Method and device for extracting understory terrain based on multi-baseline SAR data

Abstract

The present invention discloses a method and device for extracting under-forest terrain based on multi-baseline SAR data. The method includes: decomposing the SAR image dataset, performing tomographic imaging on the decomposed signals using spectral reconstruction to obtain a forest tomographic structure profile; projecting the main lobe signals in the structure profile to obtain a reconstructed three-dimensional forest point cloud; after performing elevation normalization processing on the three-dimensional forest point cloud using a rough digital terrain model, extracting the elevation-normalized ground point cloud in the three-dimensional forest point cloud, and superimposing the elevation-normalized ground point cloud on the rough digital terrain model to obtain a ground point cloud; projecting the ground point cloud onto a two-dimensional grid to form an under-forest terrain model. By adopting the above technical solution, a three-dimensional forest point cloud is reconstructed based on the forest tomographic structure profile, and then the ground point cloud therein is extracted, and the under-forest terrain is obtained through the ground point cloud, which significantly improves the accuracy of under-forest terrain extraction and overcomes the problem of obvious breakpoints and jumps in the under-forest terrain.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology, and in particular, to a method and device for extracting understory terrain based on multi-baseline SAR data. Background Art

[0002] The understory terrain is the focus of forest vertical structure parameters and forestry surveys, and plays an important guiding role in forest growth potential and wood production. In terms of remote sensing research, there are currently various forms and methods for remote sensing of the understory terrain. In terms of remote sensing types, there are mainly optical remote sensing, LiDAR, and synthetic aperture radar (SAR, Synthetic Aperture Radar).

[0003] Tomographic SAR is an advanced method for extracting vertical forest structure parameters by analyzing the height distribution differences of multiple baseline observations, thereby providing vertical resolution and three-dimensional detection within the forest. However, in the actual application of this method, the understory terrain information is usually extracted by screening pixels of the forest tomographic profile by setting a threshold, and the parts that do not meet the threshold are eliminated, which easily eliminates the information representing the true ground surface, resulting in discontinuous backscattering power distribution in the azimuth and range directions within the forest tomographic profile. The understory terrain obtained based on the forest tomographic profile after threshold screening processing will have obvious breakpoints and jumps. Summary of the Invention

[0004] Object of the Invention: The present invention provides a method and device for extracting understory terrain based on multi-baseline SAR data, aiming to solve the technical problems in the prior art that in the actual application of tomographic SAR, the backscattering power distribution of the forest tomographic profile is discontinuous, and the understory terrain will have obvious breakpoints and jumps.

[0005] Technical Solution: The present invention provides a method for extracting understory terrain based on multi-baseline SAR data, including: obtaining a multi-baseline dataset of a target area collected, synthesizing an SAR image dataset, and correcting the interference phase error in the SAR image dataset; decomposing the signals in the SAR image dataset, performing tomographic imaging on the decomposed signals using spectral reconstruction to obtain a forest tomographic structure profile; removing the sidelobe signals in the structure profile and retaining the main lobe signals in the structure profile; obtaining a reconstructed forest three-dimensional point cloud by projecting the main lobe signals in the structure profile; generating a rough digital terrain model representing the terrain undulation changes, using the rough digital terrain model to perform elevation normalization processing on the forest three-dimensional point cloud, and then using cloth simulation filtering to extract the elevation-normalized ground point cloud in the forest three-dimensional point cloud, and superimposing the elevation-normalized ground point cloud on the rough digital terrain model to obtain the ground point cloud in the forest three-dimensional point cloud; projecting the ground point cloud onto a two-dimensional grid to form an understory terrain model.

[0006] Specifically, obtain the P-band multi-baseline dataset of the target area collected by the aircraft.

[0007] Specifically, remove the phase error, flat earth phase, and terrain phase in the SAR image dataset.

[0008] Specifically, perform SKP signal decomposition on the SAR image dataset; use Capon spectral reconstruction to suppress the noise of the SAR echo signal and improve the resolution of tomography in the height direction.

[0009] Specifically, set the size of the sliding window, determine the position of the maximum power corresponding to the maximum backscattering coefficient within the sliding window in the structural profile; at the position of the maximum power, extract the pixels corresponding to the backscattering signals in the upper and lower ranges of the window through the sliding window; perform sidelobe signal elimination and main lobe signal extraction on the structural profile pixel by pixel in the azimuth or range direction.

