Satellite autonomous posture determination method, system and storage medium

Abstract

The present invention provides a method for determining the autonomous posture of a satellite, by transmitting the remote sensing image taken by an optical camera and the template image retrieved from the knowledge graph to a deep convolutional network to obtain matching feature information; according to the matching feature information, the optical camera imaging model, motion equation and observation equation are brought into the determined model to solve, thereby obtaining the satellite attitude state. The method of the present invention can effectively improve the efficiency and accuracy of satellite attitude and orbit estimation, thereby realizing real-time posture determination of the satellite in an offline state, and improving the stability and reliability of the satellite operation and control system.

Description

Technical Field

[0001] The present invention relates to the field of satellite control, and in particular to a method, system and storage medium for determining satellite autonomous positioning and attitude based on visual SLAM (Simultaneous Localization and Mapping) technology. Background Art

[0002] For a long time, space has been a sacred place for human exploration and yearning. Since the beginning of the 21st century, mankind has made remarkable achievements in science and technology, politics, society, finance and economy, which has provided a solid foundation for mankind to explore the universe and move towards space. At the same time, the progress of high-tech represented by aerospace technology provides strong support for a country's economic construction and prosperity. Therefore, countries around the world have placed the development of aerospace industry technology in an extremely important position.

[0003] Satellite navigation is a key link in the operation of spacecraft. With the continuous development of satellite navigation technology, a lot of research has been carried out on the autonomous orbit determination of satellites. At present, the autonomous orbit determination methods that are already in use include: autonomous orbit determination method based on onboard GPS, which uses global, multi-observation data and low-cost onboard GPS measurement to process onboard GPS observation data in real time on orbit, obtain high-precision orbit parameters of satellites, and realize autonomous real-time orbit determination of low-orbit satellites; autonomous orbit determination method based on star sensors, which uses star sensors to retrieve background star data, and uses related algorithms to calculate the position and attitude of the satellite, thereby realizing autonomous orbit determination; autonomous orbit determination method based on magnetometer, which uses star sensors, earth sensors and magnetometers as measurement sensors, uses related optimization algorithms for state estimation, and realizes autonomous orbit determination of satellites. With the rapid development of science and technology in the space field and the rapid increase in the number of spacecraft, the above methods are gradually unable to meet the performance requirements of accuracy, intelligence, autonomy, stability and reliability. Summary of the invention

[0004] The present invention provides a method for determining satellite autonomous positioning and attitude. The method is applied to satellite autonomous navigation and can effectively improve the efficiency and accuracy of satellite attitude and orbit estimation, thereby realizing real-time attitude determination of the satellite in an offline state and improving the stability and reliability of the satellite operation and control system.

[0005] The present invention provides a method for determining an autonomous posture of a satellite, comprising: acquiring a remote sensing image taken by an optical camera; searching in a knowledge graph based on the remote sensing image to obtain a template image; transmitting the remote sensing image and the template image to a deep convolutional network to obtain matching feature information; determining an imaging model, a motion equation and an observation equation of the optical camera; and solving the motion equation and the observation equation according to the matching feature information to obtain a satellite attitude state.

[0006] Furthermore, the step of transmitting the remote sensing image and the template image to the deep convolutional network to obtain matching feature information includes: transmitting the remote sensing image and the template image to the preprocessing layer of the deep convolutional network; extracting multiple sets of feature information through the preprocessing layer; building a feature space converter; performing normalization operations based on the multiple sets of feature information obtained and through the feature space converter; transmitting the obtained normalized feature information to the deep convolutional network; and performing matching operations on the normalized feature information to obtain matching feature information.

[0007] Furthermore, the step of extracting multiple groups of feature information through the preprocessing layer also includes: extracting corresponding feature information according to different convolution kernels in the preprocessing layer; and superimposing the corresponding feature information extracted by different convolution kernels to obtain multiple groups of feature information.

[0008] Furthermore, the step of determining the optical camera imaging model, motion equation and observation equation includes: acquiring a light beam through an optical camera and projecting an image on a camera plane; obtaining a motion equation corresponding to the satellite motion according to the motion changes of the satellite, feature information extracted by the optical camera and noise obtained by the satellite during the motion; and obtaining an observation equation corresponding to the motion equation according to the satellite's observation data and the noise obtained by the satellite during the observation process.

