Satellite high-precision image pair matching and reconstruction method and system based on task planning

Abstract

The invention discloses a satellite high-precision image pair matching and reconstruction method and system based on task planning, and belongs to the technical field of satellite autonomous task planning. The method comprises the following steps: generating a plurality of three-dimensional image pairs based on a current remote sensing image, and calculating characteristic parameters of the three-dimensional image pairs based on satellite orbit data corresponding to the current remote sensing image; on the basis of the parameters of the three-dimensional image pair and the observation time window of the current remote sensing image, new satellite orbit data are acquired in combination with target requirements, a plurality of new observation time windows are obtained through orbit visibility calculation, and a plurality of new remote sensing images are generated according to the new observation time windows; and obtaining an optimal new three-dimensional image pair based on the track related data and constraints of the new remote sensing image and the DSM error of the new three-dimensional image pair. According to the invention, efficient and high-quality satellite observation scheme selection can be realized, so that more high-precision three-dimensional surface reconstruction images can be obtained.

Description

Technical Field

[0001] The present invention belongs to the technical field of satellite autonomous mission planning, and in particular relates to a satellite high-precision image pair matching and reconstruction method and system based on mission planning. Background Art

[0002] With the continuous development of remote sensing technology, three-dimensional Earth surface reconstruction technology has been widely used in fields such as disaster prevention and mitigation, emergency preparedness, and urban planning. By acquiring photogrammetric images with appropriate intersection angles and applying geometric analysis, a stereoscopic view of a specific area can be achieved. Unmanned aerial vehicle (UAV) aerial imagery can achieve three-dimensional reconstruction of an area, but compared to UAVs, satellite imagery not only covers a wider geographic area but also captures Earth observation data in a shorter time, allowing the generation of a three-dimensional reconstruction of any target area worldwide. By applying stereo matching algorithms to generate dense three-dimensional point clouds and digital surface models (DSMs), high-spatial-resolution satellite imagery is obtained, from which high-quality three-dimensional scene information can be reconstructed.

[0003] Planning coordinated remote sensing satellite missions requires comprehensive consideration of multiple factors, including satellite orbit design, acquisition timing, and coordination strategies with other satellites. By accurately calculating and simulating different time windows, optimal acquisition times and angles can be determined to minimize the impact of external factors, such as lighting variations, on image quality. A sound coordination strategy not only ensures the coordination of multiple satellites in observation missions, avoiding mission conflicts and resource waste, but also improves overall observation efficiency.

[0004] In Earth satellite observation, remote sensing imagery acquisition primarily relies on satellite push-broom imaging, a method prone to errors when performing three-dimensional stereo reconstruction. After acquiring an emergency mission, when images need to be selected, their quality is affected by a variety of factors, including the intersection angle between images and the solar altitude, which can lead to significant uncertainty and unpredictability. Currently, the workflow for three-dimensional reconstruction following satellite remote sensing imaging is typically an "acquire first, select later" model, where satellite images are first acquired and then appropriate images are selected to form stereo pairs. However, this method not only fails to achieve high-precision three-dimensional reconstruction of the Earth's surface, but also increases unnecessary remote sensing data acquisition and transmission. Summary of the Invention

[0005] The present invention provides a method and system for matching and reconstructing satellite high-precision image pairs based on mission planning. By developing a quality assessment model for high-precision stereo image pairs and applying a multi-objective optimization algorithm, efficient and high-quality satellite observation scheme selection is achieved, thereby obtaining more high-precision three-dimensional surface reconstructed images.

[0006] To achieve the above objectives, the technical implementation scheme of the present invention includes the following contents.

