GK2A data spatio-temporal context feature-based wildfire real-time monitoring method

Abstract

The invention discloses a GK2A data spatio-temporal context feature-based wildfire real-time monitoring method, and belongs to the technical field of remote sensing. And constructing a window by taking the potential fire point as a center, and determining the size of a background window according to whether the number of effective pixels in the window meets a corresponding formula or not. And identifying whether the potential fire point pixel is a real fire point pixel or not through an absolute brightness temperature change test and a spatio-temporal context fire point judgment formula by utilizing the background information of the fire point. And finally, according to the land coverage type, the fire point pixel confidence coefficient, the normalized vegetation index and the normalized combustion index, further screening the generated fire point data so as to reduce the false alarm rate of the wildfire monitoring algorithm. According to the method, a complete wildfire monitoring process is designed, internal parameters are adjusted, the sensitivity of GK2A data in early wildfire monitoring is improved, missing detection of fire point pixels is effectively reduced, and the accuracy of a wildfire monitoring algorithm is improved.

Description

Technical Field

[0001] The invention belongs to the field of remote sensing technology and relates to a wildfire real-time monitoring method based on geostationary meteorological satellite Geo-Kompsat-2A data. Background Art

[0003] In recent years, with the rapid development of remote sensing technology, the application of different types of remote sensing satellites in the field of wildfire monitoring has made significant progress. Among them, geostationary meteorological satellites can provide near-real-time, high-temporal resolution ground information, thereby identifying wildfires more quickly, and then provide decision-making suggestions for relevant departments to respond to rescue in a timely manner.

[0004] Wildfire monitoring methods based on geostationary meteorological satellites mainly rely on satellite sensors to capture infrared and thermal infrared radiation characteristics related to wildfires. By analyzing radiation changes, the location and intensity of fire points can be accurately identified. This type of method includes traditional spatial context algorithms and machine learning models to detect heat signals generated by wildfires and other natural or artificial heat sources. However, despite the significant advantages of geostationary meteorological satellites in temporal resolution, their monitoring technology also faces certain challenges. First, the insufficient spatial resolution of current satellites limits the accurate identification of the location of fire sources. Second, the existing wildfire monitoring algorithms have a high false alarm rate and low accuracy, which seriously affects the effect of real-time monitoring. In addition, cloud cover and atmospheric conditions may weaken the performance of sensors, thereby reducing the accuracy of monitoring. Therefore, we should make full use of the data of the new generation of high-performance geostationary meteorological satellite GK2A, deeply explore the spatial, spectral and temporal characteristics of wildfire pixels, develop real-time wildfire monitoring technology based on GK2A data, and build a fast cloud mask algorithm suitable for the southwest region to improve monitoring capabilities and effectively support wildfire management. Summary of the invention

[0005] The present invention provides a real-time wildfire monitoring method based on spatial, spectral and temporal features and applicable to Geo-Kompsat-2A satellite data in spatiotemporal context, so as to achieve accurate and efficient monitoring of wildfires in the southwest region. The method adopts a random forest algorithm to remove water and cloud pixels that have a greater impact on wildfire determination through water cloud mask. At the same time, the concept of background window is introduced, and the determination conditions are set according to different wildfire types to screen out all potential fire point pixels. The potential fire point pixels are determined by spatiotemporal context information, and the false alarm rate is reduced in four ways, and finally the wildfire point distribution results are generated.

[0006] In order to ensure that the wildfire monitoring algorithm is not interfered by clouds and water, it is necessary to mask out the cloud and water pixels in the GK2A image. The cloud pixel removal uses the random forest algorithm to build a high-precision, low false alarm rate cloud detection model. The random forest reduces the risk of overfitting and improves the model's generalization ability for various cloud pixel characteristics by introducing randomness between multiple decision trees. In addition, the random forest algorithm provides feature importance assessment to help understand which spectral features are most critical for cloud pixel identification, thereby improving the interpretability and accuracy of the algorithm. The removal of water pixels is based on the water mask algorithm and is determined according to the relevant formula.

