Short-range single-station active target location method based on multi-modal deep neural network

Abstract

The present application relates to the technical field of ionosphere and deep learning cross application, and particularly relates to a short-wave single-station active target positioning method based on a multi-modal deep neural network. The present application uses ionospheric parameter data and meteorological parameter data related to the intensity of solar and geomagnetic activity to construct a regional annual ionospheric electron concentration grid file, and generates a large amount of simulation data corresponding to the relationship between the group path and the geographical position based on a three-dimensional ray tracing algorithm, and generates the corresponding electron concentration profile by using a bilinear interpolation method. A multi-modal single-station active target positioning deep neural network model based on dynamic MLP is established, which can realize the demand of long-distance target positioning based on an end-to-end network model, and perform high-precision active target positioning based on the target positioning deep neural network model, and the positioning error can be converged to about 40 km.

Description

Technical Field

[0001] The present invention relates to the technical field of cross-application of ionosphere and deep learning, and in particular to a shortwave single-station active target positioning method based on a multimodal deep neural network. Background Art

[0002] Currently, the primary means of detecting distant aerial targets is still over-the-horizon radar systems, which rely on shortwave signals reflected from the ionosphere. These systems exploit the ionosphere's reflection and refraction of shortwave signals, as well as backscatter, to detect distant targets.

[0003] Shortwave communication refers to the use of radio waves with a wavelength of 10 to 100 meters. Shortwave communication boasts advantages such as easy networking, strong survivability, long communication distances, low cost, and easy maintenance. It plays an important role in the field of radio communications, particularly in military, meteorological, transportation, navigation, and broadcasting applications. Traditional shortwave single-station active positioning technology relies primarily on ionospheric vertical measurement information and ionospheric models to locate distant maneuvering targets. First, an ionogram is acquired through real-time detection at a vertical measurement station. The International Reference Ionospheric Reflection (IRI) model is then used to construct a background field model. Finally, the virtual height of the equivalent reflection point is calculated to determine the positioning result.

[0004] However, in existing shortwave communication technology, since it is impossible to obtain real-time data on changes in the ionosphere when distant shortwave signals are transmitted through the ionosphere, the model finally constructed has a large deviation from the actual situation of the ionosphere, so the positioning accuracy of long-distance detection targets is low. Summary of the Invention

[0005] The present invention provides a shortwave single-station active target positioning method based on a multimodal deep neural network, which is used to solve the defects introduced in the prior art, such as many procedural errors and low positioning accuracy, and realize high-precision shortwave single-station active positioning.

[0006] The present invention provides a shortwave single-station active target positioning method based on a multimodal deep neural network, comprising the following steps:

[0007] Step 1: obtaining ionospheric parameter data;

[0008] Step 2: Calculate the meridional and zonal winds at each point in three-dimensional space using the horizontal wind field model, and combine them with the annual accumulated day, universal time, longitude, latitude, altitude, and ionospheric parameter data to form a data sample for electron concentration prediction;

[0009] Step 3: predicting the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0010] Step four, the ionosphere shortwave three-dimensional ray tracing algorithm is used to calculate the propagation group path of shortwave in the ionosphere in different transmission parameters at each station position in three-dimensional space;

[0011] Step five, determine the electron density profile corresponding to the propagation group path; according to the propagation group path and the electron density profile, obtain the data set for multi-modal model training;

[0012] Step six, a shortwave single-station active target positioning network based on dynamic MLP is constructed, which is used to obtain a shortwave single-station active target positioning result according to the trajectory information parameters of shortwave ray propagation and the electron density profile; the shortwave single-station active target positioning network is trained offline using the training set to obtain the shortwave single-station active target positioning network based on dynamic MLP;

[0013] Step seven, the shortwave single-station active target positioning network is used to obtain the shortwave single-station active target positioning result.

[0014] According to the shortwave single-station active target positioning method based on the multi-modal deep neural network provided by the application, the ionosphere parameters include solar wind, solar sunspot number, geomagnetic Kp index, geomagnetic Dst index, geomagnetic Ap index and solar radiation flux.

[0015] According to the shortwave single-station active target positioning method based on the multi-modal deep neural network provided by the application, the electron density profile corresponding to the propagation group path is determined, including the following steps:

[0016] For each station position, the profile is determined along the preset azimuth angle, and a plurality of electron density profiles are obtained by using the bilinear interpolation method;

[0017] The propagation path parameters of the propagation group path and the plurality of electron density profiles are paired to determine the electron density profile corresponding to the propagation group path.

