A method for determining regional 3D atmospheric duct inversion results

Abstract

The present application provides a method for determining the inversion result of a regional 3D atmospheric waveguide, comprising: obtaining a radar PPI image database; obtaining a measured sea clutter PPI image; determining a radial 2D atmospheric waveguide inversion result based on the measured sea clutter PPI image; determining a target image database based on the radar PPI image database and the radial 2D atmospheric waveguide inversion result; and determining the inversion result of the regional 3D atmospheric waveguide based on the target image database and the measured sea clutter PPI image. The method solves the technical problems in the prior art that the inversion of a local three-dimensional atmospheric waveguide is limited in real time and the correlation of the waveguide structure between directions is not considered.

Description

Technical Field

[0001] The present application relates to the field of atmospheric waveguides, and in particular to atmospheric waveguide inversion. Background Art

[0002] Using radar sea clutter data to invert atmospheric duct is a highly nonlinear inverse problem. In recent years, great breakthroughs have been made in this field. For example, the physical inversion method has basically solved the real-time problem of inversion and the error problem of atmospheric duct parameterization model. However, in the process of practice, it is found that there are still some problems that need to be overcome, as described below:

[0003] The radar PPI image is a sector-scan echo map (a 360° full-range scan can be divided into several sector-shaped areas). To obtain the information of the local three-dimensional atmospheric waveguide, the radial waveguide structures in different azimuths can be inverted separately (i.e., N×2D mode, N represents the number of azimuth sampling). On the one hand, this will increase the inversion time and affect the real-time performance of the inversion; on the other hand, since the inversion between different azimuths is performed independently, the correlation of the waveguide structure between azimuths is not considered, which may cause the waveguide structure to be discontinuous between two adjacent azimuths. How to solve these two problems is the key to the commercial application of real-time inversion of regional three-dimensional atmospheric waveguides using radar sea clutter data. Summary of the invention

[0004] The present application provides a method for determining a regional 3D atmospheric waveguide inversion result, so as to solve the problems in the prior art of insufficient real-time performance of 3D three-dimensional atmospheric waveguide inversion and failure to consider the correlation of waveguide structures between directions.

[0005] To achieve the above objectives, the present application provides a method for determining a regional 3D atmospheric duct inversion result, comprising:

[0006] Obtain radar PPI image database;

[0007] Obtain measured sea clutter PPI images;

[0008] Determine a radial 2D atmospheric waveguide inversion result based on the measured sea clutter PPI image;

[0009] Determine a target image database based on the radar PPI image database and the radial 2D atmospheric waveguide inversion result;

[0010] Based on the target image database and the measured sea clutter PPI image, a 3D atmospheric waveduct inversion result of the region is determined.

[0011] Optionally, determining a radial 2D atmospheric waveduct inversion result based on the measured sea clutter PPI image includes:

[0012] Extracting a radial backscatter power profile based on the measured sea clutter PPI image;

[0013] Obtain a 2D atmospheric waveguide inversion model based on deep learning;

[0014] Based on the radial backscattered power profile and the deep learning-based 2D atmospheric waveduct inversion model, the radial 2D atmospheric waveduct inversion result is determined.

[0015] Optionally, the construction process of the 2D atmospheric waveguide inversion model based on deep learning is:

[0016] Based on the atmospheric duct parameters and sea surface attribute parameters, determine the sensitivity of environmental parameters and the coupling relationship between them;

[0017] Based on the sensitivity of the environmental parameters and the coupling relationship between them, a radial 2D environmental database is constructed;

[0018] Based on the radial 2D environment database, determining a radial sea clutter backscatter power profile database;

[0019] Based on the 2D environment database and the radial sea clutter backscatter power profile database, construct a radial 2D atmospheric waveguide inversion model;

[0020] Based on the radial 2D atmospheric waveguide inversion model, a 2D atmospheric waveguide inversion model based on deep learning is determined.

[0021] Optionally, the radar PPI image database is constructed as follows:

[0022] Based on the radial 2D environment database; constructing a local 3D environment database;

[0023] Based on the local 3D environment database, a radar PPI image database is determined.

[0024] Optionally, the step of determining the sensitivity of environmental parameters and the coupling relationship between them based on the atmospheric duct parameters and the sea surface attribute parameters is as follows:

[0025] The atmospheric duct parameters and sea surface attribute parameters are analyzed by using a full parameter sensitivity analysis method to obtain the sensitivity of the environmental parameters and the coupling relationship between them.

