Power transmission channel flammable risk remote sensing inversion method and device, electronic equipment and storage medium

Abstract

The invention discloses a power transmission channel flammable risk remote sensing inversion method. The method comprises the following steps: extracting and correcting corresponding features of a remote sensing image and DEM data; screening features meeting a first preset condition; inputting the features into a target CNN classification model for combustible material type classification; calculating a spectrum-scattering collaborative factor; screening features meeting a second preset condition; dividing the features into feature matrixes corresponding to combustible material types; inputting each feature matrix into a corresponding target gradient boosting tree regression model, and outputting the predicted moisture content and the predicted load of the combustible; and marking the region where the pixel meeting the preset risk condition is located as a high-risk region. According to the method, through a collaborative architecture of classification and regression of the double-stage model, coupling interference of type classification and parameter inversion is solved, the weak sensitive parameter inversion capability is enhanced, and the method is adaptive to a complex scene of a power transmission channel. The invention further discloses a device for implementing the method, electronic equipment and a computer readable storage medium.

Description

Technical Field

[0001] This invention relates to the field of aerospace satellite technology, and in particular to a method, apparatus, electronic device, and storage medium for remote sensing inversion of the combustible risk of power transmission channels. Background Technology

[0002] With the continuous expansion of the power grid, transmission lines inevitably need to traverse complex geographical environments such as mountainous areas and forests. These areas have dense vegetation and high combustible material loads, making transmission lines constantly vulnerable to wildfires. Once a wildfire occurs, it can easily lead to line tripping, equipment damage, and even large-scale power outages, posing a serious threat to the safe and stable operation of the power grid. The three elements that cause wildfires include combustible materials, ignition sources, and meteorological conditions. Among these, combustible materials are the material basis for wildfires, and their type, moisture content, and load are the core parameters determining the fire risk level. Accurately, comprehensively, and dynamically acquiring these parameters of combustible materials along transmission lines is a key technical prerequisite for achieving accurate risk classification and early warning of wildfires, guiding differentiated prevention and control, and ensuring the safe operation of the power grid. Existing remote sensing inversion technology for combustible materials has the following drawbacks: Insufficient accuracy of inversion of weakly sensitive parameters: Combustible material moisture content and load are weakly sensitive parameters in remote sensing. Traditional single-source data inversion models cannot effectively capture their complex relationship with electromagnetic wave signals, resulting in large inversion errors and failing to meet the needs of refined monitoring of power transmission channels. The limitations of a single data source are significant: existing technologies mostly rely on single SAR data or optical data. SAR data is easily affected by terrain scattering interference, while optical data is greatly affected by cloud and rain weather, and it is difficult to simultaneously retrieve multiple parameters such as combustible material type, moisture content, and load. Summary of the Invention

[0003] To address the aforementioned problems in the existing technology, this invention provides a remote sensing inversion method, apparatus, electronic device, and storage medium for flammability risk assessment of power transmission channels. The technical problem to be solved by this invention is achieved through the following technical solution: The first aspect of this invention provides a remote sensing inversion method for ignition risk assessment of power transmission channels, comprising the following steps: Extract the corresponding features of remote sensing images and DEM data of the target power transmission channel area, and correct the spectral reflectance and backscattering coefficient of the remote sensing images according to the topographic features of the corresponding features to obtain corrected remote sensing features; Features that meet the first preset conditions are selected from the corrected remote sensing features and the terrain features and then standardized to obtain the first standardized optimal feature set. The multispectral image of the remote sensing image and the corresponding first standardized optimal feature set are input into the target CNN classification model for classification, and the combustible material type of each pixel is output. The spectral-scattering synergy factor is determined based on the NDMI value and the corrected backscattering coefficient of the corrected remote sensing features; Features that satisfy the second preset condition are selected from the spectral-scattering synergistic factor, the corrected remote sensing features, and the terrain features, and then standardized to obtain the second standardized optimal feature set. The second standardized optimal feature set is divided into feature matrices corresponding to the combustible material type according to the combustible material type; Each feature matrix is ​​input into the target gradient boosting tree regression model for each combustible material type, and the predicted moisture content and predicted load of combustible material for each pixel are output. The areas where the predicted moisture content and predicted load of the combustible material meet the preset risk conditions are marked as high-risk areas.

[0004] In one embodiment of the present invention, the remote sensing image includes: multispectral image and SAR fully polarimetric image; the corresponding features include: target optical spectral features, target SAR scattering features and target terrain features; the corrected remote sensing features include: corrected optical spectral features and corrected SAR scattering features. The corresponding features of the remote sensing image and DEM data of the target power transmission channel area are extracted. Based on the topographic features of these corresponding features, the spectral reflectance and backscattering coefficient of the remote sensing image are corrected to obtain corrected remote sensing features, including: Multispectral images, SAR all-polarization images and DEM data of the target power transmission channel area are preprocessed and target optical spectral features, target SAR scattering features and target terrain features are extracted. Based on the target optical spectral characteristics and target terrain characteristics, with A′=Acosθ+ksinθ as the objective function, the least squares method is used to fit the terrain correction coefficient k corresponding to the combustible material type; where A represents the reflectance in the target optical spectral characteristics, θ represents the slope angle calculated from the target terrain characteristics, and A′ represents the measured reflectance obtained using a ground spectrometer. The corrected reflectance is determined based on the reflectance in the target optical spectral characteristics, the slope angle calculated from the terrain characteristics, and the terrain correction coefficient k, so as to obtain the corrected optical spectral characteristics. Determine the normal direction of each pixel based on the DEM data; The local incident angle of each pixel center point is determined based on the normal direction and the satellite orbit direction of the SAR fully polarimetric image; The corrected backscattering coefficient of each pixel is determined based on the preset reference incident angle, the local incident angle, and the backscattering coefficient of the target SAR scattering feature, so as to obtain the corrected SAR scattering feature.

[0005] In one embodiment of the present invention, the formula for calculating the corrected reflectivity is: B′=Acosθ+ksinθ Wherein, B′ represents the corrected reflectivity; The formula for calculating the corrected backscattering coefficient is as follows: σ′=σ(cosθ0 / cosθ i ) Where σ′ represents the corrected backscattering coefficient, θ0 represents the preset reference incident angle, and θ i The local incident angle is represented by σ, and the backscattering coefficient of the target SAR scattering feature is represented by σ.

