A forest fire overfire extraction method, device and medium

Abstract

The application discloses a forest fire extraction method and device and a medium, and comprises the following steps: acquiring first full polarization data and second full polarization data; using a gray level co-occurrence matrix to extract texture features of HV polarization channel data of the first full polarization data and the second full polarization data to generate a first feature image with the texture features; calculating backscattering intensity changes of HV polarization channel data of the first full polarization data and HV polarization channel data of the second full polarization data to generate a second feature image with the backscattering intensity changes; calculating a polarization burning index difference value of the first full polarization data and the second full polarization data to generate a third feature image with the polarization burning index difference value; stacking the first feature image, the second feature image and the third feature image at a pixel level to obtain a multi-dimensional feature vector; and inputting the feature vector into a trained classifier to obtain a fire-burned area distribution map. The method has high extraction accuracy and strong robustness.

Description

Technical Field

[0001] This invention belongs to the field of fire damage extraction technology, specifically relating to a method, apparatus and medium for extracting fire damage from forest fires. Background Technology

[0002] Forest fires are among the most destructive natural disasters globally, posing a serious threat to the ecological environment, biodiversity, human life and property, and regional climate. After a fire breaks out, a rapid and accurate assessment of the burned area, burning intensity, and damage is crucial for disaster relief deployment, ecological restoration planning, and the development of future fire prevention strategies.

[0003] With the development of remote sensing technology, acquiring surface information using satellites or aerial platforms has become the mainstream method for monitoring forest fires and their impacts. Optical remote sensing, such as using data from satellites like Landsat, Sentinel-2, and MODIS, can effectively identify burned areas by calculating vegetation indices such as the Normalized Burn Ratio (NBR) and its variation (dNBR). However, optical remote sensing is highly susceptible to atmospheric conditions such as clouds, rain, and smoke, especially during and after fires when severe weather often accompanies them, making it difficult to obtain clear, continuous, and effective data, thus limiting its application in emergency monitoring. Furthermore, optical sensors may experience signal saturation in areas with dense vegetation cover or extremely high levels of burning, affecting the accuracy of identification. Thermal infrared remote sensing is mainly used to detect active fire points and high-temperature areas, but it has limitations in accurately delineating post-fire boundaries and assessing the degree of burning.

[0004] Synthetic Aperture Radar (SAR), as an active microwave remote sensing technology, possesses all-weather, all-day operation capabilities. It can penetrate clouds, fog, and smoke, and is sensitive to changes in surface structure and dielectric properties, thus exhibiting unique advantages in forest fire monitoring. Traditional SAR technology primarily utilizes backscatter intensity information to distinguish between burned and unburned areas. Fires cause reductions in vegetation biomass, changes in surface cover, and alterations in soil moisture, all of which affect the backscatter coefficient in SAR images. However, relying solely on intensity information is sometimes insufficient to accurately differentiate areas with varying degrees of burning. It is also susceptible to interference from factors such as terrain undulations, surface roughness, and sensor viewing angle, and the presence of speckle noise further impacts extraction accuracy.

[0005] To improve the performance of SAR in fire monitoring, polarimetric SAR (PolSAR) and interferometric SAR (InSAR) technologies have been introduced. PolSAR can provide rich information on the scattering mechanisms of ground objects, such as volume scattering, surface scattering, and secondary scattering. By analyzing changes in scattering mechanisms before and after a fire—for example, the reduction in volume scattering due to vegetation burning—fire traces can be identified more precisely. InSAR technology utilizes coherence information; the destruction of vegetation structure and changes in the ground surface caused by fire usually lead to a significant decrease in the coherence of radar signals, which is also used as a basis for extracting burned areas.

[0006] However, using PolSAR and PolSAR for fire zone extraction still has the following problems:

[0007] 1. Relying on only one or a few features such as intensity, polarization parameters or coherence is often insufficient to fully reflect the impact of fire on complex forest environments. The distinguishing ability of a single feature may decrease under different vegetation types, terrain conditions or combustion patterns, leading to misclassification or omission.