[0026] Optionally, constructing a radial 2D environment database based on the sensitivity of the environmental parameters and the coupling relationship between them comprises:

[0027] Based on the sensitivity of the environmental parameters and the coupling relationship between them, a first environmental sensitivity parameter is determined; wherein the first environmental sensitivity parameter is an environmental parameter whose sensitivity meets a first threshold condition, and the first threshold condition is used to judge the sensitivity of the environmental parameter;

[0028] Based on the first environmental sensitivity parameter, a radial 2D environmental database is constructed.

[0029] Optionally, it is characterized in that, based on the 2D environment database and the radial sea clutter backscatter power profile database, a radial 2D atmospheric waveguide inversion model based on deep learning is determined; specifically:

[0030] The 2D environment database and the radial sea clutter backscatter power profile database are divided into a training sample set and a test sample set according to a preset ratio, and the 2D atmospheric waveguide inversion model is constructed based on the training sample set;

[0031] Based on the radial 2D atmospheric waveguide inversion model, a 2D atmospheric waveguide inversion model based on deep learning is determined; specifically:

[0032] Based on the test sample set and the radial 2D atmospheric waveguide inversion model, the deep learning-based radial 2D atmospheric waveguide inversion model is determined.

[0033] Optionally, determining a target image database based on the radar PPI image database and the radial 2D atmospheric waveguide inversion result includes:

[0034] Based on the radial 2D atmospheric waveguide result, an image sample corresponding to the radial 2D atmospheric waveguide result is determined from the radar PPI image database, and the image sample is used as the target image database.

[0035] Optionally, determining the regional 3D atmospheric waveduct inversion result based on the target image database and the measured sea clutter PPI image includes:

[0036] Processing the measured sea clutter PPI image into a grayscale image with the same pixels as the target image database area to obtain a grayscale processed measured sea clutter PPI image;

[0037] Based on the grayscale-processed measured sea clutter PPI image and the target image database, a 3D atmospheric waveduct inversion result of the region is determined.

[0038] Optionally, determining the regional 3D atmospheric waveduct inversion result based on the measured sea clutter PPI image after the grayscale processing and the target image database includes:

[0039] Matching the grayscale values ​​of the measured sea clutter PPI image after the grayscale processing with the target image database;

[0040] Determine whether the error of the position gray value matching result meets a second threshold condition;

[0041] If so, a sample PPI image corresponding to the grayscale value of the position in the target image database is determined, and the environmental parameters corresponding to the sample PPI image are used as the 3D atmospheric waveguide inversion result of the region.

[0042] This application has the following beneficial effects:

[0043] In this application, the radial 2D atmospheric inversion result is determined by measuring the sea clutter PPI image, and then according to the 2D atmospheric inversion result, the target image database corresponding to the 2D atmospheric inversion result is determined from the radar PPI image database, and then the target image database is matched with the measured sea clutter PPI image to determine the regional 3D atmospheric waveguide inversion result. By constructing an inversion model and performing simple image matching calculations, real-time inversion can be achieved. The time consumption is mainly used in the construction of the database and the inversion network model. For the network model that has been constructed, the actual inversion calculation is instantaneous.

[0044] In addition to the above-described purposes, features and advantages, the present application also has other purposes, features and advantages. The present application will be further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0046] Figure 1 It is a flowchart of a method for determining the 3D atmospheric waveguide inversion result of the application area;

[0047] FIG2( a ) is a schematic diagram of a radar PPI image under normal refraction conditions when the elevation angle is 0° in this application;

[0048] FIG2( b ) is a schematic diagram of a radar PPI image under surface waveguide conditions when the elevation angle is 0° in this application;

[0049] Figure 3 It is an overall schematic diagram of the technical concept flow of the preferred embodiment of the present application;

[0050] Figure 4 It is a schematic diagram of a process for determining a 2D atmospheric waveguide inversion model based on deep learning in a preferred embodiment of the present application;

[0051] FIG5( a ) is an abstract conceptual diagram of a deep neural network of the present application;

[0052] Figure 5(b) is a schematic diagram of the construction process of the 2D atmospheric waveguide inversion network model based on deep learning in this application. DETAILED DESCRIPTION

[0053] The embodiments of the present application are described in detail below with reference to the accompanying drawings; however, the present application can be implemented in many different ways as defined and covered by the claims.