[0007] A method for matching and reconstructing satellite high-precision images based on mission planning, the method comprising:

[0008] generating a plurality of stereo image pairs based on a current remote sensing image, and calculating characteristic parameters of the stereo image pairs based on satellite orbit data corresponding to the current remote sensing image;

[0009] Based on the parameters of the stereo image pair and the observation time window of the current remote sensing image, new satellite orbit data is acquired in combination with target requirements, a plurality of new observation time windows are obtained through orbit visibility calculation, and a plurality of new remote sensing images are generated according to the new observation time windows;

[0010] An optimal new stereo image pair is obtained based on orbit-related data and constraints of the new remote sensing image and a DSM error of the new stereo image pair; wherein the new stereo image pair is generated based on the two new remote sensing images.

[0011] Furthermore, the characteristic parameters of the stereo image pair include: the average value of the intersection angle of the two current remote sensing images, the maximum side swing angle in the two current remote sensing images, the smaller solar altitude angle in the two current remote sensing images, the solar altitude angle difference in the two current remote sensing images, the solar azimuth angle difference in the two current remote sensing images, the comprehensive spatial angle difference of the solar altitude angle and azimuth angle in the two current remote sensing images, and the date interval between the two current remote sensing images.

[0012] Furthermore, a DSM error of the new stereo image pair is obtained based on a regression model; wherein the training process of the regression model includes:

[0013] Obtaining a training data set consisting of remote sensing images, and combining the remote sensing image data in pairs to perform three-dimensional surface reconstruction;

[0014] After the three-dimensional reconstruction, a plurality of stereo image pair samples are formed, and a reconstruction accuracy value of each stereo image pair sample is obtained;

[0015] According to the parameters of the remote sensing image, the characteristic parameters X of the stereo image pair samples are extracted k , k is a positive integer;

[0016] Calculate each feature parameter X k Pearson correlation coefficient with the reconstruction accuracy value;

[0017] According to the Pearson correlation coefficient, the characteristic parameter X k Filter out the feature parameters that are highly correlated with the reconstruction accuracy and obtain the feature set X′;

[0018] Perform different feature transformations on the feature parameters in the feature set X′ to capture nonlinear relationships and perform data expansion to obtain the feature parameter set X″;

[0019] Perform Lasso feature selection on all feature parameter sets X″ to obtain feature parameter set X″′;

[0020] Performing catBoost regression modeling based on the feature parameter set X″′ and the reconstruction accuracy value of the stereo image pair sample to obtain an initial regression model;

[0021] Take the coefficient of determination R 2 The current regression model is evaluated using the root mean square error and MAE indicators, and the model is iterated according to the evaluation results until the fitting accuracy of the current regression model reaches the set threshold.

[0022] Furthermore, the calculation of each characteristic parameter X k Pearson correlation coefficients with reconstruction accuracy values, including:

[0023] Get the characteristic parameter X k The set of observation values ​​and the set of reconstruction accuracy values ​​corresponding to the observation values, and calculate the average observation value and the average reconstruction accuracy

[0024] Based on the average observation and the average reconstruction accuracy Calculate the characteristic parameter X k Pearson correlation coefficient with the reconstruction accuracy value.

[0025] Furthermore, the orbit-related data of the new remote sensing image includes: the roll angle of the new remote sensing image and the intersection angle of any two new remote sensing images;

[0026] The obtaining of an optimal new stereo image pair based on the orbit-related data and constraints of the new remote sensing image and the DSM error of the new stereo image pair comprises:

[0027] Based on the roll angle of the new remote sensing image, the intersection angle of any two new remote sensing images, and the DSM error of the new stereo image pair, a multi-objective problem planning is generated, and an objective function Obj of the multi-objective problem planning is constructed; wherein the objectives include: the number of tasks, the reconstructed DSM error, and the timeliness of the tasks;

[0028] Setting constraints for solving the objective function; wherein the constraints include: a mission time span constraint, a maneuvering time limit, a swing angle constraint, and an intersection angle constraint. The mission time span constraint is used to specify that the observation time window of each mission must be within the entire mission cycle. The maneuvering time limit is used to ensure that the satellite has sufficient maneuvering time to adjust its attitude for two consecutive observations. The swing angle constraint is used to limit the satellite swing angle to be less than a maximum range. The intersection angle constraint is used to limit the intersection angle value of the new stereo imaging pair to be within a valid range.

