An interferometric tomographic SAR forest canopy height estimation optimization method and system

Abstract

The application provides an interferometric tomography SAR forest canopy height estimation optimization method and system, and belongs to the technical field of forest canopy height estimation. In the first stage, envelope fitting is performed on the relative reflectivity signal of the vertical structure of the forest to obtain initial envelope lines of the forest canopy and the ground, and the corresponding relative reflectivity loss threshold K is determined by referring to the LiDAR height. In the second stage, the signal continuity threshold is used to filter out the noise interference at the top and bottom of the forest profile canopy to obtain more reasonable envelope lines. In the third stage, the K value is optimized based on the canopy density segmentation, the forest canopy density is graded, different grades are given different K values of the forest canopy and the ground, the relative reflectivity envelope line is extracted, and finally the more accurate forest height is obtained by subtracting the height index corresponding to the envelope lines of the canopy and the ground. The three-stage error optimization method effectively improves the estimation accuracy of the forest canopy height.

Description

Technical Field

[0001] This invention relates to the field of forest canopy height estimation technology, and specifically to an optimization method and system for estimating forest canopy height using interferometric tomography SAR. Background Technology

[0002] Forest height is a crucial indicator of forest growth and health, and a key parameter for estimating forest biomass and carbon storage. However, ground-based forest height measurements are costly and challenging due to factors such as transportation, topography, and the complexity of forest structure. In recent years, the development of UAVs and airborne LiDAR technology has provided excellent technical means for accurate measurement of forest tree height; however, the high cost of data acquisition makes large-scale forest height measurements difficult. Synthetic Aperture Radar (SAR) is an active remote sensing technology with all-weather, all-day operation, strong penetration, and a large observation range. It can penetrate most forest branches and leaves to reach the ground, thus obtaining vertical structure information of forests through SAR signals, compensating for the shortcomings of other remote sensing methods.

[0003] Forest parameters are extracted using synthetic aperture radar (SAR). Commonly used forest canopy height inversion models include the random volume on ground (RVoG) model, inversion methods utilizing the phase difference between the canopy and the ground, and TomoSAR. Representative models of the RVOG model include the three-stage RVOG inversion method and the maximum likelihood estimation algorithm. However, this model, based on an interferometric complex coherence model, has excessively high requirements for initial value settings, resulting in poor practical performance. Among inversion methods utilizing the phase difference between the canopy and the ground, the classic ESPRIT method obtains the phase center of the canopy and the ground and uses the difference between these phase centers to invert canopy height. However, due to the complexity of forest structure, errors exist in the phase center between the canopy and the ground surface. TomoSAR has been widely used for acquiring forest vertical structure. TomoSAR technology obtains three-dimensional structure by separating the reflected signals of target objects along the elevation direction, showing significant advantages in acquiring forest canopy height. In 2000, Reigber et al. first used L-band airborne datasets to obtain the three-dimensional structure of forests through Fast Fourier Transform (FFT) technology, achieving the first successful experiment in forest applications. This laid the foundation for many researchers to apply tomographic SAR to forests.

[0004] Currently, after acquiring forest vertical structure information using TomoSAR, the envelope method is commonly used to extract the envelope of the relative reflectivity signal of the forest vertical structure, thereby obtaining the forest canopy height. However, to implement TomoSAR technology, SAR data preprocessing, temporal decoherence, image reconstruction, and result analysis are required. Data preprocessing includes flat-ground phase removal, terrain phase correction, and phase calibration of SAR data. Due to differences in terrain, target features, and sensors, even after relevant preprocessing, some residual phase effects will remain. Temporal decoherence needs to consider the baseline time interval. Beamforming methods include non-parametric methods (Beamforming, Capon, etc.) and parametric methods (Music, WSF, etc.). The difference between parametric and non-parametric methods is that non-parametric methods do not require prior knowledge of the number of scatterers or scattering mechanisms, and can directly process the signal. In practical applications, due to various influencing factors such as terrain, baseline time interval, vegetation complexity, and uncertainties in beamforming methods, SAR data suffers from phase focusing errors and residual phase effects, causing the envelope to fall outside the vertical structure information of the forest. Furthermore, the penetration of microwaves through different forest covers amplifies the vertical information range of the forest, leading to an overestimation of forest canopy height.

[0005] In summary, after performing terrain phase correction and phase compensation on the data, some residual phase effects still exist due to various influencing factors such as terrain, baseline time interval, vegetation complexity, and uncertainties in beamforming methods. Therefore, how to further improve the accuracy of TomoSAR forest height estimation is a technical problem that urgently needs to be solved. Summary of the Invention

[0006] The purpose of this invention is to provide an optimization method and system for estimating forest canopy height using interferometric tomography SAR, so as to improve the accuracy of forest height inversion by at least reducing the relative reflection noise interference of TomoSAR.

[0007] To achieve the above objectives, this invention provides an optimization method for estimating forest canopy height using interferometric tomography SAR, the method comprising:

[0008] L-band UAVSAR multi-baseline data were read, and a multi-baseline InSAR covariance matrix was constructed based on the ground phase compensation results. The relative reflectance of the forest vertical profile was reconstructed using Capon spectral estimation.

