Plateau complex climate environment under airborne millimeter wave sar disaster identification system and method

Abstract

The application discloses a highland complex climate environment airborne millimeter wave SAR disaster identification system and method, relates to the technical field of disaster identification, and comprises a data acquisition module, a meteorological correction module, a SAR imaging module, a feature extraction module and a lightweight disaster identification module.The meteorological correction module is used for constructing a SAR signal highland meteorological correction model, correcting airborne millimeter wave SAR echo data, and obtaining first corrected millimeter wave SAR data.The SAR imaging module is used for correcting the first corrected millimeter wave SAR data based on target region DEM data correction, and generating a millimeter wave SAR image.The feature extraction module is used for constructing a scattering-texture-terrain multi-dimensional coupling feature set.The lightweight disaster identification module is used for identifying and outputting a disaster region segmentation graph and a pre-disaster and post-disaster change region graph.The disaster identification system realizes real-time correction of SAR data by integrating meteorological correction into a SAR imaging chain, improves the accuracy of SAR imaging, and improves the precision and real-time performance of highland disaster identification under complex terrain conditions by using the constructed multi-dimensional coupling feature set and the lightweight disaster identification model.

Description

Technical Field

[0001] This invention relates to the field of disaster identification technology, and more specifically, to an airborne millimeter-wave SAR disaster identification system and method for complex climate environments in high-altitude regions. Background Technology

[0002] Plateau regions (such as the Qinghai-Tibet Plateau) are important national ecological barriers and strategic locations. They have complex geological structures and fragile ecological environments, and are prone to natural disasters such as landslides, debris flows, and barrier lakes. These disasters pose a serious threat to the safety of people's lives and property and the operation of major projects. There is an urgent need for efficient and accurate disaster identification and monitoring technologies to respond quickly to natural disasters occurring in plateau regions.

[0003] However, existing disaster monitoring technologies have significant limitations when applied to disaster identification on plateaus. For example, optical remote sensing is heavily dependent on lighting and weather conditions, while plateaus are covered by clouds, fog, rain and snow all year round, resulting in a large number of blind spots for optical remote sensing on plateaus, which cannot meet the emergency needs of all-weather and all-time. Although spaceborne synthetic aperture radar (SAR) has all-weather capabilities, its long revisit period, limited spatial resolution and fixed orbit make it difficult to achieve rapid response and detailed observation of sudden disasters.

[0004] For millimeter-wave SAR-related disaster identification technologies, their universality and practicality face multiple bottlenecks in high-altitude environments: First, millimeter-wave signals are easily affected by low air pressure, uneven water vapor, and strong convective precipitation in high-altitude areas, resulting in severe attenuation and phase disturbances, leading to blurred SAR images and / or geometric distortion, directly affecting the accuracy of subsequent information extraction. For example, Chinese patent CN119360224A, when processing radar data from permafrost regions in high-altitude areas, only uses a general radar preprocessing procedure without considering the impact of complex environments such as low temperature and low air pressure on radar imaging accuracy, making its radar imaging accuracy susceptible to weather conditions. Second, the dramatic topographic relief in high-altitude areas easily causes geometric distortions such as SAR image overlay and shadows, making it difficult to discern the boundaries and shapes of disaster bodies. Accurate interpretation is required; thirdly, the backscattering characteristics of snow, glaciers, bare rock and disaster-damaged bodies are similar, and traditional interpretation methods based on single amplitude information are prone to misjudgment and omission. For example, Chinese patent CN119961783A only extracts spatial features of SAR data based on convolutional neural networks, ignoring the distinction between easily confused features such as plateau snow, glaciers and bare rock, and also failing to consider the strong support of other features in SAR data for disaster identification, resulting in insufficient model generalization ability; fourthly, existing algorithms are mostly designed for specific regions or single disaster types, lacking adaptability to plateau scenarios, with weak generalization ability, and most algorithms rely on expert manual interpretation, with low levels of automation and intelligence, failing to meet the timeliness requirements of disaster emergency response. Summary of the Invention

[0005] The purpose of this invention is to provide an airborne millimeter-wave SAR disaster identification system and method for complex climate environments in high-altitude areas, so as to solve the problems of low imaging accuracy and insufficient disaster identification accuracy of traditional disaster identification technologies in complex climate environments in high-altitude areas as mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments includes:

[0008] The data acquisition module is used to acquire airborne millimeter-wave SAR echo data, plateau meteorological parameters of the flight area, and digital elevation model (DEM) data of the target area;

[0009] The meteorological correction module is used to construct a high-altitude meteorological correction model for SAR signals based on the high-altitude meteorological parameters of the flight area, and to correct the airborne millimeter-wave SAR echo data based on the high-altitude meteorological correction model for SAR signals to obtain the first corrected millimeter-wave SAR data.

[0010] The SAR imaging module is used to correct the first corrected millimeter-wave SAR data based on the digital elevation model (DEM) data of the target area to obtain the second corrected millimeter-wave SAR data, and generate a millimeter-wave SAR image based on the second corrected millimeter-wave SAR data.

[0011] The feature extraction module is used to extract amplitude statistical features, polarization decomposition features, coherence / interference features and texture features from the second corrected millimeter-wave SAR data, extract key terrain factors based on the target area digital elevation model (DEM) data, and construct a scattering-texture-terrain multidimensional coupled feature set based on the amplitude statistical features, polarization decomposition features, coherence / interference features, texture features and key terrain factors.

[0012] The lightweight disaster identification module is used to identify disaster areas and pre- and post-disaster change areas based on the scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image, and output disaster area segmentation map and pre- and post-disaster change area map.

[0013] The principle of the airborne millimeter-wave SAR disaster identification system in the complex climate environment of the plateau is as follows: The data acquisition module simultaneously acquires airborne millimeter-wave SAR echo data, plateau meteorological parameters, and digital elevation model (DEM) data of the target area; Addressing the attenuation and scattering interference of SAR signals caused by the complex weather conditions of the plateau, the meteorological correction module constructs a plateau-specific SAR signal meteorological correction model using the acquired plateau meteorological parameters and introduces it into the SAR imaging processing chain to perform the first correction on the airborne millimeter-wave SAR echo data, obtaining first-corrected millimeter-wave SAR data with reduced meteorological influence; Considering the large topographic relief in the plateau region, the SAR imaging module combines the DEM data to perform terrain correction on the first-corrected millimeter-wave SAR data, correcting the signal distortion caused by the terrain, and generating second-corrected millimeter-wave SAR data. Finally, the millimeter-wave SAR image generated from the second-corrected millimeter-wave SAR data can greatly improve the imaging accuracy and precision. Subsequently, the feature extraction module extracts multiple SAR features, such as amplitude statistical features, from the second-corrected millimeter-wave SAR data and extracts key terrain factors from the DEM data. The multiple SAR features and key terrain factors are fused to construct a scattering-texture-terrain multidimensional coupled feature set, providing high-quality input for the subsequent disaster identification network. Finally, the lightweight disaster identification module uses the scattering-texture-terrain multidimensional coupled feature set and the high-quality millimeter-wave SAR image as input to complete the segmentation of plateau disaster areas and the identification of pre- and post-disaster change areas, outputting intuitive disaster area segmentation maps and post-disaster change area maps, achieving high efficiency, timeliness, and accuracy in disaster identification under the complex climate environment of the plateau.

