Atmosphere outside the polarization light field continuation model based on neural network learning

Abstract

The application provides an atmospheric out-of-layer polarized light field continuation model calculation method based on neural network learning, comprising the following steps: S1, collecting spatial polarized light field data and obtaining to-be-calculated data; S2, building a neural network model and adjusting weights and biases; S3, training the neural network model by taking the spatial polarized light field data as training samples, inputting the to-be-calculated data into the trained neural network model, and calculating a predicted polarization azimuth and a predicted degree of polarization through forward propagation; and S4, obtaining a position vector and a first sun vector of a to-be-observed point, calculating a measured polarization azimuth and a measured degree of polarization, and calculating a polarization azimuth error percentage and a degree of polarization error percentage to check the accuracy of the neural network model. The application can improve the high-precision measurement and analysis capability of the atmospheric out-of-layer polarized light field, and provide accurate and reliable data support for a satellite navigation system.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of space multi-physical field calculation, in particular to a polarized light field continuation model calculation method based on neural network learning outside the atmosphere. BACKGROUND

[0002] When sunlight passes through the atmosphere, the gas molecules and aerosol particles in the atmosphere cause scattering and absorption of incident light, etc., so that the atmospheric polarized light field in the sky forms a relatively stable distribution over time and space. In nature, sandworms, bees, locusts, mantis shrimps and other organisms can perceive and use the atmospheric polarized light field by using their unique eye structure. The polarization vision phenomenon of organisms inspires scholars to study the bionic technology of realizing navigation and positioning by using the atmospheric polarized light field. The bionic polarization light navigation technology has the characteristics of strong anti-interference, no cumulative error and strong concealment, and is suitable for long-time and long-distance navigation in complex environment, and has broad prospects in the navigation field.

[0003] When observing the polarized light outside the atmosphere, the satellite obtains the polarized light signals reflected by the atmosphere and the ground through observation. These polarization characteristics are closely related to the attitude and orbit of the satellite, and can provide important information for navigation. The traditional calculation method is usually based on the Rayleigh scattering model and the Mie scattering model and mathematical equations for derivation, but there are certain limitations in dealing with complex light field distribution and nonlinear problems, and it is difficult to quickly learn and predict the continuation of the spatial polarized light field. The learning ability for complex nonlinear relationships is poor, which may lead to inaccurate continuation results of the spatial polarized light field. In order to overcome the limitations of the traditional derivation method based on physical model and mathematical equation in dealing with complex light field distribution and nonlinear problems, the present application aims to provide model support and verification basis for satellite autonomous navigation, and to provide accurate and reliable data support for satellite navigation system, improve the high-precision measurement and analysis capability of the polarized light field outside the atmosphere, and has important practical significance and wide application prospect. SUMMARY

[0004] The technical problem to be solved by the present application is to improve the high-precision measurement and analysis capability of the polarized light field outside the atmosphere, and to provide accurate and reliable data support for satellite navigation system. In order to overcome the defects of the above prior art (or related technology), the present application provides a polarized light field continuation model calculation method based on neural network learning outside the atmosphere.

[0005] The present application provides a polarized light field continuation model calculation method based on neural network learning outside the atmosphere, comprising:

[0006] Step S1, data acquisition: collecting spatial polarized light field data and obtaining data to be calculated;

[0007] Step S2, model construction: build an input layer, a hidden layer and an output layer to obtain a neural network model, and set a corresponding activation function and an initialization method to adjust weights and biases of the neural network model;

[0008] Step S3, model training: train the neural network model with the spatial polarization light field data as training samples, and after the training is completed, input the to-be-calculated data into the trained neural network model, and calculate a predicted polarization azimuth and a predicted polarization degree through forward propagation;

[0009] Step S4, model verification: obtain a position vector and a first sun vector of a to-be-observed point, obtain a measured polarization azimuth and a measured polarization degree of the to-be-observed point according to the position vector and the first sun vector, obtain a polarization azimuth error percentage according to the predicted polarization azimuth and the measured polarization azimuth, and obtain a polarization degree error percentage according to the predicted polarization degree and the measured polarization degree, so as to verify the accuracy of the neural network model.