[0018] According to the shortwave single-station active target positioning method based on the multi-modal deep neural network provided by the application, the shortwave single-station active target positioning network based on dynamic MLP includes a global image semantic feature extraction module, an additional information extraction module and a feature fusion module;

[0019] According to the trajectory information parameters of shortwave ray propagation and the electron density profile, the shortwave single-station active target positioning result is obtained, including the following steps:

[0020] The trajectory information parameters are input into the additional information extraction module to obtain the additional information features output by the additional information extraction module;

[0021] Inputting the electron concentration profile into a global image semantic feature extraction module to obtain a global image semantic feature output by the global image semantic feature extraction module;

[0022] Inputting the additional information features and the global image semantic features into a feature fusion module to obtain fusion features output by the feature fusion module;

[0023] According to the fusion features, a shortwave single-station active target positioning result is obtained.

[0024] According to the shortwave single-station active target positioning method based on multimodal deep neural network provided by the present invention, the shortwave single-station active target positioning network is subjected to offline supervised training using a training set to obtain a shortwave single-station active target positioning network based on dynamic MLP, specifically:

[0025] The shortwave single-station active target positioning network is trained offline with supervision using the training set. The root mean square error is used as the loss metric function to measure the error between the predicted target position and the actual simulated target position. The training is iterated until the loss function converges, and a shortwave single-station active target positioning network based on dynamic MLP is obtained.

[0026] The present invention also provides a shortwave single-station active target positioning device based on a multimodal deep neural network, comprising:

[0027] Ionospheric parameter acquisition module, used to obtain ionospheric parameter data;

[0028] The data sample module is used to calculate the meridional wind and zonal wind at each point in three-dimensional space using the horizontal wind field model, and to form a data sample for electron concentration prediction with the annual accumulated day, universal time, longitude, latitude, altitude and ionospheric parameter data;

[0029] An electron concentration prediction module is used to predict the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0030] The group path calculation module is used to calculate the shortwave propagation group path in the ionosphere at each station position in three-dimensional space with different transmission parameters using the ionospheric shortwave three-dimensional ray tracing algorithm;

[0031] A dataset construction module is used to determine the electron concentration profile corresponding to the propagation group path; based on the propagation group path and the electron concentration profile, a dataset for multimodal model training is obtained;

[0032] The target positioning network module is used to construct a shortwave single-station active target positioning network based on a dynamic MLP, wherein the shortwave single-station active target positioning network based on the dynamic MLP is used to obtain a shortwave single-station active target positioning result based on the trajectory information parameters and the electron concentration profile of the shortwave ray propagation; the shortwave single-station active target positioning network is trained offline with supervision using a training set to obtain a shortwave single-station active target positioning network based on the dynamic MLP;

[0033] The target positioning module is used to obtain the shortwave single-station active target positioning result using the shortwave single-station active target positioning network.

[0034] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and runnable on the processor. When the processor executes the program, it implements any of the above-described shortwave single-station active target positioning methods based on a multimodal deep neural network.

[0035] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the shortwave single-station active target positioning methods based on a multimodal deep neural network as described above.

[0036] The present invention also provides a computer program product, comprising a computer program, which, when executed by a processor, implements any of the above-described shortwave single-station active target positioning methods based on a multimodal deep neural network.

[0037] The shortwave single-station active target positioning method based on a multimodal deep neural network provided by the present invention utilizes ionospheric parameter data and meteorological parameter data that characterize the intensity of solar and geomagnetic activity to construct a regional annual ionospheric electron concentration grid file, and generates a large amount of simulation data of the correspondence between group paths and geographical locations based on a three-dimensional ray tracing algorithm, and generates the corresponding electron concentration profile using a bilinear interpolation method; a multimodal single-station active target positioning deep neural network model based on a dynamic MLP is established, which can meet the needs of long-distance target positioning based on an end-to-end network model, and perform high-precision active target positioning based on the target positioning deep neural network model, with the positioning error converging to about 40km. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the present invention or the prior art, a brief introduction is given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1It is a flow chart of a shortwave single-station active target positioning method based on a multimodal deep neural network provided by the present invention;

[0040] Figure 2 Schematic diagram of changes in solar wind speed, geomagnetic Kp index, sunspot number, geomagnetic Dst index, geomagnetic disturbance amplitude Ap, and solar radiation flux F10.7 during the period 2001-2022 according to an embodiment of the present invention;