[0014] Preferably, in order to fully understand the complex climate environment of the plateau region and thus provide plateau-specific meteorological parameters for correcting the general SAR signal correction model, the plateau meteorological parameters of the flight area include: precipitation intensity, cloud liquid water content, atmospheric water vapor content, atmospheric liquid water path, atmospheric temperature vertical profile, atmospheric pressure vertical profile, and atmospheric humidity vertical profile of the flight area; the plateau meteorological correction model of the SAR signal includes: a path integral attenuation model and a phase perturbation correction model corrected by plateau meteorological parameters.

[0015] Preferably, the method for obtaining the first corrected millimeter-wave SAR data includes:

[0016] Aligning the plateau meteorological parameters of the flight area with the airborne millimeter-wave SAR echo data in time and space generates a plateau meteorological parameter distribution field along the round-trip propagation path of the radar signal. Through time and space alignment, plateau meteorological data consistent with the time and space of the radar signal propagation path is provided for the correction of the path integral attenuation model, reducing the signal correction deviation caused by the mismatch between the meteorological parameter field and the time and space distribution of the echo data.

[0017] A path integral attenuation model and a phase perturbation correction model are constructed. Based on the plateau meteorological parameter distribution field, the path integral attenuation model and the phase perturbation correction model are modified to obtain the path integral attenuation model and the phase perturbation correction model modified by plateau meteorological parameters. The unique low-pressure effect and uneven humidity distribution of the plateau have a special attenuation effect on the propagation of radar signals. By modifying the general path integral attenuation model and the phase perturbation correction model in combination with the unique and complex meteorological distribution field of the plateau, the calculation accuracy of the correction process of SAR echo signal amplitude attenuation and phase perturbation caused by the complex meteorological conditions of the plateau can be improved, thereby enhancing the accuracy and precision of radar imaging.

[0018] Based on the path integral attenuation model and phase perturbation correction model corrected by plateau meteorological parameters, signal attenuation compensation and phase perturbation compensation are performed on the airborne millimeter-wave SAR echo data to obtain the first corrected millimeter-wave SAR data.

[0019] Preferably, the second corrected millimeter-wave SAR data is obtained in the following ways:

[0020] The target area digital elevation model (DEM) data is registered with the airborne millimeter-wave SAR echo data to obtain registered DEM data. This ensures that the DEM data and SAR echo data correspond precisely in spatial location, providing a spatially consistent data foundation for subsequent terrain-based SAR data correction and reducing correction errors caused by misalignment between terrain and echo data.

[0021] Based on the registered DEM data, key terrain factors are extracted, including: slope, aspect, local incident angle, and terrain curvature of each pixel in the millimeter-wave SAR image; terrain feature parameters that directly affect the scattering of SAR signals by plateau terrain are extracted, providing a quantitative terrain basis for subsequent terrain scattering correction.

[0022] A terrain-scattering response model is constructed, and the backscattering coefficient of the first corrected millimeter-wave SAR data is corrected based on the key terrain factors and the terrain-scattering response model to obtain the second corrected millimeter-wave SAR data. The plateau has a drastic terrain undulation. By correcting the interference of terrain undulation on the SAR echo backscattering coefficient through the terrain-scattering response model, the authenticity of the ground object scattering characteristics can be improved and the SAR signal distortion caused by terrain undulation can be reduced.

[0023] Preferably, the method for constructing the scattering-texture-terrain multidimensional coupled feature set includes:

[0024] Based on the key terrain factors corresponding to each pixel in the millimeter-wave SAR image, pixels that meet the preset terrain factor threshold for high-altitude disaster-prone areas are selected as feature pixels. By pre-screening with terrain thresholds, the spatial range of feature processing is limited, which can reduce invalid pixel data in non-disaster-prone areas of the plateau, thereby reducing the computational load and redundant information in subsequent feature processing.

[0025] The amplitude statistical features, polarization decomposition features, coherence / interference features, and texture features corresponding to each feature pixel are associated with the local incident angle, slope, aspect, and terrain curvature of that pixel, respectively. This yields amplitude statistical features labeled with local incident angle, polarization decomposition features labeled with slope, coherence / interference features labeled with aspect, and texture features labeled with terrain curvature for each feature pixel. A specific mapping relationship between each type of SAR feature and plateau terrain factors is established, so that the physical meaning of each type of feature is accurately matched with the corresponding terrain factor, thereby enhancing the coupling relationship between each feature and reducing the low processing efficiency caused by feature independence.

[0026] The amplitude statistical feature with local incident angle label, the polarization decomposition feature with slope label, the coherence / interference feature with slope aspect label, and the texture feature with terrain curvature label corresponding to each feature pixel are dimensionally concatenated to obtain the scattering-texture-terrain multidimensional coupled feature vector corresponding to each feature pixel.

[0027] The set of scattering-texture-terrain multidimensional coupled feature vectors corresponding to all the feature pixels is the scattering-texture-terrain multidimensional coupled feature set.

[0028] Preferably, the lightweight disaster identification module employs a multi-task deep learning network, which includes a shared encoder, a feature fusion module, and a multi-task decoder. The shared encoder uses a lightweight DeepLabv3+ backbone network. The feature fusion module introduces an attention mechanism, which includes a channel attention unit and a spatial attention unit. The channel attention unit is used to allocate channel weights among multiple channels of the input scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image. The spatial attention unit is used to allocate spatial weights at different spatial locations of the input millimeter-wave SAR image. The multi-task decoder includes a disaster segmentation decoding branch and a change detection decoding branch. The disaster segmentation decoding branch is used to identify the disaster area in the input feature map and output a disaster area segmentation map. The change detection decoding branch is used to compare the changed areas in the pre-disaster and post-disaster images and output a pre-disaster and post-disaster changed area map. Preferably, the lightweight DeepLabv3+ backbone network is used to adapt to the lower computing power of airborne edge devices. Through the dual-path attention and multi-task decoder architecture design, the efficiency and timeliness of the disaster identification model are further improved.