[0009] Envelope fitting was performed on the relative reflectance signal of the forest vertical profile, and the initial envelope of the forest canopy, the initial envelope of the land surface, and the initial signal loss threshold were determined by combining LiDAR height reference data. ;

[0010] The relative reflectance signal of the forest vertical profile is subjected to noise filtering based on the continuity of the vertical direction signal to remove noise interference from the top of the canopy and the bottom of the ground, so as to obtain the denoised forest vertical structure reflectance profile and a more reasonable envelope.

[0011] Pauli decomposition was performed on the fully polarimetric SAR data to obtain the canopy density index. Different canopy signal loss thresholds were assigned to pixels of different canopy density levels according to the canopy density grading results. and surface signal loss threshold ;

[0012] The relative reflectivity signal is re-extracted from the envelope, and the forest canopy height is obtained by subtracting the height index corresponding to the canopy envelope and the ground surface envelope.

[0013] Optionally, the relative reflectance of the reconstructed forest vertical profile is estimated using Capon spectral estimation, including:

[0014] Preprocessing is performed on L-band UAVSAR multi-baseline data;

[0015] The ground phase is estimated by calling the three-stage multi-baseline RVOG method, and residual phase compensation is performed on the multi-baseline observation data based on the ground phase.

[0016] Calculate the multi-baseline InSAR covariance matrix R based on the compensated multi-baseline observation data;

[0017] The multi-baseline InSAR covariance matrix R is input into the Capon beamforming power estimator to calculate the relative reflectance in the vertical direction of the vegetated area. The specific expression is as follows:

[0018] ;

[0019] In the formula, Let z represent the vertical distribution function of backscattered power estimated using the Capon algorithm; z represents the height variable in the vertical direction; a(z) represents the steering vector at height z. R represents the conjugate transpose of the steering vector a(z); R represents the covariance matrix of the multi-baseline InSAR data. Let represent the inverse of the covariance matrix R.

[0020] Optionally, the initial envelope of the forest canopy, the initial envelope of the land surface, and the signal loss threshold K are determined, including:

[0021] Based on the maximum forest canopy height in the study area Apply upper limit constraints to the inversion results;

[0022] Set initial signal loss thresholds for different phase lengths. Extract the upper and lower envelopes of the TomoSAR relative reflectivity signal;

[0023] Based on the difference between the upper and lower envelopes, forest canopy height estimates under different threshold conditions were obtained, and inversion heights greater than a certain threshold were removed. The sample;

[0024] Select the optimal initial signal loss threshold based on the RMSE between the inversion results and LiDAR-RH100. The expression is as follows:

[0025] ;

[0026] in, This represents the optimal initial signal loss threshold; This represents the height of the upper envelope or the canopy side height extracted from the upper envelope under the current initial signal loss threshold condition; This represents the lower envelope height or surface height obtained from the lower envelope under the current initial signal loss threshold condition. This represents the forest canopy height estimate obtained under the current initial signal loss threshold. Indicates the LiDAR-RH100 reference altitude; This indicates the maximum forest canopy height in the study area.

[0027] Optionally, noise filtering based on vertical signal continuity is performed on the relative reflectance signal of the forest vertical profile, including:

[0028] Set a threshold T, and use this threshold T as the maximum critical value for the upward transition of the forest canopy to noise and the downward transition of the ground surface to noise; indexed by height. Compare the relative reflectivity functions layer by layer from the top to the bottom of the cross-section. When satisfied At that time, the corresponding height index is recorded as the feedback value. ,in The feedback height is the index in the feedback sequence; for Discarding is performed on height-level layers; after filtering all height-level layers, a feedback sequence ordered by height is obtained. And calculate the height difference between adjacent feedback points. , , Select the two adjacent feedback heights that produce the maximum adjacent feedback height difference. The main signal boundary is used as the boundary, and the relative reflectance corresponding to the height index between the main signal boundaries is extracted as the normal distribution range of the forest after denoising.

[0029] Optionally, setting a threshold T includes: comparing the threshold T with the initial signal loss threshold. A unified setting is adopted to ensure that the envelope extraction threshold and the noise continuity screening threshold are used for noise removal and signal boundary discrimination under the same data standard.

[0030] Optionally, Pauli decomposition can be performed on the fully polarimetric SAR data to obtain canopy density indices, including:

[0031] Perform Pauli decomposition on fully polarimetric SAR data;

[0032] Pauli decomposition brightness was extracted as an indicator of forest canopy density.

[0033] Based on the canopy density index, the study area was divided into sparse cover, medium cover, and dense cover levels;

[0034] Output the canopy density level label for each pixel.

[0035] Optionally, different canopy signal loss thresholds can be assigned to pixels of different canopy density grading results. and surface signal loss threshold ,include:

[0036] For sparse cover, medium cover, and dense cover levels, separate threshold sets for canopy envelope extraction and land surface envelope extraction are established; based on the canopy density level of each pixel... Call the corresponding canopy signal loss threshold and surface signal loss threshold ;Use the canopy signal loss threshold obtained from the call and surface signal loss threshold The canopy envelope and the land envelope are re-extracted separately, and adaptive correction is performed on the positional offset of the envelope. The corrected height indexes of the canopy envelope and the land envelope are then output.