[0054] Three-dimensional atmospheric duct inversion is of great significance in many fields such as meteorology, communications, radar detection, military applications, and aerospace, and plays an indispensable role in improving the performance of related systems, ensuring safety, and promoting scientific research.

[0055] Radar PPI images provide data basis and features for three-dimensional atmospheric waveguide inversion. The inversion results can interpret images, correct distortions and evaluate radar performance. The two are closely related. Radar PPI images are sector-scan echo maps (360° full-range scanning can be divided into several sector-shaped areas). To obtain information about local three-dimensional atmospheric waveguides, radial waveguide structures in different azimuths can be inverted separately (i.e., N×2D mode, N represents the number of azimuth sampling). On the one hand, this will increase the inversion time and affect the real-time performance of the inversion; on the other hand, since the inversion between different azimuths is performed independently, the correlation of the waveguide structure between azimuths is not considered, which may cause discontinuity in the waveguide structure between two adjacent azimuths. How to solve these two problems is the key to the commercial application of real-time inversion of regional three-dimensional atmospheric waveguides using radar sea clutter data.

[0056] In order to solve the above problems, Figure 1 As shown, the present application provides a method for determining the results of regional 3D atmospheric waveguide inversion, and the specific steps are as follows:

[0057] Step 101: Obtain radar PPI image database;

[0058] Among them, the radar PPI (Plan Position Indicator) image database is a collection of radar plane position display images. The radar PPI image is an image that displays the radar scanning results in polar coordinates. It is centered on the radar station and presents the target information (such as the reflectivity and position of the target) detected by the radar at different azimuths and distances in a graphical way. This database is like an "image warehouse", and each PPI image in it contains the radar's detection of the target environment (such as meteorological targets, air vehicles, sea surface targets, etc.) at a specific time and within a specific scanning range. These images can be generated by different types of radars such as meteorological radars, navigation radars, and air defense radars.

[0059] In this step, after the local 3D environment database is constructed, the radar PPI image database, also known as the fan-shaped area three-dimensional space environment database, is substituted into the radar sea clutter backscatter power model in each direction, and combined to generate the corresponding grayscale image. For a given central azimuth environment parameter information, the size of the corresponding image database is immediately determined.

[0060] The specific process of building a 3D environment database mentioned here will be described in detail later and will not be described here for the time being.

[0061] Step S102: Acquire the measured sea clutter PPI image;

[0062] The measured sea clutter PPI (Plan Position Indicator) image refers to a sea clutter image obtained by actual measurement (usually using a radar system) and presented in a plane position display. Sea clutter refers to the radar echo signal caused by waves, roughness of the sea surface, foam on the sea surface and other sea surface related factors when the radar irradiates the sea surface. This PPI image is centered on the location of the radar and shows the intensity distribution of sea clutter in different azimuths and different distance ranges on a two-dimensional plane (in polar coordinate form, including azimuth and distance). Like a "map", it depicts the spatial distribution of sea clutter in the radar scanning area, and this information is directly obtained from the actual measurement scene. It should be noted that the area corresponding to the measured sea clutter PPI image in this application is determined by those skilled in the art based on actual applications, and no excessive restrictions are made here.

[0063] Due to air-sea interaction and turbulent transport, the structural characteristics of the atmospheric waveguide vary significantly over time and space, and the low-elevation radar PPI image reflects the instantaneous propagation medium information (Figure 2), which makes it possible to invert the real-time atmospheric waveguide information using the echo signal. Figure 2 shows the radar PPI image at an elevation angle of 0°. Figure 2 (a) is the normal refraction condition, and Figure 2 (b) is the surface waveguide condition.

[0064] It should be noted that the method for obtaining the measured sea clutter PPI image can be selected according to the actual situation on site, and this article does not make any specific restrictions on this.

[0065] Step S103: determining a radial 2D atmospheric waveguide inversion result based on the measured sea clutter PPI image;

[0066] The radial 2D atmospheric duct inversion result refers to the inference result about the atmospheric duct characteristics in two-dimensional space (generally including the vertical height direction and radial distance direction) obtained by processing and analyzing the relevant data along a certain radial direction (usually the ray direction centered on detection equipment such as radar).

[0067] In this step, after obtaining the measured sea clutter PPI image, it is processed accordingly to extract the central radial backscattering power profile, and the radial backscattering power profile is input into the 2D atmospheric waveguide inversion model based on deep learning, and the radial 2D atmospheric waveguide inversion result is output, for example, the central azimuth radial environmental parameter information. It should be noted that here and in the following text, for the convenience of understanding, the central azimuth radial environmental parameter information is used as an example, and the central radial of other azimuths can also be extracted. In practical applications, it is determined according to the field that needs to be inverted, and no further elaboration is given here.