[0029] Preferably, the lightweight disaster identification module trains the multi-task deep learning network using a joint loss function, which includes a disaster area segmentation loss term and a change detection loss term. The weights of each loss term are automatically adjusted based on the combined weights of task uncertainty and the terrain factors of the currently input millimeter-wave SAR image.

[0030] Preferably, to improve the accuracy and consistency of disaster identification and change detection tasks and ensure the training stability of deep learning models in complex plateau terrain scenarios, the weights of each loss term are automatically adjusted based on the comprehensive weights of task uncertainty and the terrain factors of the currently input millimeter-wave SAR image, specifically including:

[0031] Based on the dispersion of the key terrain factors in each millimeter-wave SAR image, the comprehensive weight of the terrain factors for each millimeter-wave SAR image is calculated. Considering the large differences in the undulations of plateau terrain, the terrain complexity of the region corresponding to a single SAR image is quantified by the dispersion of the terrain factors. The more complex the terrain of the image, the higher the comprehensive weight of the corresponding terrain factors. This allows the loss weight of the model training to be adapted to the recognition difficulty of different terrains, reducing the interference of the training results of simple terrain areas on the recognition effect of complex terrain areas.

[0032] Calculate the normalized values ​​of the disaster area segmentation loss term and the change detection loss term in the current training round;

[0033] Based on the comprehensive weights of terrain factors in the millimeter-wave SAR image input in the current training round, the weights of the disaster area segmentation loss term and the change detection loss term in the previous training round, and the normalized values ​​of the disaster area segmentation loss term and the change detection loss term in the current training round, the weights of the disaster area segmentation loss term and the change detection loss term in the current training round are calculated. By dynamically adjusting the loss weights of the two types of tasks, the model automatically tilts the weights towards the more difficult tasks for identifying plateau areas with complex terrain, thereby improving the model's accuracy in disaster identification and change detection under complex terrain.

[0034] Preferably, to quantify the calculation method of the loss weights of each loss term, and to enable the model to adaptively adjust the loss weights while taking into account the terrain complexity and the balance of multi-task training during the training process, the calculation formulas for the weights of the disaster area segmentation loss term and the change detection loss term in the current training round are as follows:

[0035] ;

[0036] ;

[0037] ;

[0038] ;

[0039] In the formula, and These represent the current training round and the previous training round, respectively. and These are the weight balance coefficients for the disaster area segmentation loss item and the weight balance coefficients for the change detection loss item, respectively. and These represent the weights of the disaster area segmentation loss term and the change detection loss term for the current training round, respectively. and These represent the weights of the disaster area segmentation loss term and the change detection loss term from the previous training round, respectively. The combined weights of terrain factors in the millimeter-wave SAR image input in the current training round. and These are the normalized values ​​of the disaster area segmentation loss term and the change detection loss term, respectively, in the current training round. and These represent the disaster area segmentation loss and change detection loss in the current training round, respectively.

[0040] This invention also provides an airborne millimeter-wave SAR disaster identification method for complex high-altitude climate environments, comprising the following steps:

[0041] S1: Acquire airborne millimeter-wave SAR echo data, high-altitude meteorological parameters of the flight area, and digital elevation model (DEM) data of the target area;

[0042] S2: Align the plateau meteorological parameters of the flight area with the airborne millimeter-wave SAR echo data in time and space to generate a plateau meteorological parameter distribution field along the round-trip propagation path of the radar signal; Register the digital elevation model (DEM) data of the target area with the airborne millimeter-wave SAR echo data to obtain registered DEM data;

[0043] S3: Construct a plateau meteorological correction model for SAR signals based on the plateau meteorological parameter distribution field, and correct the airborne millimeter-wave SAR echo data based on the plateau meteorological correction model to obtain the first corrected millimeter-wave SAR data;

[0044] S4: Correct the first corrected millimeter-wave SAR data based on the registered DEM data to obtain the second corrected millimeter-wave SAR data, and generate a millimeter-wave SAR image based on the second corrected millimeter-wave SAR data;

[0045] S5: Extract the amplitude statistical features, polarization decomposition features, coherence / interference features and texture features of the second corrected millimeter-wave SAR data, extract the key terrain factors of the registered DEM data, and construct a scattering-texture-terrain multidimensional coupled feature set based on the amplitude statistical features, polarization decomposition features, coherence / interference features, texture features and key terrain factors;

[0046] S6: Based on the scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image, identify the disaster area and the area of ​​change before and after the disaster, and output the disaster area segmentation map and the area of ​​change before and after the disaster.

[0047] One or more technical solutions provided by this invention have at least the following technical effects or advantages:

[0048] 1. By integrating real-time acquired plateau meteorological parameters into the millimeter-wave SAR imaging processing chain, real-time correction of millimeter-wave SAR echo data for complex plateau climate environments is achieved. Furthermore, DEM data is used to further correct the millimeter-wave SAR echo data, enhancing the millimeter-wave SAR imaging's resistance to meteorological interference and terrain distortion, thereby improving the accuracy and precision of millimeter-wave SAR imaging under the influence of complex plateau environments. Through the construction of a multi-dimensional coupled feature set of scattering-texture-terrain and a lightweight disaster identification module, real-time identification of plateau disaster areas is achieved on edge devices, improving the timeliness and accuracy of plateau disaster identification.

[0049] 2. The traditional path integral attenuation model and phase perturbation correction model are corrected by using the plateau meteorological parameter distribution field that is spatiotemporally unified with the propagation path of millimeter-wave SAR data. The corrected model is then used to correct the millimeter-wave SAR echo data, so that the signal correction process can be adapted to the complex climate environment of the plateau, thereby improving the authenticity of the corrected millimeter-wave SAR echo data.

[0050] 3. By constructing a multi-dimensional coupled feature set of scattering-texture-topography associated with topographic factors in plateau disaster-prone areas, the feature data of non-disaster-prone areas can be reduced during disaster identification, thereby improving the processing efficiency, identification accuracy and reliability of plateau disaster identification;

[0051] 4. By constructing a deep learning network with a lightweight DeepLabv3+ backbone network, introducing a dual-path attention mechanism and a multi-task decoder architecture, and combining a multi-dimensional coupled feature set of scattering-texture-terrain with high-quality millimeter-wave SAR images, we can automatically identify and segment plateau disaster areas, realize intelligent, real-time and accurate identification of plateau disaster areas, and improve the timeliness of plateau disaster identification.

