Push-broom image processing method and system based on regression model

Through a regression model-based method, a mathematical model is established using multi-projection center model and collinear equations, simulated control points and checkpoints are generated, and the best scanning line for the push-swept image is directly determined, which solves the problem of iterative calculations in the existing technology. High cost and long-term problems enable fast and efficient image processing.

CN118918204APending Publication Date: 2024-11-08SHAOXING UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410980501.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-22
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

When processing linear push-swept images, the prior art requires iterative calculation based on collinear equations, which leads to high calculation costs and long time consuming, making it difficult to meet the rapid processing needs of practical applications.

Method used

A regression model-based method is adopted to establish a mathematical model through multi-projection center model and collinear equations, and simulated control points and simulated checkpoints are generated. The regression model is used to directly determine the best scanning line for the push-swept image without iterative calculations.

Benefits of technology

It shortens processing time, improves the efficiency of push-sweep image processing, achieves better subpixel accuracy, and is suitable for real-time applications.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118918204A_ABST
    Figure CN118918204A_ABST
Patent Text Reader

Abstract

The invention discloses a push-broom image processing method and system based on a regression model, and the method comprises the steps: obtaining push-broom images collected by a plurality of sensors, and obtaining internal direction parameters and ground control points; according to the external direction parameters and the ground control points, performing space cutting of a multi-projection center model on the push-broom image to obtain the external direction parameters of each scanning line, and according to the ground control points, obtaining the average height of the research area; establishing a mathematical model of the average height of the research area and all external direction parameters by using a collinear equation, generating simulation control points and simulation check points, and inputting the simulation check points and the parameters of the regression model into the trained regression model for testing to obtain the optimal number of scanning lines; according to the method, the simulation check points and the optimal scanning line number are accurately evaluated to determine the optimal scanning line of the push-scan image, and a collinear equation does not need to be iteratively used, so that the processing time is shortened, the sub-pixel precision is better, and the push-scan image processing efficiency is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a push-broom image processing method and system based on a regression model. Background Art

[0002] Linear pushbroom cameras are becoming increasingly popular in various applications of photogrammetry and remote sensing, including global and terrain mapping, environmental monitoring, change detection, geological surveys, object detection, and three-dimensional (3D) reconstruction. Linear array pushbroom cameras can provide high-resolution panchromatic or multispectral satellite images with a wider field of view and high revisit frequency, making it easier to draw maps at different scales. However, processing linear pushbroom images is more complicated than processing frame-type images. In a linear pushbroom imaging sensor, the two-dimensional (2D) image scene is the result of capturing consecutive one-dimensional (1D) scan lines at different times, and each scan line has its own external orientation parameters (EOP). This dynamic imaging geometry complicates the process of converting 3D ground space to 2D image space. Unlike frame-type images, each image has a set of EOPs. When pushbroom images are used, the position and orientation of the center of projection vary from line to line. This variation in the EOP for each scan line makes it challenging to determine the exact exposure time required to achieve the EOP for each ground control point. This problem is called best scan line search / determination (BSS / BSD) or the exact time determination of the exposure. The object-to-image transformation based on the collinearity equation (CE) requires the EOP of the corresponding scan line, which is not available for pushbroom images. This conversion process is the basis of linear pushbroom image geometry processing, which includes orthorectification, DTM generation, image matching, epipolar resampling, and stereo measurement.

[0003] Currently, a lot of research has been done on the object-to-image space conversion problem in push-broom images, also known as BSS (Best Scanline Search). Several methods have been proposed to address this challenge, including sequential search (SS), backtracking, binary window search (BWS), Newton-Raphson (NR), central perspective plane (CPP) based, general distance prediction (GDP) based, optimal global polynomial (OGP) using genetic algorithm (GA), and artificial neural network (ANN). However, except for OGP and ANN methods, all of these methods require continuous complex iterations based on CE, which are computationally expensive and time-consuming, making them unsuitable for practical applications. Therefore, there is an urgent need to provide a fast and non-iterative push-broom image processing method. Summary of the invention

[0004] In view of this, the present invention provides a push-broom image processing method and system based on a regression model, which does not require any collinear equation iteration, thereby shortening the processing time to solve the above-mentioned technical problems, and is specifically implemented by the following technical solutions.

[0005] In a first aspect, the present invention provides a push-broom image processing method based on a regression model, comprising the following steps:

[0006] Acquire push-broom images acquired by a plurality of sensors, and obtain internal orientation parameters and ground control points of a plurality of said sensors;

[0007] Performing spatial excision of a multi-projection center model on the push-broom image according to the external direction parameters and the ground control points to obtain external direction parameters of each scanning line corresponding to the push-broom image;

[0008] Obtaining an average height of a study area in the push-broom image according to the ground control points, and establishing a mathematical model of the average height of the study area and all the external direction parameters using a collinearity equation;

[0009] Generating simulation control points in the training phase and simulation check points in the testing phase in the mathematical model, wherein the simulation control points are used to calculate the parameters of the regression model in the training phase;

[0010] Inputting the simulated checkpoint and the parameters of the regression model into the trained regression model for testing to obtain the optimal number of scan lines corresponding to the simulated checkpoint;

[0011] The simulated checkpoints and the optimal number of scan lines are evaluated for accuracy to determine optimal scan lines for the push-broom image.

