A radar detection capability intelligent representation method and system based on a UNet network model

Abstract

The application discloses a radar detection capability intelligent representation method and system based on a UNet network model, and the method comprises the following steps: S1, acquiring the detection capability related attributes of a radar to be analyzed; S2, constructing a potential field three-dimensional grid coordinate system: constructing a three-dimensional grid coordinate system covering the detection range of the radar to be analyzed; S3, data preprocessing: preprocessing the attribute data acquired in the step S1 and the direction distance component data calculated in the step S2; S4, feature coding: data splicing and fusion, forming a comprehensive coding feature vector of each grid point; S5, inputting the model and calculating to obtain a detection probability value: inputting the generated 32-dimensional comprehensive coding feature vector into the trained 3D UNet network model, and calculating to obtain the radar detection probability value. The method can effectively represent the radar detection potential field, realize one-time efficient prediction of the detection probability of all potential field points, and is efficient, real-time and high in accuracy.

Description

Technical Field

[0001] This invention relates to the fields of network modeling and computation, and electromagnetic domain technology, specifically to a method and system for intelligent characterization of radar detection capabilities based on the UNet network model. Background Technology

[0002] With the rapid development of artificial intelligence and deep learning technologies, using deep learning network models for radar detection capability analysis has become an emerging method. One of the key indicators for radar detection capability analysis is the detection probability. Existing techniques typically calculate the detection probability based on radar equations and the Swerling model. However, when it is necessary to calculate the potential field points in the potential field, this method suffers from severe time-consuming inference; and using traditional neural networks or other deep models is also difficult to efficiently solve the timeliness challenge brought about by the prediction of massive potential field points.

[0003] In summary, the existing technology mainly has the following problems:

[0004] 1. Radar Detection Potential Field Characterization Problem: Existing detection probability calculations can only characterize the detection capability of a single point in the potential field, lacking an effective method for characterizing the entire detection potential field. This is crucial for subsequent combat system analysis (such as the fusion calculation of the detection potential fields of multiple radars), but currently, no relevant literature has adequately addressed this issue.

[0005] 2. Method Interpretability Issues: Deep learning models generally exhibit "black box" characteristics, lacking interpretability. Existing technologies struggle to simultaneously and effectively address both the overall representation of the radar detection potential field and the interpretability of the model. Summary of the Invention

[0006] To address the problems existing in the prior art, the present invention aims to provide an intelligent representation method for radar detection capability based on the UNet network model. This method can effectively represent the radar detection potential field and, utilizing the structural characteristics of the UNet model, achieve a one-time, efficient prediction of the detection probability of all potential field points, possessing technical advantages of high efficiency, strong real-time performance, and high accuracy. Another objective of the present invention is to provide an intelligent representation system for radar detection capability based on the UNet network model that implements the above method.

[0007] To achieve the above objectives, this invention provides an intelligent representation method for radar detection capabilities based on the UNet network model, the method comprising the following steps:

[0008] S1. Obtain the detection capability-related attributes of the radar to be analyzed: the attributes include radar attributes, force attributes, environmental attributes, and configuration item attributes;

[0009] S2. Construct a three-dimensional grid coordinate system for the potential field: Construct a three-dimensional grid coordinate system covering the detection range of the radar to be analyzed; calculate the distance components of each grid point in the three-dimensional grid coordinate system relative to the radar and the jammer on the X, Y, and Z coordinate axes;

[0010] S3. Data Preprocessing: Preprocess the attribute data obtained in step S1 and the directional distance component data calculated in step S2;

[0011] S4. Feature Encoding: The normalized and standardized distance data of each grid point calculated in step S3 are spliced ​​and fused with the radar, troop, environment and configuration attribute data preprocessed in step S2 to form a comprehensive encoded feature vector for each grid point.

[0012] S5. Input the model and calculate the detection probability value: The generated 32-dimensional comprehensive encoded feature vector The input is fed into the trained 3D UNet network model to calculate the radar detection probability value.

[0013] Furthermore, the radar attributes include radar transmit power, radar antenna gain, radar transmit frequency, pulse repetition frequency, receiver bandwidth, and noise figure.