[0007] The technical solution of the present invention is: a wildfire real-time monitoring method based on the spatiotemporal context features of GK2A data, the method comprising:

[0008] Step 1: Collect satellite image data of the target area and train a cloud detection model. Use the trained cloud detection model to perform cloud detection on the newly collected images; then identify water pixels; finally, mask out the cloud pixels and water pixels;

[0009] Step 2: Extract potential fire points;

[0010] Pixels that meet the following formula conditions are initially identified as potential fire points:

[0011] BT 07,i >305K and BT 07,i -BT 14,i >20K

[0012] Among them, BT 07,i is the observed brightness temperature value of the 7th band of the target pixel, BT 14,i is the observed brightness temperature value of the target image in the 14th band; K is the temperature unit;

[0013] Step 3: Determine the background information of potential fire point pixels;

[0014] Determine the background window size of the potential fire point pixel, and the judgment formula is:

[0015] N v >w 2 *0.25orN v >8

[0016] Where w is the background window size, N v is the number of effective background pixels; the starting size of the background window is set to the rectangular window size surrounding the potential fire point pixels, and the window size is expanded until the judgment formula is satisfied or the set maximum window size is reached;

[0017] Step 4: Determine the fire point based on the spatiotemporal context;

[0018] Step 4.1: Perform absolute brightness temperature change test;

[0019]

[0020] in, is the brightness temperature at the current moment, is the brightness temperature change value at the previous moment. If the brightness temperature change of the target pixel exceeds 4K in a short time, it is determined that the pixel has thermal anomaly;

[0021] Step 4.2: If the judgment in step 4.1 is no, then the spatiotemporal contextual fire point information is judged:

[0022]

[0023] Among them, ΔBT 07,i is the brightness temperature difference before and after the target pixel, It is the absolute value of the average brightness temperature difference before and after the effective pixel in the background window; if it meets the spatiotemporal context fire point information judgment conditions, it is judged that the pixel has thermal anomaly;

[0024] Step 5: Eliminate false alarms;

[0025] Step 5.1: Calculate Z using the following formula 07 , Z ΔT :

[0026]

[0027] Among them, Z 07 It represents the absolute deviation of the brightness temperature value of the 7th band of the fire point pixel in the background window, Z ΔT It represents the absolute deviation of the brightness temperature difference between the 7th band and the 14th band of the fire point pixel in the background window, and ΔT represents the brightness temperature difference between the 7th band and the 14th band. represents the average value of the brightness temperature difference between the 7th and 14th bands; δ 07 Represents the standard deviation of the brightness temperature of the 7th band of the effective background pixel, δ ΔT Represents the standard deviation of the brightness temperature difference between the 7th and 14th bands of effective background pixels;

[0028] Step 5.2: Calculate confidence;

[0029] The confidence of each pixel is composed of five sub-confidence combinations labeled C1 to C5, each sub-confidence ranges from the lowest confidence 0 to the highest confidence 1;

[0030] C1=S(BT 07 ; 310K, 340K)

[0031] C2=S(Z 07 ; 2.5,6)

[0032] C3=S(Z ΔT ; 3,6)

[0033] C4=1-S(N ac ; 3,6)

[0034] C5=1-S(N aw ; 3,6)

[0035]

[0036] Among them, N aw Represents the number of water pixels adjacent to the central fire pixel, N ac Represents the number of cloud pixels adjacent to the central fire pixel; S() is the slope function;

[0037] Step 5.3: Use the Normalized Difference Vegetation Index and the Normalized Difference Burn Index to compare with the surrounding area and exclude the fire pixels below a certain threshold. The calculation method is as follows:

[0038]

[0039] in,

[0040] and Respectively represent the average value of NDVI and NBR of the effective background pixels in the background window of the fire point pixel, RMSD NDVI and RMSD NBR They represent the root mean square deviation of NDVI and NBR of effective background pixels in the background window respectively; ρ 3,i , 4,i and ρ 5,i They refer to the reflectivity of the red band, near-infrared band and short-wave infrared band respectively.