[0041] Figure 3 It is a schematic diagram of a three-dimensional electron concentration grid file visualization display of the shortwave single-station active target positioning method based on a multimodal deep neural network provided by the present invention;

[0042] Figure 4 This is one of the simulation result schematic diagrams of step 4 of the shortwave single-station active target positioning method based on a multimodal deep neural network in an embodiment of the present invention;

[0043] Figure 5 This is a second schematic diagram of simulation results of step 4 of the shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention;

[0044] Figure 6 Schematic diagram of an electron concentration profile of a shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention;

[0045] Figure 7 Schematic diagram of an electron concentration profile for multimodal model input in a shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention;

[0046] Figure 8 1 is a schematic diagram of a shortwave single-station active target positioning network structure of a shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention;

[0047] Figure 9 is a curve showing the change in the loss function of the shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention as the number of iterations increases;

[0048] Figure 10 is a curve showing the change of the root mean square error on the test set of the shortwave single-station active target positioning method based on a multimodal deep neural network in an embodiment of the present invention as a inverse of the number of iterations;

[0049] Figure 11 It is a structural schematic diagram of a shortwave single-station active target positioning device based on a multimodal deep neural network provided by the present invention;

[0050] Figure 12 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0051] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0052] The following combination Figures 1-12 The present invention describes a shortwave single-station active target positioning method based on a multimodal deep neural network.

[0053] Figure 1 : is a flow chart of a shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention, such as Figure 1 As shown, it includes steps one to seven, specifically:

[0054] Step 1: obtaining ionospheric parameter data;

[0055] Step 2: Calculate the meridional and zonal winds at each point in three-dimensional space using the horizontal wind field model, and combine them with the annual accumulated day, universal time, longitude, latitude, altitude, and ionospheric parameter data to form a data sample for electron concentration prediction;

[0056] Step 3: predicting the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0057] Step 4: Using the ionospheric shortwave three-dimensional ray tracing algorithm, the propagation group path of the shortwave in the ionosphere at each station position in three-dimensional space with different transmission parameters is calculated;

[0058] Step 5: Determine the electron concentration profile corresponding to the propagation group path; obtain a data set for multimodal model training based on the propagation group path and the electron concentration profile;

[0059] Step 6: construct a shortwave single-station active target positioning network based on dynamic MLP, wherein the shortwave single-station active target positioning network based on dynamic MLP is used to obtain the shortwave single-station active target positioning result according to the trajectory information parameters and electron concentration profile of the shortwave ray propagation; the shortwave single-station active target positioning network is trained offline with supervision using the training set to obtain the shortwave single-station active target positioning network based on dynamic MLP;

[0060] Step 7: Use the shortwave single-station active target positioning network to obtain the shortwave single-station active target positioning result.

[0061] In step one of the embodiment of the present application, the ionospheric parameters include solar wind, sunspot number, geomagnetic Kp index, geomagnetic Dst index, geomagnetic Ap index, and solar radiation flux. Specifically, the ionospheric parameter data samples of the year 2022 are screened in chronological order from the relevant ionospheric website, and the solar wind, geomagnetic Kp index, sunspot number R, geomagnetic Dst index, geomagnetic Ap index, and solar radiation flux F10.7 parameters corresponding to the eight whole points between 9 and 16 o'clock every day are selected.

[0062] In step one of the embodiment of the present application, the relevant parameters representing the intensity of solar activity and global geomagnetic activity are downloaded from the OMNIWeb official website, Figure 2 The variation diagram of the solar wind speed, geomagnetic Kp index, sunspot number, geomagnetic Dst index, geomagnetic disturbance amplitude Ap, and solar radiation flux F10.7 during the period of 2001-2022 of the embodiment of the present application. In order to preliminarily verify the effectiveness of the algorithm, the embodiment of the present application selects the solar wind, geomagnetic Kp index, sunspot number R, geomagnetic Dst index, geomagnetic Ap index, and solar radiation flux F10.7 corresponding to the eight whole points between 9 and 16 o'clock every day from all the parameters in 2022, and selects 2920 groups of data samples;

[0063] In step two of the embodiment of the present application, the meridional wind and zonal wind at each point in the specified three-dimensional space are calculated using the horizontal wind field model HWM-14, and the annual day, universal time, longitude, latitude, height, and six ionospheric parameters in the first step are used to form a data sample. Specifically, the latest version of the horizontal wind field model HWM-14 used in the embodiment of the present application can provide the average horizontal wind of the Earth's atmosphere from the ground to a height of 500 km. The input of the model includes height, geomagnetic Kp index, annual day, universal time, longitude, latitude, and other parameters, and the output is the meridional wind and zonal wind at any point, which is combined with the annual day, universal time, longitude, latitude, height, solar wind, geomagnetic Kp index, sunspot number R, geomagnetic Dst index, geomagnetic Ap index, and solar radiation flux F10.7 to form a group of data samples, and finally 419,428,800 groups of sample sets containing various parameters are obtained;