[0009] Furthermore, the step of solving the motion equation and the observation equation according to the matching feature information to obtain the satellite attitude state includes: parameterizing the motion equation and the observation equation; obtaining the probability equation of the satellite's motion state according to the conditional probability information; and solving the probability equation by adopting the maximum a posteriori probability method to obtain the satellite's posture state information.

[0010] The present invention also provides a satellite autonomous posture determination system, comprising: a remote sensing image acquisition module, used to acquire a remote sensing image taken by an optical camera; a template image acquisition module, used to search in a knowledge graph based on the remote sensing image to obtain a template image; a feature information acquisition module, used to transmit the remote sensing image and the template image to a deep convolutional network to obtain matching feature information; a model equation determination module, used to determine the optical camera imaging model, motion equation and observation equation; and a posture state acquisition module, used to solve the motion equation and the observation equation according to the matching feature information to obtain the satellite posture state.

[0011] Furthermore, the feature information obtaining module includes: an image transmission unit, used to transmit the remote sensing image and the template image to the preprocessing layer of the deep convolutional network; a feature extraction unit, used to extract multiple sets of feature information through the preprocessing layer; a converter construction unit, used to build a feature space converter; a normalization operation unit, used to perform a normalization operation based on the obtained multiple sets of feature information and through the feature space converter; a matching operation unit, used to perform a matching operation on the normalized feature information to obtain matching feature information.

[0012] Furthermore, the model equation determination module includes: a beam acquisition unit, which is used to acquire a light beam through an optical camera and project an image on a camera plane; a motion equation acquisition unit, which is used to obtain a motion equation corresponding to the satellite motion according to the motion changes of the satellite, the characteristic information extracted by the optical camera, and the noise obtained by the satellite during the motion; and an observation equation acquisition unit, which is used to obtain an observation equation corresponding to the motion equation according to the satellite's observation data and the noise obtained by the satellite during the observation process.

[0013] Furthermore, the attitude state acquisition module includes: a parameterized operation unit, which is used to perform parameterized operations on the motion equation and the observation equation; a probability equation acquisition unit, which is used to obtain the probability equation of the motion state of the satellite based on the conditional probability information; and a posture state acquisition unit, which is used to solve the probability equation by adopting the maximum a posteriori probability method to obtain the posture state information of the satellite.

[0014] The present invention also provides a storage medium, wherein a plurality of instructions are stored in the storage medium, and the instructions are suitable for being loaded by a processor to execute the method for determining the autonomous position and posture of a satellite as described above.

[0015] The embodiment of the present invention provides a method for determining the autonomous posture of a satellite, by transmitting the remote sensing image taken by an optical camera and the template image retrieved from the knowledge graph to a feature matching deep convolutional network to obtain matching feature information; according to the matching feature information, the optical camera imaging model, motion equation and observation equation are brought into the determined model to solve, thereby obtaining the satellite attitude state. The method of the present invention can effectively improve the efficiency and accuracy of satellite attitude and orbit estimation, thereby realizing real-time posture determination of the satellite in an offline state, and improving the stability and reliability of the satellite operation and control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The technical solutions and other beneficial effects of the present invention will be made apparent by describing in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0017] Figure 1 The present invention is a flowchart of a method for determining the autonomous position and posture of a satellite provided in one embodiment of the present invention.

[0018] Figure 2 A schematic diagram of an optical camera imaging model provided by an embodiment of the present invention.

[0019] Figure 3 A schematic diagram of the architecture of a multi-scale deep convolutional network provided for one embodiment of the present invention.

[0020] Figure 4 A flowchart of sub-steps of step S3 provided for one embodiment of the present invention.

[0021] Figure 5 A flowchart of sub-steps of step S302 provided for one embodiment of the present invention.

[0022] Figure 6 A flowchart of sub-steps of step S4 provided for one embodiment of the present invention.

[0023] Figure 7 A flowchart of sub-steps of step S5 provided for one embodiment of the present invention.

[0024] Figure 8 The present invention is a functional module diagram of a satellite autonomous posture determination system provided in one embodiment of the present invention.

[0025] Fig. 9 This is a functional module diagram of the feature information obtaining module provided in one embodiment of the present invention.

[0026] Fig.10 This is a functional module diagram of the feature extraction unit block provided in one embodiment of the present invention.

[0027] Fig.11This is a functional module diagram of the model equation determination module provided in one embodiment of the present invention.