[0041] Furthermore, the cloud detection model in step 1 is a random forest model. During the training process of the random forest model, the ratio of positive sample cloud pixels to negative sample non-cloud pixels is adjusted to 2:1, and the cloud pixel data set is divided into a training set and a test set at a ratio of 7:3; the random forest model parameter adjustment adopts a random search method to randomly select a parameter combination within the set parameter range;

[0042] The water pixel determination method is:

[0043] If the following conditions are met, it is judged as a water pixel:

[0044] Daytime: ρ 6,i >0.012, or, at night: abs(ρ 3,i )<0.01 and abs(ρ 4,i )<0.01

[0045] Among them, ρ 3,i , 4,i , 6,i They are the reflectances of the 3rd, 4th and 6th bands of the GK2A image, abs(·) represents the absolute value of the band reflectance, and i represents the i-th pixel.

[0046] Furthermore, in step 2, a detection window is first set, and potential fire points are detected according to the window;

[0047] The initial background window range is set to 7×7; if the number of clear sky land pixels in the window exceeds 8, the window is considered to be a valid background reference area; if the initial window does not meet this condition, the range of the background window is expanded outward with the target pixel as the center until a sufficient number of valid pixels are obtained. The maximum background window range is set to 11×11.

[0048] Furthermore, the method for further identifying a single potential fire point pixel in a small-scale wildfire event in step 2 is:

[0049] BT 07,i >BT 07,min +5K and BT 07,i -BT 14,i >(BT 07 -BT 14 ) min +5K

[0050] In the formula, BT 07,min is the minimum brightness temperature value in the background pixel, (BT 07 -BT 14 ) min It is the minimum value of the brightness temperature difference observed between the background pixels in the 7th and 14th bands; if this condition is met, it is determined to be a small-scale wildfire event.

[0051] Furthermore, the method for further identifying a large number of potential fire point pixels in step 2 is:

[0052] and

[0053] In the formula is the average brightness temperature of the 7th band in the background window, It is the average value of the brightness temperature difference between the 7th and 14th bands in the background window.

[0054] Furthermore, the method for further identifying large wildfire pixels in step 2 is:

[0055] BT 07,i >310K and and

[0056] If this condition is met, it is determined to be a large wildfire pixel.

[0057] Furthermore, the ramp function in step 5.2 is:

[0058]

[0059] Among them, x, α, and β are the elements of the corresponding positions in the ramp function.

[0060] Beneficial effects of the present invention: The present invention proposes a real-time wildfire monitoring method based on the spatiotemporal contextual features of the geostationary meteorological satellite Geo-Kompsat-2A. Specifically, it is a method that compares the brightness temperature changes of the target pixel before and after with the brightness temperature changes of other surrounding valid pixels before and after, and on this basis, uses a cloud detection model constructed by a random forest algorithm, combines the red light band and near-infrared band information, removes water and cloud pixels through a water cloud mask, and performs a series of screening on the generated fire point data to reduce the false alarm rate of the wildfire monitoring algorithm. The invention improves the sensitivity of GK2A data in early wildfire monitoring, effectively reduces the missed detection of wildfire pixels, and improves the accuracy of the wildfire monitoring algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 This is a schematic diagram of the study area in southwest my country;

[0062] In the figure, 1 evergreen coniferous forest, 2 evergreen broad-leaved forest, 3 deciduous coniferous forest, 4 deciduous broad-leaved forest, 5 mixed forest, 6 closed shrub forest, 7 open shrub forest, 8 tropical savanna, 9 tropical grassland, 10 grassland, 11 permanent wetland, 12 cultivated land, 13 urban and built-up land, 14 cultivated land / natural vegetation mosaic, 15 permanent ice and snow, 16 wasteland, 17 water body.

[0063] Figure 2 This is the cloud detection visualization result diagram for the GK2A cloud product, fixed threshold method, and random forest algorithm.