[0006] In one embodiment of the present invention, when the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set are input into the target CNN classification model for processing, the feature data output by the pooling layer is input into the fully connected layer and converted into a one-dimensional feature vector, which is then concatenated with the terrain adaptation factor for classification; wherein, the terrain adaptation factor includes the slope angle normalized value and the slope aspect encoded value.

[0007] In one embodiment of the present invention, determining the spectral-scattering synergy factor based on the NDMI value and the corrected backscattering coefficient of the corrected remote sensing features includes: The spectral-scattering synergy factor is determined based on the NDMI value of the corrected optical spectral characteristics, the corrected backscattering coefficient of HV of the corrected SAR scattering characteristics, the corrected backscattering coefficient of HH, and the corrected backscattering coefficient of VV.

[0008] In one embodiment of the present invention, the formula for calculating the spectral-scattering synergy factor is as follows: F=(NDMI×σ HV ) / (σ HH +σ VV ) Where, σ HV The corrected backscattering coefficient of HV, σ, represents the corrected SAR scattering characteristics. HH The corrected backscattering coefficient HH, representing the corrected SAR scattering characteristics, and σ VV The corrected backscattering coefficient VV represents the corrected SAR scattering characteristics.

[0009] In one embodiment of the present invention, the label vector in the training process of the target gradient boosting tree regression model is the measured moisture content and measured load of combustible material, and the training feature matrix and label vector are used as training data pairs during the training process.

[0010] A second aspect of the present invention provides a remote sensing inversion device for flammability risk assessment of power transmission channels, comprising: The extraction and correction module is used to extract the corresponding features of remote sensing images and DEM data of the target power transmission channel area, and correct the spectral reflectance and backscattering coefficient of the remote sensing image according to the topographic features of the corresponding features to obtain the corrected remote sensing features. The first screening module is used to screen features that meet the first preset conditions from the corrected remote sensing features and the terrain features and perform standardization processing to obtain the first standardized optimal feature set. The classification module is used to input the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set into the target CNN classification model for classification, and output the combustible type of each pixel; The determination module is used to determine the spectral-scattering synergy factor based on the NDMI value and the corrected backscattering coefficient of the corrected remote sensing features; The second screening module is used to screen features that meet the second preset conditions from the spectral-scattering synergistic factor, the corrected remote sensing features and the terrain features, and perform standardization processing to obtain the second standardized optimal feature set. The partitioning module is used to partition the second standardized optimal feature set into feature matrices corresponding to the combustible material type according to the combustible material type. The prediction module is used to input each feature matrix into the target gradient boosting tree regression model for each type of combustible material, and output the predicted moisture content and predicted load of combustible material for each pixel. The processing module is used to mark the areas where the predicted moisture content and predicted load of the combustible material meet the preset risk conditions as high-risk areas.

[0011] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a remote sensing inversion method for flammability risk of power transmission channels provided in the first aspect of the present invention.

[0012] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a remote sensing inversion method for flammability risk of power transmission channels provided in the first aspect of the present invention.

[0013] The beneficial effects of this invention are: This invention eliminates terrain interference by correcting for spectral reflectance and backscattering coefficients. It integrates the type identification advantages of optical data with the moisture sensing capabilities of SAR data, and strengthens the correlation between weakly sensitive parameters and remote sensing features by constructing a spectral-scattering synergistic factor. This results in a synergistic factor that combines moisture sensitivity and structural identification, enhancing the correlation between weakly sensitive parameters and remote sensing features and addressing the problem of one-sided information from single-source data. Furthermore, this invention employs a two-stage model with a classification-then-regression synergistic architecture, resolving the coupling interference between type classification and parameter inversion while enhancing the inversion capability of weakly sensitive parameters, making it suitable for complex power transmission channel scenarios.

[0014] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0015] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 A flowchart illustrating a remote sensing inversion method for flammability risk of power transmission channels provided in an embodiment of the present invention; Figure 2 A schematic diagram of a remote sensing inversion device for flammability risk of power transmission channels provided in an embodiment of the present invention. Detailed Implementation

[0017] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0018] The core parameters for remote sensing inversion of combustibles are: combustible type, combustible moisture content, and combustible load, which are the three core parameters characterizing fire hazard levels. Among them, combustible type determines the basic properties of combustion (such as the difference in burning rate between trees, shrubs, and herbs); moisture content is the "switch" for fire occurrence (flammability is significantly increased when moisture content is ≤15%), and is usually obtained indirectly by inverting vegetation moisture content through remote sensing data; and load is the "regulator" of fire intensity (high-intensity fires are more likely to occur when load is ≥5t / ha), and inversion needs to be combined with vegetation biomass estimation models.

[0019] Characteristics of multi-source remote sensing data: Optical data: It has rich spectral information and can calculate features such as NDVI (vegetation coverage), NDMI (moisture index), and EVI (enhanced vegetation index) by combining blue, green, red, near-infrared, and short-wave infrared bands. It is suitable for identifying combustible types and making preliminary moisture assessments. However, it is greatly affected by cloud and rain weather and cannot penetrate the vegetation canopy to obtain information on the lower layer of dead branches and fallen leaves.

[0020] SAR data: The backscattering coefficients of different polarization modes (HH, HV, VH, VV) reflect vegetation structure and moisture status. Among them, HV polarization is most sensitive to changes in vegetation moisture content and can penetrate part of the vegetation canopy, making it suitable for moisture content inversion. However, it is easily affected by scattering interference caused by topographic slope and aspect, and topographic correction is required.

[0021] Digital Elevation Model (DEM) data: Topographic parameters such as slope, aspect, and altitude can be extracted and used to correct the modulation effect of terrain on remote sensing signals. It is a key auxiliary data for improving the inversion accuracy in complex terrain areas.

[0022] like Figure 1 As shown, the first aspect of this invention provides a remote sensing inversion method for flammability risk assessment of power transmission channels, comprising the following steps: Step 11: Extract the corresponding features of the remote sensing image and DEM data of the target power transmission channel area, and correct the spectral reflectance and backscattering coefficient of the remote sensing image according to the topographic features of the corresponding features to obtain the corrected remote sensing features.

[0023] Step 12: Select features that meet the first preset conditions from the corrected remote sensing features and terrain features, and perform standardization processing to obtain the first standardized optimal feature set.

[0024] Step 13: Input the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set into the target CNN classification model for classification, and output the combustible material type of each pixel.

[0025] Step 14: Determine the spectral-scattering synergy factor based on the NDMI value of the corrected remote sensing features and the corrected backscattering coefficient.