[0068] Step S104: Determine a target image database based on the radar PPI image database and the radial 2D atmospheric waveguide inversion result;

[0069] In this step, after obtaining the radial 2D atmospheric waveguide inversion result, that is, the central azimuth radial environmental parameter information, M2 local three-dimensional environmental parameter radar PPI image samples corresponding to the central azimuth radial environmental parameter information are extracted from the radar PPI image database, and the radar PPI image samples here are used as the target image database for subsequent calculations.

[0070] It can be understood that since the radial 2D atmospheric waveguide inversion result here is obtained based on the image of the measured sea clutter PPI, after obtaining the radial 2D atmospheric waveguide inversion result, the image database corresponding to the 2D atmospheric waveguide inversion result is screened and determined from the radar PPI image database based on the 2D atmospheric waveguide inversion result obtained at this time, so as to use it as the target image database. Obviously, the target image database contains less content than the radar PPI image database, which obviously improves the efficiency and accuracy of the subsequent image matching.

[0071] Step S105: determining the regional 3D atmospheric waveguide inversion result based on the target image database and the measured sea clutter PPI image;

[0072] After obtaining the measured sea clutter PPI image, it is processed into a grayscale image with the same pixels as the target image database area. This is done to facilitate image matching, thereby determining the final 3D atmospheric waveguide inversion results. The regional 3D atmospheric waveguide inversion results refer to the characteristic description of the atmospheric waveguide in three-dimensional space (length, width, and height) obtained by various detection methods for a specific geographical area and using the inversion algorithm to obtain data. Atmospheric waveguides are a special meteorological phenomenon in which the refractive index distribution in the atmosphere restricts the propagation of electromagnetic waves (such as radar waves, radio waves, etc.) within a certain height range, forming a waveguide-like structure. The regional 3D atmospheric waveguide inversion results are a detailed presentation of this special structure in this specific area.

[0073] Specifically, after the measured sea clutter PPI image is gray-processed, the measured sea clutter PPI image after gray-processing is obtained, and then the measured sea clutter PPI image after gray-processing is matched with the target image database for position gray-scale value. An exhaustive method is used to determine whether the error of the position gray-scale value matching result meets the second threshold condition. If so, the sample PPI image corresponding to the position gray-scale value is determined in the target image database, and the environmental parameters corresponding to the sample PPI image are used as the local 3D atmospheric waveguide inversion result of the region. The second threshold condition mentioned here can be understood as the sample PPI image closest to the measured sea clutter PPI image after gray-scale processing. Since the matching calculation process is a simple linear calculation, even if the magnitude of the image data is too large, the global search can be completed in a short time, which also leads to the inversion result of the 3D atmospheric waveguide being real-time.

[0074] Combination Figure 3 The following contents are elaborated in detail: Figure 3 (in Chinese: EFAST stands for Extended Fourier Amplitude Sensitivity Test; RSCM stands for Radar Sea Clutter Backscatter Power Model; DNN stands for Deep Neural Network).

[0075] In one embodiment, based on the measured sea clutter PPI image, the radial 2D atmospheric waveguide inversion result is determined as follows:

[0076] After obtaining the measured sea clutter PPI image, the radial backscatter power profile is obtained through four steps: image preprocessing, target area positioning, pixel value extraction and conversion, and profile generation and fitting.

[0077] After obtaining the radial backscatter power profile, the pre-trained 2D atmospheric waveguide inversion model based on deep learning is obtained, and the radial backscatter power profile is input into the 2D atmospheric waveguide inversion model based on deep learning to obtain the radial 2D atmospheric waveguide inversion result.

[0078] It can be understood that here, after obtaining the measured sea clutter PPI image, the radial backscatter power profile is first extracted, and then the result of the radial backscatter power profile extraction is input into the pre-trained 2D atmospheric waveguide inversion model based on deep learning, so as to obtain the radial 2D atmospheric waveguide inversion result. This is done here to determine the image corresponding to the 2D environmental parameters of the measured sea clutter PPI image area in the radar PPI image database through the radial 2D inversion result, thereby reducing the workload of matching the subsequent target image database with the measured sea clutter PPI image after grayscale processing.