[0037] Optionally, the corresponding canopy signal loss threshold can be applied based on the canopy density level to which each pixel belongs. and surface signal loss threshold The expression is as follows:

[0038] ; in, Indicates coverage level The corresponding canopy signal loss threshold, Indicates coverage level The corresponding ground signal loss threshold; This represents a threshold candidate set, from which a traversal search is performed to determine... The possible values ​​of ; Represents relative reflectance or the evaluation function corresponding to relative reflectance; Indicates position At the canopy signal loss threshold The height of the canopy envelope extracted under the given conditions; Indicates position At the ground signal loss threshold The height of the surface envelope extracted under the given conditions; Indicates position LiDAR reference height at the location; Indicates the level of forest cover; Indicates the level of dense cover; Indicates a medium coverage level; Indicates the sparse coverage level; Represents the spatial location index of a cell.

[0039] Optionally, the forest canopy height can be obtained by subtracting the height indices corresponding to the canopy envelope and the ground surface envelope, specifically:

[0040] Read the corrected height index corresponding to the canopy envelope and the height index corresponding to the land surface envelope;

[0041] Perform a difference calculation between the height index corresponding to the canopy envelope and the height index corresponding to the land surface envelope to obtain the pixel-by-pixel forest canopy height. The specific expression is as follows:

[0042] ;

[0043] In the formula, Indicates position Pixel-by-pixel forest canopy height; Indicates position Using the corresponding canopy signal loss threshold Corrected canopy envelope height index; Indicates position Using the corresponding surface signal loss threshold Corrected surface envelope height index;

[0044] Spatial aggregation was performed on all pixel-by-pixel forest canopy heights to output the forest canopy height results for the study area.

[0045] To achieve the above objectives, the present invention also provides an optimization system for estimating forest canopy height using interferometric tomography SAR, the system comprising:

[0046] The relative reflectance calculation module is used to read L-band UAVSAR multi-baseline data, construct the multi-baseline InSAR covariance matrix based on the ground phase compensation results, and reconstruct the relative reflectance of the forest vertical profile using Capon spectrum estimation.

[0047] The relative reflectance loss threshold K determination module is used to perform envelope fitting on the relative reflectance signal of the forest vertical profile, and combine LiDAR height reference data to determine the initial envelope of the forest canopy, the initial envelope of the land surface, and the initial signal loss threshold. ;

[0048] The noise error elimination module is used to perform noise filtering based on the vertical signal continuity on the relative reflectance signal of the forest vertical profile, remove noise interference from the top of the canopy and the bottom of the ground, and obtain a denoised forest vertical structure reflectance profile and a more reasonable envelope.

[0049] The forest canopy density grading module is used to perform Pauli decomposition on fully polarimetric SAR data to obtain canopy density indices, and assign different canopy signal loss thresholds to pixels of different grading levels according to the canopy density results. and surface signal loss threshold ;

[0050] The forest height inversion module is used to re-extract the envelope from the relative reflectivity signal and obtain the forest canopy height by subtracting the height index corresponding to the canopy envelope and the ground surface envelope.

[0051] Beneficial Effects: Through the above technical solution, this invention proposes an optimization method and system for estimating forest canopy height using interferometric tomographic SAR. First, the ground phase is estimated using the multi-baseline RVOG three-stage method, and some residual phase is removed to construct the multi-baseline InSAR covariance matrix. Based on this, Capon spectrum estimation (beamforming power estimator) is used to reconstruct the vertical relative reflectance profile of the vegetation area. Second, envelope fitting is performed on the vertical profile relative reflectance signal, and the initial signal loss threshold is determined by traversing the signal. The initial upper and lower envelopes of the canopy and the ground surface are extracted, and the canopy height is estimated from the difference between the two envelopes. Using LiDAR-CHM as a benchmark, the optimal initial signal loss threshold is determined when the RMSE of the estimated height is minimized. The initial envelope of the canopy and the ground surface is obtained accordingly. Subsequently, to address the noise residue at the top of the canopy and below the ground surface caused by phase focusing errors, a threshold-based denoising method based on signal continuity is proposed: a threshold T is set as the maximum critical value for the transition from the canopy upwards and from the ground surface downwards to noise; this value is compared and recorded layer by layer along the height index. The feedback height sequence is calculated, and the adjacent intervals are determined. The adjacent feedback height corresponding to the largest difference between adjacent feedback heights is used to determine the main signal boundary, and the relative reflectivity between them is preserved to obtain the true profile and stable envelope. Finally, to suppress the envelope system shift caused by microwave penetration differences under different canopy density conditions, the study area is divided into three levels—sparse, medium, and dense—based on Pauli decomposition brightness as a surrogate index of canopy density. Corresponding segment K is selected for each level to adaptively correct the envelope, thereby improving the accuracy and spatial consistency of forest canopy height estimation. This invention improves the accuracy of forest height estimation by correcting the relative reflectivity envelope error of interferometric tomographic SAR, and can effectively mitigate the impact of phase focusing error on forest height estimation.

