Vectorization Acceleration Method and System for Geolocation of Optical Remote Sensing Imagery

By establishing a coordinate system for optical remote sensing images and utilizing matrix calculations, and employing a single-instruction, multi-data-stream, and vectorized design, the problem of low geolocation efficiency of optical remote sensing images was solved, achieving more efficient pixel-by-pixel localization processing.

CN118015071BActive Publication Date: 2026-11-17SHANGHAI SATELLITE ENG INST
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202410015160.7
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-01-04
Publication Date
2026-11-17
Estimated Expiration
2044-01-04

AI Technical Summary

Technical Problem

Existing technologies are inefficient in geolocation processing of optical remote sensing images, especially with large amounts of data, where pixel-by-pixel positioning speed cannot meet real-time requirements, and they rely on limited hardware parallel resources.

Method used

By establishing the optical remote sensor's body coordinate system, the Earth-fixed coordinate system, and the geodetic coordinate system, and using matrix calculation to output coordinate vectors, the number of loops is reduced by adopting a single-instruction, multi-data-stream approach and vectorized design, thus accelerating the geographic positioning of optical remote sensing images.

Benefits of technology

Without relying on parallel hardware resources, it significantly improves the processing speed of pixel-by-pixel geolocation of large data volumes and enhances the processing efficiency of optical remote sensing images.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118015071B_ABST
    Figure CN118015071B_ABST
Patent Text Reader

Abstract

This invention provides a vectorized acceleration method and system for geolocation of optical remote sensing images, including step S1: establishing a coordinate system; the coordinate system includes an optical remote sensor body coordinate system, a fixed Earth coordinate system, and a geodetic coordinate system; step S2: collecting relevant parameters, and outputting coordinate vectors based on matrix calculations. This invention, through reasonable design of the optical remote sensing image geolocation processing cycle, data storage, and processing flow, reduces the number of loops using a single-instruction, multiple-data-stream approach, and accelerates optical remote sensing image geolocation through vectorization. It can effectively improve the processing speed of pixel-by-pixel geolocation of large data volumes without relying on parallel hardware resources such as GPUs.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of optical remote sensing, and more specifically, to a vectorization acceleration method and system for geolocation of optical remote sensing images. Background Technology

[0002] Optical remote sensing has long played an irreplaceable role in fields such as topographic observation, environmental monitoring, marine exploration, and weather forecasting. With the improvement of spatial and spectral resolution of optical remote sensing images, optical remote sensing equipment is becoming increasingly rich in channels and has a larger pixel count. The application of optical remote sensing images is inseparable from high-precision geolocation. For large-scale remote sensing applications such as meteorological and climate research and environmental monitoring, remote sensing payloads operate continuously 24 hours a day, requiring pixel-by-pixel geolocation results. Pixel-by-pixel geolocation under massive amounts of data poses new challenges to the processing speed of application systems.

[0003] For geolocation of optical remote sensing images, Reference 1 (Calibration and Accuracy Verification of Wide-Span Camera of Jilin-1 Wide-Span 01 Satellite, Optics and Precision Engineering, Vol. 29, No. 8, 2021) established a rigorous geometric model of a multi-chip TDI CCD mechanical staggered pushbroom imaging system. Based on pixel-by-pixel geolocation, it realized the generation of seamless multi-chip TDI CCD images and the inter-spectral registration of multi-spectral images.

[0004] Reference 2 (Research on the Development of High Longitude Geographic Positioning Software System for Fengyun Satellite Remote Sensing Data, Journal of Remote Sensing, Vol. 12, No. 2, 2008) presents the processing flow of a geographic positioning calculation software system for remote sensing data, using the 10-channel scanning radiometer of Fengyun-1D satellite and the medium-resolution spectral imager of Fengyun-3A satellite as unstudied objects.

[0005] Reference 3 (On-orbit geometric calibration and longitude verification of Gaofen-5 visible and shortwave infrared hyperspectral images, Journal of Remote Sensing, Vol. 24, No. 4, 2020) uses a geometric calibration model based on the pointing angle of the probe to perform geographic positioning of hyperspectral resolution Earth observations under conditions without control points.

[0006] Reference 4 (Progress in solving problems in improving the image positioning and satellite wind accuracy of Fengyun-2 meteorological satellite, Journal of Nanjing University of Information Science and Technology (Natural Science Edition), Vol. 12, No. 1, 2020) established a closed imaging model of a spin geostationary meteorological satellite. In response to the mismatch between satellite attitude and scanning instruments, an image positioning algorithm based on the time series of Earth disk images was designed to automatically calculate positioning parameters.

[0007] Reference 5 (Research on High-Precision Earth Positioning Technology of Optical Remote Sensing Satellite Geometric Imaging Model, Network Information Engineering, Vol. 13, 2021) analyzes the role of the geometric imaging model and studies the satellite image stereo positioning technology under the background of ephemeris attitude for joint stereo positioning of multiple images.

[0008] Reference 6 (Research on High Longitude Earth Positioning Technology of Optical Remote Sensing Satellite Imagery, Doctoral Dissertation of Information Engineering University, 2015) established a rigorous imaging geometric model of optical remote sensing satellite imagery. Based on geometric calibration, a rational function model was constructed for correction, which effectively improved the positioning accuracy of the imagery.