[0008] 2. Simple threshold combinations or rule-based classification fail to fully explore and utilize the complementary information and inherent correlations among multi-dimensional features such as intensity, polarization, and coherence in UAVSAR data. The effectiveness of feature fusion strategies directly affects the accuracy and robustness of the final extraction results. Summary of the Invention

[0009] To address the issues of low extraction accuracy and poor robustness in existing methods, this invention provides a method, apparatus, and medium for extracting forest fire burns.

[0010] The objective of this invention is achieved through the following technical solution:

[0011] The first aspect of this invention provides a method for extracting forest fire damage, comprising the following steps:

[0012] Acquire first full-polarization data and second full-polarization data, wherein the first full-polarization data and the second full-polarization data are full-polarization data of the same area before and after the fire, respectively.

[0013] The first and second fully polarized data are preprocessed respectively to obtain the first and second fully polarized data. gamma 0 Backscattering coefficient image, the gamma 0 The backscattering coefficient image includes HH polarization channel data, VH polarization channel data, HV polarization channel data, and VV polarization channel data;

[0014] The texture features of the HV polarization channel data / VH polarization channel data of the first and second fully polarized data are extracted using the gray-level co-occurrence matrix. The change features are extracted by the ratio to generate a first feature image with texture features.

[0015] The backscattering intensity variation of the HV polarization channel data / VH polarization channel data of the first fully polarized data and the HV polarization channel data / VH polarization channel data of the second fully polarized data is calculated to generate a second feature image with backscattering intensity variation;

[0016] The polarization combustion index difference between the first fully polarized data and the second fully polarized data is calculated based on the HH polarization channel data, HV polarization channel data and VV polarization channel data, or the polarization combustion index difference between the first fully polarized data and the second fully polarized data is calculated based on the HH polarization channel data, VH polarization channel data and VV polarization channel data, so as to generate a third feature image with polarization combustion index difference;

[0017] At the pixel level, the first feature image, the second feature image, and the third feature image are stacked to obtain a multi-dimensional feature vector.

[0018] The feature vector is input into the already trained classifier to obtain a fire-affected area distribution map.

[0019] A second aspect of the present invention provides a forest fire damage extraction device, comprising a memory and a controller connected in sequence, wherein the memory stores a computer program, and the controller is used to read the computer program and execute a forest fire damage extraction method as described in the first aspect and any of its possibilities.

[0020] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform a forest fire burn extraction method as described in the first aspect and any of its possibilities.

[0021] Compared with the prior art, the present invention has at least the following advantages and beneficial effects:

[0022] This invention utilizes high-resolution, multi-polarization, and multi-temporal fully polarized data, and integrates multi-dimensional features such as texture, backscattering variation, and a newly constructed polarization combustion index to achieve accurate and robust extraction of burned areas. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is a mean texture feature map obtained in an example of the present invention;

[0025] Figure 2 This is a variance texture feature map obtained in an example of the present invention;

[0026] Figure 3 This is a contrast texture feature map obtained in an example of the present invention;

[0027] Figure 4 This is the second feature image obtained in an example of the present invention;

[0028] Figure 5 This is the third feature image obtained in an example of the present invention;

[0029] Figure 6 This is a fire-affected area distribution map obtained in an example of the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0031] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0032] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0033] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0034] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention. In addition, the terms "first," "second," etc., are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0035] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0036] The first aspect of the present invention provides a method for extracting forest fire damage, specifically, the method includes steps S01 to S05.

[0037] Step S01: Obtain the first full polarization data and the second full polarization data, wherein the first full polarization data and the second full polarization data are full polarization data of the same area before and after the fire.

[0038] The full polarization data in this step includes HH polarization channel data, VH polarization channel data, HV polarization channel data, and VV polarization channel data simultaneously, such as UAVSAR data.

[0039] The fully polarized data can be local data, data stored in the cloud, or data stored on a specific storage device.