[0064] In step three of the embodiment of the present invention, each set of parameters generated in step two is traversed and input into the three-dimensional electron concentration model of the ionosphere obtained through previous training, thus obtaining the electron concentration value at each point coordinate in the three-dimensional space. On this basis, the three-dimensional coordinates and electron concentration values ​​at each point can be processed to generate an electron concentration grid file in a specified format; the embodiment of the present invention uses the regional three-dimensional ionosphere electron concentration model to predict the data samples generated in the second step and generate electron concentration values ​​at corresponding positions, and stores them according to the electron concentration values ​​at longitude (Lng1~Lng2), latitude (Lat1~Lat2) and corresponding different altitudes (Lgt1~Lgt2km), to generate a three-dimensional electron concentration grid file that meets the input format of the three-dimensional ray tracing algorithm. The schematic diagram of the visualization of the three-dimensional electron concentration grid file is shown as follows. Figure 3 shown.

[0065] In step four of the embodiment of the present invention, in order to simulate the actual transmission process of shortwave rays in the ionosphere, the present invention first sets the parameters of the three-dimensional ray tracing algorithm according to the radar parameters (azimuth, pitch angle, transmission frequency, etc.), and then uses the three-dimensional ray tracing algorithm to calculate the correspondence between the group path value of the shortwave rays propagating in the electron concentration grid file and the geographical location of the transmitting station under the initial parameter conditions. The embodiment of the present invention uses a three-dimensional ray tracing algorithm to simulate the propagation of shortwave rays in the ionosphere. By changing the azimuth, pitch angle and transmission frequency, the correspondence between the group path of shortwave propagation and the geographical location under each set of parameters is calculated respectively. Figure 4 and Figure 5 Schematic diagrams of the changing relationship between the group path and the geographical location are given for two cases: changing the transmission frequency while keeping the azimuth and elevation angles unchanged, and changing the elevation angle while keeping the azimuth and transmission frequency unchanged. Finally, the present invention generated a total of 216,023 sets of simulation data by alternating any one of the variables.

[0066] In step 5 of the embodiment of the present invention, determining the electron concentration profile corresponding to the propagation group path includes the following steps:

[0067] For each station location, a profile is determined along a preset azimuth, and a plurality of electron concentration profiles are obtained using bilinear interpolation;

[0068] The propagation path parameters of the propagation group path are matched with a plurality of electron concentration profiles to determine the electron concentration profile corresponding to the propagation group path.

[0069] Due to factors such as the curvature of the earth's surface, solar activity, the influence of the geomagnetic field, and the change in electron density with altitude, the transmission of shortwave rays in the ionosphere is a very complex process. Therefore, the present invention considers introducing the electron concentration profile passed by shortwave rays during transmission into the shortwave single-station active target positioning model. Step 5 of the embodiment of the present invention first calculates the electron concentration profile passed by shortwave rays during propagation in the electron concentration file using the bilinear interpolation method along the specified direction within the range of azimuth angle α∈<α1,α2>, as shown in FIG. Figure 6 As shown, the electron concentration profile is saved as a pseudo-color image, as shown Figure 7 The colors in the image represent the electron concentration in the ionosphere; the darker the color, the greater the concentration. The group path information generated in step 4 is then paired with the electron concentration profile image generated in this step.

[0070] Specifically, step 5 of this embodiment of the present invention involves traversing the three-dimensional electron concentration grid file one by one, selecting station locations and rotating them along a specific azimuth at 2° intervals. During this rotation, bilinear interpolation is used to calculate the electron concentration values ​​at different locations on any cross-section and save them as pseudo-color images. By pairing the shortwave propagation path parameters and pseudo-color images based on file name and azimuth, 705,307 multimodal sample pairs are generated, resulting in a dataset for multimodal model training.