[0012] As a preferred embodiment of the above technical solution, a mathematical model of the average height of the study area and all the external direction parameters is established using a collinear equation, including:

[0013] The mathematical model is used to establish the relationship between the 2D image space and the 3D ground space. The expression of the collinear equation is:

[0014]

[0015] Where (x, y) represents the 2D image coordinates of any point, f represents the focal length, (X, Y, Z) represents the 3D ground coordinates of any point, and i represents the scan line number. and They represent the position of the projection center in the object space and the rotation angle ω, The rotation matrix element corresponding to k;

[0016] According to the dynamic characteristics of the push-broom image and expression (1), the spatial resection algorithm of the multi-projection center model is used to determine the external direction parameters of all scan lines. The corresponding expression (2) is:

[0017]

[0018] Among them, t i represents the exposure time of the i-th scan line; X0, X1, ... and k2 represent the external direction parameters of the reference scan line calculated in the spatial resection step; X s i (t),Y s i (t),…, and k s i (t) represents the external direction parameter of the i-th scan line.

[0019] As a preferred embodiment of the above technical solution, generating simulation control points in the training phase and simulation checkpoints in the testing phase in the mathematical model includes:

[0020] Creating regular grids for simulation control points and simulation check points in the image space corresponding to the mathematical model to obtain the simulation control points and the image coordinates of the simulation check points;

[0021] The image simulation points of the push-broom image are transferred to the object space using the collinearity equation, the average height of the study area and the exterior orientation parameters to obtain simulation control points and simulation check points in image coordinates and object coordinates.

[0022] As a preferred embodiment of the above technical solution, the regression model includes a polynomial regression model or a linear regression model. The independent variable in the regression model predicts the dependent variable. The linear regression includes simple linear regression and multivariate linear regression. The simple linear regression is a regression model with a single independent variable. The expression corresponding to the simple linear regression is:

[0023] y=a0+a1x (3)

[0024] Among them, y represents the dependent variable, x represents the independent variable, a0 and a1 represent the intercept and linear term values ​​respectively;

[0025] Multiple linear regression models the linear relationship between one or more independent variables and the dependent variable. The expression of multiple linear regression is:

[0026]

[0027] Where y represents a single dependent variable, (a0, ..., a m ) represents the coefficients forming the A matrix, (x 11 , ..., x nm ) represents the independent variables that form the X matrix;

[0028] Use the least squares algorithm to find the best straight line or curve that fits the data set, thereby minimizing the cumulative squared residual. The expression of the least squares algorithm is:

[0029] A=(X T X)- 1 X T Y(5).

[0030] As a preferred embodiment of the above technical solution, the polynomial regression model represents a multiple regression modeled by an n-order polynomial. When there is a curvilinear relationship between the independent variable and the dependent variable, the expression of the polynomial regression model is:

[0031] y=a0+a1x+a2x 2 +…+a m x n (6)

[0032] Among them, y represents the dependent variable, x represents the independent variable, (a0,..., a m ) is the regression coefficient of the independent variable, n represents the polynomial degree or order of the regression model; if there are multiple independent variables, the X matrix is ​​used to rewrite equation (6).

[0033] As a preferred embodiment of the above technical solution, the linear regression model includes a training phase and a testing phase. In the training phase, simulated checkpoints are used for accuracy assessment of each unknown ground control point; the simulated control points are used in the least squares algorithm to calculate the unknown parameters of the hybrid logistic regression model, and the hybrid logistic regression model uses the ground coordinates (X, Y) as input parameters and the row number (r) as the target parameter to establish the conversion between the ground space and the image space;

[0034] In the testing phase, the hybrid logistic regression model is applied to the ground coordinates (X,Y) of the simulated checkpoints and the row values ​​are estimated. The accuracy of the simulated checkpoints is evaluated using the root mean square error (RMSE) value.

[0035] As a preferred embodiment of the above technical solution, the accuracy of the simulated checkpoints and the optimal number of scan lines is evaluated to determine the optimal scan line of the push-broom image, including:

[0036] A plurality of measurement criteria are used for comparison to determine the best scan line value, the measurement criteria include a root mean square error, a maximum best scan line determination error among all simulation checkpoints, and an execution time, wherein the root mean square error and the maximum best scan line determination error among all simulation checkpoints are obtained by comparing the best scan line value obtained with the accurate scan line value obtained in the simulation phase through linear regression;

[0037] The number of simulated control points and the fitness of the model were checked to obtain a robust model.