[0079] In one embodiment, Figure 4 As shown in the figure, the construction process of the 2D atmospheric waveguide inversion model based on deep learning is:

[0080] Step S401: Determine the sensitivity of environmental parameters and the coupling relationship between them based on atmospheric duct parameters and sea surface attribute parameters;

[0081] Massive observation data is the basis for the research on the construction of inversion models based on big data information mining. In daily observations, there are very few high-precision atmospheric duct data and radar sea clutter data with regional time and space matching, so the importance of how to obtain the database required for model learning is self-evident. Under the condition of insufficient observation data, physical model simulation can be used to supplement the database. In the simulation process, not only the influence of atmospheric duct should be considered, but also the influence of sea surface properties on electromagnetic wave propagation. Therefore, a complete environmental database should include possible combinations of atmospheric duct and sea surface property parameters as much as possible. For insensitive parameters (i.e., parameters that contribute little to the propagation of electromagnetic waves), they can be approximated as constants in the local range; for highly sensitive parameters (i.e., parameters that contribute greatly to the propagation of electromagnetic waves), it is necessary to consider their changes with distance and range.

[0082] The electromagnetic wave parabolic equation propagation model is currently the most advanced tropospheric radio wave propagation model. It can accurately simulate the interaction between electromagnetic waves and the atmospheric medium and the sea surface, which provides a basis for the full parameter sensitivity analysis of electromagnetic wave propagation in the sea-air boundary layer. The influence of sea surface properties on electromagnetic wave propagation is mainly manifested in three aspects: 1) the influence on the atmospheric waveguide structure; 2) the influence of sea surface roughness on the multipath effect of electromagnetic wave propagation; 3) the influence of sea surface dielectric properties on electromagnetic wave absorption and reflection. These effects occur simultaneously and affect each other. In this application, the input parameters of the model must include atmospheric waveguide parameters and sea surface attribute parameters, the latter mainly including sea surface wind speed, wind direction, sea surface temperature and salinity (the first two parameters are related to sea surface roughness, and the latter two parameters can determine the sea surface dielectric constant in combination with the electromagnetic wave frequency). The model output is the propagation factor of the directional pattern of the sea surface that varies with distance. Substituting it into the radar equation can obtain the radar sea clutter backscattering power.

[0083] It is understandable that, in this application, after obtaining the atmospheric duct parameters and the sea surface attribute parameters, a full parameter sensitivity analysis method is used to perform sensitivity analysis on environmental parameters such as atmospheric duct parameters and sea surface attribute parameters, and then the sensitivity degree of environmental parameters and the coupling relationship between them are obtained. Specifically, after analyzing the atmospheric duct environmental parameters and the sea surface parameters, taking into account the sensitivity degree of environmental parameters and the coupling relationship between them, the first environmental sensitivity parameter can be determined. The first environmental sensitivity parameter is a parameter whose sensitivity satisfies the first threshold condition, and the first threshold condition is used to judge the sensitivity of the environmental parameter. It should be noted that the first threshold condition can be determined according to the specific sea level in the actual display, and this document does not make specific restrictions on this. In addition, the first environmental sensitivity parameter in this application does not mean that only one sensitive environmental parameter is included. As long as the environmental parameter is greater than the first threshold, it can be regarded as an environmental parameter with high sensitivity, which can be regarded as the first environmental parameter. The environmental parameter less than the first threshold is regarded as a parameter with low sensitivity, which is generally regarded as a constant. In other words, in this application, there is no specific restriction on the number of environmental parameters included in the first environmental sensitivity parameter.

[0084] It should be noted that the extended Fourier amplitude sensitive test (EFAST) can be used in this application to conduct full-parameter sensitivity analysis research. In specific usage scenarios, the full-parameter sensitivity analysis research method corresponding to the scenario can be selected according to the specific applicable scenario. This article does not make specific restrictions here.