[0014] Furthermore, force attributes include force cross-sectional area, jamming power, jammer bandwidth, and jamming frequency.

[0015] Furthermore, environmental attributes include ambient temperature, rainfall rate, and ionospheric disturbance index.

[0016] Furthermore, step S2 is implemented as follows:

[0017] First, a standardized 3D mesh coordinate system is constructed, including:

[0018] S2.1 Coordinate System Construction: Using the radar's geographical location as the origin of the coordinate system, longitude, latitude, and altitude are used... , , () indicates; construct a three-dimensional Cartesian grid coordinate system covering its detection range;

[0019] S2.2 Mesh Vertex Normalization and Alignment: Calculate the initial geographic coordinates of the bottom left vertex of the mesh;

[0020] S2.3 Grid point coordinate generation: Based on the standardized vertex coordinates and preset scale factor, generate the geometric coordinates of all 32×32×32 grid points in the 3D grid;

[0021] Then calculate the directional distance component:

[0022] S2.4 For radar: Set the radar's position coordinates in the three-dimensional grid coordinate system as ( , , ); Calculate each grid point ( , , Three-dimensional directional range components relative to the radar position:

[0023] S2.5 For the jammer: Set the position coordinates of the jammer in the three-dimensional grid coordinate system as ( , , ); Calculate each grid point ( , , The three-dimensional directional distance components relative to the jammer's position.

[0024] Furthermore, step S3 includes the following steps:

[0025] S3.1 Set normalization factor: Set the corresponding normalization factor for each attribute;

[0026] S3.2 Calculate the normalization ratio: Calculate the ratio of the true value of each attribute to its corresponding normalization factor:

[0027]

[0028] The resulting ratio is used as the encoded vector value of this attribute and is then used as input to the subsequent model.

[0029] Furthermore, in step S4, the directional distance component and attribute data of each grid point are first concatenated and fused to generate a 32-dimensional comprehensive encoded feature vector. Then, the feature vectors are normalized and standardized.

[0030] Furthermore, the specific formulas for normalization and standardization are as follows:

[0031] Normalization: Mapped to [0,1]

[0032] standardization: Mean 0, Standard deviation 1

[0033] in, , , These represent the encoded feature vectors respectively. The minimum, maximum, and mean values.

[0034] Furthermore, the creation and training process of the 3D UNet network model used in step S5 is as follows:

[0035] S110. Dataset Construction: For multiple different radar samples, repeat steps S1 to S4 to generate a set of encoded feature vectors for all grid points in the 3D grid of each radar sample; summarize the feature vector sets of all radar samples to construct the total dataset; divide the total dataset into a training set and a test set, where the training set is used for model training and the test set is used for model validation.

[0036] S120. Baseline Detection Probability Calculation: Based on the radar equation and the Swerling model, calculate the detection probability value of each grid point in the total dataset described in S110, as a baseline reference value;

[0037] S130. Model Construction and Training: Construct a 3D UNet network model; use the comprehensive encoded feature vector in the training set constructed in S110 as the model input, and use the baseline detection probability value of the corresponding grid point calculated in S120 as the supervision label. The loss function adopts L2 norm loss to train the 3D UNet network model.

[0038] S140. Model Validation: The integrated encoded feature vectors in the test set constructed in S110 are input into the 3D UNet network model trained in S130 to obtain the predicted detection probability values; the predicted values ​​are compared with the corresponding baseline detection probability values ​​calculated in S120 to verify the model's performance and generalization ability.

[0039] On the other hand, the present invention provides an intelligent representation system for radar detection capability based on the UNet network model, the system being used to implement the above-mentioned intelligent representation method for radar detection capability based on the UNet network model.

[0040] The beneficial effects of this invention are as follows:

[0041] 1. Providing an effective potential field representation method: This invention proposes to use a unified three-dimensional grid coordinate system to represent each point in the radar detection potential field, providing a basic support for the analysis of radar detection capabilities and the fusion calculation of multi-radar detection potential fields in subsequent combat systems.

[0042] 2. Achieving Efficient and Accurate Prediction: A method using the UNet model to predict the detection probability of all grid points in a potential field in a single operation is proposed. Compared to traditional neural networks or other deep models, the UNet model fully utilizes the strong spatial correlation between neighboring points in the potential field, significantly improving inference speed and prediction accuracy.