[0038] As a preferred embodiment of the above technical solution, the push-broom images are collected by at least one sensor including IKONOS, Pleiades 1A and 1B, QuickBird, SPOT6, SPOT7, WorldView1 and WorldView2.

[0039] As a preferred embodiment of the above technical solution, the push-broom image has different coverage areas, spatial resolutions and sensor characteristics.

[0040] In a second aspect, the present invention further provides a push-broom image processing system based on a regression model, which is applied to the push-broom image processing method based on a regression model, comprising:

[0041] An image acquisition module, used to acquire push-broom images collected by multiple sensors, and to acquire internal direction parameters and ground control points of multiple sensors;

[0042] a parameter determination module, configured to perform spatial excision of a multi-projection center model on the push-broom image according to the external direction parameter and the ground control point, so as to obtain the external direction parameter of each scanning line corresponding to the push-broom image;

[0043] A model building module, used for obtaining the average height of the study area in the push-broom image according to the ground control points, and establishing a mathematical model of the average height of the study area and all the external direction parameters using a collinear equation;

[0044] A model processing module, used to generate simulation control points in the training phase and simulation check points in the testing phase in the mathematical model, wherein the simulation control points are used to calculate the parameters of the regression model in the training phase;

[0045] A regression model testing module, used for inputting the simulated checkpoint and the parameters of the regression model into the trained regression model for testing, and obtaining the optimal number of scan lines corresponding to the simulated checkpoint;

[0046] An accuracy evaluation module is used to perform accuracy evaluation on the simulated checkpoints and the optimal number of scan lines to determine the optimal scan line of the push-broom image.

[0047] The present invention provides a push-broom image processing method and system based on a regression model. The method and system obtain push-broom images collected by multiple sensors and obtain internal direction parameters and ground control points of the multiple sensors. The push-broom images are spatially cut out of a multi-projection center model according to the external direction parameters and the ground control points to obtain the external direction parameters of each scanning line corresponding to the push-broom images. The average height of a study area in the push-broom images is obtained according to the ground control points. A mathematical model of the average height of the study area and all external direction parameters is established using a collinearity equation. Simulated control points in a training phase and simulated checkpoints in a test phase are generated in the mathematical model. The simulated checkpoints and the parameters of the regression model are input into the trained regression model for testing to obtain the optimal number of scanning lines corresponding to the simulated checkpoints. The simulated checkpoints and the optimal number of scanning lines are evaluated for accuracy to determine the optimal scanning line of the push-broom image. The training and testing phases of the simulated control points and the simulated checkpoints are respectively involved. There is no need to iteratively use the collinearity equation, thereby shortening the processing time, having better sub-pixel accuracy, and improving the push-broom image processing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0049] Figure 1 A flowchart of a push-broom image processing method based on a regression model provided by the present invention;

[0050] Figure 2 A schematic diagram of the source of the data set corresponding to the push-broom image provided by the present invention;

[0051] Figure 3 The BSD method provided by the present invention is compared with RMSE, time and dr max A comparative diagram of

[0052] Figure 4 This is a structural block diagram of the push-broom image processing system based on the regression model provided by the present invention. DETAILED DESCRIPTION

[0053] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.

[0054] See also Figure 1 The present invention provides a push-broom image processing method based on a regression model, comprising the following steps:

[0055] S1: acquiring push-broom images collected by a plurality of sensors, and acquiring internal direction parameters and ground control points of the plurality of sensors;

[0056] S2: performing spatial excision of a multi-projection center model on the push-broom image according to the external direction parameter and the ground control point to obtain an external direction parameter of each scanning line corresponding to the push-broom image;

[0057] S3: obtaining an average height of the study area in the push-broom image according to the ground control points, and establishing a mathematical model of the average height of the study area and all the external direction parameters using a collinearity equation;

[0058] S4: generating simulation control points in the training phase and simulation check points in the testing phase in the mathematical model, wherein the simulation control points are used to calculate the parameters of the regression model in the training phase;

[0059] S5: inputting the simulated checkpoint and the parameters of the regression model into the trained regression model for testing to obtain the optimal number of scan lines corresponding to the simulated checkpoint;

[0060] S6: Evaluate the accuracy of the simulated checkpoints and the optimal number of scan lines to determine the optimal scan lines for the push-broom image.

[0061] In this embodiment, the use of linear array push-broom images presents new challenges in photogrammetry applications when it comes to converting object coordinates to image coordinates. To address this issue, the best scan line search / determination (BSS / BSD) field focuses on obtaining the exterior orientation parameters (EOP) of each scan line, which helps solve the problem of accurately converting objects to image coordinates. Current BSS / BSD solutions are often impractical for real-time tasks due to their high time requirements and complexity, as they rely on the collinearity equation (CE) of each ground control point in an iterative process.

