High-precision target inversion method based on space-borne platform

Abstract

The invention provides a high-precision target inversion method based on a satellite platform, and relates to the technical field of target inversion methods. The invention transmits radar signals of different wavelengths to an inversion target through a satellite-borne radar, obtains a continuous reflection signal of the target to be inverted, extracts reflection features through fast Fourier transform, calculates a direction angle average value, a downward angle difference value, and a height difference distribution value according to a coordinate position of the inversion target, a satellite-borne radar coordinate position and a quantity, forms a spatial feature of the inversion target, obtains environmental data, calculates an environmental impact coefficient by calculating a refraction value and an attenuation coefficient, judges the impact of environmental changes on radar waves, forms an inversion time series data set by combining reflection features, radar signal wavelengths, spatial features, and environmental impact coefficients with an inversion target type, obtains a training data set, establishes a target inversion model, collects a real-time inversion feature set as an input of the model, and obtains a real-time inversion result.

Description

Technical Field

[0001] The present invention relates to the technical field of target inversion methods, and in particular to a high-precision target inversion method based on a satellite-borne platform. Background Art

[0002] With the continuous development of remote sensing technology, satellite-borne radar systems have played an important role in target monitoring, reconnaissance and inversion applications. The satellite-borne platform can obtain the spatial position, morphology and dynamic characteristics of the target by transmitting radar signals of different wavelengths to the target and receiving the reflected signals. However, in practical applications, satellite-borne radar faces multiple technical difficulties when inverting targets. First, how to extract accurate target features from multi-dimensional and multi-band reflected signals, especially in complex environments, where the attenuation and propagation of signals are significantly affected by atmospheric conditions and ground environment, has become an urgent problem to be solved. Secondly, the high-precision target inversion of satellite-borne radar platforms places higher requirements on the positioning accuracy, wavelength selection and environmental changes of the radar. Therefore, how to effectively integrate the reflection characteristics, spatial characteristics, environmental data and other influencing factors of the target to establish an accurate target inversion model has become a key issue in the current field of remote sensing technology.

[0003] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0004] The purpose of the present invention is to provide a high-precision target inversion method based on a satellite platform to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solutions:

[0006] A high-precision target inversion method based on a satellite-borne platform, the specific steps include:

[0007] Step 1: Use the spaceborne radar to transmit radar signals of different wavelengths to different types of inversion targets to be inverted, obtain the continuous reflection signals of the targets to be inverted for preprocessing, extract the reflection features, and make a one-to-one correspondence between the radar signal wavelength, reflection features and the inverted target type;

[0008] Step 2: Obtain the running track of each satellite-borne radar and the coordinate position of the target to be inverted, calculate the direction angle, the downward angle, and the height difference between the satellite-borne radar and the inversion target during inversion, and calculate the direction angle average value, the downward angle difference value, and the height difference distribution value according to the number of satellite-borne radars to form the spatial characteristics of the inversion target;

[0009] Step 3: According to the coordinate position between each satellite-borne radar and the inversion target, obtain the real-time environmental data of each radar transmission signal path area, calculate the environmental impact coefficient, and determine the impact of environmental changes on radar waves;

[0010] Step 4: The reflection characteristics, radar signal wavelength, spatial characteristics and environmental impact coefficient are stamped with timestamps to form an inversion feature set, and the inversion time series data set formed by the inversion target type corresponding to the inversion is processed by maximum normalization to form a training data set;

[0011] Step 5: Establish the target inversion model, take the inversion feature set as the input of the model, and use the inversion time series data set as the output of the model to train the model, collect the real-time inversion feature set as the input of the model, and obtain the real-time inversion results.

[0012] Furthermore, the specific method of obtaining the continuous reflection signal of the target to be inverted and preprocessing the reflection feature is as follows: discretely processing the received reflection signal, processing the obtained discrete signal to filter out clutter, and extracting the reflection feature by fast Fourier transform:

[0013] The specific steps of performing discrete processing on the received reflected signal and obtaining the discrete signal for processing and filtering out clutter are as follows:

[0014] The interval time of the continuous reflection signal obtained Sampling and setting Interval time is the time window, and the mean filter is used for filtering. The calculation formula of the mean filter is: in, is the radar reflection signal after filtering, for Interval time The width of the time window composed of For Radar signals at all times, For The time window width is The original radar signal.