[0009] To address the efficiency issue of geolocation algorithms for remote sensing images, Reference 7 (GPU-supported parallel processing of RPC correction for wide-span remote sensing images, Radio Engineering, Vol. 51, No. 8, 2021) uses an NVIDIA TITAN RTX GPU to allocate threads for each pixel to be calculated, thus completing the conversion from image coordinates to geographic coordinates.

[0010] Reference 8 (A Fast Orthorectification Method for Remote Sensing Images Based on Block Transformation and GPU Parallelism, Space Return and Remote Sensing, Vol. 39, No. 6, 2018) addresses the problem of excessive computational cost of pixel-by-pixel coordinate transformation in traditional orthorectification by exploring a fast rectification method using GPU parallel technology.

[0011] Reference 9 (Research and Implementation of Coordinate Transformation of Large-Scale Remote Sensing Images Based on GPU Parallel Computing, Surveying and Mapping and Spatial Geographic Information, Vol. 39, No. 3, 2016) uses GPU parallel computing to perform coordinate transformation, realizing rapid transformation of large-scale remote sensing images.

[0012] Chinese patent document CN112750089B discloses a method for dehazing optical remote sensing images based on the prior of the largest and smallest pixels in local blocks. The method includes: Step 1: Calculating a corrected atmospheric light map using latent image information, an atmospheric scattering model, and the definition of the smallest pixel in local blocks; Step 2: Calculating a corrected transmission map based on the corrected atmospheric light map obtained in Step 1; Step 3: Obtaining a preliminary dehazed optical remote sensing image based on the atmospheric scattering model; Step 4: Calculating the atmospheric light map of the preliminary dehazed optical remote sensing image obtained in Step 3 using the atmospheric scattering model and the definition of the largest pixel in local blocks; Step 5: Calculating the transmission map of the preliminary dehazed optical remote sensing image based on the atmospheric light map obtained in Step 4; Step 6: Obtaining a dehazed optical remote sensing image with enhanced contrast based on the atmospheric scattering model. However, the processing efficiency of this patent document is still not ideal. Summary of the Invention

[0013] To address the shortcomings of existing technologies, the purpose of this invention is to provide a vectorization acceleration method and system for geolocation of optical remote sensing images.

[0014] A vectorization acceleration method for geolocation of optical remote sensing images provided by the present invention includes:

[0015] Step S1: Establish a coordinate system;

[0016] The coordinate system includes the optical remote sensor body coordinate system, the Earth-fixed coordinate system, and the geodetic coordinate system;

[0017] Step S2: Collect relevant parameters, and output coordinate vectors after matrix calculation.

[0018] Preferably, step S2 includes the following sub-steps:

[0019] Step S2.1: Initialize the line-of-sight matrix of the optical remote sensor body coordinate system;

[0020] Step S2.2: Calculate the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time;

[0021] Step S2.3: Calculate the satellite position and line-of-sight matrix in the Earth-fixed coordinate system at the corresponding exposure time;

[0022] Step S2.4: Calculate the matrix of intersection points between the line of sight and the Earth;

[0023] Step S2.5: Calculate the coordinate vector in the geodetic coordinate system.

[0024] Preferably, the calculation is based on an algorithm program; the function library of the algorithm program implementation environment supports vectorized operations; the functions include trigonometric functions and arithmetic functions.

[0025] Preferably, the line-of-sight matrix of the optical remote sensor body coordinate system is initialized once; if the optical remote sensor does not undergo focusing changes while in orbit, the line-of-sight matrix remains unchanged; the optical remote sensor calculates the pointing vector of the corresponding line of sight in the optical remote sensor body coordinate system based on the system imaging geometry of the pixel to be located; all lines of sight are initialized into a line-of-sight matrix u1; the line-of-sight matrix u1 is a 3-row × N-column matrix, where N is the number of line-of-sight pointing vectors, and the N line-of-sight pointing vectors contain all pixels of all channels exposed simultaneously; the vectorization acceleration processing is performed on a cycle of one exposure, and the positioning results of all pixels of all channels exposed simultaneously are calculated at the same time.

[0026] Preferably, in step S2.2, the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time is defined as M, then the line-of-sight matrix u in the Earth-fixed coordinate system is... ecr The calculation method is as follows:

[0027] u ecr =M*u1;

[0028] Where * denotes matrix multiplication, M is a 3x3 matrix, u1 is a 3xN matrix, and u ecr It is a matrix with 3 rows and N columns.

[0029] Preferably, step S2.4 includes the following sub-steps:

[0030] Step S2.4.1: Calculate the satellite position p in the Earth-fixed coordinate system ecr Vector p after scaling:

[0031]

[0032] Where, p ecr Let p be a column vector with 3 rows and 1 column, and real numbers. ecrx p ecry p ecrz They are vectors p ecr The three elements are p, which is a column vector of 3 rows and 1 column, constant a, which is the semi-major axis of the Earth reference ellipsoid, and constant b, which is the semi-minor axis of the Earth reference ellipsoid.