[0052] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0053] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0054] Figure 1 This is a flowchart illustrating the optimization method for estimating forest canopy height using interferometric tomography SAR according to the present invention.

[0055] Figure 2 This is a schematic diagram illustrating the principle of envelope fitting and the corresponding determination of the signal loss threshold K in this invention.

[0056] Figure 3 This is a schematic diagram of the present invention based on noise removal of vertical signal continuity.

[0057] Figure 4 This is a diagram showing the noise removal results of the coverage grading method of this invention.

[0058] Figure 5 This is a diagram showing the change in canopy height before and after the optimization of this invention. Detailed Implementation

[0059] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0060] like Figure 1 As shown, this invention provides an optimization method for estimating forest canopy height using interferometric tomography SAR, the method comprising:

[0061] S1: Based on L-band UAVSAR data, the Capon spectrum estimation method is used to reconstruct the relative reflectivity of the forest vertical profile, reconstruct the forest vertical structure characteristics, and obtain the vegetation vertical structure profile. This method is based on the traditional minimum variance tomography method and achieves high-resolution imaging of target ground objects by optimizing the weight allocation of radar echo data.

[0062] S2: Fit the envelope of the relative reflectance signal in the vertical direction of the forest to obtain the initial envelope of the forest canopy and the ground surface, as well as the corresponding initial signal loss threshold. First, determine the maximum forest canopy height in the study area according to the following formula ( Set the initial signal loss threshold for different time lengths. Estimate the upper and lower envelope heights of TomoSAR to invert canopy heights under different conditions, and discard canopy heights greater than [value missing]. The sample is used to calculate the RMSE between the inversion result and LiDAR-RH100, and the initial signal loss threshold is determined when the RMSE is minimized. .

[0063] S3: Utilize vertical signal continuity to filter and eliminate noise interference from the top and bottom of the forest canopy, obtaining a more accurate envelope; first, set a threshold T as the maximum critical value for the upward transition of the forest canopy to noise and the downward transition of the ground surface to noise; then index by height. Comparing layer by layer from the top to the bottom of the cross-section, when Record the corresponding height index as the feedback value. ,when Recording continues until all height layers of the pixel have been filtered, resulting in a feedback sequence arranged in height order. Based on this, the height difference between adjacent feedback points is calculated. , , Since the effective signal span of the forest's vertical structure is significantly larger than the span of the top / bottom noise, the two adjacent feedback heights that generate the maximum adjacent feedback height difference are selected. As the boundary of the main signal, the relative reflectance corresponding to the height index between them is extracted as the normal distribution range of the forest after denoising, thereby effectively removing noise above the canopy and below the ground surface, and obtaining a stable forest vertical structure reflectance profile for subsequent height inversion.

[0064] S4: Selecting signal loss thresholds based on canopy density segments, incorporating LiDAR canopy height. Forest canopy density is graded, with different canopy density levels assigned different canopy signal loss thresholds. and surface signal loss threshold The envelope of the relative reflectance signal in the vertical direction of the forest is extracted, and the forest height is obtained more accurately by subtracting the height index corresponding to the envelope of the canopy and the ground surface.

[0065] In a preferred embodiment, in step S1, the relative reflectance of the forest in the vertical direction is reconstructed based on L-band UAVSAR data using the Capon spectral estimation method. This method is based on the traditional minimum variance tomography method and achieves high-resolution imaging of the target ground objects by optimizing the weight allocation of radar echo data.

[0066] In a preferred embodiment, in step S2, the relative reflectance signal of the forest vertical profile is fitted with an envelope to obtain the initial envelope of the forest canopy and the ground surface, as well as the corresponding initial signal loss threshold. First, determine the maximum forest canopy height in the study area according to the following formula ( Set the initial signal loss threshold for different time lengths. Estimate the upper and lower envelope heights of TomoSAR to invert canopy heights under different conditions, and discard canopy heights greater than [value missing]. The sample is analyzed, and the RMSE between the inversion result and LiDAR-RH100 is calculated. The optimal initial signal loss threshold is determined when the RMSE is minimized. .

[0067] In a preferred embodiment, in step S3, noise interference from the top and bottom of the forest canopy is filtered out using a signal continuity threshold to obtain a more accurate envelope; firstly, a threshold T is set as the maximum critical value for the upward transition of the forest canopy to noise and the downward transition of the ground surface to noise; then, it is indexed by height. Comparing layer by layer from the top to the bottom of the cross-section, when Record the corresponding height index as the feedback value. ,when Recording continues until all height layers of the pixel have been filtered, resulting in a feedback sequence arranged in height order. Based on this, the height difference between adjacent feedback points is calculated. , , Since the effective signal span of the forest's vertical structure is significantly larger than the span of the top / bottom noise, the two adjacent feedback heights that generate the maximum adjacent feedback height difference are selected. As the boundary of the main signal, the relative reflectance corresponding to the height index between them is extracted as the normal distribution range of the forest after denoising, thereby effectively removing noise above the canopy and below the ground surface, and obtaining a stable forest vertical structure reflectance profile for subsequent height inversion.