[0015] Furthermore, the method for extracting reflection features by fast Fourier transform is:

[0016] The interval time of the continuous reflection signal obtained Sampling and setting Interval time For the time window: in, is the frequency domain signal of the filtered reflected radar signal frequency components, The reflected radar signal samples, is the total number of samples, is an imaginary unit, Represents the frequency domain frequency components, .

[0017] Furthermore, the calculation formula for calculating the direction angle is: in, is the coordinate position of the spaceborne radar during inversion, is the coordinate position of the inversion target, is the relative coordinate position between the spaceborne radar and the inversion target during inversion, is the azimuth angle between the spaceborne radar and the inversion target during inversion;

[0018] The calculation steps for calculating the average value of the direction angle are:

[0019] According to the number of satellite-borne radars, the azimuth angle between each satellite-borne radar and the inversion target is calculated in turn, and the azimuth angles are sorted from small to large: in, is the maximum azimuth angle between the spaceborne radar and the inversion target during inversion, is the number of spaceborne radars involved in the inversion, The distance between the spaceborne radar and the inversion target during inversion. The direction angle of the Calculate the average value of the direction angle, , is a positive integer, .

[0020] Furthermore, the calculation formula for calculating the top-down angle is: in, is the distance between the spaceborne radar and the inversion target during inversion, It is the downward angle between the spaceborne radar and the inversion target during inversion;

[0021] The calculation formula for calculating the difference in the top-down angle is: in, is the difference in top-down angle, is the number of spaceborne radars involved in the inversion, For the The vertical viewing angle between a spaceborne radar and the inversion target during inversion, , is a positive integer, .

[0022] Furthermore, the calculation formula for calculating the height difference is: in, is the height difference between the spaceborne radar and the inversion target;

[0023] The calculation formula for calculating the height difference distribution value is: in, is the height difference distribution value, is the number of spaceborne radars involved in the inversion, For the The height difference between a spaceborne radar and the inversion target during inversion, , is a positive integer, , They are respectively the maximum and minimum values ​​of the height difference between the spaceborne radar and the inversion target during inversion.

[0024] Furthermore, the real-time environmental data includes temperature, humidity, air pressure and particle concentration;

[0025] The steps of obtaining the real-time environmental data of each radar transmission signal path area and calculating the environmental impact coefficient are as follows:

[0026] According to the temperature, humidity, air pressure and particle concentration, the refraction value and attenuation coefficient of the environmental data on the radar signal are calculated, and the environmental impact coefficient is calculated according to the refraction value and attenuation coefficient:

[0027] The calculation formulas for calculating the refraction value and attenuation coefficient of the radar signal by environmental data are: in, is the refraction value of the radar signal, is the air pressure constant, is the temperature constant, is the humidity constant, is the particle constant, is the real-time air pressure, is the real-time temperature, For real-time humidity, is the real-time particle concentration, is the attenuation coefficient of the radar signal, is the basic value of the attenuation coefficient, , are the attenuation factors related to humidity and particle concentration, respectively;

[0028] The calculation formula for calculating the environmental variation coefficient is: in, is the environmental variation coefficient.

[0029] Furthermore, the maximum normalization calculation formula is: in, is the training data set, is the largest data of each data type in the inverted time series dataset, is the smallest data of each data type in the inversion time series data set. For each data type in the inverted time series dataset data, is the number of data in each data type in the inverted time series dataset.

[0030] Furthermore, the target inversion model is based on a spatiotemporal convolutional neural network, including an input gate, a forget gate and an output gate:

[0031] The equation for the input gate is: in, is the current switch state of the input gate, represents the sigmoid function, is the weight of the input gate convolution kernel, represents the convolution operation, is the input gate bias parameter, is the output hidden state of the previous moment, is the convolution kernel weight of the output hidden state at the previous moment, is the inversion feature set in the training data set at the current moment;

[0032] The equation of the forget gate is: in, is the current switch state of the forget gate, is the weight of the forget gate convolution kernel, is the convolution kernel weight of the hidden state in the previous moment in the forget gate, represents the convolution operation, is the bias parameter of the forget gate;

[0033] Determine how much of the unit state at the previous moment needs to be retained to the current moment;

[0034] The update equation is: in, Candidate cell states, To represent the hyperbolic tangent function, is the candidate cell weight, is the candidate cell state bias parameter, is the convolution kernel weight of the hidden state at the previous moment in the current step;

[0035] in, Current cell state, is the cell state at the previous moment;

[0036] Determines the contribution of the previous moment's memory and the current candidate memory to the current unit state;

[0037] The output gate equation is: in, is the current switch state of the output gate, is the output gate weight, is the output gate bias parameter, is the convolution kernel weight of the hidden state at the previous moment in the current step, is the inversion time series dataset at the current moment;

[0038] Controls how much of the current unit state needs to be output to the current output value;

[0039] Among them, the mathematical expressions of the sigmoid function and the hyperbolic tangent function are: in, is the sigmoid function, is the input of the sigmoid function, is the hyperbolic tangent function, Input to the function.