[0062] This paper aims to develop a new BSD framework for push-broom images, which does not rely on the reuse of CE and has low computational complexity. The proposed BSD method is based on a linear regression model (LRM) and uses two types of points: simulated control points (SCOP) and simulated check points (SCP). The LRM BSD method consists of two steps: a training phase and a testing phase. In the training phase, SCOP is used to calculate the unknown parameters of the LR model. Then, in the testing phase, SCP is used to evaluate the accuracy and execution time of the method. The proposed LRM BSD method is evaluated using 10 different push-broom images, 5 million SCPs, and a limited number of SCOPs. The root mean square error (RMSE) is about 10 to the negative 9th power, indicating a very high accuracy. In addition, it is more robust than the traditional BSS / BSD method when processing various types of push-broom images, making it suitable for practical and real-time applications.

[0063] The push-broom images are acquired by at least one of the sensors selected from IKONOS, Pleiades 1A and 1B, QuickBird, SPOT6, SPOT7, WorldView 1 and WorldView2, and the push-broom images have different coverage areas, spatial resolutions and sensor characteristics. Figure 1 As shown in Figure 2, the dataset includes ten different satellite pushbroom images acquired by various sensors such as IKONOS, Pleiades 1A and 1B, QuickBird, SPOT6, SPOT7, WorldView 1 and WorldView2. The images used have different coverage areas (urban, agricultural, plains and a combination of all areas), spatial resolutions and sensor characteristics (such as Figure 2 ), which provides a more comprehensive evaluation of the proposed BSD method.

[0064] Table 1 Dataset specifications

[0065] image sensor Coverage Area Subset size (pixel) Resolution(m) SPI IKONOS Sao Paulo, Brazil 8300*8600 0.80 MP Pleiades1A Melbourne,Australia 6000*7000 0.5 AP Pleiades1B Annapolis, USA 6057*5636 0.5 JbXj QuickBird Jaipur, India 6000*6000 0.6 JS SPOT6 Jaicos, Brazil 6200*6600 1.5 AS SPOT7 Amsterdam, Netherlands 5824*6616 1.5 CS SPOT7 Curitiba, Brazil 2597*1463 6 BWV WorldView1 Boulder, USA 6000*6000 0.5 SWV WorldView2 Sydney, Australia 6000*6000 0.5 SDW WorldView2 San Diego, USA 3996*4015 0.5

[0066] Figure 2 To use datasets with different coverage areas, Figure 2 (a) in the figure represents the SPI image. Figure 2 (b) in the figure represents the MP image. Figure 2 (c) in the figure represents the AP image. Figure 2 (d) in the figure represents the JQB image. Figure 2 (e) in the figure represents a JS image. Figure 2 (f) in the figure represents the AS image. Figure 2 (g) in the figure represents the CS image. Figure 2 (h) in the figure represents the BWV image. Figure 2(i) in the figure represents the SWV image. Figure 2 (j) in FIG. 5 represents the SDWV image.

[0067] Specifically, a mathematical model of the average height of the study area and all the external directional parameters is established using collinear equations, including:

[0068] The mathematical model is used to establish the relationship between the 2D image space and the 3D ground space. The expression of the collinear equation is:

[0069]

[0070] Where (x, y) represents the 2D image coordinates of any point, f represents the focal length, (X, Y, Z) represents the 3D ground coordinates of any point, and i represents the scan line number. and They represent the position of the projection center in the object space and the rotation angle ω, The rotation matrix element corresponding to k;

[0071] According to the dynamic characteristics of the push-broom image and expression (1), the spatial resection algorithm of the multi-projection center model is used to determine the external direction parameters of all scan lines. The corresponding expression (2) is:

[0072]

[0073] Among them, t i represents the exposure time of the i-th scan line; X0, X1, ... and k2 represent the external direction parameters of the reference scan line calculated in the spatial resection step; X s i (t),Y s i (t),…, and k s i (t) represents the external direction parameter of the i-th scan line. From the perspective of photogrammetry, mathematical models are used to establish the relationship between the 2D image space and the 3D ground space. A well-known mathematical model based on the image geometry at the time of imaging is CE, which has been widely used for push-broom images.

[0074] Specifically, generating simulation control points in the training phase and simulation checkpoints in the testing phase in the mathematical model includes:

[0075] Creating regular grids for simulation control points and simulation check points in the image space corresponding to the mathematical model to obtain the simulation control points and the image coordinates of the simulation check points;

[0076] The image simulation points of the push-broom image are transferred to the object space using the collinearity equation, the average height of the study area and the external orientation parameter to obtain the simulation control points and simulation check points of the image coordinates and the object coordinates. Among them, in order to use the collinearity equation CE, a reliable external orientation parameter EOP is necessary. Therefore, considering the dynamic characteristics of the push-broom image, a spatial resection method is required to obtain the EOP of all scan lines. The MPC (Multi-Projection Center) model is used for the spatial resection step, and the expression of the MPC model equation can be found in formula (2).

