0258] An electronic device may include a processor 301 and a memory 302 storing computer program instructions.

[0259] Specifically, the processor 301 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0260] Memory 302 may include mass storage for data or instructions. For example, and not limitingly, memory 302 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. In one instance, memory 302 may include removable or non-removable (or fixed) media, or memory 302 may be a non-volatile solid-state memory. Memory 302 may be internal or external to an electronic device.

[0261] In one instance, memory 302 may be read-only memory (ROM). In one instance, the ROM may be a mask-programmed ROM, a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable PROM (EEPROM), an electrically rewritable ROM (EAROM), or flash memory, or a combination of two or more of these.

[0262] Memory 302 may include read-only memory (ROM), random access memory (RAM), disk storage media device, optical storage media device, flash memory device, electrical, optical, or other physical / tangible memory storage device. 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 fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion with high-orbit staring satellites according to the first aspect of this disclosure.

[0263] The processor 301 reads and executes computer program instructions stored in the memory 302 to achieve... Figure 1The embodiment shown illustrates a fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion using a high-orbit staring satellite.

[0264] In one example, the electronic device may also include a communication interface 303 and a bus 304. For example, Figure 3 As shown, the processor 301, memory 302, and communication interface 303 are connected through bus 304 and complete communication with each other.

[0265] The communication interface 303 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0266] Bus 304 includes hardware, software, or both, that couples components of an electronic device together. For example, and not as a limitation, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 304 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0267] This electronic device can execute the fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, as described in this application embodiment, thereby achieving a combination of... Figures 1-3 The fire monitoring method and device based on the coupling of thermal anomaly evolution and smoke diffusion by a high-orbit staring satellite are described.

[0268] Furthermore, in conjunction with the fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion using high-orbit staring satellites 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 of the fire monitoring methods based on the coupling of thermal anomaly evolution and smoke diffusion using high-orbit staring satellites in the above embodiments.

[0269] In an optional embodiment, in conjunction with the fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion of high-orbit staring satellites in the above embodiments, this application embodiment can provide a computer program product to implement it. The instructions in the computer program product are executed by the processor of an electronic device, enabling the electronic device to implement any of the fire monitoring methods based on the coupling of thermal anomaly evolution and smoke diffusion of high-orbit staring satellites in the above embodiments.

[0270] 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.

[0271] The functional blocks shown in the above 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.

[0272] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. 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.

[0273] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (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 apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, 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.

[0274] The above are merely specific embodiments 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 fire monitoring method based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, characterized in that, include: Acquire a sequence of observation images of the target area, wherein the sequence of observation images includes multiple consecutive observation images; Based on the observed image sequence, a thermal anomaly growth index is calculated for each pixel in each observed image. The thermal anomaly growth index is used to characterize the thermal anomaly growth characteristics of the target area. Smoke anomaly features are extracted from the observed image sequence to obtain smoke anomaly features; The smoke diffusion dynamics model is used to analyze the abnormal characteristics of the smoke and obtain the smoke diffusion analysis results. The smoke diffusion analysis results include the smoke centroid trajectory, smoke diffusion velocity, smoke diffusion direction, smoke area and smoke area growth rate. The preset smoke diffusion dynamics model is used to analyze the dynamic change law in the smoke diffusion process. The thermal anomaly growth index and the smoke diffusion analysis results are input into a preset spatiotemporal coupling model. The preset spatiotemporal coupling model performs spatiotemporal coupling calculations on the thermal anomaly growth index and the smoke diffusion analysis results to obtain the fire point-smoke spatiotemporal coupling index. Fire monitoring is performed on the target area based on the fire point-smoke spatiotemporal coupling index, and monitoring results are obtained.

2. The method according to claim 1, characterized in that, The step of calculating the thermal anomaly persistence growth index for each pixel in each of the observed image sequences includes: Perform thermal infrared radiation inversion on the observed images to calculate the brightness temperature value of each pixel in each observed image; For any pixel, a brightness temperature time series corresponding to each pixel is constructed based on the brightness temperature value of each pixel; Neighborhood enhancement analysis is performed on multiple pixels within a preset neighborhood window to obtain a local thermal anomaly coefficient, which is used to characterize the degree of thermal anomaly between the pixel and the background image. When the local thermal anomaly coefficient meets the preset thermal anomaly conditions, the brightness temperature change rate at adjacent moments is calculated based on the brightness temperature time series. The thermal anomaly growth index is obtained by summing the products of the brightness temperature change rate at adjacent time points and the preset time decay weight.