[0010] Specifically, pixel by pixel in the azimuth or range direction, project the pixels to the corresponding height according to the elevation information corresponding to the pixels of the backscattering signals in the structural profile, and reconstruct the three-dimensional forest point cloud.

[0011] Specifically, grid the three-dimensional forest point cloud in the horizontal direction, and use the lowest point in each grid as the reference point to obtain the corresponding rough digital terrain model.

[0012] Specifically, for each target point in the three-dimensional forest point cloud, calculate the elevation difference between the elevation value of the target point and the elevation value of the corresponding point in the rough digital terrain model, and use the elevation difference as the elevation value of the target point, so as to obtain the three-dimensional forest point cloud with normalized elevation.

[0013] Specifically, interpolate the elevation information of the ground point cloud, and then project it onto a two-dimensional grid to form an understory terrain model.

[0014] The present invention also provides an understory terrain extraction device based on multi-baseline SAR data, including an image data synthesis unit, a tomography unit, a signal extraction unit, a point cloud reconstruction unit, a ground point cloud extraction unit, and an understory terrain model generation unit, where: The image data synthesis unit is used to obtain the multi-baseline data set of the target area collected, synthesize the SAR image data set, and correct the interference phase error in the SAR image data set; The tomography unit is used to decompose the signals in the SAR image data set, perform tomography on the decomposed signals using spectral reconstruction, and obtain the forest tomography structure profile; The signal extraction unit is used to remove the sidelobe signals in the structure profile and retain the main lobe signals in the structure profile; The point cloud reconstruction unit is used to project the main lobe signals in the structure profile to obtain the reconstructed three-dimensional forest point cloud; The ground point cloud extraction unit is used to generate a rough digital terrain model representing the terrain undulation changes. After using the rough digital terrain model to perform elevation normalization processing on the three-dimensional forest point cloud, the elevation-normalized ground point cloud in the three-dimensional forest point cloud is extracted using cloth simulation filtering, and the elevation-normalized ground points are superimposed on the rough digital terrain model to obtain the ground point cloud in the three-dimensional forest point cloud;

[0015] The understory terrain model generation unit is used to project the ground point cloud onto a two-dimensional grid to form an understory terrain model.

[0016] Beneficial effects: Compared with the prior art, the present invention has the following remarkable advantages: By reconstructing the three-dimensional forest point cloud based on the forest tomography structure profile, then extracting the ground point cloud therein, and obtaining the understory terrain through the ground point cloud, the accuracy of understory terrain extraction is significantly improved, and the problem of obvious breakpoints and jumps in the understory terrain is overcome. Brief Description of the Drawings

[0017] Figure 1 is a schematic flow chart of the understory terrain extraction method based on multi-baseline SAR data provided by the present invention;

[0018] Figure 2 and Figure 3 are respectively the RGB false color map and the SLC intensity map in the SAR image data set provided by the present invention;

[0019] Figure 4 、 Figure 5 and Figure 6 are respectively the interference phases after removing the phase error, removing the flat ground phase, and removing the terrain phase in the SAR image data set provided by the present invention;

[0020] Figure 7 is a schematic diagram of the forest tomography structure profile provided by the present invention, where Figure 7 (a)and Figure 7(b) are respectively schematic diagrams of the forest tomography structure profiles in the azimuth and range directions provided by the present invention;

[0021] Figure 8 and Figure 9 are respectively schematic diagrams of the forest tomography structure profiles with sidelobe signals retained and sidelobe signals eliminated provided by the present invention;

[0022] Figure 10 is a schematic diagram of the reconstructed three-dimensional forest point cloud provided by the present invention;

[0023] Figure 11 is a schematic diagram of the principle of cloth simulation filtering provided by the present invention;

[0024] Figure 12 and Figure 13 are respectively the ground points and non-ground points extracted from the three-dimensional forest point cloud by the MCSF algorithm provided by the present invention;

[0025] Figure 14 is the understory terrain model obtained by applying the method provided by the present invention. Detailed implementation manners

[0026] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0027] Refer to Figure 1 , which is a schematic flow chart of the method for extracting the understory terrain based on multi-baseline SAR data provided by the present invention.