[0071] Figure 8 : is a schematic diagram of the shortwave single-station active target positioning network structure of the shortwave single-station active target positioning method based on a multimodal deep neural network according to an embodiment of the present invention, such as Figure 8 As shown, in step six of the embodiment of the present invention, the shortwave single-station active target positioning network based on dynamic MLP includes a global image semantic feature extraction module, an additional information extraction module and a feature fusion module;

[0072] According to the trajectory information parameters and electron concentration profile of shortwave ray propagation, the shortwave single-station active target positioning results are obtained, including the following steps:

[0073] Inputting the trajectory information parameters into the additional information extraction module to obtain additional information features output by the additional information extraction module;

[0074] Inputting the electron concentration profile into a global image semantic feature extraction module to obtain a global image semantic feature output by the global image semantic feature extraction module;

[0075] Inputting the additional information features and the global image semantic features into a feature fusion module to obtain fusion features output by the feature fusion module;

[0076] According to the fusion features, a shortwave single-station active target positioning result is obtained.

[0077] In an embodiment of the present invention, a multimodal single-station active target positioning deep neural network model based on dynamic MLP is constructed, comprising three parts: a global image semantic feature extraction module based on a deep residual network, an azimuth, frequency, and other additional information extraction module based on a nonlinear multilayer perceptron, and a dynamic MLP-based image feature and additional information feature fusion module. The global image semantic feature extraction module and the additional information extraction module constitute a backbone network, which, as the main structure of the model, respectively extracts feature maps from the original ionospheric electron concentration pseudo-color image and performs high-dimensional feature mapping of additional information. The dynamic MLP network fully utilizes the potential for mutual information complementation between image features and additional information features, fusing image features and additional information features before the classifier to generate better weights. Compared to a backbone network with a single input source, the fusion of multi-source information significantly improves the accuracy of shortwave single-station active positioning compared to a backbone network with a single input source.

[0078] Furthermore, the training set is used to perform offline supervised training on the shortwave single-station active target localization network, and a shortwave single-station active target localization network based on dynamic MLP is obtained, specifically:

[0079] The shortwave single-station active target positioning network is trained offline with supervision using the training set. The root mean square error is used as the loss metric function to measure the error between the predicted target position and the actual simulated target position. The training is iterated until the loss function converges, and a shortwave single-station active target positioning network based on dynamic MLP is obtained.

[0080] In order to quantitatively analyze the results of the shortwave single-station active target positioning method based on the multimodal deep neural network, the present invention refers to the root mean square error (RMSE), which is defined as follows:

[0081]

[0082] Where m represents the number of samples; y i represents the truth value; Represents the predicted value.

[0083] In this embodiment, after constructing a multimodal shortwave single-station active target positioning deep neural network, with the help of the multimodal dataset constructed in the present invention, 548,144 samples are randomly divided into training samples and 157,163 samples are tested, and offline supervised training of the deep neural network is performed. In the training phase, the number of network iterations Epoch is set to 300, the number of batch training samples is 128, the optimizer uses the stochastic gradient descent algorithm, the learning rate is 0.001, and the objective function is the root mean square error. The graphics card uses Nvidia 3090, and the development environment uses the PyTorch 1.7 deep learning framework. The curve of the loss function changing with the iteration round is shown as follows: Figure 9 As shown in the figure, the curve of the root mean square error on the test set changes with the number of iterations is as follows Figure 10 As shown in the figure, the loss function converges to a lower value, and the root mean square error fluctuates greatly in the first 100 rounds of iteration, and then converges to a lower level.

[0084] During the testing phase, the multimodal samples in the test set were input into the trained shortwave single-station active target positioning network model to verify the accuracy of the model in the target positioning task. It can be seen that the shortwave single-station active target positioning method according to the embodiment of the present invention achieved a test result on 157,163 samples in the test set: the RMSE converged to approximately 40 km.

[0085] Experimental results show that the shortwave single-station active target positioning method based on multimodal deep neural network proposed in this invention verifies the effectiveness of the end-to-end network model for shortwave single-station target positioning, avoids various errors introduced in the multi-stage shortwave single-station target positioning such as ionospheric background field construction and three-dimensional ray tracing in traditional methods, and provides a new technical approach for shortwave single-station active target positioning.

[0086] In summary, the shortwave single-station active target positioning method based on a multimodal deep neural network in an embodiment of the present invention utilizes ionospheric parameter data and meteorological parameter data that characterize the intensity of solar and geomagnetic activity to construct a regional annual ionospheric electron concentration grid file, and generates a large amount of simulation data of the correspondence between group paths and geographical locations based on a three-dimensional ray tracing algorithm, and generates corresponding electron concentration profiles using a bilinear interpolation method; a multimodal single-station active target positioning deep neural network model based on dynamic MLP is established, which can meet the needs of long-distance target positioning based on an end-to-end network model, and perform high-precision active target positioning based on the target positioning deep neural network model, and the positioning error can converge to about 40km.