[0026] Step 15: Select features that meet the second preset conditions from the spectral-scattering synergy factor, corrected remote sensing features, and terrain features, and perform standardization processing to obtain the second standardized optimal feature set.

[0027] Step 16: Divide the second standardized optimal feature set into feature matrices corresponding to the combustible material type according to the combustible material type.

[0028] Step 17: Input each feature matrix into the target gradient boosting tree regression model corresponding to each combustible material type, and output the predicted moisture content and predicted combustible material load for each pixel.

[0029] Step 18: Mark the areas where the predicted moisture content and predicted load of combustibles meet the preset risk conditions as high-risk areas.

[0030] In this embodiment, terrain interference is eliminated through correction using spectral reflectance and backscattering coefficient. It integrates the type identification advantages of optical data with the moisture sensing capabilities of SAR data, and strengthens the correlation between weakly sensitive parameters and remote sensing features by constructing a spectral-scattering synergistic factor. This forms a synergistic factor that combines moisture sensitivity and structural identification, enhancing the correlation between weakly sensitive parameters and remote sensing features and addressing the problem of one-sided information from single-source data. In this embodiment, the dual-stage model's classification-then-regression synergistic architecture not only solves the coupling interference between type classification and parameter inversion but also enhances the inversion capability of weakly sensitive parameters, adapting to complex power transmission channel scenarios.

[0031] Based on the first aspect of the present invention, the second aspect of the present invention provides a more detailed description of a remote sensing inversion method for the flammability risk of power transmission channels. The second aspect of the present invention provides a remote sensing inversion method for the flammability risk of power transmission channels, comprising the following steps: Step 1: Multi-source data acquisition and preprocessing Step 21: Preprocess the multispectral images, SAR fully polarimetric images and DEM data of the target power transmission channel area and extract the target optical spectral features, target SAR scattering features and target terrain features.

[0032] Data Acquisition: Optical data: Acquire high-resolution multispectral satellite images (spatial resolution ≤5m) along the power transmission corridor, including more than 5 bands including blue, green, red, near-infrared, and short-wave infrared, for the purpose of extracting vegetation type and spectral characteristics; SAR data: Acquire fully polarimetric SAR images (C-band or L-band), including four polarization modes: HH, HV, VH, and VV, to capture SAR scattering features related to vegetation water content; Topographic data: Acquire a 30m resolution digital elevation model (DEM); Measured data: Typical sample plots (covering combustible materials such as trees, shrubs, herbs, and fallen leaves) were set up along the power transmission channel. The moisture content (by drying and weighing method), load (by quadrat harvesting method), and type information of combustible materials were measured and used as model training and validation data.

[0033] Optical data preprocessing: Radiometric calibration, atmospheric correction, geometric fine correction, super-resolution, and resampling are performed to ensure that the original data is free from interference such as atmospheric scattering and sensor errors. Indices such as NDVI, EVI, and NDMI are used to calculate and extract the vegetation spectral feature set. SAR data preprocessing: radiometric calibration, multi-look filtering, terrain correction (range-Doppler correction based on DEM), and extraction of scattering feature sets such as backscattering coefficients and polarization decomposition parameters (e.g., entropy, alpha angle); Topographic data: Extract topographic features such as slope, aspect, and elevation.

[0034] Step 22, terrain adaptation preprocessing optimization: Calculate the slope of the sample points based on the DEM data, perform slope correction optimization on the spectral characteristics of the optical data, and perform local incident angle correction optimization on the backscattering coefficient of the SAR data to improve the elimination of terrain interference on remote sensing signals.

[0035] Specifically, step 22 includes steps 221-224: steps for slope correction optimization of the spectral features of the optical data. Step 221: Determine the terrain correction coefficient k by fitting measured data.

[0036] In this step, the spectral reflectance of vegetation in multispectral images is measured using a ground-based spectrometer and used as the true value.

[0037] Step 222: Based on multispectral image analysis, obtain combustible material areas (such as pine forest areas, shrubland areas, and herbaceous areas) in the target power transmission channel area, and simultaneously acquire the reflectance A (the spectral reflectance of blue, green, red, near-infrared, and short-wave infrared bands extracted pixel by pixel in step 21) and the slope angle θ calculated from the topographic features of the combustible material areas.

[0038] In this step, the slope angle θ is calculated based on 30m resolution digital elevation model (DEM) data. The slope distribution map along the power transmission channel is generated pixel by pixel using the slope analysis tool of GIS software.

[0039] Step 223: Using A′=Acosθ+ksinθ as the objective function, the least squares method is used to fit the terrain correction coefficient k corresponding to different combustible material areas.

[0040] Here, different terrain correction coefficients are fitted to different areas based on the type of combustible material. For example, k=0.12 for the Pinus tabuliformis forest area, k=0.08 for the shrubland area, and k=0.05 for the herbaceous area. The fitting error is controlled within ±0.01.

[0041] Here, data for each type of combustible material are fitted and calculated separately.

[0042] Step 224: Determine the corrected reflectance based on the reflectance in the target optical spectral characteristics, the slope angle calculated from the target terrain characteristics, and the terrain correction coefficient k, so as to obtain the corrected optical spectral characteristics.

[0043] In this step, the corrected reflectance B′ of each pixel is calculated using the formula B′=Acosθ+ksinθ, based on the terrain correction coefficient k for different combustible material areas. B′=Acosθ+ksinθ is used to eliminate the interference of terrain slope on the spectral reflectance of optical satellite data.

[0044] In the calculation, the slope angle θ is converted from degrees (°) to radians (rad) to ensure the accuracy of the trigonometric function calculation. Acosθ is used to correct for changes in vegetation light area caused by slope. For example, when A=0.6 and θ=30°, Acosθ=0.6×0.866≈0.5196. ksinθ is used to compensate for shading and differences in slope reflectivity caused by slope. For example, when k=0.12 and θ=30°, ksinθ=0.12×0.5=0.06.

[0045] The calculated corrected reflectance, together with other features in the target optical spectral characteristics, constitutes the corrected optical spectral characteristics.