[0029] Combined with the constraints, the objective function Obj is optimized and matched using the Grey Wolf algorithm to obtain the best new stereo image pair.

[0030] Furthermore, the objective function Obj=ω1·Obj1+ω2·(-Obj2)+ω3·(-Obj3); wherein ω1, ω2, and ω3 are the first weight, the second weight, and the third weight respectively, and the first objective function Obj1=∑ i,j x i,j , the second objective function Obj2=∑ i ∑ j DSM(i,j)·x i,j , the third objective function Obj3=∑ i ∑ j max(te i ,te j )·x i,j , x i,j represents the decision variable for selecting a new stereo image pair (i, j), x i,j =1 means selected, x i,j =0 means not selected, i and j represent the new remote sensing image i and the new remote sensing image j respectively, DSM(i,j) represents the DSM error of the new stereo image pair (i,j), te i and te j represent the imaging time of the new remote sensing image i and the new remote sensing image j respectively.

[0031] A satellite high-precision image pair matching and reconstruction system based on mission planning, the system comprising:

[0032] A stereo image pair generation module, configured to generate a plurality of stereo image pairs based on a current remote sensing image, and calculate characteristic parameters of the stereo image pairs based on satellite orbit data corresponding to the current remote sensing image;

[0033] A new remote sensing image acquisition module is configured to acquire new satellite orbit data based on the parameters of the stereo image pair and the observation time window of the current remote sensing image, in combination with target requirements, obtain a number of new observation time windows through orbit visibility calculation, and generate a number of new remote sensing images according to the new observation time windows;

[0034] A new stereo image pair generation module is used to obtain an optimal new stereo image pair based on orbit-related data and constraints of the new remote sensing image and a DSM error of the new stereo image pair; wherein the new stereo image pair is generated based on two new remote sensing images.

[0035] An electronic device comprises: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements any one of the above-mentioned mission planning-based high-precision satellite image pair matching and reconstruction methods.

[0036] A computer-readable storage medium having computer program instructions stored thereon, wherein the computer program instructions, when executed by a processor, implement any of the above-mentioned mission planning-based satellite high-precision image pair matching and reconstruction methods.

[0037] A computer program product, when running on a computer device, enables the computer device to execute any one of the above-mentioned mission planning-based satellite high-precision image pair matching and reconstruction methods.

[0038] Compared with the prior art, the present invention has at least the following beneficial effects:

[0039] 1. By developing a quality assessment model for high-precision stereo image pairs, the accuracy of 3D surface reconstruction is significantly improved. Traditional methods often rely on manual selection or simple automated algorithms to determine the optimal stereo image pair, which is not only time-consuming but also difficult to ensure the consistency and high quality of the results. The quality assessment model proposed in this paper comprehensively considers multiple key factors (such as intersection angle, solar angle difference, incidence angle, and acquisition time difference) and can accurately predict the quality of the DSM generated by each pair of stereo images, thereby ensuring the high accuracy and reliability of the final 3D reconstruction results.

[0040] 2. The present invention utilizes a multi-objective optimization algorithm to achieve efficient task planning, effectively reducing redundant remote sensing and data transmission costs. The existing acquisition-selection mode usually performs screening after image acquisition, resulting in waste of resources and low efficiency. In contrast, the present invention searches and selects the optimal stereo image pair through a metaheuristic algorithm before image acquisition, and combines the quality estimation model to predict the quality value, thereby optimizing the task execution sequence and resource allocation. This method not only greatly improves the quality and efficiency of three-dimensional reconstruction, but is also particularly suitable for fields such as emergency management in emergency mapping, and can provide high-quality geographic information support in a short time. Experimental results show that the method of the present invention has shown significant advantages on multiple real data sets, verifying its feasibility and superiority in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flow chart of a method for matching and reconstructing satellite high-precision images based on mission planning. DETAILED DESCRIPTION