[0028] Fig.12 This is a functional module diagram of a posture state obtaining module provided in one embodiment of the present invention. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0030] The disclosure below provides many different embodiments or examples to realize different structures of the present invention. In order to simplify the disclosure of the present invention, the parts and settings of specific examples are described below. Of course, they are only examples, and the purpose is not to limit the present invention. In addition, the present invention can repeat reference numbers and / or reference letters in different examples, and this repetition is for the purpose of simplicity and clarity, which itself does not indicate the relationship between the various embodiments and / or settings discussed. In addition, the present invention provides various specific examples of processes and materials, but those of ordinary skill in the art can be aware of the application of other processes and / or the use of other materials.

[0031] With the increasing development of simultaneous positioning and mapping (SLAM) technology and the continuous improvement of the resolution of optical payloads to the ground, the method of autonomous orbit determination and attitude determination of satellites based on visual SLAM technology has brought a new development direction for autonomous navigation of satellites. Compared with traditional satellite navigation methods, the method of autonomous orbit determination and attitude determination of satellites based on visual SLAM technology has obvious advantages: on the one hand, the optical image itself contains rich information, and the image vision field based on the visual SLAM method has made great progress, making autonomous navigation of satellites based on visual SLAM technology possible; on the other hand, the use of artificial intelligence methods based on deep learning enables satellites to have higher autonomous performance, effectively reducing dependence on ground stations and saving manpower and material resources.

[0032] like Figure 1 As shown, an embodiment of the present invention provides a method for determining the autonomous posture of a satellite based on SLAM, comprising the following steps S1 to S9.

[0033] Combination Figure 2 As shown, step S1, obtaining a remote sensing image taken by an optical camera.

[0034] Step S2: Based on the remote sensing image, search in the knowledge graph to obtain a template image.

[0035] Step S3, transmit the remote sensing image and the template image to the deep convolutional network (such as Figure 3 As shown in Figure 1, convolutional neural network (CNN), dense features, and sparse features are used to obtain matching feature information. Figure 4 As shown, step S3 specifically includes the following steps: S301 to S306.

[0036] Step S301, transmitting the remote sensing image and the template image to the preprocessing layer of the deep convolutional network.

[0037] Step S302, extracting multiple groups of feature information through a preprocessing layer.

[0038] Further, if Figure 5 As shown, step S302 specifically includes the following steps:

[0039] Step S3021, extracting corresponding feature information according to different convolution kernels in the preprocessing layer.

[0040] Step S3022, the corresponding feature information extracted by different convolution kernels is superimposed to obtain multiple sets of feature information. Considering the differences in scale and position between the target features on the remote sensing image taken by the optical camera and the target features on the retrieved template image, the captured remote sensing image and the retrieved template image are first transmitted to the preprocessing layer of the deep convolutional network for multi-scale feature extraction. This layer has feature convolution kernels of multiple scales. Multiple copies of the remote sensing image will be transmitted, and feature extraction operations will be performed synchronously on convolution kernels of different scales. At the same time, convolution kernels of different scales are not independent. Different features are superimposed on multi-scale features through upsampling, and finally multiple sets of feature information with wide applicability are obtained.

[0041] Step S303, building a feature space converter.

[0042] Step S304, based on the obtained multiple sets of feature information, a normalization operation is performed through a feature space converter. After the preprocessing layer, considering that the multiple sets of feature information are obtained, although each set of feature information has wide applicability, there is a problem that different sets of features have different feature spaces. Therefore, a feature space normalization converter is introduced in the second stage of the deep convolutional network to perform feature space normalization operations on multiple sets of features.

[0043] Step S305, transmitting the obtained normalized feature information to the deep convolutional network.

[0044] Step S306: performing a matching operation on the normalized feature information to obtain matching feature information.

[0045] Specifically, after obtaining the normalized feature information, considering the sparsity and density of the normalized feature information, the feature information is subjected to sparse and dense synchronous matching convolution operations, and different operations introduce different feature matching algorithms. Then, different matching operations are superimposed to screen out matching features with strong robustness.

[0046] Step S4, determine the optical camera imaging model, motion equation and observation equation. Figure 6 As shown, step S4 specifically includes the following steps S401 to S404.

[0047] Step S401, acquiring a light beam through an optical camera and projecting an image on a camera plane.

[0048] Step S402, obtaining a motion equation corresponding to the satellite motion according to the motion change of the satellite, the feature information extracted by the optical camera, and the noise obtained by the satellite during the motion.