[0043] 3. Enhanced Model Interpretability: An innovative data-driven modeling approach is adopted, using the detection probability baseline values ​​calculated by the radar equation and the Swerling model as supervised information to train the deep learning model. This enables the model to approximate the results of traditional physical calculation methods, effectively improving the model's interpretability. Attached Figure Description

[0044] Figure 1 This is a schematic diagram of the 3D UNet network model architecture according to the present invention;

[0045] Figure 2 This is a schematic diagram of the grid coordinates according to the present invention;

[0046] Figure 3 This is a schematic diagram of the vector concatenation process;

[0047] Figure 4 This is a schematic diagram of the generated point cloud. Detailed Implementation

[0048] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] In the description of this invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0050] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0051] The following combination Figures 1-4 Specific embodiments of the present invention will be described in detail below. It should be understood that the specific embodiments described herein are for illustrative and explanatory purposes only and are not intended to limit the present invention.

[0052] This invention provides an intelligent representation method for radar detection capability based on the UNet network model, the method comprising the following steps:

[0053] S1. Obtain radar detection capability analysis related attributes: The attributes include radar attributes, troop attributes, environmental attributes, and configuration item attributes;

[0054] S2. Construct a three-dimensional grid coordinate system for the potential field: Construct a three-dimensional grid coordinate system covering the detection range of the radar to be analyzed; calculate the distance components of each grid point in the three-dimensional grid coordinate system relative to the radar and the jammer on the X, Y, and Z coordinate axes; the grid points are potential field points;

[0055] S3. Data preprocessing: Preprocess the attribute data obtained in step S1 and the directional distance component data calculated in step S2;

[0056] S4. Feature Encoding: The normalized / standardized distance data of each grid point calculated in step S3 is spliced ​​and fused with the radar, troop, environment and configuration attribute data preprocessed in step S2 to form a comprehensive encoded feature vector for each grid point;

[0057] S5. Input the model and calculate the detection probability value: The generated 32-dimensional comprehensive encoded feature vector The input is fed into the trained 3D UNet network model to calculate the radar detection probability value.

[0058] Specifically, the detailed technical content of each step is as follows:

[0059] S1. Obtain the detection capability-related attributes of the radar to be analyzed: These attributes include radar attributes, troop attributes, environmental attributes, and configuration item attributes; the radar to be analyzed is the opponent's radar. The attributes in the table below are known parameters obtained based on the radar model and jammer model, which can be directly obtained and used in subsequent steps. Radar parameters are obtained from data; environmental node parameters can be imported from meteorological stations and geographic information system databases.

[0060] In one specific embodiment, the detection capability-related attributes of the radar to be analyzed include four main categories: radar attributes, troop attributes, environmental attributes, and configuration item attributes, as shown in Table 1 below:

[0061] Table 1: Detection capability-related attributes of the radar to be analyzed

[0062]

[0063] S2. Construct a three-dimensional grid coordinate system for the potential field: Construct a three-dimensional grid coordinate system covering the detection range of the radar to be analyzed; calculate the distance components of each grid point (potential field point) in the three-dimensional grid coordinate system relative to the radar to be analyzed and the jammer on the X, Y, and Z coordinate axes; the jammer is our equipment, and there is one jammer. The jammer is used to interfere with the enemy radar and reduce the detection value of the potential field.

[0064] Specifically, the first step is to construct a standardized three-dimensional mesh coordinate system, and the specific steps are as follows:

[0065] S2.1 Coordinate System Construction: Based on radar geographic location (longitude lon, latitude lat, altitude h), using ( , , () represents the origin of the coordinate system, and a three-dimensional Cartesian grid coordinate system covering its detection range is constructed. Here, the scale factor corresponding to longitude is denoted by (). (0.2), the latitude scale factor corresponding to the Y-axis direction. (0.2), height scale factor corresponding to the Z-axis direction. (20000), with a grid resolution of 32×32×32.