[0068] In a preferred embodiment, in step S4, a signal loss threshold is selected based on canopy density segmentation, taking into account LiDAR canopy height. Forest canopy density is graded, and different canopy densities are assigned different canopy signal loss thresholds. and surface signal loss threshold The envelope of the relative reflectance signal in the vertical direction of the forest is extracted, and the forest height is obtained more accurately by subtracting the height index corresponding to the envelope of the canopy and the ground surface.

[0069] Therefore, this embodiment proposes an optimization method for estimating forest canopy height using interferometric tomography SAR. By correcting the relative reflectivity envelope error of interferometric tomography SAR, the accuracy of forest height estimation is improved, which can effectively mitigate the impact of phase focusing error on forest height estimation.

[0070] To more clearly explain this application, a specific example of the optimization method for estimating forest canopy height using interferometric tomography SAR is provided below. For example... Figure 2 As shown, this embodiment constructs a forest height inversion method and system based on L-band UAVSAR multi-baseline fully polarimetric data and LiDAR-CHM constraints, consisting of "vertical profile reconstruction—envelope extraction—denoising correction—hierarchical threshold adaptation". First, the ground phase is estimated using the multi-baseline RVOG three-stage method, and some residual phase is removed to construct the multi-baseline InSAR covariance matrix. Based on this, Capon spectrum estimation (beamforming power estimator) is used to reconstruct the vertical relative reflectance profile of the vegetation area. Second, envelope fitting is performed on the vertical profile relative reflectance signal, and the initial signal loss threshold is traversed. The initial upper and lower envelopes of the canopy and the ground surface are extracted, and the canopy height is estimated from the difference between the two envelopes. Using LiDAR-CHM as a benchmark, the optimal initial signal loss threshold is determined when the RMSE of the estimated height is minimized. The initial envelope of the canopy and the ground surface is obtained accordingly. Subsequently, to address the noise residue at the top of the canopy and below the ground surface caused by phase focusing errors, a threshold-based denoising method based on signal continuity is proposed: a threshold T is set as the maximum critical value for the transition from the canopy upwards and from the ground surface downwards to noise; this value is compared and recorded layer by layer along the height index. Feedback height sequence Calculate adjacent intervals , , The boundary of the main signal is determined by the adjacent feedback height corresponding to the maximum adjacent feedback height difference, and the relative reflectivity between them is preserved to obtain the true profile and stable envelope. Finally, to suppress the envelope system shift caused by microwave penetration differences under different canopy density conditions, the study area is divided into three levels—sparse, moderate, and dense—based on Pauli decomposition brightness as a proxy index for canopy density, and corresponding canopy signal loss thresholds are selected for each level. and surface signal loss threshold Adaptive correction of the forest canopy height envelope improves the accuracy and spatial consistency of forest canopy height estimation. This invention improves the accuracy of forest biomass estimation by correcting the relative reflectivity envelope error of interferometric tomographic SAR, effectively mitigating the impact of phase focusing errors on forest height estimation.

[0071] Regarding step S1: Based on L-band UAVSAR data, the Capon spectral estimation method is used to reconstruct the relative reflectivity of the forest vertical profile, reconstruct the forest vertical structure characteristics, and obtain the vegetation vertical structure profile. This method is based on the traditional minimum variance tomography method and achieves high-resolution imaging of target features by optimizing the weight allocation of radar echo data. The specific implementation steps are as follows:

[0072] (1) Data processing:

[0073] The L-band airborne multi-baseline PolInSAR data and lidar validation data were both derived from publicly available datasets from the AfriSAR project. The PolInSAR datasets underwent polarization calibration, baseline fine registration, and spectral filtering, and were provided in single-look complex format, with each orbit containing SLC data for four polarization channels. The test area had eight orbits; the data underwent multi-look processing, and the RVOG three-stage method was used to estimate the ground phase to remove some phase errors and improve phase focusing. The multi-baseline InSAR covariance matrix was also calculated.

[0074] (2) Capon beamforming power estimator:

[0075] The Capon beamforming power estimator is used to estimate the relative reflectance of the vegetation area in the vertical direction. The Capon spectral estimator is a common nonparametric method in tomographic analysis, capable of obtaining an infinite vertical profile of vegetation without any prior knowledge of the statistical properties of the data. This method is an improved algorithm based on the minimum variance criterion of conventional beamforming. It uses optimal weighting vectors to perform spatial filtering on the signals of each array element to suppress noise interference and enhance the desired signal. The spectral estimation formula is as follows:

[0076] ;

[0077] In the formula, The vertical distribution function of backscattered power is estimated using the Capon algorithm. Let z represent the steering vector at height z, and R represent the covariance matrix of the multi-baseline InSAR data.