[0049] The satellite moves in orbit, and its motion equation is extracted into a mathematical equation: k =f(x k-1 ,u k ,w k ), where x k is the satellite motion change from k-1th to kth time, u k is the input of the satellite-borne sensor (here, the features extracted by the onboard optical camera), w k This is the noise introduced during this movement.

[0050] Step S403, obtaining an observation equation corresponding to the equation of motion based on the satellite's observation data and the noise obtained by the satellite during the observation process.

[0051] Specifically, the satellite moves in orbit, and its position is characterized by displacement and rotation angle, that is, The parameters x, y and z here represent the displacement of the three coordinate axes of the coordinate system, the parameter θ represents the rotation angle, the parameter x subscript k is the change of satellite motion from the k-1th to the kth time, and the input parameters are the change of two time intervals and the rotation angle. At this time, the motion equation is expressed as: Finally, the above equations of motion and observation equations are combined, that is, At this time, w k and v k , j is the noise of the equation of motion and the observation equation. Z subscript (k, j) represents the observation equation generated by the j-th observation landmark point y subscript j seen during the motion of x subscript k.

[0052] Step S404, obtaining coordinate information of an observation point and position information of the optical camera, and constructing a set of constraint equations for the observation point.

[0053] Specifically, consider an observation point P in space, which is also the following feature point. The homogeneous coordinates of this point are P = (X, Y, Z, 1) T , and on the pixel plane, the corresponding projection point coordinates are P'=(u1,v1,1) T ; At this time, the position and posture of the spaceborne optical camera is represented by R,t, where R is the rotation matrix and t is the translation vector. According to the coordinates of the observation point and the camera position and posture information, the equation is constructed: Here, u subscript 1 and v subscript 1 represent the horizontal and vertical coordinates of the pixel plane, respectively, and X, Y, and Z represent the homogeneous components of the three axes in the spatial coordinate system.

[0054] Using the last row of the augmented matrix consisting of the rotation matrix and the translation vector, eliminating s, we get two constraint equations: Furthermore, to simplify writing, the row vector of T is defined as: From the above, we can see that each pair of feature points provides two constraint equations, so the constraint equations for N pairs of feature points are: N represents multiple, such as 1 pair of feature points, 2 pairs of feature points, 3 pairs of feature points...N pairs of feature points (8 pairs are commonly used), and P is an observation point in space.

[0055] Continue reading Figure 1 , step S5, solving the motion equation and the observation equation according to the matching feature information to obtain the satellite attitude state.

[0056] Combination Figure 7 As shown, step S5 specifically includes the following steps S501 to S503.

[0057] Step S501, parameterize the motion equation and observation equation to obtain sz k,j =K(R k y j +t k ), where K is the intrinsic parameter of the optical camera, s is the distance of the pixel point, and it is assumed that the noise of this implementation method satisfies the Gaussian distribution with a mean of zero, that is, w k ~N(0,R k ),v k ~N(0,Q k,j ), inferring the information of the pose x through the noisy data z and u, which constitutes a state estimation problem.

[0058] Step S502: Obtain a probability equation for the motion state of the satellite based on the conditional probability information.

[0059] Given the input data u and the observation data z, solve the state of the pose x and the landmark point y, introduce the Bayesian theorem, and get

[0060] Step S503, solving the probability equation by adopting the maximum a posteriori probability method to obtain the satellite's position and posture state information.

[0061] Specifically, the optimization method is used to maximize the posterior probability of a state, that is, Here, arg max refers to the maximum a posteriori probability, and the subscript MAP is also the value corresponding to the maximum a posteriori probability. Based on the above Gaussian distribution assumption, corresponding to a certain observation: z k,j =h(y j ,x k )+v k,j , and the noise satisfies v k ~N(0,Q k,j ), so the conditional probability of this observation is: P(z j,k |x k ,y j )=N(h(y i ,x k ),Q k,j ). Consider expanding the high-dimensional Gaussian distribution x~N(μ,Σ) and taking its negative logarithm, that is, Considering the above negative logarithmic equation, since the logarithm is monotonically increasing, maximizing the original function is equivalent to minimizing the negative logarithmic function. When minimizing the above equation, considering that the x value is independent of the first term on the right, only the quadratic term on the right is considered and brought into the observation equation, that is, Further assuming that the motion and observation at each moment are independent, the joint distribution can be factorized: Redefine each actual motion and the error between the observation and the model: e u,k =x k -f(x k-1 ,u k ),e z,j,k =z k,j -h(x k ,y j ). Convert the above minimization equation into a joint product form, that is This is converted into a least squares problem. Finally, the least squares equation mentioned above can be used to solve the position and state of the onboard optical camera, and then the satellite's orbit and other information can be calculated. Among the above parameters, μ and Σ are the mean and variance of the Gaussian distribution respectively; det() is used to solve the determinant of a square matrix; Q and the subscript (k, j) indicate that the noise variance in the normal distribution is determined by the parameters (k, j), (k, j) see the above explanation. E and the subscript (u, k) represent the error between each actual movement and the observation and the model.