[0052] 5. By incorporating the comprehensive weight of terrain factors into the calculation of the weights of each loss term in the multi-task joint loss function, the weights of each loss term are dynamically adapted to the characteristics of the plateau terrain and the loss feedback during the training process. This ensures the training stability of the disaster identification model under the influence of complex plateau terrain, thereby improving the adaptive and generalization capabilities of the disaster identification model and the accuracy of the disaster identification results. Attached Figure Description

[0053] The accompanying drawings, which are provided to further illustrate embodiments of the invention and constitute a part of this invention, are not intended to limit the scope of the invention.

[0054] Figure 1 This is a schematic diagram of the structure of the airborne millimeter-wave SAR disaster identification system in the complex climate environment of the plateau region in this invention;

[0055] Figure 2 This is a flowchart illustrating the airborne millimeter-wave SAR disaster identification method for complex high-altitude climate environments in this invention.

[0056] Figure 3 This is a comparison image of the original millimeter-wave SAR image and the corrected millimeter-wave SAR image after correction using the SAR signal plateau meteorological correction model and DEM data of this invention;

[0057] in, Figure 3Sub-images (a), (c), (e), and (g) are the original SAR images of different plateau regions that have severe overlay and shadowing without being corrected using the SAR signal plateau meteorological correction model of this invention. Sub-images (b), (d), (f), and (h) are the SAR images of the same plateau regions that have been corrected using the SAR signal plateau meteorological correction model of this invention, respectively. Detailed Implementation

[0058] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that, where there is no conflict, the embodiments of the present invention and the features thereof can be combined with each other.

[0059] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0060] Example 1

[0061] Please refer to Figure 1 This invention provides an airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments, comprising: a data acquisition module, a meteorological correction module, a SAR imaging module, a feature extraction module, and a lightweight disaster identification module; wherein, the data acquisition module is used to simultaneously acquire airborne millimeter-wave SAR echo data, high-altitude meteorological parameters of the flight area, and digital elevation model (DEM) data of the target area; wherein, the data acquisition module includes: a high-altitude long-endurance unmanned aerial vehicle (UAV) platform for carrying airborne equipment to fly over the target high-altitude area; a Ka / W-band airborne millimeter-wave SAR, mounted on the UAV platform, for acquiring airborne millimeter-wave SAR echo data; and an airborne meteorological radar, mounted on the UAV platform, for acquiring flight data. The system includes: regional precipitation intensity and cloud liquid water content; an airborne microwave radiometer mounted on an UAV platform to measure atmospheric water vapor content and atmospheric liquid water path; a GPS radiosonde to acquire vertical profiles of atmospheric temperature, atmospheric pressure, and atmospheric humidity; a ground-based meteorological station to acquire reference values ​​for ground temperature, air pressure, and humidity parameters to calibrate the measurement results of the airborne equipment; and an airborne LiDAR to acquire digital elevation model (DEM) data of the target area. The data acquisition module uses GPS timestamps and spatial coordinate registration to achieve time synchronization and spatial registration of airborne millimeter-wave SAR echo data, plateau meteorological parameters of the flight area, and DEM data of the target area.

[0062] The meteorological correction module is used to construct a high-altitude meteorological correction model for SAR signals based on high-altitude meteorological parameters of the flight area. Based on this model, amplitude and phase corrections are performed on the airborne millimeter-wave SAR echo data to obtain the first corrected millimeter-wave SAR data. Specifically:

[0063] A linear interpolation algorithm is used to interpolate discrete meteorological observation data to generate a continuous plateau meteorological parameter distribution field along the round-trip propagation path of the airborne millimeter-wave SAR signal; a cubic spline interpolation method is used to interpolate the meteorological parameters in time according to the reception time of the airborne millimeter-wave SAR echo data.

[0064] A path integral attenuation model is constructed, comprising oxygen absorption attenuation, water vapor absorption attenuation, cloud attenuation, and precipitation attenuation terms. Based on the plateau meteorological parameter distribution field, the oxygen absorption attenuation, water vapor absorption attenuation, cloud attenuation, and precipitation attenuation terms are corrected respectively, resulting in a path integral attenuation model corrected for plateau meteorological parameters. The expression for the path integral attenuation model is:

[0065] ;

[0066] in, This represents the total attenuation of the millimeter-wave SAR signal. , , and These are the oxygen absorption attenuation term, water vapor absorption attenuation term, cloud layer attenuation term, and precipitation attenuation term, respectively; among which, the oxygen absorption attenuation term... The correction method is as follows: based on the vertical atmospheric pressure profile, when calculating using the ITU-R P.676 standard model, actual air pressure values ​​at different altitudes are used instead of standard sea-level air pressure calculations. Water vapor absorption attenuation term The correction method is as follows: calculation based on the true values ​​of the vertical profiles of atmospheric temperature, atmospheric pressure, and atmospheric humidity. Cloud attenuation term The correction method is as follows: based on the measured values ​​of atmospheric liquid water path and effective radius, the cloud attenuation model of ITU-R P.840 is used for calculation. Rainfall attenuation term The correction method is as follows: the rainfall rate R is calculated using real-time precipitation intensity;

[0067] A phase perturbation correction model is constructed. Based on the vertical profiles of atmospheric temperature, atmospheric pressure, and atmospheric humidity, the atmospheric refractive index is calculated using the Smith-Weintraub formula. Then, the atmospheric refractive index profile along the radar signal propagation path is obtained by inversion. Based on the atmospheric refractive index profile and the phase perturbation correction model, the phase delay field along the round-trip propagation path of the radar signal is estimated using the tropospheric delay and turbulence structure function.

[0068] The path integral attenuation model corrected by plateau meteorological parameters is used to perform amplitude compensation on the airborne millimeter-wave SAR echo data to obtain amplitude-compensated millimeter-wave SAR echo data.

[0069] The phase delay field is used as a two-dimensional phase screen. Before SAR azimuth compression, the corresponding phase delay value is subtracted from the amplitude-compensated millimeter-wave SAR echo data to obtain the first corrected millimeter-wave SAR data, thereby achieving pre-compensation for phase error and reducing the accumulation of phase distortion during imaging.