[0040] Step S02: Preprocess the first fully polarized data and the second fully polarized data respectively to obtain the first fully polarized data and the second fully polarized data. gamma 0 Backscattering coefficient image, the gamma 0 The backscattering coefficient image includes HH polarization channel data, VH polarization channel data, HV polarization channel data, and VV polarization channel data.

[0041] Specifically, preprocessing includes multi-looking processing, radiometric calibration, and geocoding and terrain correction.

[0042] Multi-view processing involves selecting a window of a certain size during the UAVSAR image imaging process, calculating the average intensity in the azimuth and range directions, and then performing incoherent stacking. This effectively suppresses speckle noise, improves the radiometric resolution of the image, and facilitates subsequent feature extraction.

[0043] Radiometric calibration converts UAVSAR data into physically meaningful backscattering coefficient values. Specifically, it employs gamma zero (γ...) calibration. 0 The calibration value represents the radar cross-section per unit scattering area in the direction perpendicular to the slant range, which can better compensate for the changes in radiation caused by local terrain slope.

[0044] Topographic correction and geocoding utilizes a digital elevation model (DEM) to eliminate geometric distortions caused by topographic undulations, such as overlay and perspective shrinkage. It also transforms the image from a slant range coordinate system to a geographic coordinate system, such as a UTM, generating geocoded images with pixel-to-pixel backscatter coefficients corresponding to the Earth's surface.

[0045] This yields two time phases before and after the two fires. gamma 0 Backscattering coefficient image, gamma 0 The backscattering coefficient image includes HH polarization channel data, VH polarization channel data, HV polarization channel data, and VV polarization channel data.

[0046] Step S03: Multi-dimensional feature extraction.

[0047] Since VH and HV polarization channel data are more sensitive to changes in vegetation scattering, this step focuses on utilizing either VH or HV polarization channel data, and combining them for feature extraction. Specifically, this step includes steps S031 to S033.

[0048] Step S031: Extract the texture features of the HV polarization channel data / VH polarization channel data of the first and second fully polarized data using the gray-level co-occurrence matrix, and extract the change features by ratio to generate a first feature image with texture features.

[0049] Changes in vegetation canopy structure and volume are the main changes caused by fire. HV polarization channel data and VH polarization channel data are most sensitive to changes in vegetation canopy structure and volume. Therefore, this step extracts texture features from the full polarization data before and after the fire, namely the first full polarization data and the second full polarization data, HV polarization channel data or VH polarization channel data.

[0050] Taking HV polarization channel data as an example, the method for generating the first feature image is as follows:

[0051] First, low-pass filtering is applied to the HV polarization channel data of the first and second fully polarized data to reduce the impact of residual speckle noise on texture calculation. Specifically, Boxcar or Gaussian filtering can be used, and the filter window size can be selected based on a trade-off between noise level and ground feature detail, such as 3x3 or 5x5.

[0052] Next, the gray level co-occurrence matrix (GLCM) of the filtered HV polarization channel data is calculated. The GLCM is a statistical matrix that describes the recurrence pattern of image gray levels in spatial orientation and can effectively quantify the texture features of ground objects.

[0053] Window Size: Set to 7x7 pixels. This window size is designed to balance local representativeness and statistical stability of texture features.

[0054] Offset: Sets the distance d = 1 pixel. Represents the distance between neighboring pixels in calculating co-occurrence relationships.

[0055] Gray Quantization Level: Set to 64 levels. Compresses the original grayscale values, reducing computation and highlighting macroscopic textures.

[0056] Directions: To capture the anisotropy that the fire area may exhibit or to eliminate directionality dependence, GLCM is calculated in four directions: 0°, 45°, 90°, and 135°. Then, the texture feature values ​​calculated in these four directions are averaged to obtain rotation-invariant texture metrics.

[0057] Next, the ratio of the gray-level co-occurrence matrices of the first fully polarized data and the second fully polarized data is calculated to obtain the first change feature.

[0058] Then, the first change feature is extracted using a 3x3 Sobel operator to obtain the second change feature.