3. The method according to claim 2, characterized in that, Neighborhood enhancement analysis is performed on multiple pixels within a preset neighborhood window to obtain local thermal anomaly coefficients, including: Each pixel undergoes neighborhood thermal anomaly enhancement processing to obtain multiple target enhanced pixels; Based on each of the target augmented pixels, a neighborhood window is constructed, the neighborhood window including multiple target augmented pixels; Based on the multiple target enhancement pixels within the neighborhood window, calculate the mean and standard deviation of the neighborhood background brightness temperature; A local thermal anomaly coefficient is constructed based on the mean and standard deviation of the background brightness temperature in the neighborhood.

4. The method according to claim 2 or 3, characterized in that, After obtaining the thermal anomaly growth index by summing the products of the brightness temperature change rate at adjacent time points and a preset time decay weight, the method further includes: Obtain the number of brightness temperatures and the time interval based on the brightness temperature data in the brightness temperature time series; Based on the ratio of the brightness temperature quantity to the time interval, a thermal anomaly persistence index is determined, which is used to constrain the continuous growth characteristics of thermal anomalies. If both the thermal anomaly persistence index and the thermal anomaly growth index meet the fire point identification criteria, the target pixel is identified as the target candidate region.

5. The method according to claim 1, characterized in that, Before calculating the thermal anomaly persistence growth index for each pixel in each of the observed images, the method further includes: The observed image sequence is subjected to data correction to obtain a corrected observed image sequence; the data correction includes: radiometric correction, atmospheric correction and geometric correction. The corrected observation image sequence is divided to obtain the first phase image and the image to be registered. The image to be registered includes all other observation images in the corrected observation influence sequence except for the first phase image. Feature points are extracted from the first temporal image and the image to be registered to obtain a feature point set, which includes a multidimensional feature descriptor for each feature point. The multidimensional feature descriptors of the first temporal image and the multidimensional feature descriptors of the image to be registered are matched to obtain an initial matching set, which includes multiple point pairs with the highest matching degree. Based on multiple point pairs in the initial matching set, an affine transformation model is constructed; The image to be registered is mapped to a preset coordinate system according to the affine transformation model to obtain a spatially registered image; When the spatially registered images meet the temporal analysis conditions, a spatiotemporal registration observation image sequence is constructed based on the spatially registered images under a preset temporal sequence. The temporal analysis conditions include that the registration accuracy is within the preset registration accuracy range. Based on the observed image sequence, the thermal anomaly persistence growth index for each pixel in each observed image is calculated, including: Based on the spatiotemporal registration observation image sequence, the thermal anomaly growth index of each pixel in each observation image is calculated.

6. The method according to claim 1, characterized in that, The step of extracting smoke anomaly features from the observed image sequence to obtain smoke anomaly features includes: Based on the observed image sequence, the smoke spatial texture features and thermal anomaly correlation features of the smoke in each observed image are extracted; The smoke spatial texture features and the thermal anomaly correlation features are fused using dual-flow features to obtain dual-flow features; The dual-flow features are input into a preset temporal modeling model. The preset temporal modeling model performs temporal feature analysis on the dual-flow features to obtain the temporal feature sequence corresponding to the smoke. The temporal feature sequence includes historical and future information in the smoke diffusion process, which is used to characterize the diffusion trend, morphological changes and motion patterns of the smoke target in the observed image sequence. The temporal feature sequence is decoded at resolution using a preset decoding network to obtain the probability of each pixel in the observed image belonging to a smoke region; Based on the aforementioned anomaly probability, a smoke segmentation mask is determined; The temporal feature sequence and the smoke segmentation mask are identified as smoke anomaly features.