[0087] The shortwave single-station active target positioning device based on a multimodal deep neural network provided by the present invention is described below. The shortwave single-station active target positioning device based on a multimodal deep neural network described below and the shortwave single-station active target positioning method based on a multimodal deep neural network described above can be referenced to each other.

[0088] Figure 11 This is a schematic diagram of the structure of the shortwave single-station active target positioning device based on the multimodal deep neural network provided by the present invention, such as Figure 11 Shown, including:

[0089] The ionospheric parameter acquisition module 1110 is used to obtain ionospheric parameter data;

[0090] The data sample module 1120 is used to calculate the meridional wind and zonal wind at each point in the three-dimensional space using the horizontal wind field model, and to form a data sample for electron concentration prediction with the annual accumulated day, universal time, longitude, latitude, altitude and ionospheric parameter data;

[0091] An electron concentration prediction module 1130 is configured to predict the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0092] The group path calculation module 1140 is used to calculate the shortwave propagation group path in the ionosphere at each station position in three-dimensional space with different transmission parameters using an ionospheric shortwave three-dimensional ray tracing algorithm;

[0093] The data set construction module 1150 is used to determine the electron concentration profile corresponding to the propagation group path; based on the propagation group path and the electron concentration profile, a data set for multimodal model training is obtained;

[0094] The target positioning network module 1160 is used to construct a shortwave single-station active target positioning network based on a dynamic MLP, wherein the shortwave single-station active target positioning network based on the dynamic MLP is used to obtain a shortwave single-station active target positioning result based on the trajectory information parameters and the electron concentration profile of the shortwave ray propagation; the shortwave single-station active target positioning network is trained offline with supervision using a training set to obtain a shortwave single-station active target positioning network based on the dynamic MLP;

[0095] The target positioning module 1170 is configured to obtain a shortwave single-station active target positioning result using a shortwave single-station active target positioning network.

[0096] It can be seen that the shortwave single-station active target positioning device based on a multimodal deep neural network in the embodiment of the present invention can use ionospheric parameter data and meteorological parameter data that characterize the intensity of solar and geomagnetic activity to construct a regional ionospheric electron concentration grid file throughout the year, and generate a large amount of simulation data of the correspondence between group paths and geographical locations based on a three-dimensional ray tracing algorithm, and generate corresponding electron concentration profiles using a bilinear interpolation method; a multimodal single-station active target positioning deep neural network model based on dynamic MLP is established, which can meet the needs of long-distance target positioning based on an end-to-end network model, and perform high-precision active target positioning based on a target positioning deep neural network model.

[0097] Figure 12 An example of a physical structure diagram of an electronic device is shown below. Figure 12 As shown, the electronic device may include: a processor 1210, a communication interface 820, a memory 1230 and a communication bus 840, wherein the processor 1210, the communication interface 1220, and the memory 830 communicate with each other via the communication bus 1240. The processor 1210 may call the logic instructions in the memory 1230 to execute a shortwave single-station active target positioning method based on a multimodal deep neural network, comprising the following steps:

[0098] Step 1: obtaining ionospheric parameter data;

[0099] Step 2: Calculate the meridional and zonal winds at each point in three-dimensional space using the horizontal wind field model, and combine them with the annual accumulated day, universal time, longitude, latitude, altitude, and ionospheric parameter data to form a data sample for electron concentration prediction;

[0100] Step 3: predicting the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0101] Step 4: Using the ionospheric shortwave three-dimensional ray tracing algorithm, the propagation group path of the shortwave in the ionosphere at each station position in three-dimensional space with different transmission parameters is calculated;

[0102] Step 5: Determine the electron concentration profile corresponding to the propagation group path; obtain a data set for multimodal model training based on the propagation group path and the electron concentration profile;

[0103] Step 6: construct a shortwave single-station active target positioning network based on dynamic MLP, wherein the shortwave single-station active target positioning network based on dynamic MLP is used to obtain the shortwave single-station active target positioning result according to the trajectory information parameters and electron concentration profile of the shortwave ray propagation; the shortwave single-station active target positioning network is trained offline with supervision using the training set to obtain the shortwave single-station active target positioning network based on dynamic MLP;

[0104] Step 7: Use the shortwave single-station active target positioning network to obtain the shortwave single-station active target positioning result.