[0059] The Sobel operator is used to extract edge features from the selected texture features to obtain more significant features.

[0060] Then, the texture features are extracted by low-pass filtering of the second change feature.

[0061] Using the above steps, two transformation feature extractions are employed to extract texture features that are indicative of fire damage identification from the gray-level co-occurrence matrix. These texture features can be selected from at least three of the following: mean, variance, contrast, homogeneity, correlation, entropy, second moment, angular second moment / energy, and dissimilarity.

[0062] Preferably, the texture features include mean, variance, and contrast. The mean reflects the average level of grayscale within the window; the variance measures the dispersion of grayscale values ​​within the window and is related to texture roughness; the contrast measures the weighted sum of the distances between matrix element values ​​and their diagonals, reflecting the intensity of local image variations and the depth of texture grooves.

[0063] Finally, a first feature image with texture features can be generated. This first feature image contains multi-band texture features of the aforementioned texture features.

[0064] The method for extracting texture features from VH polarization channel data is the same as that for HV polarization channel data.

[0065] Step S032: Calculate the backscattering intensity variation of the HV polarization channel data / VH polarization channel data of the first fully polarized data and the HV polarization channel data / VH polarization channel data of the second fully polarized data to generate a second feature image with backscattering intensity variation.

[0066] Fires typically result in significant vegetation loss and a substantial reduction in volumetric scattering, which in turn causes a sharp decrease in the backscattering intensity of HV or VH polarization channel data. This change can be quantified using the ratio method or logarithmic ratio.

[0067] Similar to step S031, this step can also calculate the backscattering intensity change based on HV polarization channel data or VH polarization channel data.

[0068] Similarly, we will use HV polarization channel data as an example for a detailed explanation:

[0069] First, low-pass filtering is applied to the HV polarization channel data of the first fully polarized data and the HV polarization channel data of the second fully polarized data to reduce the impact of residual speckle noise on texture calculation.

[0070] Next, the changes in backscattering intensity of the HV polarization channel data before and after the fire are calculated. This step can be performed using either the ratio method or the logarithmic difference method.

[0071] The calculation method when using the ratio method is as follows:

[0072] Feature_RatioChange=PostFire_HV_γ 0 / PreFire_HV_γ 0 ,

[0073] In the formula, PostFire_HV_γ 0 PreFire_HV_γ is the backscattering coefficient of the HV polarization channel data characterizing the second fully polarized data. 0 The backscattering coefficients of the HV polarization channel data characterize the first fully polarized data, and Feature_RatioChange represents the change in the backscattering coefficients.

[0074] The calculation method when using the logarithmic ratio method is as follows:

[0075] Feature_RatioChange =10 * log10(PostFire_HV_γ 0 -10 * log10(PreFire_HV_γ) 0 ).

[0076] The method for calculating the backscattering coefficient variation based on VH polarization channel data is the same as that for HV polarization channel data.

[0077] Finally, a second feature image with changes in the backscattering coefficient can be generated.

[0078] Step S033: Calculate the polarization combustion index difference between the first fully polarized data and the second fully polarized data based on the HH polarization channel data, HV polarization channel data and VV polarization channel data, or calculate the polarization combustion index difference between the first fully polarized data and the second fully polarized data based on the HH polarization channel data, VH polarization channel data and VV polarization channel data, so as to generate a third feature image with polarization combustion index difference.

[0079] Fire causes drastic changes in vegetation structure and dielectric constant, which in turn alters the relative relationships between HH, HV, VV, or HH, VH, VV. The difference in polarization combustion index can quantify this comprehensive change in polarization scattering characteristics caused by fire.

[0080] The third feature image can be obtained in two ways. The first is based on HH polarization channel data, HV polarization channel data, and VV polarization channel data; the second is based on HH polarization channel data, VH polarization channel data, and VV polarization channel data.