[0046] It should be noted that traditional optical correction uses fixed empirical formulas and does not consider the heterogeneity of the slope of the transmission channel, resulting in a 15%-20% deviation in reflectance after correction. The optical spectral slope correction formula in this embodiment (B′=Acosθ+ksinθ) fits a dynamic coefficient k to measured data. Based on measured plots along the line with slopes from 0° to 60°, the least squares method is used to fit k values ​​of 0.12, 0.08, and 0.05 for the Pinus tabuliformis forest area, shrubland area, and herbaceous area, respectively, completely solving the local distortion problem of the traditional method of "correcting all terrains with the same coefficient." Simultaneously, the formula integrates physical mechanisms and measured data: Acosθ corrects for changes in illuminated area, and ksinθ compensates for differences in shadows and slope reflectance, reducing the deviation between the corrected reflectance and the measured value from ±12% to within ±3%. Ultimately, topographic interference in optical reflectance is reduced by 35%, and the accuracy of vegetation identification in mountainous areas with slopes >25° is improved by 28%, solving the problem of "different spectra for the same object, and the same spectrum for different objects."

[0047] Specifically, step 22 also includes steps 225-227: a step of optimizing the SAR data backscattering coefficients by local incident angle correction, used to eliminate the influence of local incident angle differences on the SAR data backscattering coefficients. The calculation process needs to combine satellite parameters and terrain data to ensure the consistency of the scattered signals. Step 225: Obtain the backscattering coefficients of the four polarization modes HH, HV, VH and VV extracted in step 21, in dB or linear values ​​(linear values ​​are used uniformly in the calculation).

[0048] Reference incident angle (θ0): The fixed value is 20°. This angle has been verified by actual measurement to be the most sensitive to changes in the moisture content of combustibles and is suitable for satellite observation conditions in most power transmission channel areas.

[0049] Satellite orbital parameters (such as azimuth and zenith angle) are extracted from SAR image header files. At the same time, based on 30m resolution DEM data, the normal direction of each pixel is calculated using GIS software.

[0050] Step 226: Combining the satellite orbit direction and the pixel normal direction, calculate the local incident angle θ using spherical trigonometry formulas. i The accuracy reaches ±0.5°, and the value range is usually 10°-45°.

[0051] Step 227, based on the preset reference incident angle θ0 and the local incident angle θ i The corrected backscattering coefficient σ′ for each pixel is determined by the backscattering coefficient σ of the target SAR scattering feature, so as to obtain the corrected SAR scattering feature.

[0052] In this step, according to the formula σ′=σ(cosθ0 / cosθ) i Calculate the corrected backscattering coefficient σ′.

[0053] Angle conversion: Convert θ0 (20°) to θ i All values ​​are converted to radians (rad). For example, θ0 = 20° is approximately 0.349 rad after conversion. i =30° after conversion ≈0.523rad, cosθ0≈0.9397, cosθ i When σ = 0.02 (linear value), σ′ = 0.02 × 1.085 ≈ 0.0217.

[0054] The calculated corrected backscattering coefficient, together with other features of the target SAR scattering characteristics, constitutes the corrected SAR scattering characteristics.

[0055] Calculation result verification and adjustment Select sample plots of the same vegetation type but different slopes to verify the coefficient of variation of σ′ after correction. It needs to be reduced from the original 0.42 to below 0.15. If the dispersion of σ′ for the same vegetation type is still high, θ needs to be rechecked. i The calculation accuracy is ensured to match the satellite orbital parameters with the DEM data, and the value of θ0 is adjusted when necessary (it needs to be verified by measured data simultaneously).

[0056] In this embodiment, the SAR backscattering coefficient is sensitive to the incident angle, and the signal difference for the same ground object can reach 30%-50%. Since the power transmission channel is distributed along the mountain range, the incident angle varies drastically, and traditional global normalization correction is insufficient to eliminate local differences. The SAR local incident angle correction formula in this embodiment is (σ′=σ(cosθ0 / cosθ)). i ), Calculate the local incident angle θ pixel by pixel based on high-resolution DEMi By combining satellite orbit parameters, the accuracy reaches ±0.5°, far superior to the traditional ±3° error. Field measurements verified that setting the reference incident angle θ0 to 20° (highest sensitivity to water content and suitable for most areas) reduced the coefficient of variation of the HV polarization scattering coefficient for the same vegetation from 0.42 to 0.15 after correction, resolving the "scattering signal dispersion" problem. Ultimately, the SAR incident angle influence elimination rate exceeded 80%, and the consistency with optical data fusion improved by 40%.

[0057] In this embodiment, the reflectivity and backscattering coefficient are dual-corrected using shared DEM topographic parameters, ensuring spatial consistency of the correction. For example, in a 30° slope Pinus tabuliformis forest area, the correlation between the corrected optical near-infrared reflectivity and the SARHV polarization coefficient reaches 0.78, laying the foundation for the construction of the spectral-scattering synergistic factor and increasing the water content inversion R² from 0.65 to over 0.85. Furthermore, key parameters (k value, θ0) have been optimized through field feedback. Validated in multiple plots, the k value fitting error is ±0.01, and the SAR scattering coefficient deviation from the measured value is within ±5%, balancing accuracy and engineering practicality. This embodiment eliminates interference from the data source, achieving a breakthrough in scale compared to traditional techniques, and constructs a correction system adapted to power transmission channels. This is a core innovation for high-precision inversion of combustibles in complex terrain.

[0058] Step 2: Multi-source feature optimization and fusion Step 23: Select features that meet the first preset conditions from the corrected remote sensing features and terrain features, and perform standardization processing to obtain the first standardized optimal feature set.

[0059] The first preset condition is that the correlation with the type of combustible material, the moisture content of the combustible material, or the amount of combustible material is greater than or equal to 0.3 and the correlation ranking is within the first preset number.

[0060] In this step, corrected optical spectral features, corrected SAR scattering features, and topographic features are integrated to form an initial feature set (containing 30+ feature variables). Feature selection is then performed on the initial feature set using a two-step method: correlation analysis followed by random forest importance ranking to select the optimal features. Step 1: Use Pearson correlation analysis to screen features with a correlation greater than or equal to 0.3 with the type of combustible material, the moisture content of the combustible material, or the load of the combustible material, and remove features with a correlation less than 0.3 with the combustible material parameters (type, moisture content, load).

[0061] Step 2: Calculate the importance scores of the filtered features using the random forest algorithm, and select the top 15 features ranked from high to low importance scores (such as NDMI normalized difference moisture index, VV polarization backscattering coefficient, polarization decomposition entropy value, slope-corrected near-infrared reflectance, etc.) to construct a fusion feature set; perform Z-score standardization on the fusion feature set to eliminate dimensional differences and prepare it for model input.

[0062] Step 3: Construction of the Two-Stage Collaborative Inversion Model Step 24: Input the multispectral image, which has undergone radiometric calibration, atmospheric correction, geometric fine correction, super-resolution, and resampling preprocessing, and the corresponding first standardized optimal feature set into the target CNN classification model for classification, and output the combustible type of each pixel.