[0042] In order to make the purpose, scheme and advantages of the present invention more clearly understood, the present invention is further described in detail by taking experiments conducted on real data sets as an example. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0043] The present invention's method for matching and reconstructing satellite high-precision images based on mission planning uses multiple pairs of stereo images to perform three-dimensional reconstruction. For each pair of stereo images, the accuracy error of the digital surface model (DSM) is calculated. Based on the shooting dates of these images and the corresponding satellite orbit data, the visible windows of each time period are calculated, and a series of visible time windows are constructed. In order to further provide a basis for mission planning optimization, the present invention establishes a regression prediction model under small sample conditions to predict the DSM accuracy error under specific windows. On this basis, a mission planning arrangement is formulated according to dimensions such as timeliness and accuracy priority, and the optimal visible window is selected for image reconstruction. The proposed method can optimize the use of satellite resources by optimizing satellite shooting missions and scientific mission planning, thereby effectively improving the quality and accuracy of 3D reconstruction.

[0044] Specifically, if Figure 1 As shown, the satellite high-precision image pair matching and reconstruction method of the present invention includes the following steps.

[0045] Step 1: Generate several stereo image pairs based on the current remote sensing image, and calculate the characteristic parameters of the stereo image pairs based on the satellite orbit data corresponding to the current remote sensing image.

[0046] In this embodiment, the current remote sensing images are first reconstructed according to a three-dimensional surface reconstruction method, and the current remote sensing images are combined in pairs to form a plurality of stereo image pairs.

[0047] Afterwards, the characteristic parameters of the stereo image pair are obtained through orbital calculation based on the current remote sensing image: the average value of the intersection angle of the two images, the maximum side swing angle of the two images, the smaller solar altitude angle of the two images, the difference in solar altitude angles between the two images, the difference in solar azimuth angles between the two images, the comprehensive spatial angle difference between the solar altitude angles and azimuth angles in the two images, and the date interval between the two images.

[0048] Step 2: Based on the parameters of the stereo image pair and the observation time window of the current remote sensing image, new satellite orbit data is obtained in combination with target requirements, several new observation time windows are obtained through orbit visibility calculation, and several new remote sensing images are generated according to the new observation time windows.

[0049] The shooting time of the current remote sensing image is T={T1,T2,...,T n}, and use the orbital dynamics model to recalculate the satellite's orbital data in combination with the target requirements to generate a new observation time window W = {W1, W2, ..., W m}, each window W k Corresponding to one imaging opportunity, several new remote sensing images are generated.

[0050] Step 3: Obtain an optimal new stereo image pair based on the orbit-related data and constraints of the new remote sensing image and the DSM error of the new stereo image pair; wherein the new stereo image pair is generated based on the two new remote sensing images.

[0051] Step 3.1: Generate a multi-objective problem plan and construct the objective function Obj of the multi-objective problem plan.

[0052] (1) The parameters and variables of the multi-objective planning model are as follows:

[0053] ts i / te i : start / end time of imaging time window i;

[0054] t start / t end : The start / end time of the entire task;

[0055] mt k : The maneuvering time required for satellite k;

[0056] α i : the side swing angle of imaging time window i;

[0057] α max: maximum satellite side swing angle;

[0058] DSM(i,j): DSM error of the stereo pair;

[0059] θ ij : Intersection angle of imaging pair (i, j);

[0060] θ min / θ max : minimum / maximum intersection angle;

[0061] x ij : The decision variable for whether to select the stereo imaging pair (i, j). ij =1 if (i,j) is selected; otherwise x ij =0.

[0062] The present invention obtains the DSM error of the stereo imaging pair by constructing a regression model, and the construction of the regression model includes the following steps.

[0063] (A) Implementing a 3D reconstruction algorithm based on a multi-view remote sensing image dataset to generate multiple sets of high-precision stereo image pairs with overlapping areas, and using their reconstruction accuracy as an evaluation metric for the reconstruction results.