[0033] Step S2.4.2: Calculate the line-of-sight matrix u in the Earth-fixed coordinate system. ecr Matrix u after scaling transformation:

[0034]

[0035] Where u is a 3-row × N-column matrix, u x u y u z Let represent the first, second, and third rows of matrix u, respectively, and u x u y u z All are row vectors with 1 row × N columns; u ecr Given a 3xN matrix, u ecrx u ecry u ecrz Representing matrix u respectively ecr The first, second, and third rows, and u ecrx u ecry u ecrz All are row vectors with 1 row × N columns;

[0036] Step S2.4.3: Calculate the line-of-sight distance matrix d:

[0037]

[0038] Where, matrix f pu Matrix f uu All are row vectors with 1 row × N columns; fpp The modulus of vector p is a real number; / represents element-wise division of a matrix, and exponentiation of a matrix means raising the corresponding element to the power of the matrix. This indicates that the corresponding element in the matrix is ​​raised to the power of 2.

[0039] Matrix f pu The calculation method is as follows:

[0040] f pu =p T *u;

[0041] Where * denotes matrix multiplication, p T f represents the transpose of vector p, which is 1 row × 3 columns; pu A row vector of 1 row × N columns; matrix f uu For matrix u 2 The result of row-wise summation is matrix u. 2 This represents the result of raising the corresponding elements of matrix u to the second power. 2 The matrix f has the same size as matrix u, 3 rows × N columns; after summing the rows, matrix f... uu A row vector consisting of 1 row and N columns;

[0042] Step S2.4.4: Calculate the intersection matrix q of the line of sight and the Earth:

[0043]

[0044] Where · denotes element-wise multiplication of matrices, and q is a 3xN matrix. x q y q z Let represent the first, second, and third rows of matrix q, respectively, and q x q y q z All are row vectors with 1 row and N columns.

[0045] Preferably, in step S2.4.3, the matrix is ​​calculated. r to the power of 0.5 0.5 When |r| is calculated, the absolute value of all elements in matrix r is taken. 0.5 .

[0046] Preferably, step S2.5 includes the following sub-steps:

[0047] Step S2.5.1: Calculate the longitude vector λ:

[0048] λ=arctan2(q y ,q x );

[0049] Where arctan2 represents the arctangent function in the fourth quadrant, λ is a vector of 1 row × N columns, and q x q y q z All are row vectors with 1 row × N columns;

[0050] Step S2.5.2: Calculate the latitude vector

[0051]

[0052] Where e represents the eccentricity of the Earth's reference ellipsoid, which is a constant; the arithmetic operation / , when performing operations on matrices, represents dividing corresponding elements in the matrix. A vector with 1 row and N columns; longitude vector λ and latitude vector λ. The elements in the array correspond one-to-one with N, which represents the positioning result of the line of sight.

[0053] Preferably, the calculation of the line-of-sight matrix and the calculation of the transformation matrix are performed based on the imaging geometry model.

[0054] A vectorization acceleration system for geolocation of optical remote sensing images provided by the present invention includes:

[0055] Module M1: Establish coordinate system;

[0056] The coordinate system includes the optical remote sensor body coordinate system, the Earth-fixed coordinate system, and the geodetic coordinate system;

[0057] Module M2: Collects relevant parameters and outputs coordinate vectors after matrix calculation.

[0058] Preferably, module M2 includes the following sub-modules:

[0059] Module M2.1: Initializes the line-of-sight matrix of the optical remote sensor's body coordinate system;

[0060] Module M2.2: Calculates the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time;

[0061] Module M2.3: Calculates the satellite position and line-of-sight matrix in the Earth-fixed coordinate system at the corresponding exposure time;

[0062] Module M2.4: Calculates the matrix of intersection points between the line of sight and the Earth;

[0063] Module M2.5: Calculates coordinate vectors in the geodetic coordinate system.

[0064] Preferably, the calculation is based on an algorithm program; the function library of the algorithm program implementation environment supports vectorized operations; the functions include trigonometric functions and arithmetic functions.

[0065] Preferably, the line-of-sight matrix of the optical remote sensor body coordinate system is initialized once; if the optical remote sensor does not undergo focusing changes while in orbit, the line-of-sight matrix remains unchanged; the optical remote sensor calculates the pointing vector of the corresponding line of sight in the optical remote sensor body coordinate system based on the system imaging geometry of the pixel to be located; all lines of sight are initialized into a line-of-sight matrix u1; the line-of-sight matrix u1 is a 3-row × N-column matrix, where N is the number of line-of-sight pointing vectors, and the N line-of-sight pointing vectors contain all pixels of all channels exposed simultaneously; the vectorization acceleration processing is performed on a cycle of one exposure, and the positioning results of all pixels of all channels exposed simultaneously are calculated at the same time.

[0066] Preferably, in module M2.2, the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time is defined as M, then the line-of-sight matrix u in the Earth-fixed coordinate system is... ecr The calculation method is as follows:

[0067] u ecr =M*u1;

[0068] Where * denotes matrix multiplication, M is a 3x3 matrix, u1 is a 3xN matrix, and u ecr It is a matrix with 3 rows and N columns.

[0069] Preferably, module M2.4 includes the following sub-modules:

[0070] Module M2.4.1: Calculate the satellite position p in the Earth-fixed coordinate system ecr Vector p after scaling:

[0071]

[0072] Where, p ecr Let p be a column vector with 3 rows and 1 column, and real numbers. ecrx p ecry p ecrz They are vectors p ecr The three elements are p, which is a column vector of 3 rows and 1 column, constant a, which is the semi-major axis of the Earth reference ellipsoid, and constant b, which is the semi-minor axis of the Earth reference ellipsoid.