[0105] In addition, the logic instructions in the above-mentioned memory 1230 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0106] On the other hand, the present invention also provides a computer program product, which includes a computer program. The computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the shortwave single-station active target positioning method based on a multimodal deep neural network provided by the above methods, including the following steps:

[0107] Step 1: obtaining ionospheric parameter data;

[0108] Step 2: Calculate the meridional and zonal winds at each point in three-dimensional space using the horizontal wind field model, and combine them with the annual accumulated day, universal time, longitude, latitude, altitude, and ionospheric parameter data to form a data sample for electron concentration prediction;

[0109] Step 3: predicting the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0110] Step 4: Using the ionospheric shortwave three-dimensional ray tracing algorithm, the propagation group path of the shortwave in the ionosphere at each station position in three-dimensional space with different transmission parameters is calculated;

[0111] Step 5: Determine the electron concentration profile corresponding to the propagation group path; obtain a data set for multimodal model training based on the propagation group path and the electron concentration profile;

[0112] Step 6: construct a shortwave single-station active target positioning network based on dynamic MLP, wherein the shortwave single-station active target positioning network based on dynamic MLP is used to obtain the shortwave single-station active target positioning result according to the trajectory information parameters and electron concentration profile of the shortwave ray propagation; the shortwave single-station active target positioning network is trained offline with supervision using the training set to obtain the shortwave single-station active target positioning network based on dynamic MLP;

[0113] Step 7: Use the shortwave single-station active target positioning network to obtain the shortwave single-station active target positioning result.

[0114] In another aspect, the present invention further provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the shortwave single-station active target positioning method based on a multimodal deep neural network provided by the above methods, comprising the following steps:

[0115] Step 1: obtaining ionospheric parameter data;

[0116] Step 2: Calculate the meridional and zonal winds at each point in three-dimensional space using the horizontal wind field model, and combine them with the annual accumulated day, universal time, longitude, latitude, altitude, and ionospheric parameter data to form a data sample for electron concentration prediction;

[0117] Step 3: predicting the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample;

[0118] Step 4: Using the ionospheric shortwave three-dimensional ray tracing algorithm, the propagation group path of the shortwave in the ionosphere at each station position in three-dimensional space with different transmission parameters is calculated;

[0119] Step 5: Determine the electron concentration profile corresponding to the propagation group path; obtain a data set for multimodal model training based on the propagation group path and the electron concentration profile;

[0120] Step 6: construct a shortwave single-station active target positioning network based on dynamic MLP, wherein the shortwave single-station active target positioning network based on dynamic MLP is used to obtain the shortwave single-station active target positioning result according to the trajectory information parameters and electron concentration profile of the shortwave ray propagation; the shortwave single-station active target positioning network is trained offline with supervision using the training set to obtain the shortwave single-station active target positioning network based on dynamic MLP;

[0121] Step seven, using a short monostatic active target positioning network to obtain a short monostatic active target positioning result.

[0122] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0123] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software and the necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0124] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A shortwave single-station active target positioning method based on multimodal deep neural network, characterized in that: The following steps are involved: Step 1: obtaining ionospheric parameter data; Step 2: Calculate the meridional and zonal winds at each point in three-dimensional space using the horizontal wind field model, and combine them with the annual accumulated day, universal time, longitude, latitude, altitude, and ionospheric parameter data to form a data sample for electron concentration prediction; Step 3: predicting the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample; Step 4: Using the ionospheric shortwave three-dimensional ray tracing algorithm, the propagation group path of the shortwave in the ionosphere at each station position in three-dimensional space with different transmission parameters is calculated; Step 5: determine the electron concentration profile corresponding to the propagation group path; According to the propagation group path and electron concentration profile, a data set for multimodal model training is obtained; Step 6: construct a shortwave single-station active target positioning network based on dynamic MLP, wherein the shortwave single-station active target positioning network based on dynamic MLP is used to obtain a shortwave single-station active target positioning result according to the trajectory information parameters and electron concentration profile of shortwave ray propagation; The shortwave single-station active target localization network is trained offline with supervision using the training set, and a shortwave single-station active target localization network based on dynamic MLP is obtained. Step 7: Use the shortwave single-station active target positioning network to obtain the shortwave single-station active target positioning result; The shortwave single-station active target positioning network based on dynamic MLP includes a global image semantic feature extraction module, an additional information extraction module and a feature fusion module; According to the trajectory information parameters and electron concentration profile of shortwave ray propagation, the shortwave single-station active target positioning results are obtained, including the following steps: Inputting the trajectory information parameters into the additional information extraction module to obtain additional information features output by the additional information extraction module; Inputting the electron concentration profile into a global image semantic feature extraction module to obtain a global image semantic feature output by the global image semantic feature extraction module; Inputting the additional information features and the global image semantic features into a feature fusion module to obtain fusion features output by the feature fusion module; According to the fusion features, a shortwave single-station active target positioning result is obtained; The training set is used to perform offline supervised training on the shortwave single-station active target localization network, and the shortwave single-station active target localization network based on dynamic MLP is obtained, specifically: The shortwave single-station active target positioning network is trained offline with supervision using the training set. The root mean square error is used as the loss metric function to measure the error between the predicted target position and the actual simulated target position. The training is iterated until the loss function converges, and a shortwave single-station active target positioning network based on dynamic MLP is obtained.