[0064] The present invention firstly reconstructs the remote sensing image I={I1,I2,...,I n} to reconstruct the image; then combine the current remote sensing images in pairs to form multiple stereo image pairs, denoted as P = {(I i ,I j )|i≠j,i,j∈[1,n]}; Finally, for each stereo image pair (I i ,I j ) Apply the 3D surface reconstruction algorithm to generate the corresponding digital surface model DSM and obtain the reconstruction accuracy data Q corresponding to each current stereo image pair ij .

[0065] (B) Evaluation of the correlation between each characteristic parameter of the stereo image pair and the reconstruction accuracy.

[0066] According to the parameters of the original image composed of the stereo pair (I i ,I j ), extract its corresponding parameter set X={X1,X2,...,X m}, where m represents the number of parameters, and for each parameter X k and the reconstruction accuracy value Q ij The Pearson correlation coefficient r(x k ,Q ij ) to evaluate the correlation between each feature parameter and reconstruction accuracy.

[0067] In one embodiment, for each stereo image pair, the feature parameter set X = {X1, X2, ..., X m}, where each characteristic parameter X k ={x k1 ,x k2 ,...,x kn}, x ki is the kth characteristic parameter X k Thus, the reconstruction accuracy value set Q corresponding to the feature parameter set X is Q={q1,q2,...,q n}.

[0068] Specifically, this embodiment first calculates the characteristic parameter X k The average value of the observation and the average reconstruction accuracy Among them, q i is the i-th value in the reconstruction accuracy value set Q:

[0069]

[0070]

[0071] Then, based on the mean of the observed values and the average reconstruction accuracy Calculate the characteristic parameter X k Pearson correlation coefficient with reconstruction accuracy in,

[0072] (C) According to the Pearson correlation coefficient, the characteristic parameters X'={X'1, X'2, ..., X' l}, where l≤m.

[0073] Set the correlation threshold τ, select the feature parameter with the absolute value of the correlation coefficient greater than τ as the highly correlated feature parameter, where X'={X k ||r(X k ,Q)|>τ}.

[0074] (D) For the highly correlated feature parameters X' that are screened out, logarithmic transformation is applied to these selected parameters to capture nonlinear relationships while performing data augmentation to generate a new feature set X".

[0075] (E) Perform Lasso feature selection on all feature sets X” and select the feature subset X”’={x1”’, x2”’, ..., x p ”'}.

[0076] Lasso regression is used for feature selection to identify and retain key features that contribute significantly to the model's predictive ability. The calculation formula is as follows:

[0077]

[0078] Where Y represents the target variable or response variable, C=[c1,...c d ] T is the regression coefficient vector; c k is the regression coefficient of the kth feature, λ||C||1 is a regression-specific penalty term using the L1 norm, which encourages some coefficients to become 0, achieving feature selection. λ is a regularization parameter that controls the strength of the penalty. A larger λ will more severely compress the coefficients, even reducing some unimportant feature coefficients to 0 completely, achieving feature selection.

[0079] Lasso (Least Absolute Shrinkage and Selection Operator) is an efficient regression analysis method widely used in feature selection and model simplification. It introduces an absolute value-based penalty term (i.e., L1 regularization) on the basis of the traditional least squares method. It can not only effectively shrink the model coefficients, but also directly reduce some unimportant feature coefficients to zero, thereby realizing automatic feature selection. This feature makes Lasso very suitable for processing high-dimensional data sets, while reducing overfitting and improving the interpretability and predictive performance of the model. According to current feature selection, most feature expansions are selected based on correlation coefficients. Compared with the original features, the data-enhanced features have a higher selection coefficient in the model, indicating that data enhancement has a significant positive effect on improving the quality of modeling data.

[0080] (F) Using feature subset X''={x1'', x2'', ..., x p ”'} and reconstruction accuracy value Q ij Perform catBoost regression modeling.