[0010] Compared with the prior art, the atmospheric layer outside polarization light field continuation model calculation method based on neural network learning has the following advantages:

[0011] In the present application, the spatial polarization light field data and the to-be-calculated data are obtained through step S1, the neural network model is built through step S2, the neural network model is trained through step S3, and the error percentage is calculated through step S4. In the present application, the neural network model is established to perform continuation calculation on the polarization light field. The deep learning technology can quickly learn and predict the azimuth and polarization degree after the spatial polarization light field is continued. Compared with the traditional method, it is more efficient, and can learn complex nonlinear relationships, so it can provide more accurate spatial polarization light field continuation results. In addition, the parameters can be adjusted through the training data, so as to adapt to different light field data characteristics, have a certain adaptability, and provide theoretical support and systematic solution for the spatial polarization light field calculation problem.

[0012] In one possible implementation, the step S2 includes:

[0013] Step S21, building the input layer, the hidden layer and the output layer to obtain the neural network model, and setting a hidden layer parameter to specify the number of neurons in the hidden layer;

[0014] Step S22, selecting a hyperbolic tangent function as the activation function, and using a uniform distribution initialization method to initialize the weights of the neural network model in a random number uniformly distributed in the range of [-a, a].

[0015] In a possible implementation, the spatially corrected light field data comprises a sun azimuth vector, an orbit parameter, an observation point latitude and longitude, and polarization information of a to-be-observed point, and the step S3 comprises:

[0016] In step S31, a corresponding loss function and an optimization algorithm are configured for the neural network model.

[0017] In step S32, the sun azimuth vector, the orbit parameter, and the observation point latitude and longitude are taken as inputs, and the polarization information of the to-be-observed point is taken as output, and the neural network model is trained.

[0018] In step S33, the to-be-calculated data is input into the trained neural network model for feature extraction and nonlinear calculation, to obtain the predicted polarization azimuth angle and the predicted polarization degree after light field continuation.

[0019] In a possible implementation, the step S4 comprises:

[0020] In step S41, the position vector of the to-be-observed point, the first sun vector in the body coordinate system, and the current time date are obtained, and the Julian date of the day is obtained according to the current time date and a pre-set historical time date.

[0021] In step S42, based on the Julian date, a first coordinate conversion matrix from the geocentric inertial coordinate system to the satellite orbit coordinate system is calculated, and a second coordinate conversion matrix from the satellite orbit coordinate system to the body coordinate system is calculated.

[0022] In step S43, the position vector and the first sun vector are cross-multiplied to obtain the included angle of the to-be-observed point and the sun relative to the observation point, and the measured polarization degree is obtained according to the included angle.

[0023] In step S44, the satellite orbit coordinate system is selected as the observation coordinate system, the second sun vector in the satellite orbit coordinate system is obtained according to the first coordinate conversion matrix, the position vector of the to-be-observed point is converted to the spherical coordinate system to calculate the measured polarization azimuth angle.

[0024] In step S45, the polarization azimuth angle error percentage is obtained according to the predicted polarization azimuth angle and the measured polarization azimuth angle, and the polarization degree error percentage is obtained according to the predicted polarization degree and the measured polarization degree, to verify the accuracy of the neural network model.

[0025] In a possible implementation, in the step S41, the Julian date is obtained according to the following calculation formula:

[0026] ;

[0027] wherein,

[0028] JD represents the Julian day;

[0029] represents the current time date;

[0030] represents the historical time date.

[0031] In a possible implementation, in the step S42, the first coordinate conversion matrix is obtained by the following calculation formula:

[0032] ;

[0033] wherein,

[0034] represents the first coordinate conversion matrix;

[0035] represents the coordinate rotation matrix of rotation along the y axis;

[0036] represents the coordinate rotation matrix of rotation along the x axis;

[0037] represents the coordinate rotation matrix of rotation along the z axis;

[0038] represents the orbital inclination;

[0039] represents the right ascension of the ascending node;

[0040] represents the latitude depression angle.