7. The method according to claim 6, characterized in that, The process of using a pre-defined smoke diffusion dynamics model to analyze abnormal smoke characteristics and obtain smoke diffusion analysis results includes: Anomaly detection is performed on the smoke segmentation mask to obtain anomaly detection results; When the anomaly detection result is determined to be without anomaly, the smoke centroid is extracted based on the smoke segmentation mask to obtain the smoke centroid trajectory, which includes the smoke centroid coordinates at any time. Based on the centroid coordinates of the smoke at any two adjacent moments, determine the smoke displacement distance, smoke diffusion velocity, and smoke diffusion direction; Obtain the total number of smoke pixels and the actual ground surface area at the target time; The first smoke area at the target time and the second smoke area at the previous time are determined by multiplying the total number of smoke pixels at the target time with the actual ground surface area. Based on the first smoke area and the second smoke area, determine the smoke area growth rate at the target time. The trajectory of the smoke centroid, the smoke diffusion velocity, the smoke diffusion direction, the smoke area, and the smoke area growth rate are determined as the results of the smoke diffusion analysis.

8. The method according to claim 1, characterized in that, Before inputting the thermal anomaly growth index and the smoke diffusion analysis results into the preset spatiotemporal coupling model, the method further includes: Obtain concurrent meteorological wind field data for the target area; Spatial registration and temporal matching are performed on the meteorological wind field data to obtain a wind field dataset that is spatiotemporally aligned with the observed image sequence. The wind field dataset includes regional wind speed and regional wind direction. Based on the wind direction and smoke diffusion direction in the area, directional consistency detection is performed to obtain the directional detection result; Based on the wind speed and smoke diffusion rate in the area, a speed consistency detection is performed to obtain the speed detection result; Based on the direction detection results and the velocity detection results, a wind field consistency index is constructed; The wind field consistency index is continuously averaged over time to obtain the average wind field consistency index. Based on the average wind field consistency index and the preset wind field verification threshold, the wind field constraint results are determined; When the wind field constraint result is wind-constrained, the thermal anomaly growth index and smoke diffusion analysis results are input into a preset spatiotemporal coupling model, including: The wind field consistency index, the thermal anomaly growth index, and the smoke diffusion analysis results are input into a preset spatiotemporal coupling model.

9. The method according to claim 4, characterized in that, After determining the target pixel as a target candidate region, the method further includes: The brightness temperature time series of each pixel in the target candidate region is normalized and smoothed to generate the thermal waveform of each pixel; Extract the waveform morphology features of the thermal waveform and construct a thermal waveform feature vector; the waveform morphology features include at least one of the following: rising phase features, peak phase features, falling phase features, overall waveform features, and periodic and repetitive features; The thermal waveform feature vector is matched with a preset multi-type heat source waveform template library to identify the heat source type of the thermal anomaly candidate region; the waveform template library contains standard waveforms of at least two heat source types and corresponding waveform feature parameter ranges. Based on the identified heat source type, the target candidate areas are screened, and non-fire type thermal anomaly areas are excluded from subsequent calculations.

10. A fire monitoring device based on the coupling of thermal anomaly evolution and smoke diffusion from a high-orbit staring satellite, characterized in that, include: An acquisition module is used to acquire an observation image sequence of a target area, wherein the observation image sequence includes multiple consecutive observation images; The calculation module is used to calculate the thermal anomaly growth index of each pixel in each of the observed images based on the observed image sequence. The thermal anomaly growth index is used to characterize the thermal anomaly growth characteristics of the target area. The extraction module is used to extract smoke anomaly features from the observed image sequence to obtain smoke anomaly features; The analysis module is used to perform smoke diffusion analysis on the abnormal characteristics of the smoke using a preset smoke diffusion dynamics model, and obtain the smoke diffusion analysis results. The smoke diffusion analysis results include the smoke centroid trajectory, smoke diffusion velocity, smoke diffusion direction, smoke area, and smoke area growth rate. The preset smoke diffusion dynamics model is used to analyze the dynamic change law during the smoke diffusion process. The input module is used to input the thermal anomaly growth index and the smoke diffusion analysis results into a preset spatiotemporal coupling model. The preset spatiotemporal coupling model performs spatiotemporal coupling calculation on the thermal anomaly growth index and the smoke diffusion analysis results to obtain the fire point-smoke spatiotemporal coupling index. The monitoring module is used to monitor the target area for fire based on the fire point-smoke spatiotemporal coupling index and obtain monitoring results.

Images