[0040] Compared with the prior art, the beneficial effects of the present invention are as follows: a satellite-borne radar transmits radar signals of different wavelengths to an inversion target, obtains a continuous reflection signal of the target to be inverted, extracts reflection features through fast Fourier transform, calculates a direction angle average value, a downward angle difference value, and a height difference distribution value according to the coordinate position of the inversion target, the coordinate position of the satellite-borne radar, and the number, forms a spatial feature of the inversion target, obtains environmental data, calculates an environmental impact coefficient by calculating a refraction value and an attenuation coefficient, judges the impact of environmental changes on radar waves, forms an inversion time series data set by combining reflection features, radar signal wavelengths, spatial features, and environmental impact coefficients with the inversion target type, obtains a training data set, establishes a target inversion model, collects a real-time inversion feature set as an input to the model, and obtains a real-time inversion result;

[0041] According to the present invention, the spaceborne radar transmits signals of different wavelengths and uses fast Fourier transform to extract the target reflection characteristics, so that the spatial position and characteristics of the target can be accurately inverted. By calculating the data of multiple dimensions such as the spatial characteristics of the target and the environmental impact coefficient, the signal attenuation and environmental interference problems existing in the traditional method can be effectively overcome. This method not only enhances the accuracy of target inversion, but also automatically adjusts the calculation of the inversion model according to environmental changes, ensuring the efficiency and reliability of target inversion. In addition, the inversion feature set collected in real time can provide accurate data for target identification and dynamic monitoring, and provide strong technical support for the target identification and monitoring tasks of the spaceborne platform. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic diagram of the overall method flow of the present invention. DETAILED DESCRIPTION

[0043] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with specific embodiments.

[0044] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0045] Example

[0046] See also Figure 1 , the present invention provides a technical solution:

[0047] A high-precision target inversion method based on a satellite-borne platform, the specific steps include:

[0048] Step 1: Use the spaceborne radar to transmit radar signals of different wavelengths to different types of inversion targets to be inverted, obtain the continuous reflection signals of the targets to be inverted for preprocessing, extract the reflection features, and make a one-to-one correspondence between the radar signal wavelength, reflection features and the inverted target type;

[0049] When detecting and inverting targets through different wavelengths, multi-scale detection can be achieved. Radar signals of different wavelengths will produce different reflection characteristics when interacting with targets, especially in terms of the intensity, scattering mode and penetration ability of the reflected signal. Short-wavelength radar signals can usually provide higher spatial resolution and are used for target identification with rich details; while long-wavelength signals have strong penetration ability and can penetrate clouds, rain, snow and certain ground objects, and are suitable for detection of more complex environments or ground targets. Therefore, using multiple wavelength signals for inversion can achieve multi-scale and multi-angle detection and obtain target information more comprehensively. Radar signals of different wavelengths have different response characteristics for different types of inversion targets (such as ground objects, buildings, natural terrain, etc.). Using radar signals of multiple wavelengths can complement each other's shortcomings, thereby improving the accuracy and robustness of the inversion results. For example, for some targets with weak reflection intensity, long-wavelength signals can provide stronger reflection signals, supplement the shortcomings of short-wavelength signals, and improve the accuracy of detection.

[0050] Radar reflection signals usually contain noise caused by environmental factors, system errors, clutter or other unrelated sources. The mean filter smoothes the data by calculating the average value in the signal neighborhood, effectively suppressing random noise and reducing the impact of clutter. For radar signals, this processing can remove high-frequency noise components, making the true target characteristics of the reflected signal more obvious, thereby improving the inversion accuracy. The basic principle of the mean filter is to perform local averaging on the signal, which can effectively smooth the data while maintaining the overall trend and shape of the signal. In target inversion, the application of the mean filter helps to retain the global information of the target and avoid the loss of important reflection features due to noise or clutter, especially the main outline, size, shape and other features of the target.