[0028] In the embodiment of the present invention, a multi-baseline data set of the target area collected is obtained, an SAR image data set is synthesized, and the interferometric phase error in the SAR image data set is corrected.

[0029] Refer to Figure 2 and Figure 3 , which are respectively the RGB false color map and the single-look complex data SLC (Single Look Comple) intensity map of the SAR image data set provided by the present invention.

[0030] In the embodiment of the present invention, a P-band multi-baseline data set of the target area collected by an aircraft is obtained.

[0031] In a specific implementation, the SAR radar system can be installed on an aircraft or a drone, and radar images of ground targets (the target forest area) are obtained through the movement of the aircraft. During the propagation and reflection of radar signals, due to various factors, the phase or amplitude information thereof changes, thus losing the original coherence (the phase relationship between signals becomes unstable or cannot be kept consistent). Multi-baseline means using multiple radar antennas or multiple observation positions (usually through multiple flights or multiple satellites) to obtain multiple images of the same target area, and processing through the interference relationship between these images. Therefore, in SAR image processing, decoherence and interference phase errors will lead to a decline in image quality and reduce the accuracy of interferometric measurement.

[0032] In a specific implementation, the P-band dataset refers to radar remote sensing data collected within the P-band (the frequency range is about 300 MHz to 1 GHz, and the wavelength is about 30 cm to 1 m). The P-band radar data has the characteristic of strong penetration ability, can penetrate the vegetation canopy, and is suitable for monitoring forest biomass and the terrain under the forest.

[0033] In a specific implementation, in the embodiment of the present invention, the flight altitude of the carrier platform is about 4000 meters, the flight speed is about 80 m / s, the heavy-track flight interval is about 30 minutes, the average spatial baseline is about 60 meters, and a total of 6 scenes of images are obtained in this flight mission. Figure 2 It is an RGB false color map under the Pauli basis decomposition of the main image. Figure 3 It is the intensity map of the single-look complex data SLC (Single Look Comple).

[0034] Refer to Figure 4 、 Figure 5 and Figure 6 They are the interferometric phases after removing the phase error, removing the flat-earth phase, and removing the terrain phase from the SAR image dataset provided by the present invention respectively.

[0035] In the embodiment of the present invention, the phase error, flat-earth phase, and terrain phase in the SAR image dataset are removed.

[0036] In a specific implementation, data preprocessing is performed on the obtained SAR image dataset to correct the interference error caused by external factors of the airborne SAR system, and a spatial baseline with higher accuracy and a corrected interferometric phase are obtained; in the embodiment of the present invention, the corrected interferometric phase is as shown in the figure, wherein, Figure 4 It is the interferometric phase after removing the phase error. Figure 5 It is the interferometric phase after removing the flat-earth phase. Figure 6 It is the interferometric phase after removing the terrain phase.

[0037] In the embodiments of the present invention, the SAR image dataset is subjected to signal decomposition, and spectral reconstruction is used to perform tomography on the decomposed signals to obtain the forest tomography structure profile.

[0038] Refer to Figure 7 (a), which is a schematic diagram of the azimuthal forest tomography structure profile provided by the present invention. Refer to Figure 7 (b), which is a schematic diagram of the range forest tomography structure profile provided by the present invention.

[0039] In the embodiments of the present invention, the SAR image dataset is subjected to SKP signal decomposition; Capon spectral reconstruction is used to suppress the noise of the SAR echo signal and improve the resolution of tomography in the height direction.

[0040] In a specific implementation, SKP (Kronecker Product Sum) decomposition is an effective signal processing method that can decompose complex scattering mechanisms into multiple simple scattering components. Through SKP decomposition, the scattering characteristics of the forest canopy and the ground can be extracted, providing basic data for the reconstruction of the forest vertical structure.

[0041] In a specific implementation, in the embodiments of the present invention, the airborne radar system uses the HH polarization channel. The HH polarization channel means that the radar system emits electromagnetic waves in the horizontal direction and receives the echo signal in the horizontal direction. The microwave signal of the HH polarization channel is sensitive to the scattering characteristics of the forest underlying surface and is suitable for monitoring the forest understory terrain.