[0070] Specifically, the SAR imaging module is used to correct the first corrected millimeter-wave SAR data based on the digital elevation model (DEM) data of the target area to reduce terrain distortion, thereby obtaining second corrected millimeter-wave SAR data, and generating a millimeter-wave SAR image based on the second corrected millimeter-wave SAR data;

[0071] The ICP registration algorithm is used to spatially register the DEM data of the target area with the airborne millimeter-wave SAR echo data. With the imaging geometric parameters of the airborne millimeter-wave SAR echo data as constraints, the spatial overlap between the DEM data and the SAR echo data is ≥99% through iterative optimization to obtain the registered DEM data.

[0072] A window analysis method was adopted, using a 3×3 pixel window. Based on the registered DEM data, key terrain factors for each pixel were extracted. The key terrain factors included: slope, aspect, local incident angle, and terrain curvature for each pixel. The slope was calculated using the maximum slope drop method, the aspect was calculated using the azimuth method, the local incident angle was calculated based on the geometric parameters of the millimeter-wave SAR image and the slope and aspect of the pixel, and the terrain curvature was calculated using a quadratic surface fitting algorithm.

[0073] A terrain-scattering response model is constructed based on radar equations and surface scattering theory. Based on the key terrain factors and the terrain-scattering response model, the backscattering coefficients of the first-corrected millimeter-wave SAR data for each pixel are corrected point-by-point to reduce radiation distortion caused by terrain undulations, thus obtaining the second-corrected millimeter-wave SAR data. The expression for the terrain-scattering response model is as follows:

[0074] ;

[0075] In the formula, The backscattering coefficient is corrected for topographic factors. The backscattering coefficients of the first corrected millimeter-wave SAR data, As a reference incident angle, the value is taken from the center of the image. The angle of incidence is the local angle of incidence, and n is the terrain adaptive parameter, where n=1.5 for bare rock areas, n=2.0 for vegetation-covered areas, and n=1.2 for snow-covered areas.

[0076] The second-corrected millimeter-wave SAR data is processed using a back projection imaging algorithm to generate a millimeter-wave SAR image.

[0077] The feature extraction module is used to extract amplitude statistical features, polarization decomposition features, coherence / interference features and texture features of the second corrected millimeter-wave SAR data, extract key terrain factors based on the target area digital elevation model (DEM) data, and construct a scattering-texture-terrain multidimensional coupled feature set based on the amplitude statistical features, polarization decomposition features, coherence / interference features, texture features and key terrain factors.

[0078] Among them, the amplitude statistical features are calculated based on the HH channel gray values ​​of the second-corrected millimeter-wave SAR data using a 5×5 pixel sliding window. The amplitude statistical features include: mean, variance, coefficient of variation, kurtosis and skewness.

[0079] Among them, the polarization decomposition features are based on the coherence matrix of fully polarimetric SAR data, and the scattering entropy H, average scattering angle α and inverse entropy A are extracted using the Cloude-Pottier polarization decomposition algorithm.

[0080] Among them, the coherence / interference characteristics are obtained by calculating the interference coherence coefficient using time-series data or dual-antenna data;

[0081] Among them, the texture features are based on the HH channel of the millimeter-wave SAR image, and the contrast, correlation, energy, texture entropy and homogeneity are extracted by the gray-level co-occurrence matrix (GLCM). In the calculation, the direction of GLCM is 0°, 45°, 90° and 135°, the distance is 1 pixel, the gray level is 16, and the mean of the four directions is finally taken as the texture feature value.

[0082] The methods for constructing the scattering-texture-terrain multidimensional coupled feature set include:

[0083] Based on the key terrain factors corresponding to each pixel in the millimeter-wave SAR image, pixels that meet the preset threshold for terrain factors prone to plateau disasters are selected as feature pixels; where the preset threshold for terrain factors prone to plateau disasters is: slope value ∈ [15°, 60°], local incident angle ∈ [30°, 70°].

[0084] The amplitude statistical feature, polarization decomposition feature, coherence / interference feature and texture feature corresponding to each feature pixel are associated with the local incident angle, slope, aspect and terrain curvature of the pixel, respectively, to obtain the amplitude statistical feature with local incident angle label, the polarization decomposition feature with slope label, the coherence / interference feature with aspect label and the texture feature with terrain curvature label for each feature pixel.

[0085] The 5-dimensional amplitude statistical feature with local incident angle label, the 3-dimensional polarization decomposition feature with slope label, the 1-dimensional coherence / interference feature with slope aspect label, and the 5-dimensional texture feature with terrain curvature label corresponding to each feature pixel are sequentially concatenated to obtain a 14-dimensional scattering-texture-terrain multidimensional coupled feature vector corresponding to each feature pixel.

[0086] The set of 14-dimensional scattering-texture-terrain multidimensional coupled feature vectors corresponding to all the feature pixels is the scattering-texture-terrain multidimensional coupled feature set.

[0087] The lightweight disaster identification module employs a multi-task deep learning network, including a shared encoder, a feature fusion module, and a multi-task decoder. The shared encoder uses a lightweight DeepLabv3+ backbone network, with MobileNetV2 replacing the original Xception network to reduce the number of parameters. The feature fusion module introduces a CBAM attention mechanism, including channel attention units and spatial attention units. The channel attention unit performs channel dimension compression and activation on the input 14-dimensional scattering-texture-terrain multi-dimensional coupled feature set and the 4-channel millimeter-wave SAR image, generating channel weights through the Sigmoid function to enhance the features of key channels. Inter-attention units are used to perform global average pooling and max pooling on the channel-weighted feature map to generate a spatial weight map with the same size as the input feature map, so as to suppress the interference of the background region. The multi-task decoder includes a disaster segmentation decoding branch and a change detection decoding branch. The disaster segmentation decoding branch uses transposed convolution to upsample the fused feature map, combines the shallow features of the encoder, and reduces the feature dimension to two classes, disaster area and non-disaster area, through 1×1 convolution, and outputs a disaster area segmentation map. The change detection decoding branch extracts change features through difference operation based on the input fused feature maps of the pre-disaster and post-disaster periods, and outputs a pre-disaster and post-disaster change area map including the changed area and the non-changed area through 3×3 convolution and upsampling.

[0088] The training methods for the multi-task deep learning network include:

[0089] We collected airborne millimeter-wave SAR data, optical images, and field verification data of historical disaster areas in plateau regions. We then performed detailed annotations on the disaster categories and disaster area outlines that included samples from the disaster areas. We constructed a sample library containing at least 10,000 samples, which was divided into training and validation sets at a ratio of 9:1.