[0051] In this embodiment, the specific method of obtaining the continuous reflection signal of the target to be inverted for preprocessing and extracting the reflection feature is: discretely processing the received reflection signal, processing the obtained discrete signal to filter out clutter, and extracting the reflection feature through fast Fourier transform:

[0052] The specific steps of performing discrete processing on the received reflected signal and obtaining the discrete signal for processing and filtering out clutter are as follows:

[0053] The interval time of the continuous reflection signal obtained Sampling and setting Interval time is the time window, and the mean filter is used for filtering. The calculation formula of the mean filter is: in, is the radar reflection signal after filtering, for Interval time The width of the time window composed of For Radar signals at all times, For The time window width is The original radar signal.

[0054] In this embodiment, the method for extracting reflection features by fast Fourier transform is:

[0055] The interval time of the continuous reflection signal obtained Sampling and setting Interval time For the time window: in, is the frequency domain signal of the filtered reflected radar signal frequency components, The reflected radar signal samples, is the total number of samples, is an imaginary unit, Represents the frequency domain frequency components, .

[0056] Fast Fourier transform can convert time domain signals into frequency domain signals, thereby revealing the spectral characteristics of the signal. For radar reflection signals, fast Fourier transform can extract the frequency components, amplitude characteristics and phase information of the reflection signals, and then extract the spatial characteristics of the target such as the shape, size, and orientation of the target. This information is difficult to obtain directly in the time domain, but after being converted into frequency domain signals through fast Fourier transform, the reflection characteristics of the target can be analyzed more intuitively and accurately. The frequency domain information processed by fast Fourier transform can provide clearer reflection characteristics, thereby helping to better identify and distinguish different types of targets. For example, the size, shape, surface roughness and other characteristics of the target will show different spectral characteristics in the frequency domain. Using fast Fourier transform to extract these characteristics can more effectively perform target classification and feature recognition.

[0057] Step 2: Obtain the running trajectory of each satellite-borne radar and the coordinate position of the target to be inverted, calculate the azimuth angle, downward angle, and height difference between the satellite-borne radar and the inversion target during inversion, and calculate the average azimuth angle, downward angle difference, and height difference distribution value based on the number of satellite-borne radars to form the spatial characteristics of the inversion target.

[0058] In this embodiment, the calculation formula for calculating the direction angle is: in, is the coordinate position of the spaceborne radar during inversion, is the coordinate position of the inversion target, is the relative coordinate position between the spaceborne radar and the inversion target during inversion, It is the direction angle between the spaceborne radar and the inversion target during inversion.

[0059] The calculation steps for calculating the average value of the direction angle are:

[0060] According to the number of satellite-borne radars, the azimuth angle between each satellite-borne radar and the inversion target is calculated in turn, and the azimuth angles are sorted from small to large: in, is the maximum azimuth angle between the spaceborne radar and the inversion target during inversion, is the number of spaceborne radars involved in the inversion, The distance between the spaceborne radar and the inversion target during inversion. The direction angle of the Calculate the average value of the direction angle, , is a positive integer, .

[0061] The azimuth angle refers to the azimuth between the spaceborne radar and the target, which defines the horizontal angle relationship between the radar and the target. The change of the azimuth angle mainly affects the spatial resolution of the target and the projection of the target on the horizontal plane.

[0062] When the azimuth angle distribution is wide, multiple observations of the target can be made from different directions, which helps capture the detailed geometric information of the target. The richer the geometric information in the inversion process, the higher the accuracy of the inversion result. If the azimuth angle distribution is uneven, some directions may lack sufficient observation angles, resulting in ambiguity or large errors in the inversion result.

[0063] The uniform value of the azimuth angle reflects that multiple satellite-borne radars are evenly distributed in the azimuth angles around the target, and multiple observations are made from different directions. When the azimuth angles of the radars are evenly distributed, the reflected signals about the target can be obtained from multiple directions, making the spatial information of the target more complete and avoiding information blind spots in certain directions. The uniform distribution of azimuth angles makes the geometric constraints obtained during target inversion more abundant and reliable, thereby effectively improving the inversion accuracy.

[0064] In this embodiment, the calculation formula for calculating the top-down angle is: in, is the distance between the spaceborne radar and the inversion target during inversion, It is the bird's-eye view between the spaceborne radar and the inversion target during inversion.