[0078] Regarding step S2: The relative reflectance signal of the forest vertical profile is fitted with an envelope to obtain the initial envelopes of the forest canopy and the ground surface, as well as the corresponding initial signal loss thresholds. ;like Figure 2 As shown, Figure 2 A schematic diagram for envelope fitting and reflectance loss threshold determination (the upper figure shows envelope fitting, and the lower figure shows the schematic diagram for determining the reflectance loss threshold). First, the maximum forest canopy height in the study area is determined according to the following formula ( Set the initial signal loss threshold for different time lengths. Estimate the upper and lower envelope heights of TomoSAR to invert canopy heights under different conditions, and discard canopy heights greater than [value missing]. The sample is used to calculate the RMSE between the inversion result and LiDAR-RH100. The K value is determined when the RMSE is minimized. The specific implementation steps are as follows:

[0079] ;

[0080] in, This represents the optimal initial signal loss threshold; This represents the height of the upper envelope or the canopy side height extracted from the upper envelope under the current initial signal loss threshold condition; This represents the lower envelope height or surface height obtained from the lower envelope under the current initial signal loss threshold condition. This represents the forest canopy height estimate obtained under the current initial signal loss threshold. Indicates the LiDAR-RH100 reference altitude; This indicates the maximum forest canopy height in the study area.

[0081] Regarding step S3: Noise interference from the top and bottom of the forest canopy is removed using a signal continuity threshold filter to obtain a more accurate envelope, such as... Figures 3-4 As shown, Figure 3 This is a schematic diagram of noise removal based on vertical signal continuity (the top image shows the original envelope; the bottom image shows a schematic diagram of noise removal based on vertical continuity checks). Figure 4 A schematic diagram illustrating the changes in the envelope before and after optimization is provided, including the following implementation steps:

[0082] First, a threshold T is set as the maximum critical value for the upward transition of noise from the forest canopy and the downward transition of noise from the ground surface; then indexed by height. Comparing layer by layer from the top to the bottom of the cross-section, when Record the corresponding height index as the feedback value. ,when Recording continues until all height layers of the pixel have been filtered, resulting in a feedback sequence arranged in height order. Based on this, the height difference between adjacent feedback points is calculated. , , Since the effective signal span of the forest's vertical structure is significantly larger than the span of the top / bottom noise, the two adjacent feedback heights that generate the maximum adjacent feedback height difference are selected. As the boundary of the main signal, the relative reflectance corresponding to the height index between them is extracted as the normal distribution range of the forest after denoising, thereby effectively removing noise above the canopy and below the ground surface, and obtaining a stable forest vertical structure reflectance profile for subsequent height inversion.

[0083] Regarding step S4: Optimization of K-value selection based on canopy density segmentation, the specific implementation steps are as follows:

[0084] Based on LiDAR canopy height, signal loss thresholds were selected according to canopy density segmentation. First, Pauli decomposition was performed on the fully polarimetric SAR data. The brightness of the Pauli decomposition was used as an indicator of canopy density, dividing the study area into three cover levels: sparse, moderate, and dense. Corresponding canopy signal loss thresholds were then applied to the extraction of canopy and surface envelopes under different cover levels. and surface signal loss threshold This allows for adaptive correction of envelope position offset and calculation of the upper and lower envelope intervals to obtain forest height.

[0085]

[0086] in, Indicates coverage level The corresponding canopy signal loss threshold, Indicates coverage level The corresponding ground signal loss threshold; This represents the threshold candidate set, from which a traversal search is performed to determine... The possible values ​​of ; Represents relative reflectance or the evaluation function corresponding to relative reflectance; Indicates position At the canopy signal loss threshold The height of the canopy envelope extracted under the given conditions; Indicates position At the ground signal loss threshold The height of the surface envelope extracted under the given conditions; Indicates position LiDAR reference height at the location; Indicates the level of forest cover; Indicates the level of dense cover; Indicates a medium coverage level; Indicates the sparse coverage level; Represents the spatial location index of a cell, such as Figure 5 As shown, Figure 5 The results show the forest height predictions before and after optimization.

[0087] In a preferred embodiment, an optimization system for estimating forest canopy height using interferometric tomography SAR is also proposed, the system comprising:

[0088] The relative reflectance calculation module is used to read L-band UAVSAR multi-baseline data, construct the multi-baseline InSAR covariance matrix based on the ground phase compensation results, and reconstruct the relative reflectance of the forest vertical profile using Capon spectrum estimation.

[0089] The relative reflectance loss threshold K determination module is used to perform envelope fitting on the relative reflectance signal of the forest vertical profile, and combine LiDAR height reference data to determine the initial envelope of the forest canopy, the initial envelope of the ground surface, and the signal loss threshold K;

[0090] The noise error elimination module is used to perform noise filtering based on the vertical signal continuity on the relative reflectance signal of the forest vertical profile, remove noise interference from the top of the canopy and the bottom of the ground, and obtain a denoised forest vertical structure reflectance profile and a more reasonable envelope.

[0091] The forest canopy density grading module is used to perform Pauli decomposition on fully polarimetric SAR data to obtain canopy density indices, and assign different canopy signal loss thresholds to pixels of different grading levels according to the canopy density results. and surface signal loss threshold ;

[0092] The forest height inversion module is used to re-extract the envelope from the relative reflectivity signal and obtain the forest canopy height by subtracting the height index corresponding to the canopy envelope and the ground surface envelope.