[0041] In a possible implementation, in the step S42, the second coordinate conversion matrix is obtained by the following calculation formula:

[0042] ;

[0043] wherein,

[0044] represents the second coordinate conversion matrix;

[0045] represents the coordinate rotation matrix of rotation along the x axis;

[0046] represents the coordinate rotation matrix of rotation along the y axis;

[0047] represents the coordinate rotation matrix of rotation along the z axis;

[0048] denotes a roll angle of the body coordinate system relative to the satellite orbit coordinate system;

[0049] denotes a pitch angle of the body coordinate system relative to the satellite orbit coordinate system;

[0050] denotes a yaw attitude angle of the body coordinate system relative to the satellite orbit coordinate system.

[0051] In one possible implementation, the measured polarization degree is obtained in the step S43 by the following calculation formula:

[0052]

[0053] wherein,

[0054] denotes the measured polarization degree;

[0055] denotes a preset maximum polarization degree;

[0056] denotes the included angle.

[0057] In one possible implementation, the measured polarization azimuth angle is obtained in the step S44 by the following calculation formula:

[0058]

[0059] wherein,

[0060] denotes the measured polarization azimuth angle;

[0061] denotes the second sun vector in the satellite orbit coordinate system;

[0062] denotes an x coordinate of the sun in the satellite orbit coordinate system;

[0063] denotes a y coordinate of the sun in the satellite orbit coordinate system;

[0064] denotes a z coordinate of the sun in the satellite orbit coordinate system;

[0065] denotes the first coordinate conversion matrix;

[0066] ​​represents a third sun vector in the geocentric inertial coordinate system;

[0067] represents a position vector of the point to be observed in the body coordinate system;

[0068] represents the second coordinate conversion matrix;

[0069] represents a position vector of the point to be observed in the body coordinate system;

[0070] represents an x coordinate of the point to be observed in the satellite orbit coordinate system;

[0071] represents a y coordinate of the point to be observed in the satellite orbit coordinate system;

[0072] represents a z coordinate of the point to be observed in the satellite orbit coordinate system;

[0073] r represents a distance from an origin to a target point in a spherical coordinate system;

[0074] represents a pitch angle of the body coordinate system relative to the satellite orbit coordinate system;

[0075] represents a roll angle of the body coordinate system relative to the satellite orbit coordinate system;

[0076] x represents an x coordinate converted to the spherical coordinate system;

[0077] y represents a y coordinate converted to the spherical coordinate system;

[0078] z represents a z coordinate converted to the spherical coordinate system;

[0079] represents a pitch angle of the sun vector in the spherical coordinate system;

[0080] represents a roll angle of the sun vector in the spherical coordinate system;

[0081] represents a pitch angle of the point to be observed in the spherical coordinate system;

[0082] represents a roll angle of the point to be observed in the spherical coordinate system.

[0083] In a possible implementation, in the step S45, the polarization azimuth angle error percentage and the polarization degree error percentage are obtained by the following calculation formulas:

[0084] ;

[0085] wherein,

[0086] represents the polarization degree error percentage;

[0087] represents the predicted polarization degree;

[0088] represents the measured polarization degree;

[0089] represents the polarization azimuth error percentage;

[0090] represents the predicted polarization azimuth;

[0091] represents the measured polarization azimuth. BRIEF DESCRIPTION OF DRAWINGS

[0092] Figure 1 is a step flowchart of the present application;

[0093] Figure 2 is a structural schematic diagram of the neural network model of the present application;

[0094] Figure 3 is a specific flowchart of step S2 of the present application;

[0095] Figure 4 is a specific flowchart of step S3 of the present application;

[0096] Figure 5 is a specific flowchart of step S4 of the present application;

[0097] Figure 6 is a simulation system interface schematic diagram of the present application;

[0098] Figure 7 is a simulation example diagram of the present application. DETAILED DESCRIPTION

[0099] First, those skilled in the art should understand that these embodiments are only used to explain the technical principles of the embodiments of the present application, and are not intended to limit the protection scope of the embodiments of the present application. Those skilled in the art can adjust them as needed in order to adapt to specific application occasions.

[0100] The present application will be further described in detail below in combination with the drawings and specific embodiments.