[0063] In this step, the CNN classification model needs to be trained first to obtain the target CNN classification model. The training process is as follows: First, obtain the dataset. Use the same method as in step 21 to obtain the training dataset. For example, select a typical section of the power transmission channel (longitude 115°~117°, latitude 40°~42°). This area traverses complex terrain such as mountains, hills, and forests. There are abundant types of combustible materials along the line (trees are mainly Pinus tabuliformis and Larch, shrubs are mainly Hippophae rhamnoides and Vitex negundo, and herbs are mainly Stipa and Leymus chinensis). Wildfires occur frequently.

[0064] Optical data: Acquired multispectral images (spatial resolution 5m, including 5 bands) in June 2025 (vegetation growing season). SAR data: Acquire SAR fully polarimetric images (C-band, spatial resolution 10m), with an imaging time difference of ≤7 days from optical images; Topographic data: DEM data (30m resolution) was used. Measured data: In June 2025, 50 typical sample plots (20m×20m each) were set up in this area. The types of combustibles, moisture content (drying method), and load (harvesting method) were measured. The latitude, longitude, slope, and aspect of the sample plots were recorded.

[0065] The training data is preprocessed and features are extracted according to the preprocessing method in step 21. Then, terrain adaptation preprocessing optimization is performed according to step 22. Next, 15 core features (including NDMI, VV polarization coefficient, entropy value, and near-infrared reflectance after slope correction) are selected according to the method in step 23. The multispectral imagery, after radiometric calibration, atmospheric correction, geometric fine correction, super-resolution, and resampling preprocessing, and its corresponding core features are used as the final training dataset. The training dataset is input into a lightweight CNN (convolutional neural network) for training. The measured sample plot type data is used as the true label, and the training and validation sets are divided in a 7:3 ratio. An optimizer is used to train the model until the validation set accuracy is ≥92%. Here, during training, the feature data output from the pooling layer is input into the fully connected layer, converted into a one-dimensional feature vector, and concatenated with terrain adaptation factors (slope angle normalized value and slope aspect encoded value) for classification, outputting the combustible material type (trees, shrubs, herbs, fallen leaves and branches, bare ground). After training, the target CNN classification model is obtained.

[0066] Step 25: Determine the spectral-scattering synergy factor based on the NDMI value of the corrected remote sensing features and the corrected backscattering coefficient.

[0067] Specifically, the spectral-scattering coordination factor is determined based on the NDMI value of the corrected optical spectral characteristics, the corrected backscattering coefficients of HV, HH, and VV of the corrected SAR scattering characteristics. The spectral-scattering coordination factor is calculated using the following formula: F=(NDMI×σ HV ) / (σ HH +σ VV ) Where, σ HV The corrected backscattering coefficient of HV, representing the corrected SAR scattering characteristics, and σ HH The corrected backscattering coefficient HH, representing the corrected SAR scattering characteristics, and σ VV VV represents the corrected backscattering coefficient for correcting SAR scattering characteristics.

[0068] Here, when calculating NDMI, an NDMI layer is generated to ensure the value range and invalid values ​​from cloud shadows and bare ground areas are removed. From the corrected SAR scattering features, backscattering coefficient layers corresponding to HH, HV, and VV polarizations are extracted separately to ensure no terrain scattering residue. The two types of data are then aligned for consistency. Resolution unification: Both the NDMI layer (optical data) and the SAR polarization coefficient layer (SAR data) were resampled to 10m (to match the resolution of the final inversion result). Spatial coordinate matching: Based on latitude and longitude coordinates, the two types of data are aligned at the pixel level (error ≤ 0.5 pixels) to avoid spatial misalignment that could lead to deviations in F-value calculation.

[0069] For each pixel to be calculated along the power transmission channel, the above parameters are extracted to calculate the spectral-scattering synergistic factor F.

[0070] In this step, numerator = NDMI × σ HV To enhance the correlation of moisture signals by coupling the macroscopic moisture characteristics (NDMI) of optical data with the sensitive moisture characteristics (σHV) of SAR data; denominator = σ HH +σ VV To integrate the structural features of SAR data, the interference of vegetation structure on moisture signals is normalized; the F-value is calculated to eliminate structural interference by division and retain the core correlation of moisture features.

[0071] Preferably, outlier handling is required during this step (to ensure the validity of the F-value): Minimum value handling of denominator: If the denominator is ≤0.001dB (caused by noise), replace it with the "average value of the denominator of the same type of combustible material area" (e.g., take the average value of the denominator of the surrounding 10 tree pixels in the tree area). F-value boundary clipping: If a pixel's F > 0.5 or F < -0.5, clip to the boundary value (0.5 or -0.5) and mark it as the area to be checked; Spatial continuity check: If the difference between the F value of a pixel and the mean value of the surrounding 3×3 window is >0.3, replace it with the mean value of the window to ensure that the spatial distribution of the F value conforms to the natural law of vegetation moisture.

[0072] After the calculation is completed, the effective F values ​​of all pixels are integrated to generate a "co-factor F feature layer" with a resolution of 10m, which serves as one of the core input features of the subsequent regression model.

[0073] Step 26: Select features that meet the second preset conditions from the spectral-scattering synergy factor, corrected remote sensing features, and terrain features, and perform standardization processing to obtain the second standardized optimal feature set.

[0074] The second preset condition is that the correlation with the moisture content or load of combustibles is greater than or equal to 0.3 and the correlation ranking is within the first preset number.

[0075] In this step, the spectral-scattering synergy factor, corrected optical spectral features (such as near-infrared reflectance after slope correction, EVI), corrected SAR scattering features (such as polarization decomposition entropy value), and topographic features (slope and aspect coding values) are integrated to form an initial feature set (multiple features).

[0076] The initial feature set is screened by firstly filtering features that are correlated with the moisture content or load of combustibles by 0.3 or greater using Pearson correlation analysis, and then removing features that are correlated with the moisture content or load of combustibles by less than 0.3.

[0077] Step 2: Calculate the importance scores of the filtered features using the Random Forest algorithm, and select the top 15 features ranked from highest to lowest importance score (F-values ​​are usually ranked in the top 3 to ensure core status). For these 15 core features containing F-values, Z-score standardization is applied: Χstd=(Χ-μ) / σ; Where Χ is the original feature value (e.g., F), μ is the mean of the feature in the set, and σ is the standard deviation of the feature in the set.