[0073] Module M2.4.2: Calculate the line-of-sight matrix u in the Earth-fixed coordinate system ecr Matrix u after scaling transformation:

[0074]

[0075] Where u is a 3-row × N-column matrix, u x u y u zLet represent the first, second, and third rows of matrix u, respectively, and u x u y u z All are row vectors with 1 row × N columns; u ecr Given a 3xN matrix, u ecrx u ecry u ecrz Representing matrix u respectively ecr The first, second, and third rows, and u ecrx u ecry u ecrz All are row vectors with 1 row × N columns;

[0076] Module M2.4.3: Calculate the line-of-sight distance matrix d:

[0077]

[0078] Where, matrix f pu Matrix f uu All are row vectors with 1 row × N columns; f pp The modulus of vector p is a real number; / represents element-wise division of a matrix, and exponentiation of a matrix means raising the corresponding element to the power of the matrix. This indicates that the corresponding element in the matrix is ​​raised to the power of 2.

[0079] Matrix f pu The calculation method is as follows:

[0080] f pu =p T *u;

[0081] Where * denotes matrix multiplication, p T f represents the transpose of vector p, which is 1 row × 3 columns; pu A row vector of 1 row × N columns; matrix f uu For matrix u 2 The result of row-wise summation is matrix u. 2 This represents the result of raising the corresponding elements of matrix u to the second power. 2 The matrix f has the same size as matrix u, 3 rows × N columns; after summing the rows, matrix f... uu A row vector consisting of 1 row and N columns;

[0082] Module M2.4.4: Calculate the matrix q of the intersection points of the line of sight and the Earth:

[0083]

[0084] Where · denotes element-wise multiplication of matrices, and q is a 3xN matrix. x q y qz Let represent the first, second, and third rows of matrix q, respectively, and q x q y q z All are row vectors with 1 row and N columns.

[0085] Preferably, in module M2.4.3, the matrix is ​​calculated. r to the power of 0.5 0.5 When |r| is calculated, the absolute value of all elements in matrix r is taken. 0.5 .

[0086] Preferably, module M2.5 includes the following sub-modules:

[0087] Module M2.5.1: Calculate the longitude vector λ:

[0088] λ=arctan2(q y ,q x );

[0089] Where arctan2 represents the arctangent function in the fourth quadrant, λ is a vector of 1 row × N columns, and q x q y q z All are row vectors with 1 row × N columns;

[0090] Module M2.5.2: Calculate the latitude vector

[0091]

[0092] Where e represents the eccentricity of the Earth's reference ellipsoid, which is a constant; the arithmetic operation / , when performing operations on matrices, represents dividing corresponding elements in the matrix. A vector with 1 row and N columns; longitude vector λ and latitude vector λ. The elements in the array correspond one-to-one with N, which represents the positioning result of the line of sight.

[0093] Preferably, the calculation of the line-of-sight matrix and the calculation of the transformation matrix are performed based on the imaging geometry model.

[0094] Compared with the prior art, the present invention has the following beneficial effects:

[0095] 1. This invention, through the rational design of the processing cycle, data storage and processing flow of optical remote sensing images for geolocation, reduces the number of loops by using a single instruction multiple data stream approach, and accelerates the geolocation of optical remote sensing images through vectorization. It can effectively improve the processing speed of pixel-by-pixel geolocation of large amounts of data without relying on parallel hardware resources such as GPUs.

[0096] 2. This invention proposes a vectorization acceleration method, which accelerates the geolocation of optical remote sensing images by rationally designing the storage and processing flow in the geolocation processing of optical remote sensing images, reducing the number of loops by using a single instruction multiple data stream approach, and achieving acceleration of geolocation of optical remote sensing images through software methods.

[0097] 3. The method provided by this invention is reasonable, simple to calculate, easy to implement, and has no hardware dependency. It can be widely applied to the vectorization acceleration of geographic positioning of optical remote sensing images.

[0098] Other beneficial effects of the present invention will be explained in detail through the introduction of specific technical features and technical solutions in specific embodiments. Those skilled in the art should be able to understand the beneficial technical effects brought about by these technical features and technical solutions through the introduction of these technical features and technical solutions. Attached Figure Description

[0099] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0100] Figure 1 This is a flowchart of the method of the present invention.

[0101] Figure 2 This is a schematic diagram illustrating the time required for geolocation of an optical remote sensing image using the traditional pixel-by-pixel geolocation method in an embodiment of the present invention.

[0102] Figure 3 This is a schematic diagram showing the time required for geolocation of a certain optical remote sensing image after using the method proposed in this invention. Detailed Implementation

[0103] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the scope of protection of the present invention.

[0104] In the geolocation process of optical remote sensing imagery, it is often necessary to geolocate a large number of pixels one by one. Under the framework of traditional serial loop computation, the geolocation time increases linearly with the number of pixels. Accelerating this process through pure software without relying on hardware parallelism can speed up execution by using a single-instruction multiple-data (SMD) approach and vectorized design. The fixed number of geolocation pixels and the absence of dependencies between pixels provide the necessary conditions for vectorized acceleration. The concept of vectorized acceleration is to accelerate the processing efficiency of batch data by using arrays, accessing contiguous memory space, and leveraging SMD.