[0085] Step S402: constructing a radial 2D environment database based on the sensitivity of the environmental parameters and the coupling relationship between them;

[0086] Through full parameter sensitivity analysis, we find out the parameters with high sensitivity, whose spatial variation has a greater impact on electromagnetic propagation. These are the parameters that need to be given priority. For parameters with low sensitivity, it can be assumed that they are approximately constants within the local range, which meets the needs of actual applications. Because the purpose of obtaining atmospheric waveguide information is to predict the propagation characteristics of electromagnetic waves and the spatial distribution law of electromagnetic fields, and the small disturbances of parameters with low sensitivity within the local range have little effect on the propagation of electromagnetic waves, and can be regarded as noise. In other words, in step S201, according to the sensitivity of environmental parameters and the coupling relationship between each other, the first environmental sensitivity parameter can be determined, and based on the first environmental sensitivity parameter, a radial 2D environmental database can be constructed. The specific process is:

[0087] First, build the environmental database at the emission source. The first environmental sensitivity parameter is , is the number of parameters. The LHS (Latin Hypercube Sampling) method is used to sample each parameter. In its variable range Sampling is performed to generate a The size of the data sample set ,in The size is determined by the sampling interval divided by the LHS method. The LHS method can extract data with equal probability in each interval, which can effectively avoid the large number of repeated sampling in traditional methods and can well cover the range defined by the parameters. It is currently a better small sample sampling method.

[0088] Secondly, according to the above generated The size of the data sample set , to construct an environmental database (radial 2D environmental database) where the central position changes with distance. The actual meteorological and hydrological environmental parameters are continuous in space, and usually the values ​​will not change dramatically within a small scale. For the first sensitivity parameter , its change in distance can be described as a Markov chain process, that is, the distance The value at the distance Assuming the distance range (usually desirable , in km) parameter changes satisfy Gaussian random disturbance (i.e., the disturbance within 1 km has a variance of Normal distribution). The size of the data sample set All the first environmental sensitivity parameters in the perturbation with distance are carried out simultaneously, and the Monte Carlo method is used to run Second-rate( is on the order of 106), we can get a The size of the database is the radial 2D environment database. The principal component analysis is performed on the change of the parameters in the distance direction, and the weight coefficient corresponding to the main eigenvector is taken as the parameter that changes with the distance.

[0089] Step S403: determining a radial sea clutter backscatter power profile database based on the radial 2D environment database;

[0090] After the radial 2D environment database is constructed as described above, when the radial sea clutter is determined and the scattering power profile database is to be obtained, the radar pattern propagation factor needs to be calculated first. The radar pattern propagation factor can be calculated according to the following formula (1);

[0091] (1)

[0092] in: represents the complex electromagnetic field quantity; Indicates distance; Indicates height; Indicates distance The complex electromagnetic field quantity at Indicates distance The complex electromagnetic field quantity at ; Representing imaginary units ; represents the vacuum wave number; , represents the atmospheric corrected refractive index, Represents the radar pattern propagation factor.

[0093] It should be noted that after constructing the radial 2D environmental database, the first sensitive environmental parameter is determined. Afterwards, the calculation data corresponding to the above formula (1) can be determined. This determination process is a common technical means in this field, so a detailed description of this specific process is omitted.

[0094] After obtaining the radar pattern propagation factor, substitute it into the formula (2) of the radar sea clutter backscatter power model to obtain the radial sea clutter backscatter power profile database.

[0095] From the radar equation, we know that the backscatter power of sea clutter received by the radar is It can be expressed as a function of environmental parameters and radar system parameters:

[0096] (2)

[0097] in: Indicates distance; Indicates the radar transmission power; represents the antenna gain; Indicates the wavelength of electromagnetic waves; Indicates the antenna beam width; It represents the propagation speed of electromagnetic waves; represents the normalized sea surface reflectivity; Indicates the distance that electromagnetic waves reach The grazing angle at represents the radar pattern propagation factor. In the above formula, except , and Except for the variables related to the environment, the other variables are radar system parameters or constants. The change with distance can be calculated by using the surface wave spectrum estimation method; the normalized sea surface reflectivity It is related to factors such as the radar's transmission frequency, grazing angle, and sea surface roughness (or sea state). Figure 3 ); Radar pattern propagation factor The electromagnetic wave parabolic equation propagation mode can be used for accurate simulation calculation (here is the formula (1) mentioned above. The lower boundary condition of the electromagnetic wave parabolic equation propagation mode is determined by the medium properties of the sea surface and the roughness of the sea surface.

[0098] The advantages of deep learning-based radial two-dimensional atmospheric duct environmental parameter inversion are: 1) Real-time inversion is possible. The time consumption is mainly used in the construction of the database and inversion network model. For the already constructed network model, the actual inversion calculation is instantaneous. 2) In addition to obtaining radial atmospheric duct parameters, sea surface attribute parameters can also be obtained.