[0042] In a specific implementation, the spectral reconstruction method is used to suppress the noise of the SAR echo signal and improve the resolution of tomography in the height direction. For example, the scattering characteristics of the forest are complex, and the noise of the SAR echo signal is relatively serious; and due to the limitations of the data acquisition hardware, environment and other conditions, the spatial resolution of the original SAR signal is difficult to meet the requirements for extracting forest structure parameters. Therefore, selecting an appropriate spectral analysis method is of great significance for extracting the forest vertical structure parameters. The forest tomography structure profile is as shown in Figure 7 (a) and Figure 7 (b) (HH polarization channel, profile position: azimuth 150, range 150).

[0043] In the embodiments of the present invention, the sidelobe signals in the structure profile are removed, and the main lobe signals in the structure profile are retained.

[0044] Refer to Figure 8 and Figure 9 , which are schematic diagrams of the forest tomography structure profiles with sidelobe signals retained and sidelobe signals eliminated provided by the present invention.

[0045] In the embodiments of the present invention, the main lobe signal is extracted by using the sliding window method, and the process includes: setting the size of the sliding window, determining the maximum power position corresponding to the maximum backscattering coefficient within the sliding window in the structural profile; at the maximum power position, extracting the pixels corresponding to the backscattering signals within the upper and lower ranges of the window through the sliding window; and performing sidelobe signal elimination and main lobe signal extraction on the structural profile pixel by pixel in the azimuth direction or the range direction.

[0046] In specific implementation, the main lobe is the area where the signal energy is most concentrated, while the sidelobes are the energy distributions in other directions outside the main lobe. In SAR imaging, these sidelobes will capture the scattered energy around the target and form stray signals in the image. In the forest vertical structural profile, the sidelobe signals may mask the weak scattering signals inside the forest (such as small branches, vegetation canopies, etc.), thus affecting the accurate analysis of the forest structure. For example, the sidelobes of strong scattering points (such as tree trunks) may extend to adjacent vegetation areas, resulting in misjudgment of the signals in these areas.

[0047] In specific implementation, in the embodiments of the present invention, when extracting the main lobe signal from the forest tomography profile under HH polarization, the size of the upper-edge filtering window is set to 6, and the size of the lower-edge filtering window is set to 3. The extracted main lobe signal is the most complete. The forest profile diagram (HH polarization) after eliminating the sidelobe signal is as Figure 5 shown, where Figure 8 is the tomography profile retaining the sidelobe signal, Figure 9 is the tomography profile after eliminating the sidelobe signal.

[0048] In specific implementation, the backscattering coefficient is the reflection intensity after the interaction between the radar signal and the target, and is usually used to characterize the scattering characteristics of the target. The reason for determining the maximum power position corresponding to the maximum backscattering coefficient is that in the forest tomography profile, the stronger the HH polarization backscattering power, the closer the distribution of its corresponding elevation is to the ground surface, which is more conducive to extracting the terrain part.

[0049] In specific implementation, within each sliding window, find the maximum value of the backscattering coefficient and its position, which usually corresponds to the main lobe signal of the target. Centered on the maximum power position, retain the effective pixels within the upper and lower ranges of the window. The backscattering coefficients of these pixels usually contain the useful information of the target.

[0050] In the embodiments of the present invention, the reconstructed forest three-dimensional point cloud is obtained by projecting the main lobe signal in the structural profile.

[0051] Refer to Figure 10 , which is a schematic diagram of the reconstructed forest three-dimensional point cloud provided by the present invention.

[0052] In specific implementation, in the existing solution, a pre-set threshold is used to screen and process pixels of the forest tomography profile to extract terrain information, and the understory terrain is directly generated based on the processed forest tomography profile. The problem is that when using the threshold to process the forest tomography profile, that is, directly eliminating some parts that do not meet the threshold. However, for example, the reflectance after attenuation due to vegetation influence on the ground surface and near the ground surface may be very low. If the threshold is directly used to cut off, then the information representing the real ground surface in this part will be eliminated, which also leads to the discontinuous distribution of the backscattering power in the azimuth angle and range direction within the forest tomography profile, and obvious breakpoints and jumps appear in the obtained understory terrain.