[0081] Specifically, this will be explained using HH polarization channel data, HV polarization channel data, and VV polarization channel data as examples:

[0082] First, calculate the polarization combustion index for the first and second fully polarized data. The calculation method is as follows:

[0083] PBR _Pre = (3.5 * HV_γ 0 _Pre) / (HH_γ 0 _Pre + 2 * HV_γ 0 _Pre+ VV_γ 0 _Pre)

[0084] PBR _Post = (3.5 * HV_γ 0 _Post) / (HH_γ 0 _Post + 2 * HV_γ 0 _Post +VV_γ 0 _Post)

[0085] In the formula, PBR_Pre is the polarization combustion index of the first fully polarized data, and HH_γ 0 _Pre is the backscattering coefficient HV_γ of the HH polarization backscattering intensity of the first fully polarized data, normalized to the ground plane after topographic radiative correction. 0 _Pre is the backscattering coefficient of the HV polarization backscattering intensity of the first fully polarized data, normalized to the ground plane after topographic radiative correction, and VV_γ 0 _Pre is the backscattering coefficient of the VV polarization backscattering intensity of the first fully polarized data, normalized to the ground plane after topographic radiative correction.

[0086] PBR_Post is the polarization combustion index of the second full polarization data, HH_γ 0 _Post represents the backscattering coefficient HV_γ of the polarization backscattering intensity of the second fully polarized data, normalized to the ground plane after orographic correction. 0 _Post represents the backscattering coefficient of the HV polarization backscattering intensity of the second fully polarized data, normalized to the ground plane after orographic correction, and VV_γ 0 _Post is the backscattering coefficient of the VV polarization backscattering intensity of the second fully polarized data, normalized to the ground plane after topographic radiative correction.

[0087] The Polarimetric Burn Ratio (PBR) is designed to enhance sensitivity to changes in scattering mechanisms caused by fire. The numerator emphasizes the HV component, which is sensitive to volumetric scattering, while the denominator incorporates information from all polarization channels, forming a normalized measure. The coefficients are optimized based on experimental or theoretical analysis to amplify the difference in PBR values ​​before and after a fire. This specific form is intended to distinguish between burning and unburned areas better than a single channel or a simple ratio, especially under complex surface conditions.

[0088] Next, the polarization combustion indices of the first and second fully polarized data are low-pass filtered to smooth noise and highlight regional variations. Specifically, a 5x5 pixel window low-pass filter can be applied to the calculated PBR_Pre and PBR_Post respectively.

[0089] Finally, the difference between the polarization combustion index of the filtered first fully polarized data and the polarization combustion index of the second fully polarized data is calculated to obtain the polarization combustion index difference value.

[0090] Specifically, Feature_dPBR = Filtered_PBR_Post - Filtered_PBR_Pre.

[0091] In the formula, Feature_dPBR is the difference in polarization combustion index, Filtered_PBR_Post is the polarization combustion index of the filtered second fully polarized data, and Filtered_PBR_Pre is the polarization combustion index of the filtered first fully polarized data.

[0092] The method for calculating the polarization combustion index difference based on HH polarization channel data, VH polarization channel data, and VV polarization channel data is the same as the method described above, except that the corresponding HV polarization channel data is replaced with the corresponding VH polarization channel data.

[0093] Step S04: Stack the first feature image, the second feature image, and the third feature image at the pixel level to obtain a multi-dimensional feature vector.

[0094] The three feature images obtained in step S03 are stacked at the pixel level to form a multi-dimensional feature vector space, where each pixel is described by all its corresponding feature values.

[0095] Step S05: Input the feature vector into the already trained classifier to obtain the fire-affected area distribution map.

[0096] The classifier used in this step can be any one of the following: decision tree classifier, support vector machine classifier, Naive Bayes classifier, K-nearest neighbor classifier, random forest (RF) classifier, or neural network classifier.

[0097] Preferably, a Random Forest classifier is selected. RF is an ensemble learning algorithm that has good processing capabilities for high-dimensional feature data, is not prone to overfitting, is insensitive to noise and missing data, can effectively handle nonlinear relationships, and can evaluate the importance of each feature, making it very suitable for fusing multi-source remote sensing features for classification tasks.