[0105] In mathematics, one-dimensional arrays are conventionally called vectors, two-dimensional data are called matrices, and three-dimensional and higher-dimensional arrays are called tensors. However, in computer storage, regardless of the number of dimensions, they are actually stored in one dimension. Therefore, operations on vectors, matrices, and tensors can also be collectively referred to as vectorized operations.

[0106] Reference Figure 1 As shown, a vectorization acceleration method for geolocation of optical remote sensing images includes:

[0107] To achieve vectorization acceleration, the algorithm's implementation environment's function library must first support vectorized operations, including trigonometric functions and arithmetic functions (including addition, subtraction, multiplication, division, and exponentiation). Secondly, the data block unit for vectorization acceleration processing needs to be clearly defined. In this invention, a single exposure is used as the cycle, with all pixels in all channels exposed simultaneously calculating their positioning results. First, the line-of-sight matrix of the optical remote sensor's body coordinate system is calculated. This calculation only requires initialization once, and the line-of-sight matrix remains unchanged as long as the optical remote sensor remains in orbit without focusing or other changes. For the optical remote sensor, regardless of whether it's an area array or linear array imaging system, and whether it's single-channel or multi-channel, the pointing vector (unit vector) of the corresponding line of sight in the optical remote sensor's body coordinate system can be calculated based on the system's imaging geometry for the pixel to be positioned. All lines of sight are initialized into a line-of-sight matrix u1, which is a 3xN matrix where N is the number of line-of-sight pointing vectors. These N line-of-sight pointing vectors contain all pixels from all channels exposed simultaneously.

[0108] Then, based on the exposure time, calculate the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at that exposure time. The calculation of the line-of-sight matrix and the transformation matrix needs to be performed according to a rigorous imaging geometry model.

[0109] If the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time is M, then the line-of-sight matrix u in the Earth-fixed coordinate system... ecr The calculation method is as follows:

[0110] u ecr =M*u1;

[0111] Where * denotes matrix multiplication, M is a 3x3 matrix, u1 is a 3xN matrix, and u ecr It is a 3xN matrix. In addition, the satellite position p in the Earth-fixed coordinate system also needs to be calculated. ecr .

[0112] The next step is to design the process of calculating the intersection point with Earth based on the satellite position and line of sight in the Earth-fixed coordinate system as a vectorizable and executable workflow.

[0113] First, calculate the satellite position p in the Earth-fixed coordinate system.ecr The vector p after scaling.

[0114]

[0115] Where, p ecr Let p be a column vector with 3 rows and 1 column, and real numbers. ecrx p ecry p ecrz They are vectors p ecr The three elements are p, which is a column vector of 3 rows and 1 column, a, a constant, and b, which is the semi-major axis of the Earth reference ellipsoid.

[0116] Calculate the line-of-sight matrix u in the Earth-fixed coordinate system ecr The matrix u after scaling transformation

[0117]

[0118] Where u is a 3-row × N-column matrix, u x u y u z These represent the row vectors of the 1st, 2nd, and 3rd rows of matrix u, which are 1 row × N columns. ecr Given a 3xN matrix, u ecrx u ecry u ecrz Representing matrix u respectively ecr The first, second, and third rows are row vectors, which are 1 row × N columns.

[0119] Calculate the line-of-sight distance matrix d.

[0120]

[0121] Where, matrix f pu Matrix f uu Both are row vectors with 1 row × N columns, f pp The modulus of vector p is a real number; / denotes element-by-element division of a matrix; and exponentiation of a matrix means raising the powers of corresponding elements in the matrix to the nearest power. This indicates that the corresponding element in the matrix is ​​raised to the power of 2.

[0122] Matrix f pu The calculation method is as follows

[0123] f pu =p T *u;

[0124] Where * denotes matrix multiplication, p T f represents the transpose of vector p, with dimensions of 1 row × 3 columns. puA row vector with 1 row and N columns. Matrix f uu For matrix u 2 The result of row-wise summation is matrix u. 2 This represents the result of raising the corresponding element of matrix u to the power of 2, i.e., matrix u 2 The matrix f has the same size as matrix u, 3 rows × N columns. After summing the rows, matrix f is... uu It is a row vector with 1 row and N columns.

[0125] Considering the fault tolerance of the method, for the matrix Calculate its 0.5 power r 0.5 To calculate |r|, you can first take the absolute value of all elements in matrix r. 0.5 This avoids the situation where outliers cause the entire real matrix to become a complex matrix. Outliers can be checked in the program by checking whether the elements in matrix r are greater than or equal to zero.

[0126] Calculate the matrix q, which represents the intersection of the line of sight with the Earth.

[0127]

[0128] Where · denotes element-wise multiplication of matrices, and q is a 3xN matrix. x q y q z These represent the row vectors of the first, second, and third rows of matrix q, which are 1 row × N columns.

[0129] Once the intersection matrix q is obtained, the geographic latitude and longitude can be calculated using library functions that support vectorization operations.

[0130] The method for calculating the longitude vector λ is as follows:

[0131] λ=arctan2(q y ,q x );

[0132] Where arctan2 represents the arctangent function in the four quadrants, and λ is a vector with 1 row and N columns.