[0090] A multi-task deep learning network is trained using a joint loss function that includes a disaster area segmentation loss term and a change detection loss term. The expression for the joint loss function is as follows:

[0091] ;

[0092] In the formula, For the joint loss value, and These are losses due to disaster area segmentation and losses due to change detection. and These are the loss weights for disaster area segmentation and change detection, respectively. At the beginning of training... , ;

[0093] The disaster area segmentation loss uses a combination of Dice Loss and Focal Loss to address class imbalance and edge detail issues, while the change detection loss uses a combination of binary cross-entropy loss and Dice Loss to balance pixel-level classification and region consistency. The expressions for the disaster area segmentation loss and the change detection loss are as follows:

[0094] ;

[0095] ;

[0096] In the formula and These are the Dice sub-loss items in the disaster area segmentation loss. and Focal sub-loss term The weights, at the beginning of training, , , and These are the binary cross-entropy sub-loss terms in the change detection loss. and Dice sub-loss term The weights, at the beginning of training, , ;

[0097] In each training round, the loss weights for disaster area segmentation and change detection are automatically adjusted based on the combined weights of task uncertainty and terrain factors of the current input millimeter-wave SAR image, specifically as follows:

[0098] The coefficient of variation method is used to calculate the comprehensive weight of the four key terrain factors in each input millimeter-wave SAR image, and the calculation formula is as follows:

[0099] ;

[0100] In the formula, t represents the current training round. The combined weights of terrain factors are the input terrain factors from the millimeter-wave SAR images in the previous training rounds. , , and These are the coefficients of variation for slope, aspect, local incident angle, and terrain curvature of the millimeter-wave SAR images input in the previous training rounds. , , and These are the weights of the coefficients of variation for slope, aspect, local incident angle, and topographic curvature, respectively. , , and These can be determined based on prior knowledge. , , , ;

[0101] The normalized values ​​of the disaster area segmentation loss term and the change detection loss term are calculated in the current training round using the following formula:

[0102] ;

[0103] ;

[0104] In the formula, and These are the normalized values ​​of the disaster area segmentation loss term and the change detection loss term, respectively, in the current training round. and These represent the disaster area segmentation loss value and the change detection loss value from the previous training rounds, respectively.

[0105] Based on the comprehensive weights of terrain factors in the millimeter-wave SAR image input in the current training round, the weights of the disaster area segmentation loss term and the change detection loss term in the previous training round, and the normalized values ​​of the disaster area segmentation loss term and the change detection loss term in the current training round, the weights of the disaster area segmentation loss term and the change detection loss term in the current training round are calculated. The calculation formulas for the weights of the disaster area segmentation loss term and the change detection loss term in the current training round are as follows:

[0106] ;

[0107] ;

[0108] In the formula, Indicates the previous training round. and These are the weight balance coefficients for the disaster area segmentation loss item and the weight balance coefficients for the change detection loss item, respectively. and These represent the weights of the disaster area segmentation loss term and the change detection loss term for the current training round, respectively. and These are the weights of the disaster area segmentation loss term and the change detection loss term from the previous training round, respectively. , Initial weights , ;

[0109] The multi-task deep learning network was trained using the stochastic gradient descent (SGD) optimizer with a training batch size of 10, a learning rate of 0.002, a momentum of 0.9, and 200 training epochs. The MultiStepLR scheduler was used to halve the learning rate at epochs 20, 40, ..., 160 and 180.

[0110] The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments also includes a visualization output module. This module visualizes the output of the lightweight disaster identification module, providing support for emergency command and decision-making. Specifically:

[0111] A visualization platform based on ArcGIS Engine is developed to overlay disaster area segmentation maps, pre- and post-disaster change area maps, and basic geographic data to generate thematic maps. The basic geographic data includes administrative divisions, water systems, roads, and settlements. The disaster area segmentation map uses a hierarchical coloring system, with disaster areas marked in red and disaster types labeled (landslides, debris flows, landslide dams, and other disasters). High-risk areas are marked in orange, low-risk areas in yellow, and non-disaster areas in green. The classification level is determined based on the disaster confidence map, pre- and post-disaster change area map, and key topographic factor data output by the lightweight disaster identification module. Specifically, a disaster confidence level ≥ 85 and key topographic factor values ​​falling within the preset plateau disaster susceptibility range are required. Pixels that meet the topographic factor threshold and belong to the changed area in the pre- and post-disaster change area map are identified as disaster areas; pixels with a disaster confidence level between 60 and 84 and whose key topographic factor values ​​fall within the preset plateau disaster-prone topographic factor threshold, or pixels with a disaster confidence level ≥ 85 but not belonging to the pre- and post-disaster change area, are identified as high-risk areas; pixels with a disaster confidence level between 30 and 59, or whose at least one key topographic factor value falls within the preset plateau disaster-prone topographic factor threshold, are identified as low-risk areas; pixels with a disaster confidence level < 30 and do not meet the preset plateau disaster-prone topographic factor threshold are identified as non-disaster areas; the pre- and post-disaster change area map uses binary coloring, with changed areas colored black and non-changed areas transparent.

[0112] Please refer to Figure 2 The present invention provides an airborne millimeter-wave SAR disaster identification method for complex high-altitude climate environments, comprising the following steps:

[0113] S1: The data acquisition module synchronously acquires airborne millimeter-wave SAR echo data, plateau meteorological parameters of the flight area, and digital elevation model (DEM) data of the target area.

[0114] S2: The data acquisition module aligns the plateau meteorological parameters of the flight area with the airborne millimeter-wave SAR echo data in time and space to generate a plateau meteorological parameter distribution field along the round-trip propagation path of the radar signal; and registers the digital elevation model (DEM) data of the target area with the airborne millimeter-wave SAR echo data to obtain registered DEM data.

[0115] S3: The meteorological correction module constructs a plateau meteorological correction model for SAR signals based on the plateau meteorological parameter distribution field, and corrects the airborne millimeter-wave SAR echo data based on the plateau meteorological correction model to obtain the first corrected millimeter-wave SAR data.

[0116] S4: The SAR imaging module corrects the first corrected millimeter-wave SAR data based on the registered DEM data to obtain the second corrected millimeter-wave SAR data, and generates a millimeter-wave SAR image based on the second corrected millimeter-wave SAR data;

[0117] S5: The feature extraction module extracts the amplitude statistical features, polarization decomposition features, coherence / interference features and texture features of the second corrected millimeter-wave SAR data, extracts the key terrain factors of the registered DEM data, and constructs a scattering-texture-terrain multidimensional coupled feature set based on the amplitude statistical features, polarization decomposition features, coherence / interference features, texture features and key terrain factors;

[0118] S6: The lightweight disaster identification module identifies disaster areas and pre- and post-disaster change areas based on the scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image, and outputs disaster area segmentation map and pre- and post-disaster change area map.