[0065] The calculation formula for calculating the difference in the top-down angle is: in, is the difference in top-down angle, is the number of spaceborne radars involved in the inversion, For the The vertical viewing angle between a spaceborne radar and the inversion target during inversion, , is a positive integer, .

[0066] The downward angle refers to the altitude angle of the spaceborne radar relative to the target, that is, the angle between the straight line from the radar to the target and the ground plane. The downward angle affects the radar's observation range and observation angle of the target.

[0067] When the downward angle is small, the distance between the radar and the target is far, the propagation attenuation of the signal is large, and the resolution of the acquired reflected signal is low, resulting in reduced inversion accuracy. When the downward angle is large, the distance between the radar and the target is close, the reflected signal is strong, the resolution is high, and the target features can be extracted more accurately, thereby improving the inversion accuracy.

[0068] Uniform distribution of the downward angle means that the downward angle of the spaceborne radar can be evenly distributed at different observation periods or positions, and observations can be made from different altitudes. The evenly distributed downward angle can prevent all radars from being at extreme high or low angles, thereby ensuring the balanced signal quality of each radar and avoiding excessive attenuation of the reflected signal at a single angle. The uniform distribution of different downward angles can provide more comprehensive target information and avoid the lack of effective observation of the target at a certain altitude, especially in terms of the local characteristics of the target.

[0069] In this embodiment, the calculation formula for calculating the height difference is: in, is the height difference between the spaceborne radar and the inversion target.

[0070] The calculation formula for calculating the height difference distribution value is: in, is the height difference distribution value, is the number of spaceborne radars involved in the inversion, For the The height difference between a spaceborne radar and the inversion target during inversion, , is a positive integer, , They are respectively the maximum and minimum values ​​of the height difference between the spaceborne radar and the inversion target during inversion.

[0071] The altitude difference refers to the height difference between the satellite platform and the inversion target, which affects the radar's line of sight, signal propagation path, and ground reflection characteristics.

[0072] When the height difference between the satellite platform and the target is small, the reflected signal of the target will be stronger and the signal attenuation will be less, so more detailed information can be obtained and the inversion accuracy can be improved. When the height difference between the platform and the target is large (that is, the target is farther away), the attenuation of signal propagation is more serious, and due to the viewing angle problem, it may result in lower signal quality and reduced inversion accuracy. Uniform distribution of height difference means that the height difference between the satellite platform and the target can be evenly distributed, that is, the height change of the platform will not be too drastic, avoiding observations that are too far or too close. Uniform distribution of height difference can ensure that the signal propagation path between the radar platform and the target is relatively stable, avoiding problems such as excessive signal attenuation or unclear signal reflection caused by extreme height difference.

[0073] Step 3: According to the coordinate position between each satellite-borne radar and the inversion target, obtain the real-time environmental data of each radar transmission signal path area, calculate the environmental impact coefficient, and determine the impact of environmental changes on radar waves.

[0074] When radar signals propagate in the atmosphere, the refractive index of the signal is affected by factors such as temperature, humidity, and air pressure. The refractive index of air is usually closely related to humidity. The higher the humidity, the higher the refractive index, and the propagation path of the radar wave may shift. Even a small change in the refractive index can cause the propagation angle of the radar signal to shift, thereby affecting the signal's arrival time, propagation path, and the strength of the received signal.

[0075] The propagation of radar waves in the atmosphere will be attenuated by factors such as atmospheric composition and weather conditions such as precipitation, clouds, and aerosols. In particular, humidity and precipitation will significantly increase the attenuation coefficient of radar waves. Attenuation causes the received signal strength to weaken, which may reduce the quality of the signal and affect the accuracy of target recognition.

[0076] The refraction and attenuation effects of the environment on radar signal propagation are factors that cannot be ignored. They directly affect the high-precision target inversion of the space-borne platform. By acquiring and processing environmental data in real time, calculating the refraction value and attenuation coefficient, and performing environmental compensation, the target inversion accuracy can be significantly improved, the error can be reduced, and the adaptability and data reliability of the system can be improved. These benefits make the high-precision target inversion method more effective and accurate in practical applications, especially in dynamically changing environments, and can better meet the needs of high-precision monitoring.