[0133] Dimension vector The calculation method is as follows:

[0134]

[0135] Where e represents the eccentricity of the Earth's reference ellipsoid, which is a constant. The arithmetic operation / , when performing operations on matrices, represents dividing corresponding elements within the matrix. A vector with 1 row and N columns. Longitude vector λ and latitude vector. The elements in the array correspond one-to-one with N, which represents the positioning result of the line of sight.

[0136] This invention improves the processing speed of pixel-by-pixel geolocation of large amounts of data by rationally designing the processing cycle, data storage and processing flow of optical remote sensing images, reducing the number of loops by using a single instruction multiple data stream approach, and accelerating the geolocation of optical remote sensing images through vectorization. It can do so without relying on hardware parallel resources such as GPUs.

[0137] The above are basic embodiments of the present invention. The technical solution of the present invention will be further described below through a preferred embodiment.

[0138] Example 1

[0139] The method of the present invention will be verified using a satellite optical remote sensing image. The satellite image has 1024×1024 pixels in a single channel and an exposure period of 0.625s. If the pixel-by-pixel geolocation of a single frame image cannot be completed within 0.625s, it will cause data congestion.

[0140] This embodiment uses Python 3.8.8 as the software language, and the vectorized function library is implemented based on NumPy, which supports matrix and mathematical operations. (See also...) Figure 2 As shown, on a PC with an 11th Gen Intel Core i7-11800H processor and 64GB of RAM, the traditional pixel-by-pixel positioning method is used; refer to Figure 3 As shown, after using the vectorization acceleration of the present invention, the processing time per frame is significantly reduced, which greatly improves the efficiency of geolocation. The processing time per frame is also much shorter than the exposure cycle of optical remote sensing instruments.

[0141] Reference Figure 2 As shown, the raw remote sensing data acquired by the fast imager channel 7 includes 4 linear arrays, each containing 256 pixels. Figure 3 This is a reconstructed image obtained after executing the technical process of this invention. Since the original data in the image is from a rapid imager scanning a local area of ​​the Earth, data exists in local areas of the nominal grid for a period of time. The areas without data are displayed as black in the image. Figure 3 The continuous black area at the middle edge. As shown in the figure, the signal-to-noise ratio of the reconstructed image is significantly improved, laying a high-precision foundation for subsequent meteorological secondary product inversion.

[0142] The present invention also provides a vectorization acceleration system for geolocation of optical remote sensing images. The vectorization acceleration system for geolocation of optical remote sensing images can be implemented by executing the process steps of the vectorization acceleration method for geolocation of optical remote sensing images. That is, those skilled in the art can understand the vectorization acceleration method for geolocation of optical remote sensing images as a preferred embodiment of the vectorization acceleration system for geolocation of optical remote sensing images.

[0143] Specifically, a vectorization acceleration system for geolocation of optical remote sensing images includes:

[0144] Module M1: Establish coordinate system;

[0145] The coordinate system includes the optical remote sensor body coordinate system, the Earth-fixed coordinate system, and the geodetic coordinate system;

[0146] Module M2: Collects relevant parameters and outputs coordinate vectors after matrix calculation.

[0147] Module M2 includes the following sub-modules:

[0148] Module M2.1: Initializes the line-of-sight matrix of the optical remote sensor's body coordinate system;

[0149] Module M2.2: Calculates the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time;

[0150] Module M2.3: Calculates the satellite position and line-of-sight matrix in the Earth-fixed coordinate system at the corresponding exposure time;

[0151] Module M2.4: Calculates the matrix of intersection points between the line of sight and the Earth;

[0152] Module M2.5: Calculates coordinate vectors in the geodetic coordinate system.

[0153] The calculations are based on an algorithm program; the function library of the algorithm program implementation environment supports vectorized operations; the functions include trigonometric functions and arithmetic functions.

[0154] The line-of-sight matrix of the optical remote sensor's body coordinate system is initialized once; if the optical remote sensor does not undergo focusing changes while in orbit, the line-of-sight matrix remains unchanged; the optical remote sensor calculates the pointing vector of the corresponding line of sight in the optical remote sensor's body coordinate system based on the system imaging geometry of the pixel to be located; all lines of sight are initialized into a line-of-sight matrix u1; the line-of-sight matrix u1 is a 3-row × N-column matrix, where N is the number of line-of-sight pointing vectors, and the N line-of-sight pointing vectors contain all pixels of all channels exposed simultaneously; vectorization acceleration processing is performed on a cycle of one exposure, and the positioning results of all pixels of all channels exposed simultaneously are calculated at the same time.

[0155] In module M2.2, the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time is defined as M. Then, the line-of-sight matrix u in the Earth-fixed coordinate system is... ecr The calculation method is as follows:

[0156] u ecr =M*u1;

[0157] Where * denotes matrix multiplication, M is a 3x3 matrix, u1 is a 3xN matrix, and u ecr It is a matrix with 3 rows and N columns.

[0158] Module M2.4 includes the following sub-modules:

[0159] Module M2.4.1: Calculate the satellite position p in the Earth-fixed coordinate system ecr Vector p after scaling:

[0160]

[0161] Where, p ecr Let p be a column vector with 3 rows and 1 column, and real numbers. ecrx p ecry p ecrz They are vectors p ecr The three elements are p, which is a column vector of 3 rows and 1 column, constant a, which is the semi-major axis of the Earth reference ellipsoid, and constant b, which is the semi-minor axis of the Earth reference ellipsoid.