[0101] Reference is made to Figure 1The embodiment of the application discloses a polarized light field continuation model calculation method based on neural network learning outside the atmosphere, comprising:

[0102] Step S1, data acquisition: collecting spatially polarized light field data and obtaining to-be-calculated data;

[0103] Step S2, model construction: building an input layer, a hidden layer and an output layer to obtain a neural network model, and setting a corresponding activation function and an initialization method to adjust the weight and bias of the neural network model;

[0104] Step S3, model training: taking the spatially polarized light field data as a training sample to train the neural network model, after the training is completed, inputting the to-be-calculated data, i.e., the current to-be-calculated spatially polarized light field data, into the trained neural network model, and calculating a predicted polarization azimuth and a predicted polarization degree through forward propagation;

[0105] Step S4, model verification: obtaining a position vector and a first sun vector of an observation point, obtaining a measured polarization azimuth and a measured polarization degree of the observation point according to the position vector and the first sun vector, obtaining a polarization azimuth error percentage according to the predicted polarization azimuth and the measured polarization azimuth, and obtaining a polarization degree error percentage according to the predicted polarization degree and the measured polarization degree, so as to verify the accuracy of the neural network model.

[0106] Referring to Figure 3 , step S2 comprises:

[0107] Step S21, building an input layer, a hidden layer and an output layer to obtain a neural network model, and setting a hidden layer parameter to specify the number of neurons in the hidden layer;

[0108] The number of neurons in the hidden layer is specified by setting the hidden layer parameter, the design of the hidden layer affects the learning ability and expression ability of the network, too few hidden layer neurons may not be able to learn complex patterns, and too many may lead to overfitting, and the appropriate size of the hidden layer can be selected according to the specific problem and data characteristics, and the optimal parameter configuration is determined through cross-validation and other methods;

[0109] Step S22, selecting a hyperbolic tangent function as an activation function, and using a uniform distribution initialization method to initialize the weight of the neural network model in a random number uniformly distributed in the range of [-a, a];

[0110] The hyperbolic tangent (tanh) is selected as the activation function, which helps to enhance the nonlinear representation ability of the neural network model, the hyperbolic tangent function has an S-shaped curve, and the input range is mapped to [-1, 1], compared with the Sigmoid function, the output mean of the tanh function is 0, which can converge faster in training, and has stronger expression ability;

[0111] The weights are initialized with a uniform distribution, specifically, the uniform distribution initializes the weights as random numbers uniformly distributed in the range [-a, a], where the calculation method of a is usually related to the size of the network layer, and a common calculation formula is to determine the size of a according to the number of neurons of the current layer and the next layer, and the formula is expressed as , wherein and represent the number of neurons of the current layer and the next layer, respectively.

[0112] Referring to Figure 4 , step S3 comprises:

[0113] Step S31, configure the corresponding loss function and optimization algorithm for the neural network model;

[0114] Among them, the selection of the loss function needs to select an appropriate loss function before training the neural network model, which is used to measure the difference between the model prediction output and the actual target output, here the loss function adopts the mean square error algorithm, once the loss function is determined, the gradient of the loss function to the network parameters (weights and biases) is calculated by the back propagation algorithm, the gradient represents the change rate of the loss function at the current parameter value, and the gradient can be used to adjust the parameters to minimize the loss function value;

[0115] Optimization algorithm: gradient descent is one of the most commonly used optimization algorithms, which updates the network parameters according to the direction of the gradient, so that the loss function gradually decreases, and the standard gradient descent algorithm is adopted, which can accelerate the training process and improve the convergence performance;

[0116] Step S32, taking the sun azimuth vector, orbital parameters, and observation point latitude and longitude as input, and taking the polarization information of the target observation point as output, training the neural network model;

[0117] Among them, the training process of iterative training is an iterative process, which will continue until the predetermined training rounds or convergence conditions are reached, each round of training includes forward propagation to calculate the output and loss, backward propagation to calculate the gradient, and updating the parameters according to the optimization algorithm, which continuously optimizes the prediction ability of the neural network model;

[0118] Data preparation is a key step in model training, which mainly includes preparing and processing training data: solar azimuth vector, orbital parameters, and observation point latitude and longitude (input), polarization information of the target point (output);

[0119] Load the trained neural network model into the computing environment for subsequent data processing and prediction operations;

[0120] Step S33, input the to-be-calculated data into the trained neural network model for feature extraction and nonlinear calculation, to obtain the predicted polarization azimuth and the predicted degree of polarization after the light field is extended;

[0121] The preprocessed spatial polarization light field data, i.e., the to-be-calculated data, is input into the neural network model, and the prediction output of the neural network model is calculated through forward propagation. In this stage, the neural network model automatically extracts features and performs complex nonlinear calculation to obtain the distribution result after the light field is extended.