[0098] During classifier training, representative training samples are selected, including pixels known as "burnt areas" and "unburnt areas." These samples can be obtained through field measurements, visual interpretation of high-resolution optical imagery, or other reliable reference data. The RF model is trained using these training samples and their corresponding multidimensional feature vectors. The trained RF model is then applied to the fused feature imagery of the entire study area to classify each pixel, predicting its probability of belonging to a "burnt area" or "unburnt area," or directly providing a classification label, generating a binary map or probability map of the burned area.

[0099] Specifically, this step first inputs the feature vector into a trained random forest classifier to obtain a binary map or probability map, where the probability map includes the probability that each point is an overburned area.

[0100] If the graph is probabilistic, it needs to be converted into a binary graph. The conversion is done by setting a threshold. The appropriateness of the threshold setting directly affects the accuracy of the generated binary graph. The optimal threshold can be determined using methods such as ROC curve analysis and the Otsu's method (maximum inter-class variance). For example, the threshold can be set to any value between 0.5 and 0.6.

[0101] Finally, spatial filtering is performed on the binary image to obtain a fire-affected area distribution map. Specifically, majority filtering and morphological operations can be used to achieve spatial filtering, eliminating salt-and-pepper noise, smoothing boundaries, and improving the spatial continuity and regularity of the classification results. Morphological operations include, but are not limited to, opening operations to remove small noise patches and closing operations to fill internal holes.

[0102] The method of this invention is used to extract forest fire damage. It integrates high-resolution texture information provided by UAVSAR, temporal backscatter intensity changes, and specially designed PBR changes. It fully leverages the all-weather, high-resolution fine observation capabilities of UAVSAR, effectively processes high-dimensional features to improve classification accuracy and model robustness, adapts to different forest types and fire severity scenarios, and provides strong support for emergency response and post-disaster management.

[0103] To facilitate the explanation of the technical effects of this solution, UAVSAR data before and after a fire in a certain area are used as an example. In this example, the first feature image, the second feature image, and the third feature image are all HV polarization channel data.

[0104] like Figure 1 , 2 As shown in Figure 3, these are the mean texture feature map, variance texture feature map, and contrast texture feature map of the UAVSAR data after the fire, extracted using the gray-level co-occurrence matrix for this area.

[0105] like Figure 4 As shown, this is the second feature image of the region; as Figure 5 As shown, this is the third feature image of the region; as... Figure 6 The image shown is a fire-affected area distribution map of the region, i.e., a binary map of the fire-affected areas.

[0106] A second aspect of this invention provides a forest fire damage extraction device, comprising a memory and a controller connected in sequence. The memory stores a computer program, and the controller is used to read the computer program and execute a forest fire damage extraction method as described in the first aspect and any of its possibilities. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; the controller may not be limited to using a microcontroller of the STM32F105 series. Furthermore, the computer device may also include, but is not limited to, a power supply unit, a display screen, and other necessary components.

[0107] A third aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, perform a forest fire burn extraction method as described in the first aspect and any of its possibilities.

[0108] The operating principle of the apparatus and medium provided in the second and third aspects of the present invention is the same as that in the first aspect, and will not be repeated here.

[0109] Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for extracting forest fire burns, characterized in that, Includes the following steps: Acquire first full-polarization data and second full-polarization data, wherein the first full-polarization data and the second full-polarization data are full-polarization data of the same area before and after the fire, respectively. The first and second fully polarized data are preprocessed respectively to obtain the first and second fully polarized data. γ 0 Backscattering coefficient image, the γ 0 The backscattering coefficient image includes HH polarization channel data, VH polarization channel data, HV polarization channel data, and VV polarization channel data; The texture features of the HV polarization channel data / VH polarization channel data of the first and second fully polarized data are extracted using the gray-level co-occurrence matrix. The variation features are extracted by ratio to generate a first feature image with texture features. The backscattering intensity variation of the HV polarization channel data / VH polarization channel data of the first fully polarized data and the HV polarization channel data / VH polarization channel data of the second fully polarized data is calculated to generate a second feature image with backscattering intensity variation; The polarization combustion index difference between the first fully polarized data and the second fully polarized data is calculated based on the HH polarization channel data, HV polarization channel data and VV polarization channel data, or the polarization combustion index difference between the first fully polarized data and the second fully polarized data is calculated based on the HH polarization channel data, VH polarization channel data and VV polarization channel data, so as to generate a third feature image with polarization combustion index difference; At the pixel level, the first feature image, the second feature image, and the third feature image are stacked to obtain a multi-dimensional feature vector. The feature vector is input into the already trained classifier to obtain a fire-affected area distribution map.