[0081] (G) Determination of evaluation indicators for the model.

[0082] Take the coefficient of determination R 2 The model is evaluated using the root mean square error (RMSE) and MAE indicators. Based on the current model results, the model is evaluated and the model is iterated until the fitting accuracy of the current model reaches a certain threshold.

[0083]

[0084] (2) Model objective function and constraints.

[0085] The present invention describes the planning of multi-objective problems such as the number of tasks, the reconstructed DSM error, and the timeliness of tasks as follows:

[0086]

[0087] Among them, max(te i ,te j ) is the latest end time of the matching time window pair (i, j). According to the above formal objectives, Obj1 (the number of tasks to be completed) needs to be maximized, Obj2 (DSM error) and Obj3 (total end time) need to be minimized. The overall goal can be calculated as the weighted sum of the above objectives to maximize

[0088] Obj=ω1·Obj1+ω2·(-Obj2)+ω3·(-Obj3)

[0089] When calculating the above equation, the targets are normalized to the same scale to avoid any one target value dominating. In order to provide a feasible stereo imaging solution, the following constraints need to be met when optimizing the above objective function:

[0090] Task time span constraint: The observation time window of each task must be within the entire task cycle.

[0091]

[0092] Maneuver time limit: This is to ensure that the satellite has enough maneuver time to adjust its attitude for two consecutive observations.

[0093]

[0094] Swing angle constraint: The satellite swing angle must be smaller than the maximum range.

[0095]

[0096] Intersection angle constraint: Stereo imaging pair (i, j)(x ij =1) needs to be within the valid range.

[0097]

[0098] Step 3.2: Combined with the constraints, the objective function Obj is optimized and matched using the Grey Wolf Algorithm to obtain the best new stereo image pair.

[0099] Because this problem is a combinatorial optimization problem, it is necessary to select the optimal solution from a large number of image pairs while performing consistency checks to ensure that the solution satisfies the constraints. The Gray Wolf Optimizer (GWO) is a metaheuristic algorithm widely used in combinatorial optimization problems. By simulating gray wolf hunting, the GWO algorithm can efficiently perform a global search in a complex solution space, avoiding local optimization.

[0100] In the GWO algorithm, the potential solutions to the problem are viewed as the positions of a pack of gray wolves in the search space. The pack consists of a leadership hierarchy consisting of four roles: α (the first leader), β (the second leader), δ (the third leader), and ω (the subordinates), representing the current optimal solution, the second-best solution, the third-best solution, and the remaining solutions, respectively.

[0101] In addition to the leadership level, the GWO algorithm also simulates the hunting mechanism, which is mainly divided into three stages: finding prey, surrounding prey, and attacking prey, to better adapt to the challenging problems of unknown search space. In the GWO algorithm, optimization (hunting) is guided by α, β, and δ, followed by ω. ij is a binary decision variable, indicating whether to select an observation imaging pair during planning. Therefore, it contains all x ij The optimization of position X is actually the search process for prey (optimal pairing combination). The optimization of X in GWO can be expressed mathematically as:

[0102] D=|C·X p (t)-X(t)|

[0103] X(t+1)=X p (t)-A·D

[0104] Where t is the current iteration number, Xp represents the location of the prey (optimal solution), and X(t) represents the location of the gray wolf in generation t, which is the solution currently being searched. A and C are coefficient vectors, calculated as follows:

[0105] A=2a·r1-a·1

[0106] C=2·r2

[0107] Where r1 and r2 are both random vectors in [0, 1] to improve the accessibility of any available position, a decreases linearly from 2 to 0 during the iteration, and 1 is a vector of all 1s. According to the above formula, A is a random vector in [-a, a].