[0122] Referring to Figure 5 , step S4 includes:

[0123] Step S41, the position vector of the to-be-observed point, the first sun vector in the body coordinate system, and the current time date are obtained, and the Julian date of the day is obtained according to the current time date and the pre-set historical time date;

[0124] Here, the Julian date of the day at the time is calculated by the following formula:

[0125] ;

[0126] Wherein, is the current time date, is the historical time date, which is January 1, 1970;

[0127] Step S42, based on the Julian date, the first coordinate conversion matrix from the geocentric inertial coordinate system to the satellite orbit coordinate system is calculated, and the second coordinate conversion matrix from the satellite orbit coordinate system to the body coordinate system is calculated;

[0128] First, the first coordinate conversion matrix from the geocentric inertial coordinate system to the satellite orbit coordinate system is calculated, which can be obtained by three times of coordinate rotation along three axes:

[0129] ;

[0130] Wherein,

[0131] is the orbit inclination;

[0132] is the ascending node right ascension;

[0133] is the latitude depression angle, which is the sum of the true perihelion angle and the perihelion angle;

[0134] Then, the second coordinate conversion matrix from the satellite orbit coordinate system to the body coordinate system is calculated, which can be obtained by three times of coordinate rotation along three axes. Here, the rotation is performed in the order of 3-2-1:

[0135] ;

[0136] wherein are the roll angle, the pitch angle and the yaw attitude angle of the satellite body coordinate system relative to the satellite orbit coordinate system at the measurement time, respectively;

[0137] Step S43, the cross product of the position vector and the first sun vector is calculated to obtain the angle between the observation point and the sun relative to the observation point, and the measured polarization degree is obtained according to the angle;

[0138] The angle between the observation point P and the sun S relative to the observation point O, i.e. the sensor installed on the satellite, is The cross product of the first sun vector in the body coordinate system and the position vector of the observation point P can obtain , and further obtain the polarization degree of the scattered light at the observation point P, wherein , The sun vector in the geocentric inertial coordinate system can be solved by Julian day, which is not extended here;

[0139] The measured polarization degree is obtained by the following calculation formula:

[0140] ;

[0141] The maximum polarization degree is 1 in the ideal condition, but in fact the value will be affected by various factors such as weather conditions, generally less than 1, here the value is temporarily taken as 0.7;

[0142] Step S44, the satellite orbit coordinate system is selected as the observation coordinate system, the second sun vector in the satellite orbit coordinate system is obtained according to the first coordinate conversion matrix, and the position vector of the observation point is converted to the spherical coordinate system to calculate the measured polarization azimuth angle;

[0143] The satellite orbit coordinate system is selected as the observation coordinate system, and the second sun vector in the satellite orbit coordinate system is:

[0144] ;

[0145] The position vector of the observation point P is:

[0146] ;

[0147] Convert to spherical coordinate system:

[0148] ;

[0149] The measured polarization azimuth angle is:

[0150] ;

[0151] Step S45, obtaining the polarization azimuth angle error percentage according to the predicted polarization azimuth angle and the measured polarization azimuth angle, and obtaining the polarization degree error percentage according to the predicted polarization degree and the measured polarization degree, for checking the accuracy of the neural network model;

[0152] The error percentage of the predicted polarization degree relative to the measured polarization degree at the current time:

[0153] ;

[0154] Wherein, the predicted polarization degree is represented by , and the measured polarization degree is represented by .

[0155] The error percentage of the predicted polarization azimuth angle relative to the measured polarization azimuth angle at the current time:

[0156] ;

[0157] Wherein, the predicted polarization degree is represented by , and the measured polarization degree is represented by .