[0077] In this embodiment, the real-time environmental data includes temperature, humidity, air pressure and particle concentration;

[0078] The steps of obtaining the real-time environmental data of each radar transmission signal path area and calculating the environmental impact coefficient are as follows:

[0079] According to the temperature, humidity, air pressure and particle concentration, the refraction value and attenuation coefficient of the environmental data on the radar signal are calculated, and the environmental impact coefficient is calculated according to the refraction value and attenuation coefficient:

[0080] The calculation formulas for calculating the refraction value and attenuation coefficient of the radar signal by environmental data are: in, is the refraction value of the radar signal, is the air pressure constant, is the temperature constant, is the humidity constant, is the particle constant, is the real-time air pressure, is the real-time temperature, For real-time humidity, is the real-time particle concentration, is the attenuation coefficient of the radar signal, is the basic value of the attenuation coefficient, , are the attenuation factors related to humidity and particle concentration, respectively.

[0081] The calculation formula for calculating the environmental variation coefficient is: in, is the environmental variation coefficient.

[0082] Step 4: The reflection characteristics, radar signal wavelength, spatial characteristics and environmental impact coefficient are time-stamped to form an inversion feature set, and the inversion time series data set formed by the corresponding inversion target type during inversion is formed through maximum normalization processing to form a training data set.

[0083] In this embodiment, the maximum normalization calculation formula is: in, is the training data set, is the largest data of each data type in the inverted time series dataset, is the smallest data of each data type in the inversion time series data set. For each data type in the inverted time series dataset data, is the number of data in each data type in the inverted time series dataset.

[0084] Maximum normalization is a common preprocessing method, which is mainly used to scale the data range to the interval [0, 1]. Its core idea is to normalize the original data by the maximum value to ensure that each feature is compared and processed at the same scale. In the preprocessing stage, it plays a role in unifying the data scale, which can help the neural network train more efficiently and stably, and avoid performance problems caused by inconsistent feature scales. Especially in the target inversion task, ensuring the consistency of the scales of various features enables the model to better learn the impact of each feature on the target, which helps to improve the inversion precision and accuracy.

[0085] Step 5: Establish the target inversion model, take the inversion feature set as the input of the model, and use the inversion time series data set as the output of the model to train the model, collect the real-time inversion feature set as the input of the model, and obtain the real-time inversion results.

[0086] The spatiotemporal convolutional neural network combines the advantages of convolutional neural networks and long short-term memory networks, and is particularly suitable for processing spatiotemporal data. Traditional long short-term memory networks can capture long-term dependencies in time series, while convolutional layers are good at processing spatial features. Combining the two, the spatiotemporal convolutional neural network can simultaneously learn complex patterns in time series and spatial data, which is very suitable for target inversion tasks in space-borne platforms, especially in the inversion process of dynamic targets, which can not only process changes in the time dimension, but also capture spatial details.

[0087] The convolutional layer can effectively extract local features in spatial data, such as the local pattern of radar echoes. By applying convolutional neural networks, spatiotemporal convolutional neural networks can automatically extract spatial features in reflected signals, such as the shape, position, and contour of the target, without the need for manual feature extraction. This enables the network to accurately invert target types in complex environments.

[0088] The long short-term memory network is good at processing the long-term dependence of time series data. In the target inversion task, the state changes of the target, such as the movement and size change of the target, are time-continuous. Moreover, time has many effects on the high-precision target inversion based on the satellite platform, mainly reflected in seasonal changes, day and night differences, atmospheric conditions, sensor performance, selection of time series data, and dynamic changes of the target. In order to improve the accuracy and reliability of target inversion, the time series needs to be considered. The spatiotemporal convolutional neural network can effectively capture the evolution law of the target in the time dimension through its built-in memory mechanism, thereby providing more accurate predictions for target inversion.

[0089] By incorporating multiple factors such as reflection characteristics, radar signal wavelength, spatial characteristics and environmental impact coefficients into the target inversion model, the model can more comprehensively understand the various characteristics of the target, thereby improving the inversion accuracy and robustness. Using a convolutional long short-term memory neural network as an inversion model can effectively combine spatiotemporal information, extract local spatial features and capture dynamic changes in time series, so that the target inversion has higher accuracy and stronger robustness in complex spatiotemporal environments.