[0064] Figure 3 This is a graph of pixel brightness temperature change and change rate.

[0065] Figure 4 Flowchart for wildfire monitoring in spatiotemporal context.

[0066] Figure 5 This is the spatial distribution map of GK2A algorithm fire points and VIIRS fire points in the study area in southwest my country. DETAILED DESCRIPTION

[0067] The present invention will be further described below in conjunction with the accompanying drawings.

[0068] (1) Cloud water detection

[0069] In order to ensure that the wildfire real-time monitoring algorithm of the present invention is not interfered by clouds and water during the calculation process, the cloud pixels and water pixels are masked out using the random forest algorithm and the discriminant formula respectively. Figure 1 ) The ratio of positive samples (cloud pixels) to negative samples (non-cloud pixels) was adjusted to 2:1, and the cloud pixel dataset was divided into training and test sets at a ratio of 7:3. The model parameter adjustment adopts a random search method to randomly select parameter combinations within the set parameter range, so as to screen out the near-optimal parameter configuration in a short time. The results are shown in Figure 2. Figure 2 As shown. The elimination of water pixels is based on the following judgment formula:

[0070] (ρ 6,i >0.012) or at night

[0071] Night: (abs(ρ 3,i )<0.01)and(abs(ρ 4,i )<0.01)

[0072] Among them, ρ 3,i , 4,i , 6,i They are the reflectances of the 3rd, 4th and 6th bands of the GK2A image respectively.

[0073] (2) Potential fire point extraction

[0074] The peak of energy released when vegetation burns is roughly concentrated in the 3.9μm area, while the energy released by wildfires in the far infrared band (i.e., the thermal infrared range) is relatively weak. Based on this spectral characteristic, the following formula is used to identify potential fire point pixels:

[0075] BT 07,i >305k&BT 07,i -BT 14,i >20k

[0076] In the formula, BT 07,i is the observed brightness temperature value of the 7th band of the target pixel, BT 14,i is the observed brightness temperature value of the target image in band 14. The two thresholds of 305 and 20 are empirical thresholds based on comprehensive considerations of factors such as the spectral characteristics of wildfire pixels and the intensity of wildfires in the southwestern research area.

[0077] The initial background window range is set to 7 × 7. If the number of clear sky land pixels in the window exceeds 8, the window is considered to be a valid background reference area. If the initial window does not meet this condition, the range of the background window is expanded outward with the target pixel as the center until a sufficient number of valid pixels are obtained. The maximum background window range is set to 11 × 11.

[0078] The potential fire point discrimination formula combining spectral and spatial characteristics is as follows:

[0079] BT 07,i >BT 07,min +5K&BT 07,i -BT 14,i >(BT 07 -BT 14 ) min +5K

[0080] In the formula, BT 07,min is the minimum brightness temperature value in the background pixel, (BT 07 -BT 14 ) min is the minimum value of the brightness temperature difference observed between the background pixel in band 7 and band 14. This discriminant formula can effectively identify a single potential fire pixel in a small-scale wildfire event.

[0081]

[0082] In the formula is the average brightness temperature of the 7th band in the background window, is the average value of the brightness temperature difference between the 7th band and the 14th band of the background window. This discriminant formula can efficiently identify and extract a large number of potential fire point pixels.

[0083]

[0084] This discriminant formula is used to determine whether a large wildfire exists and identify target pixels as potential fire points.

[0085] (3) Determination of fire point background information

[0086] Determine the background window size of the potential fire point pixel, and the judgment formula is:

[0087] N v >w 2 *0.25orN v >8

[0088] Where w is the background window size, N vis the number of effective background pixels. The initial size of the background window is set as a 11×11 rectangular window surrounding the potential fire point pixels, and the window size is expanded until the judgment formula is satisfied or the 21×21 window size is reached.