[0077] As mentioned before, the proposed method consists of two key phases: a training step and a testing step, i.e., utilizing two different sets of points: simulated control points (SCOP) and simulated check points (SCP). While both sets are generated in a similar way, SCOP uses a limited number of points in the training phase, while SCP uses a much larger number of points in the testing phase. To generate the points, a regular grid is created in the image space for SCOP and SCP, respectively, providing them with image coordinates based on their position in the grid. The image simulated points are then transferred to the object space using the CE, the average height of the study area (determined by the true GCPs), and the EOP of all scan lines obtained through the MPC model. This results in a set of SCOP and SCPs with known image and object coordinates, which are ready for processing and evaluation by the BSD method.

[0078] It should be understood that by acquiring push-broom images collected by multiple sensors, and acquiring multiple internal directional parameters and ground control points of the sensors, the push-broom images are spatially cut out of the multi-projection center model according to the external directional parameters and the ground control points, and the external directional parameters of each scanning line corresponding to the push-broom images are obtained. The average height of the study area in the push-broom images is obtained according to the ground control points. A mathematical model of the average height of the study area and all external directional parameters is established using collinearity equations. Simulated control points in the training phase and simulated checkpoints in the testing phase are generated in the mathematical model. The parameters of the simulated checkpoints and the regression model are input into the trained regression model for testing to obtain the optimal number of scanning lines corresponding to the simulated checkpoints. The accuracy of the simulated checkpoints and the optimal number of scanning lines is evaluated to determine the optimal scanning line of the push-broom image. The training and testing phases of the simulated control points and the simulated checkpoints are respectively involved. There is no need to iteratively use collinearity equations, thereby shortening the processing time, having better sub-pixel accuracy, and improving the push-broom image processing efficiency.

[0079] Optionally, the regression model includes a polynomial regression model or a linear regression model. The independent variable in the regression model predicts the dependent variable. Linear regression includes simple linear regression and multivariate linear regression. Simple linear regression is a regression model with a single independent variable. The expression corresponding to simple linear regression is:

[0080] y=a0+a1x(3)

[0081] Among them, y represents the dependent variable, x represents the independent variable, a0 and a1 represent the intercept and linear term values ​​respectively;

[0082] Multiple linear regression models the linear relationship between one or more independent variables and the dependent variable. The expression of multiple linear regression is:

[0083]

[0084] Where y represents a single dependent variable, (a0, ..., a m ) represents the coefficients forming the A matrix, (x 11 , ..., x nm ) represents the independent variables that form the X matrix;

[0085] Use the least squares algorithm to find the best straight line or curve that fits the data set, thereby minimizing the cumulative squared residual. The expression of the least squares algorithm is:

[0086] A=(X T X)- 1 X T Y(5).

[0087] In this embodiment, the polynomial regression model represents a multiple regression modeled by an n-order polynomial. When there is a curvilinear relationship between the independent variable and the dependent variable, the expression using the polynomial regression model is:

[0088] y=a0+a1x+a2x 2 +…+a m x n (6)

[0089] Among them, y represents the dependent variable, x represents the independent variable, (a0,..., a m ) is the regression coefficient of the independent variable, n represents the polynomial degree or order of the regression model; if there are multiple independent variables, the X matrix is ​​used to rewrite equation (6).

[0090] Specifically, the linear regression model includes a training phase and a testing phase. In the training phase, simulated checkpoints are used for each unknown ground control point for accuracy assessment. The simulated control points are used in the least squares algorithm to calculate the unknown parameters of the hybrid logistic regression model. The hybrid logistic regression model uses the ground coordinates (X, Y) as input parameters and the row number (r) as the target parameter to establish the transformation between the ground space and the image space.

[0091] In the testing phase, the hybrid logistic regression model is applied to the ground coordinates (X, Y) of the simulated checkpoints and the row values ​​are estimated. The accuracy of the simulated checkpoints is evaluated using the root mean square error (RMSE) value.