[0099] Step S404: constructing a radial 2D atmospheric waveguide inversion model based on deep learning based on the 2D environment database and the radial sea clutter backscatter power profile database;

[0100] Deep Neural Network (DNN) is the most basic network structure in deep learning. It is simple to train and has fast learning convergence speed, which is suitable for prediction problems. Based on the radial 2D environment database and the radial sea clutter backscatter power profile database, this application intends to use the DNN method to construct the network model required for radial two-dimensional atmospheric waveguide inversion, that is, the radial 2D inversion model based on deep learning mentioned above.

[0101] As shown in FIG5 , the construction process of the radial 2D inversion model based on deep learning is as follows: the radial 2D environment database and the radial sea clutter backscattering power profile database are constructed by calling the database 1 (referring to database 1 in FIG5 ), which is divided into two parts according to a preset ratio, namely, a training sample set and a test sample set. For the convenience of example, the test sample set occupies two-thirds of the ratio, and the remaining one-third is used as the test sample set. First, a network model required for inversion is obtained by learning the training sample set, and then the accuracy of the model is tested by the test sample set, and the corresponding DNN network hyperparameters are adjusted, and finally a high-precision inversion model suitable for this application is obtained. The key to constructing a high-precision inversion model through DNN is how to set the network hyperparameters, such as: network structure, activation function, loss function, learning rate, optimization algorithm, etc. In this application, the Tensorflow framework is proposed to be used. First, various types of hyperparameters are tried, and then the parameters are gradually adjusted in combination with the test results to find the optimal network hyperparameters that meet the requirements of this application.

[0102] It can be understood that the present application constructs a 2D atmospheric waveguide inversion model based on a training sample set of a preset proportion of database 1, and then continuously tests the accuracy of the constructed 2D atmospheric waveguide inversion model based on a test sample set in the remaining proportion of database 1. After adjusting to obtain an optimal network hyperparameter, the radial 2D atmospheric waveguide inversion model corresponding to the network hyperparameter can be used as the 2D atmospheric waveguide inversion model based on deep learning of the present application.

[0103] In one embodiment, the construction process of the radar PPI image database is as follows:

[0104] Step S501: constructing a local 3D environment database based on the radial 2D environment database;

[0105] Construct a three-dimensional space environment database for the fan-shaped area (i.e., a local 3D environment database). Based on the radial 2D environment database, a similar method is used to construct the radial 2D environment database. For the first environmental sensitivity parameter , its change in direction can also be described as a Markov chain process. Assuming the azimuth interval The variation of parameters within the range meets Gaussian random perturbation of The value can be determined based on the radar PPI azimuth resolution). The disturbance of the first environmental sensitivity parameter with the change of the azimuth is carried out simultaneously, and the Monte Carlo method is used to run Second-rate( The order of magnitude is 104), and the size of the local 3D environment database finally constructed is .

[0106] Step S502: Determine a radar PPI image database based on the local 3D environment database.

[0107] Since the calculation process of determining the radar PPI image database based on the local 3D environment database is the same as the calculation process of determining the radial sea clutter power profile database based on the radial 2D environment database, this article will not provide a detailed description here, but only briefly explain the specific process.

[0108] After the local 3D environment database is constructed, the 3D environment database is substituted into the radar sea clutter backscatter power model in each direction and combined to generate the corresponding grayscale image. For a given central azimuth environment parameter information, the size of the corresponding image database is .

[0109] It should be noted that the environmental parameter database constructed by the method in this application is universal, but different radar sea clutter databases will be obtained for different radar system parameters. Therefore, for radars with different working systems, it is necessary to recalculate the radar sea clutter database that matches the parameters.

[0110] In summary, the above technical solution of the present application takes into account the coupling of environmental information between different azimuths. For the inversion of local three-dimensional atmospheric waveguide information, the azimuth-by-azimuth radial (N×2D) inversion method cannot be used. The regional 3D atmospheric waveguide inversion result is determined by measuring the sea clutter PPI image and the target image database, thereby avoiding the discontinuity problem of the waveguide structure in adjacent azimuths and greatly improving the real-time performance of the inversion.

[0111] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for determining regional 3D atmospheric duct inversion results, characterized in that: include: Obtain radar PPI image database; Obtain measured sea clutter PPI images; Determining a radial 2D atmospheric waveguide inversion result based on the measured sea clutter PPI image; Determine a target image database based on the radar PPI image database and the radial 2D atmospheric waveguide inversion result; The target image database and the measured sea clutter PPI image are matched with position grayscale values ​​to determine the 3D atmospheric waveguide inversion result of the region.