[0089] (4) Determination of hotspots in spatiotemporal context

[0090] First, the absolute brightness temperature change test ( Figure 3 ):

[0091]

[0092] in is the brightness temperature at the current moment, is the brightness temperature change value of the previous moment. If the brightness temperature change of the target pixel exceeds 4K in a short time, we determine that the pixel has thermal anomaly after strictly removing water cloud pixels.

[0093] If the absolute brightness temperature change test fails, the spatiotemporal context fire point information determination is performed ( Figure 4 ), the optimal judgment formula adjusted for the southwest region is as follows:

[0094]

[0095] Where ΔBT 07,i is the brightness temperature difference before and after the target pixel, It is the absolute value of the average brightness temperature difference before and after the effective pixel in the background window.

[0096] (5) False alarm elimination

[0097] The fire point pixels were checked using the MCD12Q1.006 land cover product.

[0098] Referring to the MODIS fire point product, the confidence calculation method is as follows:

[0099]

[0100]

[0101] These two variables represent the fire point pixel in the background window BT 07 and BT 14 The absolute deviation, and we use the ramp function to calculate:

[0102]

[0103] When calculating confidence, the confidence of each pixel is composed of a combination of five sub-confidences labeled C1 to C5, each of which ranges from 0 (lowest confidence) to 1 (highest confidence).

[0104] C1=S(BT 07 ; 310K, 340K)

[0105] C2=S(Z 07 ; 2.5,6)

[0106] C3=S(Z ΔT ; 3,6)

[0107] C4=1-S(N ac ; 3,6)

[0108] C5=1-S(N aw ; 3,6)

[0109]

[0110] Where N aw Represents the number of water pixels adjacent to the central fire pixel, N ac Represents the number of cloud pixels adjacent to the central fire pixel; δ 07 Represents the standard deviation of the brightness temperature of the 7th band of the effective background pixel; δ ΔT Represents the standard deviation of the brightness temperature difference between the 7th and 14th bands of effective background pixels.

[0111] The normalized difference vegetation index and normalized burn index are compared with the surrounding area to exclude the fire pixels below a certain threshold. The calculation method is as follows:

[0112]

[0113] in and Respectively represent the average value of NDVI and NBR of the effective background pixels in the background window of the fire point pixel, RMSD NDVI and RMSD NBR Respectively represent the root mean square deviation of NDVI and NBR of effective background pixels in the background window. 3,i , 4,i and ρ 5,i They refer to the reflectivity of the red band, near-infrared band and short-wave infrared band respectively.

[0114] The present invention also introduces a potential fire point detection algorithm to comprehensively identify and extract all wildfire-related pixels and effectively exclude pixels that do not contain fire conditions. According to the specific temperature properties of vegetation burning, and combined with Planck's law, Stefan-Boltzmann's law and Wien's displacement law, the matching relationship between the energy peak of vegetation burning and the 7th band of the GK2A satellite is derived. By comparing the radiation differences between the mid-infrared and far-infrared bands, wildfires and other high-temperature sources can be effectively distinguished. In addition, the introduction of the background window concept further improves the ability to identify potential fire points. Expand the background window range with the target pixel as the center until a sufficient number of valid pixels are obtained. Since there is usually a significant difference in brightness temperature between the fire point pixels and the surrounding pixels, this method identifies potential fire points whose spectral characteristics may be ignored. The mathematical expression of Planck's law is:

[0115]

[0116] Where B(λ,T) is the radiation intensity at wavelength λ at temperature T; h is Planck's constant; c is the speed of light; k is the Boltzmann constant; T is the absolute temperature of the black body; and λ is the wavelength of the radiation.

[0117] The mathematical expression of the Stefan-Boltzmann law is:

[0118] E=σT 4

[0119] Where E is the radiant power per unit area (unit: watt per square meter, W / m 2 ), T is the absolute temperature of the black body, σ is the Stefan Boltzmann constant, which is approximately 5.67×10 -8 Wm -2 K -4 .