[0092] It should be noted that the accuracy evaluation of the simulation checkpoints and the optimal number of scan lines to determine the optimal scan line of the push-broom image includes: using multiple measurement criteria for comparison to determine the optimal scan line value, the measurement criteria include the root mean square error, the maximum optimal scan line determination error among all simulation checkpoints, and the execution time, wherein the root mean square error and the maximum optimal scan line determination error among all simulation checkpoints are obtained by comparing the obtained optimal scan line value with the accurate scan line value obtained in the simulation stage through linear regression; checking the number of simulation control points and the applicability of the model to obtain a robust model. In other words, one of the simplest and most widely used methods in statistics and machine learning algorithms is the linear regression model, which is a method for establishing a linear relationship between a dependent variable and an independent variable, in which the independent variable predicts the dependent variable. A regression model with a single independent variable is called a simple linear regression (SLR), and the expression of SLR is shown in formula (3). Polynomial regression (PR) is a multivariate regression involving modeling with a polynomial. When there is a curvilinear relationship between the independent variable and the dependent variable, the general model of PR is shown in equation (6). The linear regression model (LRM) BSD method is divided into two steps, namely the training step and the testing step. In the training step, SCOP with known ground and image coordinates is used to obtain the model parameters. Then, SCP is used for each unknown ground point for accuracy evaluation. SCOP is used to calculate the unknown parameters of the MLR model in the least squares method (LSM). The MLR model takes the ground coordinates as input parameters and the row number as the target parameter. In this way, the transformation between the ground space and the image space is established; in the testing step, the MLR model is applied to the ground coordinates of SCP and the row value is estimated; finally, the root mean square error (RMSE) value is used to evaluate the accuracy of SCP. PRM is also applied to the proposed method to obtain more accurate evaluation using SCOP and SCP. These two steps and accuracy evaluation are similar to the case of using the MLR model.

[0093] Furthermore, several measurement criteria were used to evaluate the performance of the proposed method in comparison with traditional effective parameters. These criteria include the root mean square error (RMSE), drmax (the largest BSD error among all SCPs), and execution time. The RMSE and drmax values ​​were obtained by comparing the best scan line values ​​obtained from the proposed LRM BSD method with the exact scan line values ​​obtained in the simulation phase. The number of SCOPs and the fitness of the model were checked to obtain a robust model.

[0094] See also Figure 4 The present invention also provides a push-broom image processing system based on a regression model, which is applied to the push-broom image processing method based on a regression model, comprising:

[0095] An image acquisition module, used to acquire push-broom images collected by multiple sensors, and to acquire internal direction parameters and ground control points of multiple sensors;

[0096] a parameter determination module, configured to perform spatial excision of a multi-projection center model on the push-broom image according to the external direction parameter and the ground control point, so as to obtain the external direction parameter of each scanning line corresponding to the push-broom image;

[0097] A model building module, used for obtaining the average height of the study area in the push-broom image according to the ground control points, and establishing a mathematical model of the average height of the study area and all the external direction parameters using a collinear equation;

[0098] A model processing module, used to generate simulation control points in the training phase and simulation check points in the testing phase in the mathematical model, wherein the simulation control points are used to calculate the parameters of the regression model in the training phase;

[0099] A regression model testing module, used for inputting the simulated checkpoint and the parameters of the regression model into the trained regression model for testing, and obtaining the optimal number of scan lines corresponding to the simulated checkpoint;

[0100] An accuracy evaluation module is used to perform accuracy evaluation on the simulated checkpoints and the optimal number of scan lines to determine the optimal scan line of the push-broom image.

[0101] In this example, the efficiency of the proposed BSD method in LRM and PRM modes is evaluated. The model structure is clear, the number of unknown parameters is known, and the required SCOPs can be checked to solve the model and achieve the required accuracy. The number of SCOPs considered for each image is ten, thirty, fifty, and one hundred. Five million SCPs are used for each image. The results are compared with traditional methods such as BWS, NR, ANN, and OGP methods using quantitative indicators including RMSE, computation time, dr max Compare to the number of SCOPs required.

[0102] The experimental results were obtained using Intel Core i7 hardware with a 2.90GHz processor, HD Graphics 620, and 8GB RAM. By applying four groups of SCOPs to analyze the variation in the number of SCOPs, a large number of SCPs provides a more accurate assessment of the robustness of the proposed BSD method. According to the results in Table 2, increasing the number of SCOPs has a significant impact on the RMSE and dr max The impact of the final accuracy is negligible. In addition, there is no significant difference in the computation time between the four groups, which shows that the limitation of the acquisition and number of SCOPs does not interfere with the proposed BSD method, and it achieves very high accuracy (RMSE equal to 10 to the negative 9th power) in a short time using only a small number of SCOPs. Despite the use of images with different characteristics, there is no correlation between the number of SCOPs, land cover, and the final accuracy, which confirms the success of the proposed BSD method using a variety of satellite images from all aspects.