[0108] During the iterative search process, the search population updates its position mainly by learning the solutions of α, β, and δ. Among them, the ω wolf group adjusts its position according to the distance relationship with the α, β, and δ wolves. The update formula is as follows:

[0109] D α =|C1·Xα -X|

[0110] D β =|C2·X β -X|

[0111] D δ =|C3·X δ -X|

[0112] X1=X α -A1·D α

[0113] X2=X β -A2·D β

[0114] X3=X δ -A3·D δ

[0115]

[0116] Where X is the current location of the search individual, X α 、X β and X δ where α, β, and δ are the positions of the solutions, respectively. A1, A2, A3 and C1, C2, C3 are coefficient vectors that change dynamically with iteration to balance the exploration and exploitation capabilities of the search. By continuously iteratively updating the positions of the search individuals and adjusting the combination of satellite imaging pairs, a solution that optimizes the objective function while satisfying the constraints is obtained.

[0117] In summary, the present invention obtains orbital parameters and determines imaging parameters in advance, and provides a method for predetermining a reconstruction scheme for shooting through mission planning optimization. This method can provide a high-quality three-dimensional reconstruction scheme, not only achieving high-precision three-dimensional reconstruction of the earth's surface, but also effectively reducing unnecessary remote sensing data acquisition and transmission. Especially in emergency situations where it is necessary to quickly generate a three-dimensional reconstruction scene, redundant operations often result in high costs. Therefore, the solution proposed in this patent has important practical significance for improving emergency response efficiency.

[0118] The above description is only one embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for matching and reconstructing satellite high-precision images based on mission planning, characterized in that: The method comprises: generating a plurality of stereo image pairs based on a current remote sensing image, and calculating characteristic parameters of the stereo image pairs based on satellite orbit data corresponding to the current remote sensing image; Based on the parameters of the stereo image pair and the observation time window of the current remote sensing image, new satellite orbit data is acquired in combination with target requirements, a plurality of new observation time windows are obtained through orbit visibility calculation, and a plurality of new remote sensing images are generated according to the new observation time windows; An optimal new stereo image pair is obtained based on orbit-related data and constraints of the new remote sensing image and a DSM error of the new stereo image pair; wherein the new stereo image pair is generated based on the two new remote sensing images.

2. The method according to claim 1, characterized in that The characteristic parameters of the stereo image pair include: the average value of the intersection angle of the two current remote sensing images, the maximum side swing angle in the two current remote sensing images, the smaller solar altitude angle in the two current remote sensing images, the solar altitude angle difference in the two current remote sensing images, the solar azimuth angle difference in the two current remote sensing images, the comprehensive spatial angle difference between the solar altitude angle and the azimuth angle in the two current remote sensing images, and the date interval between the two current remote sensing images.

3. The method according to claim 1, characterized in that Obtaining a DSM error of the new stereo image pair based on a regression model; wherein the training process of the regression model includes: Obtaining a training data set consisting of remote sensing images, and combining the remote sensing image data in pairs to perform three-dimensional surface reconstruction; After the three-dimensional reconstruction, a plurality of stereo image pair samples are formed, and a reconstruction accuracy value of each stereo image pair sample is obtained; According to the parameters of the remote sensing image, the characteristic parameters X of the stereo image pair samples are extracted k , k is a positive integer; Calculate each feature parameter X k Pearson correlation coefficient with the reconstruction accuracy value; According to the Pearson correlation coefficient, the characteristic parameter X k Filter out the feature parameters that are highly correlated with the reconstruction accuracy and obtain the feature set X′; Perform different feature transformations on the feature parameters in the feature set X′ to capture nonlinear relationships and perform data expansion to obtain the feature parameter set X″; Perform Lasso feature selection on all feature parameter sets X″ to obtain feature parameter set X″′; Performing catBoost regression modeling based on the feature parameter set X″′ and the reconstruction accuracy value of the stereo image pair sample to obtain an initial regression model; Take the coefficient of determination R 2 The current regression model is evaluated using the root mean square error and MAE indicators, and the model is iterated according to the evaluation results until the fitting accuracy of the current regression model reaches the set threshold.