[0162] Module M2.4.2: Calculate the line-of-sight matrix u in the Earth-fixed coordinate system ecr Matrix u after scaling transformation:

[0163]

[0164] Where u is a 3-row × N-column matrix, u x u y u z Let represent the first, second, and third rows of matrix u, respectively, and u x u y u z All are row vectors with 1 row × N columns; u ecr Given a 3xN matrix, u ecrx u ecry u ecrz Representing matrix u respectively ecr The first, second, and third rows, and u ecrx u ecry u ecrz All are row vectors with 1 row × N columns;

[0165] Module M2.4.3: Calculate the line-of-sight distance matrix d:

[0166]

[0167] Where, matrix f pu Matrix f uu All are row vectors with 1 row × N columns; f pp The modulus of vector p is a real number; / represents element-wise division of a matrix, and exponentiation of a matrix means raising the corresponding element to the power of the matrix. This indicates that the corresponding element in the matrix is ​​raised to the power of 2.

[0168] Matrix f pu The calculation method is as follows:

[0169] f pu =p T *u;

[0170] Where * denotes matrix multiplication, p T f represents the transpose of vector p, which is 1 row × 3 columns; pu A row vector of 1 row × N columns; matrix f uu For matrix u 2 The result of row-wise summation is matrix u. 2 This represents the result of raising the corresponding elements of matrix u to the second power. 2 The matrix f has the same size as matrix u, 3 rows × N columns; after summing the rows, matrix f... uu A row vector consisting of 1 row and N columns;

[0171] Module M2.4.4: Calculate the matrix q of the intersection points of the line of sight and the Earth:

[0172]

[0173] Where · denotes element-wise multiplication of matrices, and q is a 3xN matrix. x q y q z Let represent the first, second, and third rows of matrix q, respectively, and q x q y q z All are row vectors with 1 row and N columns.

[0174] In module M2.4.3, the matrix is ​​calculated. r to the power of 0.5 0.5 When |r| is calculated, the absolute value of all elements in matrix r is taken. 0.5 .

[0175] The module M2.5 includes the following sub-modules:

[0176] Module M2.5.1: Calculate the longitude vector λ:

[0177] λ=arctan2(q y ,q x );

[0178] Where arctan2 represents the arctangent function in the fourth quadrant, λ is a vector of 1 row × N columns, and q x q y q z All are row vectors with 1 row × N columns;

[0179] Module M2.5.2: Calculate the latitude vector

[0180]

[0181] Where e represents the eccentricity of the Earth's reference ellipsoid, which is a constant; the arithmetic operation / , when performing operations on matrices, represents dividing corresponding elements in the matrix. A vector with 1 row and N columns; longitude vector λ and latitude vector λ. The elements in the array correspond one-to-one with N, which represents the positioning result of the line of sight.

[0182] The calculation of the line-of-sight matrix and the calculation of the transformation matrix are performed based on the imaging geometry model.

[0183] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0184] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A vectorization acceleration method for optical remote sensing image geolocation, characterized in that, include: Step S1: Establish a coordinate system; The coordinate system includes the optical remote sensor body coordinate system, the Earth-fixed coordinate system, and the geodetic coordinate system; Step S2: Collect relevant parameters, and output coordinate vectors after matrix calculation; Step S2 includes the following sub-steps: Step S2.1: Initialize the line-of-sight matrix of the optical remote sensor body coordinate system; Step S2.2: Calculate the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time; Step S2.3: Calculate the satellite position and line-of-sight matrix in the Earth-fixed coordinate system at the corresponding exposure time; Step S2.4: Calculate the matrix of intersection points between the line of sight and the Earth; Step S2.5: Calculate the coordinate vectors in the geodetic coordinate system; The line-of-sight matrix of the optical remote sensor body coordinate system is initialized once; if the optical remote sensor does not undergo focusing changes while in orbit, the line-of-sight matrix remains unchanged. The optical remote sensor calculates the pointing vector of the corresponding line of sight in the optical remote sensor's body coordinate system based on the imaging geometry of the pixel to be located; it initializes all lines of sight into a line of sight matrix u1; the line of sight matrix u1 is a 3-row × N-column matrix, where N is the number of line of sight pointing vectors, and the N line of sight pointing vectors contain all pixels of all channels exposed at the same time; the vectorization acceleration processing is performed on a cycle of one exposure, and the positioning results of all pixels of all channels exposed at the same time are calculated simultaneously.

2. The vectorization acceleration method for optical remote sensing image geolocation according to claim 1, characterized in that, The calculations are based on an algorithm program; the function library of the algorithm program implementation environment supports vectorized operations; the functions include trigonometric functions and arithmetic functions.

3. The vectorization acceleration method for optical remote sensing image geolocation according to claim 1, wherein, The conversion matrix of the optical remote sensing body coordinate system corresponding to the exposure moment to the earth-fixed coordinate system is defined as M, and the line-of-sight matrix u in the earth-fixed coordinate system is ecr The calculation method is: ; wherein represents a matrix multiplication, M is a 3 row x 3 column matrix, ui is a 3 row x N column matrix, u ecr is a 3 row x N column matrix.