[0103] Table 2 Results of the proposed BSD method by testing different numbers of SCOPs

[0104]

[0105]

[0106] It should be noted that the comparison results of the proposed BSD method with other methods such as NR, BWSBSS, ANN and OGPBSD methods are shown in Table 3. The comparison is based on some quantitative indicators, including RMSE, computation time, drmax and the number of SCOPs for efficiency comparison. For a better understanding, Figure 3The results of this comparison are also shown. The proposed method requires significantly less time than all other methods. The NR and BWS methods require 250 and 750 times more computational time than the proposed method, respectively. The reason for the significant reduction in time is the non-iterative process used in the proposed method, regardless of whether CE is used or not. On the other hand, the NR and BWS methods require a large number of CE iterations for each ground point, and their accuracy depends on this. However, the BWS method cannot achieve sub-pixel accuracy, which is crucial for photogrammetric applications.

[0107] Table 3 Results of the proposed BSD method compared with the traditional

[0108]

[0109]

[0110]

[0111]

[0112] In addition, the RMSE and dr of the proposed method and NR method max The difference between them is negligible. The comparison of ANN and OGPBSD methods with the proposed method shows that the LRM method achieves higher RMSE by using fewer SCOPs in a shorter time. The main reason for this difference is that ANN and OGPBSD methods require more SCOPs than the LRM method in the error modeling stage. The number and distribution of these SCOPs directly affect the final accuracy of modeling and calculation time to ensure the accuracy of determining the optimal scan line.

[0113] like Figure 3 As shown, Figure 3 (a), (b) and (c) in the figure represent the BSD method of the present invention and RMSE, time and dr respectively. max Schematic diagram of the comparison. Among the compared methods, the proposed LRM and NR methods have the best RMSE. In addition, BWS, NR, OGP, ANN and the proposed method have the highest computational time in this order. NR and the proposed LRM method have the smallest d rmax Based on this evidence, the proposed LRM method is favored due to its high accuracy, low required time and minimal dr maxThe proposed method is superior to previous real-time applications. A method for object-to-image processing using satellite-pushed images via LRM in BSD is proposed. The method is based on LRM and includes training and testing phases involving SCOP and SCP, respectively. It is easy to implement and does not rely on iterative use of CE. Accuracy evaluation was performed using 10, 30, 50 and 100 sets of SCOP and 5 million SCP. The BSD process of the proposed LRM method is faster than other methods and has better sub-pixel accuracy. The LRM method is 250 times faster than the NR method, but the accuracy of the two methods is the same. The BWS method cannot obtain an accuracy better than 0.5 pixel and requires a lot of execution time. On the other hand, the LRM method obtains better accuracy than the ANN and OGP methods in a shorter time and requires fewer SCOPs. The proposed LRM method has potential applications in orthophoto generation, epipolar resampling and other similar tasks.

[0114] In all examples shown and described herein, any specific values ​​should be interpreted as merely exemplary and not as limiting, and thus other examples of the exemplary embodiments may have different values.

[0115] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0116] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the present invention. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A push-broom image processing method based on regression model, It is characterized in that The following steps are involved: Acquire push-broom images acquired by a plurality of sensors, and obtain internal orientation parameters and ground control points of a plurality of said sensors; Performing spatial excision of a multi-projection center model on the push-broom image according to the external direction parameters and the ground control points to obtain external direction parameters of each scanning line corresponding to the push-broom image; Obtaining an average height of a study area in the push-broom image according to the ground control points, and establishing a mathematical model of the average height of the study area and all the external direction parameters using a collinearity equation; Generating simulation control points in the training phase and simulation check points in the testing phase in the mathematical model, wherein the simulation control points are used to calculate the parameters of the regression model in the training phase; Inputting the simulated checkpoint and the parameters of the regression model into the trained regression model for testing to obtain the optimal number of scan lines corresponding to the simulated checkpoint; The simulated checkpoints and the optimal number of scan lines are evaluated for accuracy to determine optimal scan lines for the push-broom image.

2. The push-broom image processing method based on regression model according to claim 1, It is characterized in that The average height of the study area and all the exterior orientation parameters were mathematically modeled using collinearity equations, including: The mathematical model is used to establish the relationship between the 2D image space and the 3D ground space. The expression of the collinear equation is: Where (x, y) represents the 2D image coordinates of any point, f represents the focal length, (X, Y, Z) represents the 3D ground coordinates of any point, and i represents the scan line number. and They represent the position of the projection center in the object space and the rotation angle ω, The rotation matrix element corresponding to k; According to the dynamic characteristics of the push-broom image and expression (1), the spatial resection algorithm of the multi-projection center model is used to determine the external direction parameters of all scan lines. The corresponding expression (2) is: Among them, t i represents the exposure time of the i-th scanning line; X 0 ,X 1 ,…and and k 2 represents the external orientation parameters of the reference scan line calculated in the spatial resection step; X s i (t),Y s i (t),…, and k s i (t) represents the external direction parameter of the i-th scan line.