[0062] like Figure 8 As shown, an embodiment of the present invention further provides a satellite autonomous posture determination system 200, including: a remote sensing image acquisition module 201, a template image acquisition module 202, a feature information acquisition module 203, a model equation determination module 204 and an attitude state acquisition module 205.

[0063] The remote sensing image acquisition module 201 is used to acquire remote sensing images taken by an optical camera.

[0064] The template image acquisition module 202 is used to search in the knowledge graph based on the remote sensing image to obtain the template image.

[0065] The feature information obtaining module 203 is used to transmit the remote sensing image and the template image to the deep convolutional network to obtain matching feature information.

[0066] like Fig. 9 As shown, the feature information obtaining module 203 includes: an image transmission unit 2031, a feature extraction unit 2032, a converter construction unit 2033, a normalization operation unit 2034 and a matching operation unit 2035.

[0067] The image transmission unit 2031 is used to transmit the remote sensing image and the template image to the preprocessing layer of the deep convolutional network.

[0068] The feature extraction unit 2032 is used to extract multiple sets of feature information through the preprocessing layer; Fig.10 As shown, the feature extraction unit 2032 includes a feature information extraction subunit 20321 and a feature information superposition subunit 20322. The feature information extraction subunit 20321 is used to extract corresponding feature information according to different convolution kernels in the preprocessing layer; the feature information superposition subunit 20322 is used to superimpose the corresponding feature information extracted by different convolution kernels to obtain multiple sets of feature information.

[0069] The converter building unit 2033 is used to build a feature space converter.

[0070] The normalization operation unit 2034 is used to perform a normalization operation based on the obtained multiple groups of feature information through a feature space converter.

[0071] The matching operation unit 2035 is used to perform a matching operation on the normalized feature information to obtain matching feature information.

[0072] The model equation determination module 204 is used to determine the optical camera imaging model, motion equation and observation equation. Fig.11 As shown, the model equation determination module 204 includes: a light beam acquisition unit 2041, a motion equation obtaining unit 2042 and an observation equation obtaining unit 2043.

[0073] The light beam acquisition unit 2041 is used to acquire the light beam through an optical camera and project an image on a camera plane.

[0074] The motion equation obtaining unit 2042 is used to obtain the motion equation corresponding to the satellite motion according to the motion change of the satellite, the feature information extracted by the optical camera and the noise obtained by the satellite during the motion.

[0075] The observation equation obtaining unit 2043 is used to obtain the observation equation corresponding to the motion equation according to the satellite's observation data and the noise obtained by the satellite during the observation process.

[0076] The attitude state obtaining module 205 is used to solve the motion equation and the observation equation according to the matching feature information to obtain the satellite attitude state.

[0077] like Fig.12 As shown, the posture state obtaining module 205 includes: a parameterized operation unit 2051, a probability equation obtaining unit 2052 and a posture state obtaining unit 2053.

[0078] The parameterization operation unit 2051 is used to perform parameterization operations on the motion equation and the observation equation.

[0079] The probability equation obtaining unit 2052 is used to obtain the probability equation of the motion state of the satellite according to the conditional probability information.

[0080] The posture state obtaining unit 2053 is used to solve the probability equation by adopting the maximum a posteriori probability method to obtain the posture state information of the satellite.

[0081] The present invention also provides a storage medium, in which a plurality of instructions are stored, and the instructions are suitable for being loaded by a processor to execute the method for determining the autonomous position and posture of a satellite.

[0082] The present invention provides a method for determining the autonomous posture of a satellite, by transmitting the remote sensing image taken by an optical camera and the template image retrieved from the knowledge graph to a feature matching deep convolutional network to obtain matching feature information; according to the matching feature information, the optical camera imaging model, motion equation and observation equation are brought into the determined model to solve, thereby obtaining the satellite attitude state. The method of the present invention can effectively improve the efficiency and accuracy of satellite attitude and orbit estimation, thereby realizing real-time posture determination of the satellite in an offline state, and improving the stability and reliability of the satellite operation and control system.