[0078] Step 27: Divide the second standardized optimal feature set into feature matrices corresponding to the combustible material type according to the combustible material type.

[0079] Based on the classification results obtained in step 24, the second standardized optimal feature set is divided into feature matrices according to the type of combustible material, such as "herbaceous feature matrix" and "tree feature matrix".

[0080] Step 28: Input each feature matrix into the target gradient boosting tree regression model corresponding to each combustible material type, and output the predicted moisture content and predicted combustible material load for each pixel.

[0081] Before this step, a gradient boosting tree regression model is pre-trained to obtain the target gradient boosting tree regression model. During training, the training data is preprocessed according to step 21, then the spectral-scattering co-factor is calculated according to step 25, and then processed according to steps 26 and 27 to obtain the feature matrix. Each feature matrix is ​​matched with the corresponding measured label (e.g., the herbaceous feature matrix is ​​matched with the measured herbaceous water content / load of the sample plot), forming the training data pair (feature matrix + label vector) for the regression model. The feature matrix of each type is divided into a training set (for model learning) and a validation set (for accuracy evaluation) in a 7:3 ratio. The training set data is input, regression training is started, and the root mean square error and feature importance of the validation set are monitored in real time. After training, the feature importance ratio of the F value is checked. The F value ratio reaches 18% (second only to NDMI's 22%), indicating that the model effectively captures the contribution of the F value. Finally, the water content inversion R² ≥ 0.85.

[0082] Here, a corresponding target gradient boosting tree regression model is trained for each type of combustible material, and each type of combustible material has its own target gradient boosting tree regression model.

[0083] When making predictions for practical applications, the corresponding trained regression model is called according to the pixel type, the feature matrix is ​​input, and the predicted water content / load is output.

[0084] Step 29: Mark the areas where the predicted moisture content and predicted load of combustibles meet the preset risk conditions as high-risk areas.

[0085] In this step, the prediction results of all pixels are integrated to generate a "moisture content distribution map" and "load distribution map" with a resolution of 10m. Areas with moisture content ≤15% and load ≥5t / ha are marked as high-risk areas.

[0086] Preferably, a 10m resolution thematic map of combustible material information in terms of "type-moisture content-load" is generated based on the output of the above model, and high-risk areas (moisture content ≤15% and load ≥5t / ha) within the transmission line corridor (within 500m on both sides of the center line) are marked.

[0087] In one feasible implementation, the accuracy of the regression model is validated: Field sampling: Select multiple typical sample plots in the predicted area and measure the moisture content of combustibles (drying and weighing method) and load (quad plot harvesting method). Error calculation: Compare the measured values ​​with the model prediction values, and calculate the relative error (≤8% is acceptable) and the coefficient of determination R² (R²≥0.85 for moisture content, R²≥0.80 for load). Troubleshooting: If the accuracy is not up to standard, first check the F-value calculation process (such as whether NDMI / SAR preprocessing is thorough and whether outlier handling is omitted), and then optimize the regression model parameters.

[0088] In this embodiment, the moisture content and load of combustible materials are weakly sensitive parameters in remote sensing, and traditional single-source data struggle to capture their complex correlation with the signal. The present invention utilizes a "spectral-scattering synergistic factor (F=(NDMI×σ)" to address this issue. HV ) / (σ HH +σ VV This step leverages the advantages of deep coupling of multi-source data. It breaks away from the traditional simple data overlay model, combining the moisture index (NDMI, reflecting the macroscopic characteristics of vegetation moisture) of optical data with the polarization characteristics (σ) of SAR data. HV Sensitive to moisture, σ HH σ VV By integrating structural information through formula calculations, a synergistic factor is formed that combines moisture sensitivity and structural identification, strengthening the correlation between weakly sensitive parameters and remote sensing features and solving the problem of one-sided information from single-source data. This factor, as a core input to the regression model, increases the R² for water content inversion from 0.78 in existing technologies to over 0.86, and the R² for load inversion from 0.72 to over 0.80. Compared to traditional methods relying on a single exponent or scattering coefficient, the synergistic factor effectively integrates multi-dimensional information, capturing subtle changes in weakly sensitive parameters and providing high-precision parameter support for wildfire early warning in power transmission corridors (such as accurately identifying high-risk areas with water content ≤15%).

[0089] This invention, through innovative two-stage model architecture and collaborative factor design, not only solves the coupling interference between type classification and parameter inversion, but also enhances the inversion capability of weakly sensitive parameters, adapting to complex scenarios in power transmission channels and providing a new technical path for high-precision inversion of multiple parameters of combustibles.

[0090] This invention integrates the advantages of optical data type identification with the moisture sensing capabilities of SAR data. By constructing a "spectral-scattering synergistic factor", it strengthens the correlation between weakly sensitive parameters and remote sensing features, improving the accuracy of water content inversion by 15%-20% and the accuracy of load inversion by 12%-18%, thus solving the problem of insufficient inversion accuracy of traditional single-source data.

[0091] This invention improves adaptability to complex areas through terrain adaptive correction: by slope correction, local incident angle correction and the introduction of terrain adaptation factors into the model, the interference of mountainous terrain on remote sensing signals is effectively eliminated, the inversion error in complex terrain areas is reduced by more than 25%, and it adapts to the monitoring needs of complex terrain in power transmission channels. The two-stage model of this invention achieves multi-parameter synchronous inversion: the collaborative architecture of classification followed by regression not only ensures the accuracy of type identification, but also improves the targeting of water content and load inversion through the "type-specific" regression model, realizing the synchronous acquisition of the three core parameters without the need for multiple independent inversions, thus improving efficiency by 30%. Customized scenarios meet the monitoring needs of power transmission channels: Focusing on narrow-band monitoring scenarios of power transmission line corridors, high-resolution thematic maps are output and high-risk areas are accurately marked, providing data support for "precise positioning and targeted measures" for power grid wildfire prevention and control, and reducing the threat of wildfire transmission lines.

[0092] In one feasible implementation, an example is: A typical section of the power transmission corridor (longitude 115°~117°, latitude 40°~42°) was selected. This area traverses complex terrain such as mountains, hills, and forests. The types of combustible materials along the route are abundant (mainly Pinus tabuliformis and Larch trees, mainly Hippophae rhamnoides and Vitex negundo shrubs, and mainly Stipa argyi and Leymus chinensis herbs). Wildfires occur frequently, which meets the application scenario of this invention.