[0119] S7: The visualization output module overlays the recognition results with basic geographic data to generate and output a visualized thematic map.

[0120] Example 2

[0121] Based on Example 1, Example 2 will be described and illustrated with specific implementation cases.

[0122] Implementation Case 1: Please refer to Figure 3 , Figure 3 Sub-images (a), (c), (e), and (g) are the original SAR images from different plateau regions that have not been corrected using the SAR signal plateau meteorological correction model constructed in this invention, and thus suffer from severe overlay and shadowing. Sub-images (b), (d), (f), and (h) are the SAR images from the same plateau regions as sub-images (a), (c), (e), and (g) that have been corrected using the SAR signal plateau meteorological correction model constructed in this invention, and thus exhibit higher imaging accuracy and terrain conformity. To evaluate imaging quality, a segment of millimeter-wave SAR data acquired under moderate precipitation conditions with a rainfall rate of 5 mm / h was selected. The image quality of the millimeter-wave SAR image without correction using the SAR signal plateau meteorological correction model constructed in this invention was compared with that of the millimeter-wave SAR image after correction using the SAR signal plateau meteorological correction model constructed in this invention. The comparison results are shown in Table 1. The comparison results show that the SAR signal plateau meteorological correction model constructed in this invention can significantly suppress image blurring and noise caused by precipitation attenuation and phase perturbation, greatly improving image clarity and usability, and laying a solid foundation for subsequent feature extraction and recognition.

[0123] Table 1. Comparison of SAR image quality before and after correction using the SAR signal plateau meteorological correction model.

[0124]

[0125] Implementation Case 2: To evaluate the classification accuracy of different feature combinations for disaster bodies and easily confused features, the accuracy of disaster body identification based on different feature combinations and the scattering-texture-topography multidimensional coupled feature set constructed by this system was compared on the basis of consistent image quality. The comparison results are shown in Table 2. The comparison results show that the scattering-texture-topography multidimensional coupled feature set constructed by this system can effectively capture the surface incoherence caused by disasters due to the addition of coherence / interference features, while the coupling of topographic factors eliminates a large number of false targets that do not meet the topographic conditions from the perspective of the occurrence mechanism, thus achieving higher classification accuracy and reliability in the complex plateau terrain environment.

[0126] Table 2. Comparison of accuracy results for identifying disaster entities based on different feature combinations.

[0127]

[0128] Implementation Case 3: To evaluate the real-time performance and accuracy of the lightweight multi-task deep learning network integrating an attention mechanism constructed in this invention for disaster identification, different disaster identification deep learning models were deployed and run on the same hardware platform. Based on the same input samples, the identification accuracy, parameter count, and processing speed of different disaster identification deep learning models were compared. The comparison results are shown in Table 3. The comparison results show that the lightweight multi-task deep learning network constructed in this invention reduces the number of model parameters by 93% while sacrificing minimal accuracy, and improves the single image processing speed by about 5 times. This indicates that the lightweight multi-task deep learning network constructed in this invention can achieve near real-time identification of plateau disaster areas in environments with limited computing resources on airborne platforms, meeting the timeliness requirements of emergency response.

[0129] Table 3 Comparison of disaster identification results of different models

[0130]

[0131] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0132] 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. An airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments, characterized in that, include: The data acquisition module is used to acquire airborne millimeter-wave SAR echo data, plateau meteorological parameters of the flight area, and digital elevation model (DEM) data of the target area; The meteorological correction module is used to construct a high-altitude meteorological correction model for SAR signals based on the high-altitude meteorological parameters of the flight area, and to correct the airborne millimeter-wave SAR echo data based on the high-altitude meteorological correction model for SAR signals to obtain the first corrected millimeter-wave SAR data. The SAR imaging module is used to correct the first corrected millimeter-wave SAR data based on the digital elevation model (DEM) data of the target area to obtain the second corrected millimeter-wave SAR data, and generate a millimeter-wave SAR image based on the second corrected millimeter-wave SAR data. The feature extraction module is used to extract amplitude statistical features, polarization decomposition features, coherence / interference features and texture features from the second corrected millimeter-wave SAR data, extract key terrain factors based on the target area digital elevation model (DEM) data, and construct a scattering-texture-terrain multidimensional coupled feature set based on the amplitude statistical features, polarization decomposition features, coherence / interference features, texture features and key terrain factors. A lightweight disaster identification module is used to identify disaster areas and pre- and post-disaster change areas based on the scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image, and output disaster area segmentation maps and pre- and post-disaster change area maps. The lightweight disaster identification module employs a multi-task deep learning network, which includes a shared encoder, a feature fusion module, and a multi-task decoder. The shared encoder uses a lightweight DeepLabv3+ backbone network. The feature fusion module introduces an attention mechanism, which includes channel attention units and spatial attention units. The channel attention units are used to focus on multiple channels of the input scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image. The spatial attention unit is used to assign spatial weights to different spatial locations of the input millimeter-wave SAR image. The multi-task decoder includes a disaster segmentation decoding branch and a change detection decoding branch. The disaster segmentation decoding branch is used to identify the disaster area in the input feature map and output a disaster area segmentation map. The change detection decoding branch is used to compare the changed areas in the pre-disaster and post-disaster images and output a pre-disaster and post-disaster changed area map. The lightweight disaster recognition module trains the multi-task deep learning network using a joint loss function, which includes a disaster area segmentation loss term and a change detection loss term. The weights of each loss term are automatically adjusted based on the combined weights of task uncertainty and the terrain factors of the currently input millimeter-wave SAR image.

2. The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments according to claim 1, characterized in that, The plateau meteorological parameters of the flight area include: precipitation intensity, cloud liquid water content, atmospheric water vapor content, atmospheric liquid water path, atmospheric temperature vertical profile, atmospheric pressure vertical profile, and atmospheric humidity vertical profile; the plateau meteorological correction model of the SAR signal includes: a path integral attenuation model and a phase perturbation correction model corrected by plateau meteorological parameters.

3. The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments according to claim 2, characterized in that, The methods for obtaining the first corrected millimeter-wave SAR data include: Align the plateau meteorological parameters of the flight area with the airborne millimeter-wave SAR echo data in time and space to generate a plateau meteorological parameter distribution field along the round-trip propagation path of the radar signal; A path integral attenuation model and a phase perturbation correction model are constructed. Based on the plateau meteorological parameter distribution field, the path integral attenuation model and the phase perturbation correction model are corrected to obtain the path integral attenuation model and the phase perturbation correction model corrected by the plateau meteorological parameters. Based on the path integral attenuation model and phase perturbation correction model corrected by plateau meteorological parameters, signal attenuation compensation and phase perturbation compensation are performed on the airborne millimeter-wave SAR echo data to obtain the first corrected millimeter-wave SAR data.