[0093] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0094] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0095] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. An optimization method for estimating forest canopy height using interferometric tomography SAR, characterized in that, The method includes: L-band UAVSAR multi-baseline data were read, and a multi-baseline InSAR covariance matrix was constructed based on the ground phase compensation results. The relative reflectance of the forest vertical profile was reconstructed using Capon spectral estimation. Envelope fitting was performed on the relative reflectance signal of the forest vertical profile, and the initial envelope of the forest canopy, the initial envelope of the land surface, and the initial signal loss threshold were determined by combining LiDAR height reference data. ; The relative reflectance signal of the forest vertical profile is subjected to noise filtering based on the continuity of the vertical direction signal to remove noise interference from the top of the canopy and the bottom of the ground, so as to obtain the denoised forest vertical structure reflectance profile and a more reasonable envelope. Specifically, noise filtering based on the vertical signal continuity is performed on the relative reflectance signal of the forest vertical profile, including: Set a threshold T, and use this threshold T as the maximum critical value for the upward transition of the forest canopy to noise and the downward transition of the ground surface to noise; indexed by height. Compare the relative reflectivity functions layer by layer from the top to the bottom of the cross-section. When satisfied At that time, the corresponding height index is recorded as the feedback value. ,in The feedback height is the index in the feedback sequence; for Discarding is performed on height-level layers; after filtering all height-level layers, a feedback sequence ordered by height is obtained. And calculate the height difference between adjacent feedback points. , , Select the two adjacent feedback heights that produce the largest adjacent feedback height difference. As the boundary of the main signal, the relative reflectance corresponding to the height index between the main signal boundaries is extracted as the normal distribution range of the forest after denoising; Setting the threshold T includes: comparing the threshold T with the initial signal loss threshold. A unified setting is used to ensure that the envelope extraction threshold and the noise continuity screening threshold are used for noise removal and signal boundary discrimination under the same data standard; Pauli decomposition was performed on the fully polarimetric SAR data to obtain the canopy density index. Different canopy signal loss thresholds were assigned to pixels of different canopy density levels according to the canopy density classification results. and surface signal loss threshold Specifically, this includes: Pauli decomposition is performed on the fully polarimetric SAR data; Pauli decomposition brightness is extracted as a forest canopy density index; the study area is divided into sparse cover, medium cover, and dense cover levels based on the canopy density index; the canopy density level label of each pixel is output; canopy envelope extraction threshold sets and surface envelope extraction threshold sets are established for sparse cover, medium cover, and dense cover levels respectively; based on the canopy density level of each pixel... Call the corresponding canopy signal loss threshold and surface signal loss threshold ;Use the canopy signal loss threshold obtained from the call and surface signal loss threshold The canopy envelope and the land envelope are re-extracted separately, and adaptive correction is performed on the positional offset of the envelope. The corrected canopy envelope height index and land envelope height index are then output. The relative reflectivity signal is re-extracted from the envelope, and the forest canopy height is obtained by subtracting the height index corresponding to the canopy envelope and the ground surface envelope.

2. The method for optimizing forest canopy height estimation using interferometric tomography SAR according to claim 1, characterized in that, The relative reflectance of the reconstructed forest vertical profile was estimated using Capon spectral estimation, including: Preprocessing is performed on L-band UAVSAR multi-baseline data; The ground phase is estimated by calling the three-stage multi-baseline RVOG method, and residual phase compensation is performed on the multi-baseline observation data based on the ground phase. Calculate the multi-baseline InSAR covariance matrix R based on the compensated multi-baseline observation data; The multi-baseline InSAR covariance matrix R is input into the Capon beamforming power estimator to calculate the relative reflectance in the vertical direction of the vegetated area. The specific expression is as follows: ； In the formula, Let z represent the vertical distribution function of backscattered power estimated using the Capon algorithm; z represents the height variable in the vertical direction; a(z) represents the steering vector at height z. R represents the conjugate transpose of the steering vector a(z); R represents the covariance matrix of the multi-baseline InSAR data. Let represent the inverse of the covariance matrix R.

3. The optimization method for estimating forest canopy height using interferometric tomography SAR according to claim 1, characterized in that, Determine the initial envelope of the forest canopy, the initial envelope of the land surface, and the initial signal loss threshold. ,include: Based on the maximum forest canopy height in the study area Apply upper limit constraints to the inversion results; Set initial signal loss thresholds for different phase lengths. Extract the upper and lower envelopes of the TomoSAR relative reflectivity signal; Based on the difference between the upper and lower envelopes, forest canopy height estimates under different threshold conditions were obtained, and inversion heights greater than a certain threshold were removed. ; Select the optimal initial signal loss threshold based on the RMSE between the inversion results and LiDAR-RH100. The expression is as follows: ； in, This represents the optimal initial signal loss threshold; This represents the height of the upper envelope or the canopy side height extracted from the upper envelope under the current initial signal loss threshold condition; This represents the lower envelope height or surface height obtained from the lower envelope under the current initial signal loss threshold condition. This represents the forest canopy height estimate obtained under the current initial signal loss threshold. Indicates the LiDAR-RH100 reference altitude; This indicates the maximum forest canopy height in the study area.