[0090] In this embodiment, the target inversion model is based on a spatiotemporal convolutional neural network, including an input gate, a forget gate, and an output gate:

[0091] The equation for the input gate is: in, is the current switch state of the input gate, represents the sigmoid function, is the weight of the input gate convolution kernel, represents the convolution operation, is the input gate bias parameter, is the output hidden state of the previous moment, is the convolution kernel weight of the output hidden state at the previous moment, is the inversion feature set in the training data set at the current moment;

[0092] The equation of the forget gate is: in, is the current switch state of the forget gate, is the weight of the forget gate convolution kernel, is the convolution kernel weight of the hidden state in the previous moment in the forget gate, represents the convolution operation, is the bias parameter of the forget gate;

[0093] Determine how much of the unit state at the previous moment needs to be retained to the current moment;

[0094] The update equation is: in, Candidate cell states, To represent the hyperbolic tangent function, is the candidate cell weight, is the candidate cell state bias parameter, is the convolution kernel weight of the hidden state at the previous moment in the current step;

[0095] in, Current cell state, is the cell state at the previous moment;

[0096] Determines the contribution of the previous moment's memory and the current candidate memory to the current unit state;

[0097] The output gate equation is: in, is the current switch state of the output gate, is the output gate weight, is the output gate bias parameter, is the convolution kernel weight of the hidden state at the previous moment in the current step, is the inversion time series dataset at the current moment;

[0098] Controls how much of the current unit state needs to be output to the current output value;

[0099] Among them, the mathematical expressions of the sigmoid function and the hyperbolic tangent function are: in, is the sigmoid function, is the input of the sigmoid function, is the hyperbolic tangent function, Input to the function.

[0100] The above formulas are all dimensionless and numerical calculations. The formula is a formula for the most recent real situation obtained by collecting a large amount of data and performing software simulation. The preset parameters in the formula are set by technicians in this field according to actual conditions.

[0101] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented by software, the above embodiments may be implemented in whole or in part in the form of a computer program product. Those skilled in the art may appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein may be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed by hardware or software methods depends on the specific application and design constraints of the technical solution.

[0102] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, and may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0103] The above description is only a specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application.

Claims

1. A high-precision target inversion method based on a satellite platform, characterized in that: The specific steps include: Step 1: Use the spaceborne radar to transmit radar signals of different wavelengths to different types of inversion targets to be inverted, obtain the continuous reflection signals of the targets to be inverted for preprocessing, extract the reflection features, and make a one-to-one correspondence between the radar signal wavelength, reflection features and the inverted target type; Step 2: Obtain the running track of each satellite-borne radar and the coordinate position of the target to be inverted, calculate the direction angle, the downward angle, and the height difference between the satellite-borne radar and the inversion target during inversion, and calculate the direction angle average value, the downward angle difference value, and the height difference distribution value according to the number of satellite-borne radars to form the spatial characteristics of the inversion target; Step 3: According to the coordinate position between each satellite-borne radar and the inversion target, obtain the real-time environmental data of each radar transmission signal path area, calculate the environmental impact coefficient, and determine the impact of environmental changes on radar waves; Step 4: The reflection characteristics, radar signal wavelength, spatial characteristics and environmental impact coefficient are stamped with timestamps to form an inversion feature set, and the inversion time series data set formed by the inversion target type corresponding to the inversion is processed by maximum normalization to form a training data set; Step 5: Establish the target inversion model, take the inversion feature set as the input of the model, and use the inversion time series data set as the output of the model to train the model, collect the real-time inversion feature set as the input of the model, and obtain the real-time inversion results.

2. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The specific method of obtaining the continuous reflection signal of the target to be inverted for preprocessing and extracting the reflection feature is: discretely processing the received reflection signal, processing the obtained discrete signal to filter out clutter, and extracting the reflection feature through fast Fourier transform: The specific steps of performing discrete processing on the received reflected signal and obtaining the discrete signal for processing and filtering out clutter are as follows: The interval time of the continuous reflection signal obtained Sampling and setting Interval time is the time window, and the mean filter is used for filtering. The calculation formula of the mean filter is: in, is the radar reflection signal after filtering, for Interval time The width of the window composed of For Radar signals at all times, For The time window width is The original radar signal.

3. The high-precision target inversion method based on a satellite platform according to claim 2, characterized in that: The method for extracting reflection features by fast Fourier transform is: The interval time of the continuous reflection signal obtained Sampling and setting Interval time is the time window; in, is the frequency domain signal of the filtered reflected radar signal frequency components, The reflected radar signal samples, is the total number of samples, is an imaginary unit, Represents the frequency domain frequency components, .

4. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The calculation formula for calculating the direction angle is: in, is the coordinate position of the spaceborne radar during inversion, is the coordinate position of the inversion target, is the relative coordinate position between the spaceborne radar and the inversion target during inversion, is the azimuth angle between the spaceborne radar and the inversion target during inversion; The calculation steps for calculating the average value of the direction angle are: According to the number of satellite-borne radars, the azimuth angle between each satellite-borne radar and the inversion target is calculated in turn, and the azimuth angles are sorted from small to large: in, is the maximum azimuth angle between the spaceborne radar and the inversion target during inversion, is the number of spaceborne radars involved in the inversion, The distance between the spaceborne radar and the inversion target during inversion. The size of the direction angle, Calculate the average value of the direction angle, , is a positive integer, .

5. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The formula for calculating the top-down angle is: in, is the distance between the spaceborne radar and the inversion target during inversion, It is the downward angle between the spaceborne radar and the inversion target during inversion; The calculation formula for calculating the difference in the top-down angle is: in, is the difference in top-down angle, is the number of spaceborne radars involved in the inversion, For the The vertical viewing angle between a spaceborne radar and the inversion target during inversion, , is a positive integer, .

6. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The formula for calculating the height difference is: in, is the height difference between the spaceborne radar and the inversion target; The calculation formula for calculating the height difference distribution value is: in, is the height difference distribution value, is the number of spaceborne radars involved in the inversion, For the The height difference between a spaceborne radar and the inversion target during inversion, , is a positive integer, , They are respectively the maximum and minimum values ​​of the height difference between the spaceborne radar and the inversion target during inversion.

7. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The real-time environmental data includes temperature, humidity, air pressure and particle concentration; The steps of obtaining the real-time environmental data of each radar transmission signal path area and calculating the environmental impact coefficient are as follows: According to the temperature, humidity, air pressure and particle concentration, the refraction value and attenuation coefficient of the environmental data on the radar signal are calculated, and the environmental impact coefficient is calculated according to the refraction value and attenuation coefficient: The calculation formulas for calculating the refraction value and attenuation coefficient of the radar signal by environmental data are: in, is the refraction value of the radar signal, is the air pressure constant, is the temperature constant, is the humidity constant, is the particle constant, is the real-time air pressure, is the real-time temperature, For real-time humidity, is the real-time particle concentration, is the attenuation coefficient of the radar signal, is the basic value of the attenuation coefficient, , are the attenuation factors related to humidity and particle concentration, respectively; The calculation formula for calculating the environmental variation coefficient is: in, is the environmental variation coefficient.

8. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The maximum normalization calculation formula is: in, is the training data set, is the largest data of each data type in the inverted time series dataset, is the smallest data of each data type in the inversion time series data set. For each data type in the inverted time series dataset data, is the number of data in each data type in the inverted time series dataset.

9. The high-precision target inversion method based on a satellite platform according to claim 1, characterized in that: The target inversion model is based on a spatiotemporal convolutional neural network, including an input gate, a forget gate, and an output gate: The equation for the input gate is: in, is the current switch state of the input gate, represents the sigmoid function, is the weight of the input gate convolution kernel, represents the convolution operation, is the input gate bias parameter, is the output hidden state of the previous moment, is the convolution kernel weight of the output hidden state at the previous moment, is the inversion feature set in the training data set at the current moment; The equation of the forget gate is: in, is the current switch state of the forget gate, is the weight of the forget gate convolution kernel, is the convolution kernel weight of the hidden state in the previous moment in the forget gate, represents the convolution operation, is the bias parameter of the forget gate; Determine how much of the unit state at the previous moment needs to be retained to the current moment; The update equation is: in, Candidate cell states, To represent the hyperbolic tangent function, is the candidate cell weight, is the candidate cell state bias parameter, is the convolution kernel weight of the hidden state at the previous moment in the current step; in, Current cell state, is the cell state at the previous moment; Determines the contribution of the previous moment's memory and the current candidate memory to the current unit state; The output gate equation is: in, is the current switch state of the output gate, is the output gate weight, is the output gate bias parameter, is the convolution kernel weight of the hidden state at the previous moment in the current step, is the inversion time series dataset at the current moment; Controls how much of the current unit state needs to be output to the current output value; Among them, the mathematical expressions of the sigmoid function and the hyperbolic tangent function are: in, is the sigmoid function, is the input of the sigmoid function, is the hyperbolic tangent function, Input to the function.

Images