[0053] In the present invention, the forest three-dimensional point cloud is reconstructed based on the forest tomography structure profile, and then the ground point cloud therein is extracted from the forest three-dimensional point cloud. The understory terrain is obtained based on the ground point cloud, avoiding the above-mentioned disadvantages of using the threshold to extract terrain information in the existing solution, and significantly improving the accuracy of understory terrain extraction. The reasons why the present invention can produce good technical effects are as follows: First, it maximally retains and utilizes the forest tomography structure profile information, that is, it includes the information representing the real ground surface in this part that was cut off by the threshold. Second, it converts the retained profile information into point cloud information, utilizes the advantages of point cloud ground point filtering, and effectively combines the two.

[0054] In the embodiment of the present invention, the steps of reconstructing the forest three-dimensional point cloud include: projecting the pixels to the corresponding height according to the elevation information corresponding to the pixels of the backscattering signal in the structure profile in the azimuth direction or the range direction, and reconstructing the forest three-dimensional point cloud.

[0055] In specific implementation, in the embodiment of the present invention, the image size is 359 (range direction) × 269 (azimuth direction). In Figure 9 the selected profile is the one corresponding to range direction 100. Among them, the X-axis (ranging from 1 to 269) represents the pixels in the azimuth direction corresponding to range direction 100, and the Y-axis represents the height to which this pixel is projected onto the three-dimensional point cloud. For example, the point (100, 120, 1650) means that the height corresponding to the pixel with range direction 100 and azimuth direction 120 is 1650 meters. Then, each pixel in range direction 100 is saved in this format in sequence, and then the entire image is converted into a three-dimensional point cloud in the range direction. Finally, the complete forest three-dimensional point cloud space is reconstructed. The same applies to the azimuth direction.

[0056] In the embodiment of the present invention, a rough digital terrain model representing the terrain undulation change is generated. After using the rough digital terrain model to perform elevation normalization processing on the forest three-dimensional point cloud, the elevation-normalized ground point cloud in the forest three-dimensional point cloud is extracted by using cloth simulation filtering, and the elevation-normalized ground point cloud is superimposed (such as added) with the rough digital terrain model to obtain the ground point cloud in the forest three-dimensional point cloud.

[0057] In a specific implementation, the MCSF (Modified Cloth Simulation Filter) provided by the present invention is an improvement based on the cloth simulation filtering algorithm CSF (Cloth Simulation Filter), and combines the digital terrain model DTM (Digital Terrain Model) to make it have a high classification accuracy in areas with large terrain undulations.

[0058] In an embodiment of the present invention, the three-dimensional forest point cloud is gridded in the horizontal direction, and the lowest point in each grid is used as a reference point to obtain a corresponding rough digital terrain model.

[0059] In a specific implementation, there are various schemes for generating the rough digital terrain model. In an embodiment of the present invention, it can also be obtained through the PTDF (progressive TIN densification filtering) algorithm. The lowest point and the four corner points within the X×X grid are selected as the initial seed points to construct the initial TIN model; then the TIN model is iterated, and in each iteration process, through preset threshold parameters (mainly including the maximum angle θ, the maximum distance d, and the maximum terrain slope s), each "potential point (unclassified point)" is judged point by point; finally, the DTM is generated according to the detected ground points. This method can improve the accuracy of the rough DTM. The purpose is to retain the main terrain change features, and the CSF algorithm has a good filtering effect in the area of detailed terrain changes, so it is not advisable to overly pursue the accuracy of the rough DTM.

[0060] In an embodiment of the present invention, the rough digital terrain model is used to perform elevation normalization processing on the three-dimensional forest point cloud. The processing process includes: for each target point in the three-dimensional forest point cloud, calculate the elevation difference between the elevation value of the target point and the elevation value of the corresponding point in the rough digital terrain model, and use the elevation difference as the elevation value of the target point, so as to obtain the three-dimensional forest point cloud after elevation normalization.

[0061] Refer to Figure 11 , which is a schematic diagram of the principle of the cloth simulation filtering provided by the present invention.