[0093] Optical data: Acquired multispectral images (spatial resolution 5m, including 5 bands) in June 2025 (vegetation growing season). SAR data: Acquire SAR fully polarimetric images (C-band, spatial resolution 10m), with an imaging time difference of ≤7 days from optical images; Topographic data: DEM data (30m resolution) was used. Measured data: In June 2025, 50 typical sample plots (20m×20m each) were set up in this area. The types of combustibles, moisture content (drying method), and load (harvesting method) were measured. The latitude, longitude, slope, and aspect of the sample plots were recorded.

[0094] The model is trained according to the above steps. After training, the multi-source data of the input area is used for inversion to generate a 3D thematic map of combustible material information with a resolution of 10m. Three high-risk areas near the transmission line are marked (mixed areas of herbaceous plants and dead branches and leaves, with a moisture content of 12%-14% and a load of 5.2-6.8t / ha).

[0095] By conducting on-site verification of three high-risk areas, the relative errors between the measured moisture content and load values ​​and the inversion results were all ≤8%, and the type identification was completely accurate, verifying the high precision and practicality of the method of the present invention.

[0096] As shown in Table 1, Table 1 is the evaluation index of the inversion results of multiple sets of data in this invention.

[0097] This invention can achieve synchronous and high-precision inversion of combustible material type, moisture content, and load under complex terrain conditions, providing data support for early warning of wildfires in power transmission channels and improving the pertinence and timeliness of wildfire prevention and control in power grids.

[0098] like Figure 2 As shown, a third aspect of the present invention provides a remote sensing inversion device for the combustible risk of power transmission channels, comprising: The extraction and correction module 31 is used to extract the corresponding features of the remote sensing image and DEM data of the target power transmission channel area, and correct the spectral reflectance and backscattering coefficient of the remote sensing image according to the topographic features of the corresponding features to obtain the corrected remote sensing features. The first screening module 32 is used to screen features that meet the first preset conditions from the corrected remote sensing features and terrain features and perform standardization processing to obtain the first standardized optimal feature set. The classification module 33 is used to input the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set into the target CNN classification model for classification, and output the combustible type of each pixel; Module 34 is used to determine the spectral-scattering synergy factor based on the NDMI value and the corrected backscattering coefficient of the corrected remote sensing features; The second screening module 35 is used to screen features that meet the second preset conditions from the spectral-scattering synergy factor, corrected remote sensing features and terrain features, and perform standardization processing to obtain the second standardized optimal feature set. The partitioning module 36 is used to partition the second standardized optimal feature set into feature matrices corresponding to the combustible material type according to the combustible material type. Prediction module 37 is used to input each feature matrix into the target gradient boosting tree regression model corresponding to each combustible type, and output the predicted moisture content and predicted combustible load of each pixel. The processing module 38 is used to mark the areas where the predicted moisture content and predicted load of combustibles meet the preset risk conditions as high-risk areas.

[0099] In one embodiment of the present invention, the remote sensing image includes: multispectral image and SAR fully polarimetric image; the corresponding features include: target optical spectral features, target SAR scattering features and target terrain features; the corrected remote sensing features include: corrected optical spectral features and corrected SAR scattering features. The corresponding features of remote sensing images and DEM data of the target power transmission channel area are extracted. Based on the topographic features of the corresponding features, the spectral reflectance and backscattering coefficient of the remote sensing images are corrected to obtain the corrected remote sensing features, including: Multispectral images, SAR all-polarization images and DEM data of the target power transmission channel area are preprocessed and target optical spectral features, target SAR scattering features and target terrain features are extracted. Based on the target's optical spectral characteristics and topographic features, with A′=Acosθ+ksinθ as the objective function, the least squares method is used to fit the topographic correction coefficient k corresponding to the combustible material type; where A represents the reflectance in the target's optical spectral characteristics, θ represents the slope angle calculated from the target's topographic features, and A′ represents the measured reflectance obtained using a ground spectrometer. The corrected reflectance is determined based on the reflectance in the target optical spectral characteristics, the slope angle calculated from the terrain characteristics, and the terrain correction coefficient k, so as to obtain the corrected optical spectral characteristics. Determine the normal direction of each pixel based on the DEM data; The local incident angle of each pixel center point is determined based on the normal direction and the satellite orbit direction of the SAR fully polarimetric image; The corrected backscattering coefficient of each pixel is determined based on the preset reference incident angle, local incident angle, and backscattering coefficient of the target SAR scattering feature, so as to obtain the corrected SAR scattering feature.

[0100] In one embodiment of the present invention, the formula for calculating the corrected reflectivity is: B′=Acosθ+ksinθ Where B′ represents the corrected reflectance; The formula for calculating the corrected backscattering coefficient is: σ′=σ(cosθ0 / cosθ i ) Where σ′ represents the corrected backscattering coefficient, θ0 represents the preset reference incident angle, and θ i σ represents the local incident angle, and σ represents the backscattering coefficient of the target's SAR scattering characteristics.

[0101] In one embodiment of the present invention, when the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set are input into the target CNN classification model for processing, the feature data output by the pooling layer is input into the fully connected layer and converted into a one-dimensional feature vector, which is then concatenated with the terrain adaptation factor for classification; wherein, the terrain adaptation factor includes the slope angle normalized value and the slope aspect encoded value.

[0102] In one embodiment of the present invention, determining the spectral-scattering synergy factor based on the NDMI value of the corrected remote sensing features and the corrected backscattering coefficient includes: The spectral-scattering synergy factor is determined based on the NDMI value of the corrected optical spectral characteristics, the corrected backscattering coefficient of HV of the corrected SAR scattering characteristics, the corrected backscattering coefficient of HH, and the corrected backscattering coefficient of VV.

[0103] In one embodiment of the present invention, the formula for calculating the spectral-scattering synergy factor is as follows: F=(NDMI×σ HV ) / (σ HH +σ VV ) Where, σ HV The corrected backscattering coefficient of HV, representing the corrected SAR scattering characteristics, and σ HH The corrected backscattering coefficient HH, representing the corrected SAR scattering characteristics, and σ VV VV represents the corrected backscattering coefficient for correcting SAR scattering characteristics.

[0104] In one embodiment of the present invention, the label vector in the training process of the target gradient boosting tree regression model is the measured moisture content and measured load of combustible material, and the training feature matrix and label vector are used as training data pairs during the training process.

[0105] A third aspect of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the remote sensing inversion method for flammable risk of power transmission channels provided by the present invention.

[0106] A fourth aspect of the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the remote sensing inversion method for flammable risk of power transmission channels provided in the above-described embodiments of the present invention.