4. The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments according to claim 1, characterized in that, The methods for obtaining the second corrected millimeter-wave SAR data include: The target area digital elevation model (DEM) data is registered with the airborne millimeter-wave SAR echo data to obtain registered DEM data. Key terrain factors are extracted based on the registered DEM data. These key terrain factors include: the slope, aspect, local incident angle, and terrain curvature of each pixel in the millimeter-wave SAR image. A terrain-scattering response model is constructed, and the backscattering coefficient of the first corrected millimeter-wave SAR data is corrected based on the key terrain factors and the terrain-scattering response model to obtain the second corrected millimeter-wave SAR data.

5. The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments according to claim 4, characterized in that, The method for constructing the scattering-texture-terrain multidimensional coupled feature set includes: Based on the key terrain factors corresponding to each pixel in the millimeter-wave SAR image, pixels that meet the preset terrain factor threshold for high-altitude disaster-prone areas are selected as feature pixels. The amplitude statistical feature, polarization decomposition feature, coherence / interference feature and texture feature corresponding to each feature pixel are associated with the local incident angle, slope, aspect and terrain curvature of the pixel, respectively, to obtain the amplitude statistical feature with local incident angle label, the polarization decomposition feature with slope label, the coherence / interference feature with aspect label and the texture feature with terrain curvature label for each feature pixel. The amplitude statistical feature with local incident angle label, the polarization decomposition feature with slope label, the coherence / interference feature with slope aspect label, and the texture feature with terrain curvature label corresponding to each feature pixel are dimensionally concatenated to obtain the scattering-texture-terrain multidimensional coupled feature vector corresponding to each feature pixel. The set of scattering-texture-terrain multidimensional coupled feature vectors corresponding to all the feature pixels is the scattering-texture-terrain multidimensional coupled feature set.

6. The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments according to claim 1, characterized in that, The weights of each loss term are automatically adjusted based on a combination of task uncertainty and terrain factors of the currently input millimeter-wave SAR image, specifically including: The comprehensive weight of the terrain factors in each millimeter-wave SAR image is calculated based on the degree of dispersion of the key terrain factors in each millimeter-wave SAR image. Calculate the normalized values ​​of the disaster area segmentation loss term and the change detection loss term in the current training round; Based on the comprehensive weight of terrain factors of the millimeter-wave SAR image input in the current training round, the weight of the disaster area segmentation loss term and the weight of the change detection loss term in the previous training round, and the normalized value of the disaster area segmentation loss term and the normalized value of the change detection loss term in the current training round, calculate the weight of the disaster area segmentation loss term and the weight of the change detection loss term in the current training round.

7. The airborne millimeter-wave SAR disaster identification system for complex high-altitude climate environments according to claim 6, characterized in that, The calculation formulas for the weights of the disaster area segmentation loss term and the change detection loss term in the current training round are as follows: ； ； ； ； In the formula, and These represent the current training round and the previous training round, respectively. and These are the weight balance coefficients for the disaster area segmentation loss item and the weight balance coefficients for the change detection loss item, respectively. and These represent the weights of the disaster area segmentation loss term and the change detection loss term for the current training round, respectively. and These represent the weights of the disaster area segmentation loss term and the change detection loss term from the previous training round, respectively. The combined weights of terrain factors in the millimeter-wave SAR image input in the current training round. and These are the normalized values ​​of the disaster area segmentation loss term and the change detection loss term, respectively, in the current training round. and These represent the disaster area segmentation loss and change detection loss in the current training round, respectively.

8. A method for airborne millimeter-wave SAR disaster identification in complex high-altitude climate environments, characterized in that, Includes the following steps: S1: Acquire airborne millimeter-wave SAR echo data, high-altitude meteorological parameters of the flight area, and digital elevation model (DEM) data of the target area; S2: Align the plateau meteorological parameters of the flight area with the airborne millimeter-wave SAR echo data in time and space to generate a plateau meteorological parameter distribution field along the round-trip propagation path of the radar signal; Register the digital elevation model (DEM) data of the target area with the airborne millimeter-wave SAR echo data to obtain registered DEM data; S3: Construct a plateau meteorological correction model for SAR signals based on the plateau meteorological parameter distribution field, and correct the airborne millimeter-wave SAR echo data based on the plateau meteorological correction model to obtain the first corrected millimeter-wave SAR data; S4: Correct the first corrected millimeter-wave SAR data based on the registered DEM data to obtain the second corrected millimeter-wave SAR data, and generate a millimeter-wave SAR image based on the second corrected millimeter-wave SAR data; S5: Extract the amplitude statistical features, polarization decomposition features, coherence / interference features and texture features of the second corrected millimeter-wave SAR data, extract the key terrain factors of the registered DEM data, and construct a scattering-texture-terrain multidimensional coupled feature set based on the amplitude statistical features, polarization decomposition features, coherence / interference features, texture features and key terrain factors; S6: Employing a multi-task deep learning network, based on the scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image, the system identifies disaster areas and pre- and post-disaster change areas, outputting disaster area segmentation maps and pre- and post-disaster change area maps. The multi-task deep learning network includes a shared encoder, a feature fusion module, and a multi-task decoder. The shared encoder uses a lightweight DeepLabv3+ backbone network. The feature fusion module introduces an attention mechanism, which includes channel attention units and spatial attention units. The channel attention units are used to allocate channels among multiple channels of the input scattering-texture-terrain multidimensional coupled feature set and the millimeter-wave SAR image. The spatial attention unit is used to assign spatial weights at different spatial locations of the input millimeter-wave SAR image. The multi-task decoder includes a disaster segmentation decoding branch and a change detection decoding branch. The disaster segmentation decoding branch is used to identify disaster areas in the input feature map and output a disaster area segmentation map. The change detection decoding branch is used to compare the changed areas in the pre-disaster and post-disaster images and output a pre-disaster and post-disaster changed area map. The multi-task deep learning network is trained based on a joint loss function, which includes a disaster area segmentation loss term and a change detection loss term. The weights of each loss term are automatically adjusted based on the combined weights of task uncertainty and the terrain factors of the currently input millimeter-wave SAR image.

Images