[0120] The mathematical expression of Wien's displacement law is:

[0121] λ max =b / T

[0122] where λ max represents the maximum intensity wavelength of blackbody radiation, T represents the absolute temperature of the blackbody, and b is the Wien displacement constant, which is approximately 2.897×10 -3 mK.

[0123] The brightness temperature is calculated using the following formula:

[0124] T B =(hc / Kλ)·ln -1 (1+2hc 2 / B(λ,T)λ 5 )

[0125] Where T Bis the brightness temperature, and the other physical quantities have the same meanings as above.

[0126] In order to determine whether the potential fire point pixel obtained is a real fire point pixel, it is necessary to use the pixel information around the potential fire point pixel to estimate the radiation signal of the potential fire point pixel when there is no wildfire. The potential fire point pixel is taken as the center of the background window, and the background value is estimated based on the effective background pixels identified in the window. Pixels in the background window that contain available observation data, non-cloud and non-water pixels, and non-potential fire point pixels are considered to be effective background pixels. When there are enough effective background pixels in the background window, that is, 25% of the effective background pixels in the window or the number of effective pixels is at least 8, the next step of spatiotemporal context information calculation is performed.

[0127] In order to extract qualified real fire points from a large number of potential fire points and meet the needs of real-time wildfire monitoring, this paper proposes a spatiotemporal context wildfire monitoring algorithm. The algorithm is mainly based on the statistical characteristics of the brightness temperature values ​​of the fire point detection channels at different periods at the same time point. Wildfire monitoring is achieved by comparing the brightness temperature changes before and after the target pixel with the changes of other valid pixels in the background window before and after the moment. The core of this method is that even if there are differences in the brightness temperature values ​​of specific target pixels at different times, the statistical properties and fluctuation range of these differences are predictable. Therefore, when the observed brightness temperature value of the target pixel exceeds the maximum value of the predicted fluctuation range, it is judged to be a fire point pixel. The brightness temperature change rate is calculated by the following formula:

[0128] ΔR=ΔT / Δu

[0129] Where ΔR is the brightness temperature change rate, ΔT is the brightness temperature difference of the same pixel before and after, and Δu is the time difference before and after.

[0130] In order to reduce the incidence of misjudgment of fire points during wildfire monitoring, the present invention proposes a variety of false alarm elimination methods to screen the generated fire point data. These methods include land cover data verification, fire point confidence screening, and comparison of normalized vegetation index and normalized combustion index. By checking the pixel values ​​of the fire point coordinates identified by the GK2A satellite and the corresponding coordinates in the land cover data, it is determined whether the fire point is caused by vegetation burning. According to the spectral characteristics of the fire point pixel and the conditions of the surrounding clouds and water bodies, the confidence of the fire point pixel is calculated, and high-confidence fire points are screened out. In addition, by comparing the normalized vegetation index and normalized combustion index of the fire point pixel with the surrounding area, if the result is lower than a specific threshold, the fire point is eliminated.

[0131] Figure 1The figure shows the location of the study area in the southwest, as well as the distribution of land cover types such as forests and grasslands extracted according to the MCD12Q1.006 IGBP classification scheme. The study area specifically consists of Sichuan Province, Chongqing City, Yunnan Province and Guizhou Province in the southwest. In the figure, 1 evergreen coniferous forest, 2 evergreen broad-leaved forest, 3 deciduous coniferous forest, 4 deciduous broad-leaved forest, 5 mixed forest, 6 closed shrub forest, 7 open shrub forest, 8 tropical savanna, 9 tropical grassland, 10 grassland, 11 permanent wetland, 12 cultivated land, 13 urban and built-up land, 14 cultivated land / natural vegetation mosaic, 15 permanent ice and snow, 16 wasteland, 17 water body.

[0132] Figure 2 The remote sensing image at 04:40 (UTC) on April 28, 2023 was selected as a case, and the classification results of the three methods were presented in the form of black and white images, where the white area represents clear sky pixels and the black area represents cloud pixels. The quantitative evaluation results show that the accuracy of the random forest algorithm in cloud detection is higher than that of the GK2A cloud mask dataset and the fixed threshold method, and its classification results can serve the subsequent cloud pixel removal.