[0083] The present invention has been introduced in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the technical solutions and core ideas of the present invention. Ordinary technicians in this field should understand that they can still modify the technical solutions recorded in the aforementioned embodiments, or replace some of the technical features therein with equivalents. However, these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for determining the autonomous position and attitude of a satellite, characterized in that: include: Acquire remote sensing images taken by an optical camera; Based on remote sensing images, search in the knowledge graph to obtain template images; Transmit the remote sensing image and the template image to the deep convolutional network to obtain matching feature information; Determine the imaging model, motion equation and observation equation of the optical camera; as well as Solving the motion equation and the observation equation according to the matching feature information to obtain the satellite attitude state; The step of determining the optical camera imaging model, motion equation and observation equation comprises: The light beam is acquired by an optical camera and an image is projected on the camera plane; The motion equation corresponding to the satellite motion is obtained based on the satellite's motion changes, the characteristic information extracted by the optical camera, and the noise obtained during the satellite's motion. Based on the satellite's observation data and the noise obtained by the satellite during the observation process, the observation equation corresponding to the equation of motion is obtained; The step of solving the motion equation and the observation equation according to the matching feature information to obtain the satellite attitude state includes: Parameterize the equations of motion and observation; According to the conditional probability information, the probability equation of the satellite's motion state is obtained; By adopting the maximum a posteriori probability method, the probability equation is solved to obtain the satellite's position and state information.

2. The method for determining the autonomous position of a satellite as claimed in claim 1, characterized in that: The step of transmitting the remote sensing image and the template image to the deep convolutional network to obtain matching feature information includes: Transmitting remote sensing images and template images to the preprocessing layer of the deep convolutional network; Extract multiple sets of feature information through the preprocessing layer; Build a feature space converter; According to the obtained multiple sets of feature information, a normalization operation is performed through a feature space converter; Transmitting the obtained normalized feature information to the deep convolutional network; and A matching operation is performed on the normalized feature information to obtain matching feature information.

3. The method for determining the autonomous position of a satellite as claimed in claim 2, wherein: The step of extracting multiple sets of feature information through the preprocessing layer also includes: Extract corresponding feature information according to different convolution kernels in the preprocessing layer; The corresponding feature information extracted by different convolution kernels is superimposed to obtain multiple sets of feature information.

4. A satellite autonomous posture determination system, characterized in that: include: A remote sensing image acquisition module, used to acquire remote sensing images taken by an optical camera; The template image acquisition module is used to search in the knowledge graph based on the remote sensing image to obtain the template image; The feature information acquisition module is used to transmit the remote sensing image and the template image to the deep convolutional network to obtain matching feature information; A model equation determination module is used to determine the optical camera imaging model, motion equation and observation equation; as well as An attitude state obtaining module is used to solve the motion equation and the observation equation according to the matching feature information to obtain the satellite attitude state; The model equation determination module includes: A light beam acquisition unit, used for acquiring the light beam through an optical camera and projecting an image on a camera plane; The motion equation obtaining unit is used to obtain the motion equation corresponding to the satellite motion according to the motion change of the satellite, the characteristic information extracted by the optical camera and the noise obtained by the satellite during the motion process; An observation equation obtaining unit is used to obtain an observation equation corresponding to the motion equation according to the satellite observation data and the noise obtained by the satellite during the observation process; The posture state obtaining module includes: A parameterized operation unit, used to perform parameterized operations on the equations of motion and observation equations; The probability equation obtaining unit is used to obtain the probability equation of the satellite's motion state according to the conditional probability information; the posture state obtaining unit is used to solve the probability equation by adopting the maximum a posteriori probability method to obtain the satellite's posture state information.

5. The satellite autonomous posture determination system as claimed in claim 4, characterized in that: The feature information obtaining module includes: An image transmission unit, used for transmitting remote sensing images and template images to a preprocessing layer of a deep convolutional network; A feature extraction unit, used for extracting multiple sets of feature information through a preprocessing layer; A converter building unit, used to build a feature space converter; A normalization operation unit, used for performing a normalization operation based on the obtained multiple sets of feature information through a feature space converter; The matching operation unit is used to perform a matching operation on the normalized feature information to obtain matching feature information.

6. A storage medium, characterized in that: The storage medium stores a plurality of instructions, and the instructions are suitable for being loaded by a processor to execute the method for determining the autonomous posture of a satellite as described in any one of claims 1 to 5.

Images