4. The method according to claim 3, characterized in that The calculation of each characteristic parameter X k Pearson correlation coefficient with the reconstruction accuracy value, include: Get the characteristic parameter X k The set of observation values ​​and the set of reconstruction accuracy values ​​corresponding to the observation values, and calculate the average observation value and the average reconstruction accuracy Based on the average observation and the average reconstruction accuracy Calculate the characteristic parameter X k Pearson correlation coefficient with the reconstruction accuracy value.

5. The method according to claim 1, wherein The orbit-related data of the new remote sensing image include: the roll angle of the new remote sensing image and the intersection angle of any two new remote sensing images; The obtaining of an optimal new stereo image pair based on the orbit-related data and constraints of the new remote sensing image and the DSM error of the new stereo image pair comprises: Based on the roll angle of the new remote sensing image, the intersection angle of any two new remote sensing images, and the DSM error of the new stereo image pair, a multi-objective problem planning is generated, and an objective function Obj of the multi-objective problem planning is constructed; wherein the objectives include: the number of tasks, the reconstructed DSM error, and the timeliness of the tasks; Setting constraints for solving the objective function; wherein the constraints include: a mission time span constraint, a maneuvering time limit, a swing angle constraint, and an intersection angle constraint. The mission time span constraint is used to specify that the observation time window of each mission must be within the entire mission cycle. The maneuvering time limit is used to ensure that the satellite has sufficient maneuvering time to adjust its attitude for two consecutive observations. The swing angle constraint is used to limit the satellite swing angle to be less than a maximum range. The intersection angle constraint is used to limit the intersection angle value of the new stereo imaging pair to be within a valid range. Combined with the constraints, the objective function Obj is optimized and matched using the Grey Wolf algorithm to obtain the best new stereo image pair.

6. The method according to claim 5, characterized in that The objective function Obj=ω1·Obj1+ω2·(-Obj2)+ω3·(-Obj3); wherein ω1, ω2, ω3 are the first weight, the second weight, and the third weight respectively. The first objective function Obj1=∑ i,j x i,j , the second objective function Obj2=∑ i ∑ j DSM(i,j)*x i,j , the third objective function Obj3=∑ i ∑ j max(te i ,te j )*x i,j , x i,j represents the decision variable for selecting a new stereo image pair (i, j), x i,j =1 means selected, x i,j =0 means not selected, i and j represent the new remote sensing image i and the new remote sensing image j respectively, DSM(i,j) represents the DSM error of the new stereo image pair (i,j), te i and te j represent the imaging time of the new remote sensing image i and the new remote sensing image j respectively.

7. A satellite high-precision image pair matching and reconstruction system based on mission planning, characterized in that: The system comprises: A stereo image pair generation module, configured to generate a plurality of stereo image pairs based on a current remote sensing image, and calculate characteristic parameters of the stereo image pairs based on satellite orbit data corresponding to the current remote sensing image; A new remote sensing image acquisition module is configured to acquire new satellite orbit data based on the parameters of the stereo image pair and the observation time window of the current remote sensing image, in combination with target requirements, obtain a number of new observation time windows through orbit visibility calculation, and generate a number of new remote sensing images according to the new observation time windows; A new stereo image pair generation module is used to obtain an optimal new stereo image pair based on orbit-related data and constraints of the new remote sensing image and a DSM error of the new stereo image pair; wherein the new stereo image pair is generated based on two new remote sensing images.

8. An electronic device, characterized in that: The electronic device includes: a processor and a memory storing computer program instructions; when the processor executes the computer program instructions, it implements the mission planning-based satellite high-precision image pair matching and reconstruction method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer program instructions, which, when executed by a processor, implement the mission planning-based satellite high-precision image pair matching and reconstruction method according to any one of claims 1 to 6.

10. A computer program product, characterized in that When the computer program product is run on a computer device, the computer device is enabled to execute the mission planning-based satellite high-precision image pair matching and reconstruction method according to any one of claims 1 to 6.

Images