[0062] In a specific implementation, the process of cloth simulation filtering includes: (1) Initialization of cloth state: This process starts with the inversion of the point cloud, followed by the process of simulating a uniform cloth to cover this dataset; (2) Movement of particles: The particles within the cloth will displace in the vertical direction under the influence of gravity; (3) Collision detection: Under the influence of gravity, when the particles displace, some particles will reach below the ground surface, and at this time, the particle positions deviate from the real terrain situation; (4) Calculation of displacement driven by internal forces: This step mainly considers the internal driving factors among the particles within the cloth; (5) Calculation of the distance moved due to elevation differences between adjacent cloth points. If two adjacent nodes are both movable points and have different height values, these two points will move in opposite directions by the same distance in the vertical direction. If one of the two points is immovable, then the other point will be moved. If the two points have the same elevation, no displacement will occur, and the distance moved in these cases is called the correction displacement amount, from which the positions of the particles can be calculated. In the embodiment of the present invention, the parameters of the cloth bag simulation filtering algorithm can be set as follows: the classification threshold for ground points is 0.5 m, the cloth resolution is 0.1 m, the maximum number of iterations is 500 times, and the filtering scenario is set to flat terrain (because elevation normalization has been performed). Refer to Figure 12 , which are the ground points extracted by the MCSF algorithm. Refer to Figure 13 , which are the non-ground points extracted by the MCSF algorithm.

[0063] In the embodiment of the present invention, the elevation information of the ground point cloud is interpolated, and then projected onto a two-dimensional grid to form an underforest terrain model.

[0064] Refer to Figure 14 , which is the underforest terrain model obtained by applying the method provided by the present invention.

[0065] In a specific implementation, after separating the ground points by the MCSF algorithm, there may still be a small number of elevation missing values, which can be filled by the interpolation algorithm to complete the missing terrain values (elevation information), and then the interpolated ground point cloud data is projected onto a two-dimensional grid to form an underforest terrain model.

[0066] In a specific implementation, the generation process of the underforest terrain model includes: (1) Grid division: The entire area is divided into grids of a certain size, and each grid represents a region; (2) Point cloud projection: Each point in the ground point cloud is projected into the grid where it is located, and the grid where the point is located can be determined by calculating the minimum distance from the point to the grid boundary; (3) Formation of the underforest terrain model: In each grid, the elevation information of the point cloud is statistically analyzed, such as the highest point, the lowest point, the average elevation, etc., to form the underforest terrain model.

[0067] The present invention also provides an understory terrain extraction device based on multi-baseline SAR data, including an image data synthesis unit, a tomography unit, a signal extraction unit, a point cloud reconstruction unit, a ground point cloud extraction unit, and an understory terrain model generation unit, wherein: The image data synthesis unit is used to obtain the multi-baseline data set of the target area collected, synthesize the SAR image data set, and correct the interference phase error in the SAR image data set; The tomography unit is used to decompose the signals in the SAR image data set, perform tomography on the decomposed signals using spectral reconstruction, and obtain the forest tomography structure profile; The signal extraction unit is used to remove the sidelobe signals in the structure profile and retain the main lobe signals in the structure profile; The point cloud reconstruction unit is used to project the main lobe signals in the structure profile to obtain the reconstructed three-dimensional forest point cloud; The ground point cloud extraction unit is used to generate a rough digital terrain model representing the terrain undulation changes. After using the rough digital terrain model to perform elevation normalization processing on the three-dimensional forest point cloud, the elevation-normalized ground point cloud in the three-dimensional forest point cloud is extracted using cloth simulation filtering, and the elevation-normalized ground points are superimposed on the rough digital terrain model to obtain the ground point cloud in the three-dimensional forest point cloud; The understory terrain model generation unit is used to project the ground point cloud onto a two-dimensional grid to form an understory terrain model.

[0068] In a specific implementation, for the execution unit of the understory terrain extraction device based on multi-baseline SAR data provided by the present invention, which executes a method, step, or function, the method, step, or function it executes can refer to the understory terrain extraction method based on multi-baseline SAR data provided by the present invention.

Claims

1. A method for extracting understory terrain based on multi-baseline SAR data, characterized in that: include: Acquire the multi-baseline data set of the target area, synthesize it to obtain the SAR image data set, and correct the interference phase error in the SAR image data set; The SAR image data set is decomposed, and the decomposed signal is tomographically imaged using spectrum reconstruction to obtain the forest tomographic structure profile; Remove the side lobe signal in the structural section and retain the main lobe signal in the structural section; The reconstructed three-dimensional point cloud of the forest is obtained by projecting the main lobe signal in the structural section; Generate a coarse digital ground model that represents the undulation of the terrain, use the coarse digital ground model to perform elevation normalization on the forest 3D point cloud, use cloth simulation filtering to extract the elevation normalized ground point cloud in the forest 3D point cloud, and superimpose the elevation normalized ground point cloud on the coarse digital ground model to obtain the ground point cloud in the forest 3D point cloud; The elevation information of the ground point cloud is interpolated and then projected onto a two-dimensional grid to form an understory terrain model.