3. The push-broom image processing method based on regression model according to claim 2, It is characterized in that Generating simulation control points in the training phase and simulation checkpoints in the testing phase in the mathematical model includes: Creating regular grids for simulation control points and simulation check points in the image space corresponding to the mathematical model, respectively, to obtain image coordinates of the simulation control points and the simulation check points; The image simulation points of the push-broom image are transferred to the object space using the collinearity equation, the average height of the study area and the exterior orientation parameters to obtain simulation control points and simulation check points in image coordinates and object coordinates.

4. The push-broom image processing method based on regression model according to claim 1, It is characterized in that The regression model includes a polynomial regression model or a linear regression model. The independent variable in the regression model predicts the dependent variable. Linear regression includes simple linear regression and multivariate linear regression. Simple linear regression is a regression model with a single independent variable. The expression corresponding to simple linear regression is: and now 0 +a 1 x (3) Among them, y represents the dependent variable, x represents the independent variable, and a 0 and a 1 represent the intercept and linear term values ​​respectively; Multiple linear regression models the linear relationship between one or more independent variables and the dependent variable. The expression of multiple linear regression is: Among them, y represents a single dependent variable, (a 0 ,…,a m ) represents the coefficients forming the A matrix, (x 11 ,…,x nm ) represents the independent variables that form the X matrix; Use the least squares algorithm to find the best straight line or curve that fits the data set, thereby minimizing the cumulative squared residual. The expression of the least squares algorithm is: A=(X T X) -1 X T Y (5)。 5. The push-broom image processing method based on regression model according to claim 4, It is characterized in that The polynomial regression model represents a multiple regression modeled by an n-order polynomial. When there is a curvilinear relationship between the independent variable and the dependent variable, the expression using the polynomial regression model is: y=a 0 +a 1 x+a 2 x 2 +…+a m x n (6) Among them, y represents the dependent variable, x represents the independent variable, (a 0 ,…,a m ) is the regression coefficient of the independent variable, n represents the polynomial degree or order of the regression model; if there are multiple independent variables, the X matrix is ​​used to rewrite equation (6).

6. The push-broom image processing method based on regression model according to claim 4, It is characterized in that The linear regression model includes a training phase and a testing phase. In the training phase, simulated checkpoints are used for each unknown ground control point for accuracy assessment. The simulated control points are used in the least squares algorithm to calculate the unknown parameters of the hybrid logistic regression model. The hybrid logistic regression model uses ground coordinates (X, Y) as input parameters and row number (r) as target parameter to establish the transformation between ground space and image space. In the testing phase, the hybrid logistic regression model is applied to the ground coordinates (X,Y) of the simulated checkpoints and the row values ​​are estimated. The accuracy of the simulated checkpoints is evaluated using the root mean square error (RMSE) value.

7. The push-broom image processing method based on regression model according to claim 1, It is characterized in that Evaluating the accuracy of the simulated checkpoints and the optimal number of scan lines to determine the optimal scan lines of the push-broom image includes: A plurality of measurement criteria are used for comparison to determine the best scan line value, the measurement criteria include a root mean square error, a maximum best scan line determination error among all simulation checkpoints, and an execution time, wherein the root mean square error and the maximum best scan line determination error among all simulation checkpoints are obtained by comparing the best scan line value obtained with the accurate scan line value obtained in the simulation phase through linear regression; The number of simulated control points and the fitness of the model were checked to obtain a robust model.

8. The push-broom image processing method based on regression model according to claim 1, It is characterized in that The push-broom images are acquired by at least one sensor selected from the group consisting of IKONOS, Pleiades 1A and 1B, QuickBird, SPOT6, SPOT7, WorldView 1 and WorldView2.

9. The push-broom image processing method based on regression model according to claim 8, It is characterized in that The pushbroom images have different coverage areas, spatial resolutions and sensor characteristics.

10. A push-broom image processing system based on regression model, It is characterized in that A push-broom image processing method based on a regression model as claimed in any one of claims 1 to 9, comprising: An image acquisition module, used to acquire push-broom images collected by multiple sensors, and to acquire internal direction parameters and ground control points of multiple sensors; a parameter determination module, configured to perform spatial excision of a multi-projection center model on the push-broom image according to the external direction parameter and the ground control point, so as to obtain the external direction parameter of each scanning line corresponding to the push-broom image; A model building module, used for obtaining the average height of the study area in the push-broom image according to the ground control points, and establishing a mathematical model of the average height of the study area and all the external direction parameters using a collinear equation; A model processing module, used to generate simulation control points in the training phase and simulation check points in the testing phase in the mathematical model, wherein the simulation control points are used to calculate the parameters of the regression model in the training phase; A regression model testing module, used for inputting the simulated checkpoint and the parameters of the regression model into the trained regression model for testing, and obtaining the optimal number of scan lines corresponding to the simulated checkpoint; An accuracy evaluation module is used to perform accuracy evaluation on the simulated checkpoints and the optimal number of scan lines to determine the optimal scan line of the push-broom image.