2. The method for extracting forest fire damage according to claim 1, characterized in that: The preprocessing of the first and second fully polarized data includes: Perform multi-view processing on the first and second fully polarized data; Radiometric calibration is performed on the first and second fully polarized data. Terrain correction and geocoding are performed on the first and second fully polarized data.

3. The method for extracting forest fire damage according to claim 1, characterized in that: The process involves extracting texture features from the HV polarization channel data / VH polarization channel data of the first and second fully polarized data using a gray-level co-occurrence matrix, and extracting variation features through a ratio to generate a first feature image with texture features, including: Low-pass filtering is applied to the HV polarization channel data and VH polarization channel data of the first and second fully polarized data. Calculate the gray-level co-occurrence matrix of the filtered HV polarization channel data / VH polarization channel data; The ratio of the gray-level co-occurrence matrices of the first fully polarized data and the second fully polarized data is calculated to obtain the first variation feature; The first change feature is extracted using a 3x3 Sobel operator to obtain the second change feature. The second variation feature is low-pass filtered to extract the texture feature; Generate a first feature image with texture features.

4. The method for extracting forest fire damage according to claim 3, characterized in that: The texture features are at least three of the following: mean, variance, contrast, uniformity, correlation, entropy, second time step, second moment of angle, and dissimilarity.

5. A method for extracting forest fire damage according to claim 3, characterized in that: The texture features include mean, variance, and contrast.

6. The method for extracting forest fire damage according to claim 1, characterized in that: The backscattering intensity change between the HV polarization channel data / VH polarization channel data of the first fully polarized data and the HV polarization channel data / VH polarization channel data of the second fully polarized data is calculated as follows: The backscattering intensity changes of the HV polarization channel data / VH polarization channel data of the first fully polarized data and the HV polarization channel data / VH polarization channel data of the second fully polarized data are calculated using the ratio method or logarithmic ratio.

7. The method for extracting forest fire damage according to claim 1, characterized in that: The difference in polarization combustion index between the first full polarization data and the second full polarization data, calculated based on HH polarization channel data, HV polarization channel data, and VV polarization channel data, is as follows: The polarization combustion index of the first and second fully polarized data was calculated based on HH polarization channel data, HV polarization channel data and VV polarization channel data, respectively. The polarization combustion index of the first and second fully polarized data is low-pass filtered; The difference between the polarization combustion index of the first fully polarized data and the polarization combustion index of the second fully polarized data after filtering is used to obtain the polarization combustion index difference.

8. A method for extracting forest fire damage according to claim 1, characterized in that: The step of inputting the feature vector into a trained classifier to obtain a fire-affected area distribution map includes: The feature vector is input into a trained random forest classifier to obtain a probability map, which includes the probability that each point is an overburned area. The probability map is converted into a binary map based on a threshold. Spatial filtering is performed on the binary image to obtain a fire-affected area distribution map.

9. A forest fire burn extraction device, comprising a memory and a controller connected in sequence, wherein the memory stores a computer program, characterized in that: The controller is used to read the computer program and execute the forest fire burn extraction method according to any one of claims 1-8.

10. A computer-readable storage medium storing instructions thereon, characterized in that: When the instructions are executed on a computer, a method for extracting forest fire damage as described in any one of claims 1-8 is performed.

Images