4. The vectorization acceleration method for optical remote sensing image geolocation according to claim 1, wherein, Step S2.4 includes the following sub-steps: Step S2.4.1: Calculate satellite position p in Earth-fixed coordinate system ecr Proportionally transformed vector p: ; where p ecr is a 3-by-1 column vector, real numbers p ecrx , p ecry , p ecrz are three elements of the vector p ecr , p is a 3-by-1 column vector, constant a is a semi-major axis of an earth reference ellipsoid, and constant b is a semi-minor axis of the earth reference ellipsoid; Step S2.4.2: Calculate the line-of-sight matrix u of the earth fixed coordinate system ecr The scaled matrix u: ; Where u is a 3-row × N-column matrix, u x u y u z Let represent the first, second, and third rows of matrix u, respectively, and u x u y u z All are row vectors with 1 row × N columns; u ecr Given a 3xN matrix, u ecrx u ecry u ecrz Representing matrix u respectively ecr The first, second, and third rows, and u ecrx u ecry u ecrz All are row vectors with 1 row × N columns; Step S2.4.3: Calculate the line-of-sight distance matrix d: ; Where, matrix f pu Matrix f uu All are row vectors with 1 row × N columns; f pp The modulus of vector p is a real number; / represents element-wise division of a matrix, and exponentiation of a matrix means raising the corresponding element to the power of the matrix. This indicates that the corresponding element in the matrix is ​​raised to the power of 2. Matrix f pu The calculation method is: ; in, p represents matrix multiplication. T f represents the transpose of vector p, which is 1 row × 3 columns; pu A row vector of 1 row × N columns; matrix f uu For matrix u 2 The result of row-wise summation is matrix u. 2 This represents the result of raising the square to the power of the corresponding elements in matrix u. 2 The matrix f has the same size as matrix u, 3 rows × N columns; after summing the rows, matrix f... uu A row vector consisting of 1 row and N columns; Step S2.4.4: Calculate the intersection matrix q of the line of sight and the Earth: ; where • denotes multiplication of corresponding elements of matrices, q is a 3-by-N matrix, q x , q y , q z denote the 1st row, 2nd row, 3rd row of matrix q, respectively, and q x , q y , q z are row vectors of 1-by-N.

5. The vectorization acceleration method for geolocation of optical remote sensing images according to claim 4, characterized in that, In the step S2.4.3, the 0.5 power of the matrix is calculated and the absolute value of all elements of the matrix r is taken, i.e. the matrix is calculated.

6. The vectorization acceleration method for geolocation of optical remote sensing images according to claim 1, characterized in that, Step S2.5 includes the following sub-steps: Step S2.5.1: Calculate the longitude vector λ: ; where arctan2 denotes the four quadrant arctangent function, λ is a 1 row by N column vector, q x , q y , q z are all 1 row by N column row vectors; Step S2.5.2: Calculate the latitude vector φ: ; Where e represents the eccentricity of the Earth's reference ellipsoid, which is a constant; the arithmetic operation / represents the division of corresponding elements in the matrix when performing operations on the matrix; φ is a vector with 1 row × N columns; the elements in the longitude vector λ and the latitude vector φ are in one-to-one correspondence; N is the positioning result of the line of sight.

7. The vectorization acceleration method for geolocation of optical remote sensing images according to claim 1, characterized in that, The calculation of the line-of-sight matrix and the calculation of the transformation matrix are performed based on the imaging geometry model.

8. A vectorization acceleration system for geolocation of optical remote sensing images, characterized in that, include: Module M1: Establish coordinate system; The coordinate system includes the optical remote sensor body coordinate system, the Earth-fixed coordinate system, and the geodetic coordinate system; Module M2: Collects relevant parameters and outputs coordinate vectors after matrix calculation; The module M2 includes the following sub-steps: Module M2.1: Initializes the line-of-sight matrix of the optical remote sensor's body coordinate system; Module M2.2: Calculates the transformation matrix from the optical remote sensing body coordinate system to the Earth-fixed coordinate system at the corresponding exposure time; Module M2.3: Calculates the satellite position and line-of-sight matrix in the Earth-fixed coordinate system at the corresponding exposure time; Module M2.4: Calculates the matrix of intersection points between the line of sight and the Earth; Module M2.5: Calculates coordinate vectors in a geodetic coordinate system; The line-of-sight matrix of the optical remote sensor body coordinate system is initialized once; if the optical remote sensor does not undergo focusing changes while in orbit, the line-of-sight matrix remains unchanged. The optical remote sensor calculates the pointing vector of the corresponding line of sight in the optical remote sensor's body coordinate system based on the imaging geometry of the pixel to be located; it initializes all lines of sight into a line of sight matrix u1; the line of sight matrix u1 is a 3-row × N-column matrix, where N is the number of line of sight pointing vectors, and the N line of sight pointing vectors contain all pixels of all channels exposed at the same time; the vectorization acceleration processing is performed on a cycle of one exposure, and the positioning results of all pixels of all channels exposed at the same time are calculated simultaneously.

Citation Information

Patent Citations

  • Optical remote sensing image dehazing method based on local block maximum and minimum pixel priors

    CN112750089B

  • SPOT-6 satellite image direct geographic positioning method based on FPGA hardware

    CN113671550A