[0158] The application also provides an atmospheric out-of-layer polarized light field continuation model calculation system based on neural network learning, comprising:

[0159] A neural network module is configured to calculate the predicted polarization degree and the predicted polarization azimuth angle under the current condition;

[0160] A time algorithm module is configured to calculate the Julian day and the sidereal hour angle of the satellite;

[0161] A coordinate system conversion module is configured to convert the sun vector in the geocentric inertial coordinate system to the satellite orbit coordinate system, and convert the sun vector from the satellite orbit coordinate system to the body coordinate system;

[0162] A polarization information calculation module is configured to calculate the measured polarization degree and the measured polarization azimuth angle at the current time;

[0163] A model accuracy checking module is configured to calculate the error percentage of the predicted polarization azimuth angle and the predicted polarization degree relative to the measured polarization azimuth angle and the measured polarization degree at the current time.

[0164] In the application, all input information can be input through a system visual interface, as shown in Figure 6 , which can be customized by the user by selectively changing the input parameters, and the error percentage calculation display interface of the polarization azimuth angle and the polarization degree is as shown in Figure 7 .

[0165] In the description of the application, the description of the terms "one embodiment", "some embodiments", "in this embodiment", "specific example", or "some examples" and the like means that the specific features, mechanisms, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the application. In this specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, mechanisms, materials or characteristics described can be combined in any suitable manner in any one or more embodiments or examples. In addition, those skilled in the art can combine and combine the different embodiments or examples described in the specification and the features of the different embodiments or examples without contradiction.

[0166] The above merely describes the specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for calculating an atmospheric out-of-layer polarized light field continuation model based on neural network learning, characterized in that, The method comprises the following steps: Step S1, data acquisition: collecting spatial offset light field data and obtaining to-be-calculated data; Step S2, model construction: building an input layer, a hidden layer and an output layer to obtain a neural network model, and setting corresponding activation functions and initialization methods to adjust weights and biases of the neural network model; Step S3, model training: training the neural network model by taking the spatial offset light field data as training samples; after the training is completed, inputting the to-be-calculated data into the trained neural network model, and calculating a predicted polarization azimuth and a predicted polarization degree by forward propagation; Step S4, model verification: obtaining a position vector and a first sun vector of an observation point, obtaining a measured polarization azimuth and a measured polarization degree of the observation point according to the position vector and the first sun vector, obtaining a polarization azimuth error percentage according to the predicted polarization azimuth and the measured polarization azimuth, and obtaining a polarization degree error percentage according to the predicted polarization degree and the measured polarization degree, so as to verify the accuracy of the neural network model; The step S4 comprises: Step S41, obtaining the position vector of the observation point, the first sun vector in a body coordinate system and a current time date, and obtaining a Julian date of the day according to the current time date and a pre-set historical time date; Step S42, based on the Julian date, calculating a first coordinate conversion matrix converted from a geocentric inertial coordinate system to a satellite orbit coordinate system, and calculating a second coordinate conversion matrix converted from the satellite orbit coordinate system to the body coordinate system; Step S43, performing cross multiplication calculation on the position vector and the first sun vector to obtain an included angle of the observation point and the sun relative to the observation point, and obtaining the measured polarization degree according to the included angle; Step S44, selecting the satellite orbit coordinate system as an observation coordinate system, obtaining a second sun vector in the satellite orbit coordinate system according to the first coordinate conversion matrix, and converting the position vector of the observation point to a spherical coordinate system to calculate the measured polarization azimuth; Step S45, obtaining the polarization azimuth error percentage according to the predicted polarization azimuth and the measured polarization azimuth, and obtaining the polarization degree error percentage according to the predicted polarization degree and the measured polarization degree, so as to verify the accuracy of the neural network model.

2. The neural network learning-based model calculation method for the atmospheric out-of-layer polarized light field continuation according to claim 1, characterized in that, The step S2 comprises: Step S21, building the input layer, the hidden layer and the output layer to obtain the neural network model, and setting a hidden layer parameter to specify the number of neurons in the hidden layer; Step S22, the hyperbolic tangent function is selected as the activation function, and the weight of the neural network model is initialized in the range of random numbers uniformly distributed.

3. The neural network learning-based model calculation method for the atmospheric out-of-layer polarized light field continuation according to claim 1, characterized in that, The spatial offset light field data comprises a sun azimuth vector, orbit parameters, observation point longitude and latitude, and polarization information of an observation point, and the step S3 comprises: Step S31, configuring a corresponding loss function and an optimization algorithm for the neural network model; Step S32, taking the sun azimuth vector, the orbit parameters and the observation point longitude and latitude as input, and taking the polarization information of the observation point as output, and training the neural network model; Step S33, inputting the to-be-calculated data into the trained neural network model for feature extraction and nonlinear calculation to obtain the predicted polarization azimuth and the predicted polarization degree after light field continuation.