[0133] Figure 3 The upper and lower figures show the change and change rate of the brightness temperature of the 7th band of the same pixel in the GK2A image within a time range of 1400 minutes (about one day). At the same location in a day, when the pixel is not disturbed by clouds or there is no thermal anomaly, its brightness temperature change trend is gentle. Therefore, when the brightness temperature change of the pixel before and after exceeds a certain range, the brightness temperature change rate can be used to determine the possible existence of thermal anomalies.

[0134] Figure 4 The specific process of real-time wildfire monitoring in spatiotemporal context is demonstrated in the paper. Its core lies in the processing of time series remote sensing images and obtaining thermal anomaly information through spatiotemporal context information.

[0135] Figure 5 The figure shows the spatial distribution of fire points produced by the wildfire monitoring algorithm proposed in the present invention and the VIIRS high-confidence fire points. It can be seen that the fire points produced by the present invention and the VIIRS fire point products maintain good consistency in 6 wildfire events, the location distribution is relatively close to the VIIRS fire point products, and the overall missed detection rate is low. The wildfire fire point distribution produced after false alarm elimination is as follows: Figure 5 It can be seen that the fire points produced by the present invention and the VIIRS fire point products maintain good consistency in 6 wildfire events, the location distribution is relatively close to the VIIRS fire point products, and the overall missed detection rate is low.

Claims

1. A method for real-time monitoring of wildfires based on the spatiotemporal context features of GK2A data, the method comprising: Step 1: Collect satellite image data of the target area and train a cloud detection model. Use the trained cloud detection model to perform cloud detection on the newly collected images; then identify water pixels; finally, mask out the cloud pixels and water pixels; Step 2: Extract potential fire points; Pixels that meet the following formula conditions are initially identified as potential fire points: BT 07,i >305K and BT 07,i -BT 14,i >20K Among them, BT 07,i is the observed brightness temperature value of the 7th band of the target pixel, BT 14,i is the observed brightness temperature of the target image in the 14th band; K is the temperature unit; Step 3: Determine the background information of potential fire point pixels; Determine the background window size of the potential fire point pixel, and the judgment formula is: N v >w 2 *0.25orN v >8 Where w is the background window size, N v is the number of effective background pixels; the starting size of the background window is set to the rectangular window size surrounding the potential fire point pixels, and the window size is expanded until the judgment formula is satisfied or the set maximum window size is reached; Step 4: Determine the fire point based on the spatiotemporal context; Step 4.1: Perform absolute brightness temperature change test; in, is the brightness temperature at the current moment, is the brightness temperature change value at the previous moment. If the brightness temperature change of the target pixel exceeds 4K in a short time, it is determined that the pixel has thermal anomaly; Step 4.2: If the judgment in step 4.1 is no, then the spatiotemporal contextual fire point information is judged: Among them, ΔBT 07,i is the brightness temperature difference before and after the target pixel, It is the absolute value of the average brightness temperature difference before and after the effective pixel in the background window; if it meets the spatiotemporal context fire point information judgment conditions, it is judged that the pixel has thermal anomaly; Step 5: Eliminate false alarms; Step 5.1: Calculate Z using the following formula 07 , Z ΔT : Among them, Z 07 It represents the absolute deviation of the brightness temperature value of the 7th band of the fire point pixel in the background window, Z ΔT It represents the absolute deviation of the brightness temperature difference between the 7th band and the 14th band of the fire point pixel in the background window, and ΔT represents the brightness temperature difference between the 7th band and the 14th band. represents the average value of the brightness temperature difference between the 7th and 14th bands; δ 07 Represents the standard deviation of the brightness temperature of the 7th band of the effective background pixel, δ ΔT Represents the standard deviation of the brightness temperature difference between the 7th and 14th bands of effective background pixels; Step 5.2: Calculate confidence; The confidence of each pixel is composed of five sub-confidence combinations labeled C1 to C5, each sub-confidence ranges from the lowest confidence 0 to the highest confidence 1; C1=S(BT 07 ;310K,340K) C2=S(Z 07 ;2.5,6) C3=S(Z ΔT ;3.6) C4=1-S(N ac ;3,6) C5=1-S(N aw ;3.6) Among them, N aw Represents the number of water pixels adjacent to the central fire pixel, N ac Represents the number of cloud pixels adjacent to the central fire pixel; S() is the slope function; Step 5.3: Use the Normalized Difference Vegetation Index and the Normalized Difference Burn Index to compare with the surrounding area and exclude the fire pixels below a certain threshold. The calculation method is as follows: in, and Respectively represent the average value of NDVI and NBR of the effective background pixels in the background window of the fire point pixel, RMSD NDVI and RMSD NBR They represent the root mean square deviation of NDVI and NBR of effective background pixels in the background window respectively; ρ 3,i , 4,i and ρ 5,i They refer to the reflectivity of the red band, near-infrared band and short-wave infrared band respectively.