2. The method according to claim 1, characterized in that Determining the radial 2D atmospheric waveguide inversion result based on the measured sea clutter PPI image includes: Extracting a radial backscatter power profile based on the measured sea clutter PPI image; Obtain a 2D atmospheric waveguide inversion model based on deep learning; Based on the radial backscattered power profile and the deep learning-based 2D atmospheric waveduct inversion model, the radial 2D atmospheric waveduct inversion result is determined.

3. The method according to claim 2, characterized in that: The construction process of the 2D atmospheric waveguide inversion model based on deep learning is as follows: Based on the atmospheric duct parameters and sea surface attribute parameters, determine the sensitivity of environmental parameters and the coupling relationship between them; Based on the sensitivity of the environmental parameters and the coupling relationship between them, a radial 2D environmental database is constructed; Based on the radial 2D environment database, determining a radial sea clutter backscatter power profile database; Based on the 2D environment database and the radial sea clutter backscatter power profile database, construct a radial 2D atmospheric waveguide inversion model; Based on the radial 2D atmospheric waveguide inversion model, a 2D atmospheric waveguide inversion model based on deep learning is determined.

4. The method according to claim 3, characterized in that The construction process of the radar PPI image database is as follows: Based on the radial 2D environment database; constructing a local 3D environment database; Based on the local 3D environment database, a radar PPI image database is determined.

5. The method according to claim 3, characterized in that: The determination of the sensitivity of environmental parameters and the coupling relationship between them based on the atmospheric duct parameters and the sea surface attribute parameters is specifically as follows: The atmospheric duct parameters and sea surface attribute parameters are analyzed by using a full parameter sensitivity analysis method to obtain the sensitivity of the environmental parameters and the coupling relationship between them.

6. The method according to claim 5, characterized in that The step of constructing a radial 2D environment database based on the sensitivity of the environmental parameters and the coupling relationship between the environmental parameters comprises: Based on the sensitivity of the environmental parameters and the coupling relationship between them, a first environmental sensitivity parameter is determined; wherein the first environmental sensitivity parameter is an environmental parameter whose sensitivity meets a first threshold condition, and the first threshold condition is used to judge the sensitivity of the environmental parameter; Based on the first environmental sensitivity parameter, a radial 2D environmental database is constructed.

7. The method according to claim 6, characterized in that The step of determining a radial 2D atmospheric waveguide inversion model based on deep learning based on the 2D environment database and the radial sea clutter backscatter power profile database is as follows: The 2D environment database and the radial sea clutter backscatter power profile database are divided into a training sample set and a test sample set according to a preset ratio, and the 2D atmospheric waveguide inversion model is constructed based on the training sample set; Based on the radial 2D atmospheric waveguide inversion model, a 2D atmospheric waveguide inversion model based on deep learning is determined; specifically: Based on the test sample set and the radial 2D atmospheric waveguide inversion model, the deep learning-based radial 2D atmospheric waveguide inversion model is determined.

8. The method according to claim 1, characterized in that The step of determining a target image database based on the radar PPI image database and the radial 2D atmospheric waveguide inversion result includes: Based on the radial 2D atmospheric waveguide result, an image sample corresponding to the radial 2D atmospheric waveguide result is determined from the radar PPI image database, and the image sample is used as the target image database.

9. The method according to claim 1, characterized in that: The target image database and the measured sea clutter PPI image are matched with position grayscale values ​​to determine the regional 3D atmospheric waveguide inversion result, including: Processing the measured sea clutter PPI image into a grayscale image with the same pixels as the target image database area to obtain a grayscale processed measured sea clutter PPI image; The grayscale processed measured sea clutter PPI image and the target image database are matched with each other in position grayscale values ​​to determine the 3D atmospheric waveguide inversion result of the region.

10. The method according to claim 9, characterized in that The grayscale processed measured sea clutter PPI image and the target image database are matched with position grayscale values ​​to determine the regional 3D atmospheric waveguide inversion result, including: Matching the grayscale values ​​of the measured sea clutter PPI image after the grayscale processing with the target image database; Determine whether the error of the position gray value matching result meets a second threshold condition; If so, a sample PPI image corresponding to the grayscale value of the position in the target image database is determined, and the environmental parameters corresponding to the sample PPI image are used as the 3D atmospheric waveguide inversion result of the region.

Images