2. The method for extracting understory terrain based on multi-baseline SAR data according to claim 1, characterized in that: The acquiring of the target area multi-baseline data set includes: Acquire the P-band multi-baseline dataset of the target area acquired by the aircraft.

3. The method for extracting understory terrain based on multi-baseline SAR data according to claim 1, characterized in that: The interferometric phase error in the SAR image data set is corrected, including: Remove phase error, flat ground phase and terrain phase from SAR image datasets.

4. The method for extracting understory terrain based on multi-baseline SAR data according to claim 1, characterized in that: The signal decomposition of the SAR image data set and tomography of the decomposed signal using spectrum reconstruction include: Perform SKP signal decomposition on SAR image dataset; Capon spectrum reconstruction is used to suppress SAR echo signal noise and improve the resolution of tomography in the altitude direction.

5. The method for extracting understory terrain based on multi-baseline SAR data according to claim 1, characterized in that: The step of removing the side lobe signal in the structural section and retaining the main lobe signal in the structural section comprises: The size of the sliding window is set, and the maximum power position corresponding to the maximum backscatter coefficient within the sliding window is determined in the structural section; At the maximum power position, the pixels corresponding to the backscattered signals within the upper and lower ranges of the window are extracted through a sliding window; The sidelobe signal is eliminated and the mainlobe signal is extracted pixel by pixel on the structural profile in the azimuth or distance direction.

6. The method for extracting understory terrain based on multi-baseline SAR data according to claim 1, characterized in that: The method of obtaining a reconstructed three-dimensional forest point cloud by projecting the main lobe signal in the structural section includes: According to the elevation information corresponding to the pixels of the backscattered signal in the structural section, the pixels are projected to the corresponding heights in the azimuth or distance direction to reconstruct the three-dimensional point cloud of the forest.

7. The method for extracting understory terrain based on multi-baseline SAR data according to claim 6, characterized in that: The method of generating a rough digital ground model representing terrain fluctuations comprises: The forest 3D point cloud is gridded in the horizontal direction, and the lowest point in each grid is used as the reference point to obtain the corresponding coarse digital ground model.

8. The method for extracting understory terrain based on multi-baseline SAR data according to claim 6, characterized in that: After the elevation normalization of the forest three-dimensional point cloud using the rough digital ground model, the method includes: For each target point in the forest 3D point cloud, the elevation difference between the target point elevation value and the elevation value of the corresponding point in the rough digital ground model is calculated, and the elevation difference is used as the elevation value of the target point, thereby obtaining the forest 3D point cloud after elevation normalization.

9. A device for extracting understory terrain based on multi-baseline SAR data, characterized in that: It includes an image data synthesis unit, a tomography unit, a signal extraction unit, a point cloud reconstruction unit, a ground point cloud extraction unit and an understory terrain model generation unit, wherein: The image data synthesis unit is used to obtain the collected target area multi-baseline data set, synthesize it to obtain the SAR image data set, and correct the interference phase error in the SAR image data set; The tomographic imaging unit is used to perform signal decomposition on the SAR image data set, perform tomographic imaging on the decomposed signal using spectrum reconstruction, and obtain a forest tomographic structure profile; The signal extraction unit is used to remove the side lobe signal in the structural section and retain the main lobe signal in the structural section; The point cloud reconstruction unit is used to obtain a reconstructed three-dimensional point cloud of the forest by projecting the main lobe signal in the structural section; The ground point cloud extraction unit is used to generate a coarse digital ground model that represents the undulation of the terrain, use the coarse digital ground model to perform elevation normalization on the three-dimensional point cloud of the forest, use cloth simulation filtering to extract the elevation normalized ground point cloud in the three-dimensional point cloud of the forest, and superimpose the elevation normalized ground points on the coarse digital ground model to obtain the ground point cloud in the three-dimensional point cloud of the forest; The understory terrain model generating unit is used to interpolate the elevation information of the ground point cloud, and then project it to a two-dimensional grid to form an understory terrain model.

Images