2. A wildfire real-time monitoring method based on GK2A data spatiotemporal context features as claimed in claim 1, characterized in that: The cloud detection model in step 1 is a random forest model. During the training process of the random forest model, the ratio of positive sample cloud pixels to negative sample non-cloud pixels is adjusted to 2:1, and the cloud pixel data set is divided into a training set and a test set at a ratio of 7:3; the random forest model parameter adjustment adopts a random search method to randomly select a parameter combination within the set parameter range; The water pixel determination method is: If the following conditions are met, it is judged as a water pixel: Daytime: ρ 6,i >0.012, or, at night: abs(ρ 3,i )<0.01 and abs(ρ 4,i )<0.01 Among them, ρ 3,i , 4,i , 6,i They are the reflectances of the 3rd, 4th and 6th bands of the GK2A image, abs(·) represents the absolute value of the band reflectance, and i represents the i-th pixel.

3. A wildfire real-time monitoring method based on GK2A data spatiotemporal context features as claimed in claim 1, characterized in that: In the step 2, a detection window is first set, and potential fire points are detected according to the window; The initial background window range is set to 7×7; if the number of clear sky land pixels in the window exceeds 8, the window is considered to be a valid background reference area; if the initial window does not meet this condition, the range of the background window is expanded outward with the target pixel as the center until a sufficient number of valid pixels are obtained. The maximum background window range is set to 11×11.

4. A wildfire real-time monitoring method based on GK2A data spatiotemporal context features as claimed in claim 1, characterized in that: The method for further identifying a single potential fire point pixel in a small-scale wildfire event in step 2 is: BT 07,i > BT 07,min +5K and BT 07,i -BT 14,i > (BT 07 -BT 14 ) min +5K In the formula, BT 07,min is the minimum brightness temperature value in the background pixel, (BT 07 -BT 14 ) min It is the minimum value of the brightness temperature difference observed between the background pixels in the 7th and 14th bands; if this condition is met, it is determined to be a small-scale wildfire event.

5. A method for real-time monitoring of wildfires based on the spatiotemporal context features of GK2A data as claimed in claim 1, characterized in that: The method for further identifying a large number of potential fire point pixels in step 2 is: and In the formula is the average brightness temperature of the 7th band in the background window, It is the average value of the brightness temperature difference between the 7th and 14th bands in the background window.

6. A method for real-time monitoring of wildfires based on the spatiotemporal context features of GK2A data as claimed in claim 1, characterized in that: The method for further identifying large wildfire pixels in step 2 is: BT 07,i >310K and and If this condition is met, it is determined to be a large wildfire pixel.

7. A method for real-time monitoring of wildfires based on the spatiotemporal context features of GK2A data as claimed in claim 1, characterized in that: The ramp function in step 5.2 is: Among them, x, α, and β are the elements of the corresponding positions in the ramp function.

Images