2. The shortwave single-station active target positioning method based on multimodal deep neural network according to claim 1 is characterized in that: The ionospheric parameters include solar wind, sunspot number, geomagnetic Kp index, geomagnetic Dst index, geomagnetic Ap index, and solar radiation flux.

3. The shortwave single-station active target positioning method based on multimodal deep neural network according to claim 1 is characterized in that: Determining the electron concentration profile corresponding to the propagation group path includes the following steps: For each station location, a profile is determined along a preset azimuth, and a plurality of electron concentration profiles are obtained using bilinear interpolation; The propagation path parameters of the propagation group path are matched with a plurality of electron concentration profiles to determine the electron concentration profile corresponding to the propagation group path.

4. A shortwave single-station active target positioning device based on a multimodal deep neural network, characterized in that: include: Ionospheric parameter acquisition module, used to obtain ionospheric parameter data; The data sample module is used to calculate the meridional wind and zonal wind at each point in three-dimensional space using the horizontal wind field model, and to form a data sample for electron concentration prediction with the annual accumulated day, universal time, longitude, latitude, altitude and ionospheric parameter data; An electron concentration prediction module is used to predict the electron concentration value at each point in the three-dimensional space using a regional three-dimensional ionospheric electron concentration model based on the data sample; The group path calculation module is used to calculate the shortwave propagation group path in the ionosphere at each station position in three-dimensional space with different transmission parameters using the ionospheric shortwave three-dimensional ray tracing algorithm; A dataset construction module for determining the electron concentration profile corresponding to the propagation group path; According to the propagation group path and electron concentration profile, a data set for multimodal model training is obtained; A target positioning network module is used to construct a shortwave single-station active target positioning network based on a dynamic MLP, wherein the shortwave single-station active target positioning network based on the dynamic MLP is used to obtain a shortwave single-station active target positioning result according to trajectory information parameters and electron concentration profiles of shortwave ray propagation; The shortwave single-station active target localization network is trained offline with supervision using the training set, and a shortwave single-station active target localization network based on dynamic MLP is obtained. A target positioning module is used to obtain a shortwave single-station active target positioning result using a shortwave single-station active target positioning network; The shortwave single-station active target positioning network based on dynamic MLP includes a global image semantic feature extraction module, an additional information extraction module and a feature fusion module; According to the trajectory information parameters and electron concentration profile of shortwave ray propagation, the shortwave single-station active target positioning results are obtained, including the following steps: Inputting the trajectory information parameters into the additional information extraction module to obtain additional information features output by the additional information extraction module; Inputting the electron concentration profile into a global image semantic feature extraction module to obtain a global image semantic feature output by the global image semantic feature extraction module; Inputting the additional information features and the global image semantic features into a feature fusion module to obtain fusion features output by the feature fusion module; According to the fusion features, a shortwave single-station active target positioning result is obtained; The training set is used to perform offline supervised training on the shortwave single-station active target localization network, and the shortwave single-station active target localization network based on dynamic MLP is obtained, specifically: The shortwave single-station active target positioning network is trained offline with supervision using the training set. The root mean square error is used as the loss metric function to measure the error between the predicted target position and the actual simulated target position. The training is iterated until the loss function converges, and a shortwave single-station active target positioning network based on dynamic MLP is obtained.

5. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, it implements the shortwave single-station active target positioning method based on a multimodal deep neural network as described in any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the shortwave single-station active target positioning method based on a multimodal deep neural network as described in any one of claims 1 to 3 is implemented.

7. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the shortwave single-station active target positioning method based on a multimodal deep neural network as described in any one of claims 1 to 3 is implemented.

Images