4. The neural network learning-based model calculation method for the atmospheric out-of-sphere polarized light field continuation according to claim 1, characterized in that, In step S41, the Julian date is obtained according to the following calculation formula: ； Wherein, represents the Julian day; represents the current time of day; represents the date of the historical moment.

5. The neural network learning-based model calculation method for the atmospheric out-of-sphere polarized light field continuation according to claim 1, characterized in that, In step S42, the first coordinate conversion matrix is obtained by the following calculation formula: ； Wherein, denotes the first coordinate transformation matrix; represents a coordinate rotation matrix for rotation about the z-axis; represents a coordinate rotation matrix for rotation about the z-axis; represents a coordinate rotation matrix for rotation along the z-axis; represents a coordinate rotation matrix for rotation about the axis denotes the orbital inclination; denotes the right ascension of the ascending node; represents the latitude depression angle.

6. The neural network learning-based model calculation method for the extraterrestrial polarized light field continuation according to claim 1, characterized in that, In step S42, the second coordinate conversion matrix is obtained by the following calculation formula: ； Wherein, denotes the second coordinate transformation matrix; represents a coordinate rotation matrix for rotation along the z-axis; represents a coordinate rotation matrix for rotation about the axis represents a coordinate rotation matrix for rotation about the axis a roll angle representing the body coordinate system relative to the satellite orbital coordinate system; denotes a pitch angle of the body coordinate system with respect to the satellite orbit coordinate system; denotes a yaw attitude angle of the body coordinate system with respect to the satellite orbit coordinate system.

7. The neural network learning-based model calculation method of the extraterrestrial polarized light field continuation according to claim 1, characterized in that, In step S43, the measured polarization degree is obtained by the following calculation formula: ； Wherein, represents the measured degree of polarization; denotes a preset maximum degree of polarization; denotes the included angle.

8. The neural network learning-based model calculation method for the extraterrestrial polarized light field continuation according to claim 1, characterized in that, In step S44, the measured polarization azimuth is obtained by the following calculation formula: ； Wherein, represents the measured polarization azimuth angle; representing the second sun vector in the satellite orbital coordinate system; representing the sun in the satellite orbit coordinate system coordinates; represents the coordinates of the sun in the satellite orbit coordinate system coordinates; represents the coordinates of the sun in the satellite orbit coordinate system coordinates; denotes the first coordinate transformation matrix; denotes a third sun vector in the geocentric inertial coordinate system; represents a position vector of the point to be observed; denotes the second coordinate transformation matrix; represents a position vector of the point to be observed in the body coordinate system; represents the coordinates of the point to be observed in the satellite orbit coordinate system coordinates; representing the point to be observed in the satellite orbit coordinate system coordinates; representing the point to be observed in the satellite orbit coordinate system coordinates; denotes the distance from the origin to the target point in the spherical coordinate system; represents a pitch angle of the body coordinate system with respect to the satellite orbital coordinate system; a roll angle representing the body coordinate system relative to the satellite orbital coordinate system; denotes an inverse function computation; representing the conversion to the spherical coordinate system coordinates; representing the conversion to the spherical coordinate system coordinates; representing the conversion to the spherical coordinate system coordinates; denotes the elevation angle of the sun vector in the spherical coordinate system; denotes the roll angle of the sun vector in the spherical coordinate system; denotes the elevation angle of the point to be observed in the spherical coordinate system; denotes the roll angle of the point to be observed in the spherical coordinate system.

9. The neural network learning-based model calculation method of the extraterrestrial polarized light field continuation according to claim 1, characterized in that, In step S45, the polarization azimuth error percentage and the polarization degree error percentage are obtained by the following calculation formula: ； Wherein, represents the percentage error in the degree of polarization; representing the predicted polarization degree; represents the measured degree of polarization; represents the percentage of the azimuth angle error of the polarization; denotes the predicted polarization azimuth angle; represents the measured polarization azimuth angle.

Images