4. The optimization method for estimating forest canopy height using interferometric tomography SAR according to claim 1, characterized in that, According to the canopy density level to which each pixel belongs Call the corresponding canopy signal loss threshold and surface signal loss threshold The expression is as follows: ； in, Indicates coverage level The corresponding canopy signal loss threshold, Indicates coverage level The corresponding ground signal loss threshold; This represents a threshold candidate set, from which a traversal search is performed to determine... The possible values ​​of ; Represents relative reflectance or the evaluation function corresponding to relative reflectance; Indicates position At the canopy signal loss threshold The height of the canopy envelope extracted under the given conditions; Indicates position At the ground signal loss threshold The height of the surface envelope extracted under the given conditions; Indicates position LiDAR reference height at the location; Indicates the level of forest cover; Indicates the level of dense cover; Indicates a medium coverage level; Indicates the sparse coverage level; Represents the spatial location index of a cell.

5. The method for optimizing forest canopy height estimation using interferometric tomography SAR according to claim 4, characterized in that, The forest canopy height is obtained by subtracting the height indices corresponding to the canopy envelope and the ground surface envelope. Specifically: Read the corrected height index corresponding to the canopy envelope and the height index corresponding to the land surface envelope; Perform a difference calculation between the height index corresponding to the canopy envelope and the height index corresponding to the land surface envelope to obtain the pixel-by-pixel forest canopy height. The specific expression is as follows: ； In the formula, Indicates position Pixel-by-pixel forest canopy height; Indicates position Using the corresponding canopy signal loss threshold Corrected canopy envelope height index; Indicates position Using the corresponding surface signal loss threshold Corrected surface envelope height index; Spatial aggregation was performed on all pixel-by-pixel forest canopy heights to output the forest canopy height results for the study area.

6. An optimization system for estimating forest canopy height using interferometric tomography SAR, characterized in that, The system includes: The relative reflectance calculation module is used to read L-band UAVSAR multi-baseline data, construct the multi-baseline InSAR covariance matrix based on the ground phase compensation results, and reconstruct the relative reflectance of the forest vertical profile using Capon spectrum estimation. The relative reflectance loss threshold K determination module is used to perform envelope fitting on the relative reflectance signal of the forest vertical profile, and combine LiDAR height reference data to determine the initial envelope of the forest canopy, the initial envelope of the land surface, and the initial signal loss threshold. ; The noise error elimination module is used to perform noise filtering based on the vertical signal continuity on the relative reflectance signal of the forest vertical profile, remove noise interference from the top of the canopy and the bottom of the ground, and obtain a denoised forest vertical structure reflectance profile and a more reasonable envelope. Specifically, noise filtering based on the vertical signal continuity is performed on the relative reflectance signal of the forest vertical profile, including: Set a threshold T, and use this threshold T as the maximum critical value for the upward transition of the forest canopy to noise and the downward transition of the ground surface to noise; indexed by height. Compare the relative reflectivity functions layer by layer from the top to the bottom of the cross-section. When satisfied At that time, the corresponding height index is recorded as the feedback value. ,in The feedback height is the index in the feedback sequence; for Discarding is performed on height-level layers; after filtering all height-level layers, a feedback sequence ordered by height is obtained. And calculate the height difference between adjacent feedback points. , , Select the two adjacent feedback heights that produce the largest adjacent feedback height difference. As the boundary of the main signal, the relative reflectance corresponding to the height index between the main signal boundaries is extracted as the normal distribution range of the forest after denoising; Setting the threshold T includes: comparing the threshold T with the initial signal loss threshold. A unified setting is used to ensure that the envelope extraction threshold and the noise continuity screening threshold are used for noise removal and signal boundary discrimination under the same data standard; The forest canopy density grading module is used to perform Pauli decomposition on fully polarimetric SAR data to obtain canopy density indices, and assign different canopy signal loss thresholds to pixels of different grading levels according to the canopy density results. and surface signal loss threshold Specifically, this includes: Pauli decomposition is performed on the fully polarimetric SAR data; Pauli decomposition brightness is extracted as a forest canopy density index; the study area is divided into sparse cover, medium cover, and dense cover levels based on the canopy density index; the canopy density level label of each pixel is output; canopy envelope extraction threshold sets and surface envelope extraction threshold sets are established for sparse cover, medium cover, and dense cover levels respectively; based on the canopy density level of each pixel... Call the corresponding canopy signal loss threshold and surface signal loss threshold ;Use the canopy signal loss threshold obtained from the call and surface signal loss threshold The canopy envelope and the land envelope are re-extracted separately, and adaptive correction is performed on the positional offset of the envelope. The corrected canopy envelope height index and land envelope height index are then output. The forest height inversion module is used to re-extract the envelope from the relative reflectivity signal and obtain the forest canopy height by subtracting the height index corresponding to the canopy envelope and the ground surface envelope.

Images