[0107] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0108] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware devices.

[0109] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0110] For the device / electronic device embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.

[0111] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (devices), and computer program products according to embodiments of the invention. It will be understood that each block of 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, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0112] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0113] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0114] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for remote sensing inversion of combustible risk of power transmission corridor, characterized in that, The method comprises the following steps: extracting corresponding features of remote sensing images and DEM data of a target power transmission channel area, correcting the spectral reflectivity and backscattering coefficient of the remote sensing images according to the terrain features of the corresponding features, and obtaining corrected remote sensing features; screening features meeting a first preset condition from the corrected remote sensing features and the terrain features, and performing standardization processing to obtain a first standardized optimal feature set; inputting the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set into a target CNN classification model for classification, and outputting the combustible material type of each pixel; determining a spectrum-scattering coordination factor according to the NDMI value of the corrected remote sensing features and the corrected backscattering coefficient; screening features meeting a second preset condition from the spectrum-scattering coordination factor, the corrected remote sensing features and the terrain features, and performing standardization processing to obtain a second standardized optimal feature set; dividing the second standardized optimal feature set into feature matrices corresponding to the combustible material types according to the combustible material types; inputting each feature matrix into a target gradient boosting tree regression model of each combustible material type to output the combustible material predicted moisture content and the combustible material predicted load of each pixel; labeling the area where the pixels meeting the preset risk condition are located as a high-risk area.

2. The method of claim 1, wherein, The remote sensing image comprises a multispectral image and a SAR full polarization image; the corresponding features comprise target optical spectrum features, target SAR scattering features and target terrain features; and the corrected remote sensing features comprise corrected optical spectrum features and corrected SAR scattering features. The step of extracting corresponding features of remote sensing images and DEM data of a target power transmission channel area, correcting the spectral reflectivity and backscattering coefficient of the remote sensing images according to the terrain features of the corresponding features, and obtaining corrected remote sensing features comprises the following steps: preprocessing the multispectral image, the SAR full polarization image and the DEM data of the target power transmission channel area and extracting target optical spectrum features, target SAR scattering features and target terrain features; determining a terrain correction coefficient k corresponding to the combustible material type by adopting a least square method to fit the terrain correction coefficient k according to the target optical spectrum features and the target terrain features, wherein A represents the reflectivity in the target optical spectrum features, θ represents a slope angle calculated from the target terrain features, and A' represents a measured reflectivity obtained by using a ground spectrometer; determining a corrected reflectivity according to the reflectivity in the target optical spectrum features, the slope angle calculated from the terrain features and the terrain correction coefficient k to obtain corrected corrected optical spectrum features; determining a normal direction of each pixel according to the DEM data; determining a local incidence angle of a center point of each pixel according to the normal direction and a satellite orbit direction of the SAR full polarization image; determining a corrected backscattering coefficient of each pixel according to a preset reference incidence angle, the local incidence angle and a backscattering coefficient of the target SAR scattering features to obtain corrected corrected SAR scattering features.

3. The method of claim 2, wherein, The calculation formula of the corrected reflectivity is: B' = A cos θ + k sin θ wherein B' represents the corrected reflectance; The calculation formula of the corrected backscattering coefficient is: σ' = σ (cos θ0 / cos θ i ) where σ' denotes the corrected backscatter coefficient, θ0denotes a pre-set reference incidence angle, θ i denotes the local incidence angle, and σ denotes the backscatter coefficient of the target SAR scattering feature.

4. The method of claim 1, wherein, When the multispectral image of the remote sensing image and the corresponding first standardized optimal feature set are input into the target CNN classification model for processing, the feature data output by the pooling layer is converted into a one-dimensional feature vector after being input into the full connection layer, and then the one-dimensional feature vector is spliced with the terrain adaptation factor for classification; wherein the terrain adaptation factor includes a slope angle normalized value and a slope direction encoding value.

5. The method of claim 2, wherein, The determination of the spectrum-scattering synergy factor according to the NDMI value and the corrected backscattering coefficient of the corrected remote sensing feature includes: The determination of the spectrum-scattering synergy factor according to the NDMI value of the corrected optical spectrum feature, the corrected backscattering coefficient of HV, the corrected backscattering coefficient of HH, and the corrected backscattering coefficient of VV of the corrected SAR scattering feature.

6. The method of claim 5, wherein, The calculation formula of the spectrum-scattering synergy factor is: F = (NDMI x σ HV ) / (σ HH + σ VV ) where σ HV represents the corrected backscatter coefficient for HV of the corrected SAR scattering feature, σ HH represents the corrected backscatter coefficient for HH of the corrected SAR scattering feature, σ VV represents the corrected backscatter coefficient for VV of the corrected SAR scattering feature.

7. The method of claim 1, wherein, The label vector in the training process of the target gradient boosting tree regression model is the measured moisture content of the combustible material and the measured load of the combustible material, and the training feature matrix and the label vector are used as training data in the training process.

8. A device for remote sensing inversion of combustible risk of a power transmission corridor, characterized in that, It includes: The extraction correction module is configured to extract corresponding features of remote sensing images and DEM data of a target power transmission channel area, correct the spectral reflectance and backscattering coefficient of the remote sensing images according to terrain features of the corresponding features, and obtain corrected remote sensing features. The first screening module is configured to screen features that meet a first preset condition from the corrected remote sensing features and the terrain features, and perform standardization processing to obtain a first standardized optimal feature set. The classification module is configured to input multispectral images of remote sensing images and the corresponding first standardized optimal feature set into a target CNN classification model for classification, and output the combustible material type of each pixel. The determination module is configured to determine a spectrum-scattering synergy factor according to an NDMI value and a corrected backscattering coefficient of the corrected remote sensing feature. The second screening module is configured to screen features that meet a second preset condition from the spectrum-scattering synergy factor, the corrected remote sensing feature, and the terrain feature, and perform standardization processing to obtain a second standardized optimal feature set. The division module is configured to divide the second standardized optimal feature set into a feature matrix corresponding to the combustible material type according to the combustible material type. The prediction module is configured to input each feature matrix into a target gradient boosting tree regression model of each combustible material type to output a combustible material predicted moisture content and a combustible material predicted load of each pixel. The processing module is configured to mark a region where a pixel meets a preset risk condition as a high-risk region.

9. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor executes the program to implement the power transmission channel combustible risk remote sensing inversion method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the power transmission channel combustible risk remote sensing inversion method of any one of claims 1 to 7.

Images