[0066] S2.2 Mesh Vertex Normalization and Alignment: Calculate the initial geographic coordinates of the bottom left vertex of the mesh: longitude ,latitude Altitude And it will be standardized according to the following rules:

[0067] (1) To and Round to one decimal place and ensure that the decimal part is even (e.g., 0.0, 0.2, 0.4...).

[0068] (2) It should be greater than 0.

[0069] By setting the above rules, it is possible to ensure that the grids of different radars are aligned in the battlefield space, providing a unified spatial reference for the fusion of multi-radar detection potential fields at the operational system level. Standardized alignment is achieved by agreeing on "alignment". and The purpose of retaining one decimal place and ensuring that the decimal part is even (e.g., 0.0, 0.2, 0.4...) is to unify the potential field point positions obtained by each radar. This embodiment of the invention is for a single radar, but if used for multiple radars, the potential field point fusion between radars can be based on a unified spatial reference. For multiple radars, the potential field grid is divided with itself as the origin and a fixed radius. Based on "to..." and "Retaining one decimal place and ensuring that the decimal part is even (such as 0.0, 0.2, 0.4...)" enables the fusion of detection values ​​at the same location, i.e., potential field point.

[0070] Figure 2 The diagram shows a grid coordinate system, a two-dimensional plane with latitude and longitude. The jammer's position is the origin, and the white grid represents the radar potential field points to be analyzed. The grid size is (32, 32, 32), with latitude and longitude elevation scale intervals of 0.2, 0.2, and 20000, respectively. The black dots in the dimensional direction are the lower left vertices. The red dots are the points to be analyzed at certain locations within the radar potential field. The model needs to predict the detection probability values ​​for each potential field point within the grid.

[0071] S2.3 Grid Point Coordinate Generation: Based on the standardized vertex coordinates and preset scale factor, generate the geometric coordinates of all 32×32×32 grid points (potential field points) within the 3D grid:

[0072]

[0073] in, , These are the coordinate indices of the 3D mesh, specifically: the index corresponding to the x-direction is... The index corresponding to the y-direction is The index corresponding to the z-direction is .

[0074] Then calculate the directional distance component:

[0075] S2.4 For radar: Set the radar's position coordinates in the three-dimensional grid coordinate system as ( , , ). Calculate each grid point (potential point) ( , , Three-dimensional directional range components relative to the radar position:

[0076] (Distance in the X direction)

[0077] (Distance in the Y direction)

[0078] (Distance in the Z direction)

[0079] S2.5 For the jammer: Set the position coordinates of the jammer in the three-dimensional grid coordinate system as ( , , ). Calculate each grid point (potential point) ( , , The three-dimensional distance components relative to the jammer's position:

[0080] (Distance in the X direction)

[0081] (Distance in the Y direction)

[0082] (Distance in the Z direction)

[0083] Note: If the jammer is not present, all three-dimensional distance components are 0.

[0084] S3. Data Preprocessing: Preprocess the attribute data obtained in step S1 and the directional distance component data calculated in step S2;

[0085] Specifically, since the numerical ranges of the attributes described in step S1 (such as radar power, transmission frequency, etc.) and the directional range component data described in S2 differ significantly, a ratio calculation method based on a preset normalization factor is used for preprocessing to ensure rapid convergence of model training.

[0086] S3.1 Set Normalization Factor: Set the corresponding normalization factor (scale) for each attribute, as shown in Table 2 below:

[0087] Table 2: Normalization factors corresponding to each attribute setting

[0088]

[0089] S3.2 Calculate the normalization ratio: Calculate the ratio of the true value of each attribute to its corresponding normalization factor:

[0090]

[0091] The resulting ratio is used as the encoded vector value of this attribute and is then used as input for subsequent models.

[0092] Note: If the jammer does not exist, its jamming power, jamming frequency, and bandwidth are all set to 0.

[0093] S4. Feature Encoding: The normalized directional distance data of each grid point preprocessed in step S3 is concatenated and fused with the radar, troop, environment and configuration attribute data preprocessed in step S3 to form a comprehensive encoded feature vector for each grid point.

[0094] Specifically, the directional distance component (6-dimensional) and attribute vector (26-dimensional) of each grid point are first concatenated and fused to generate a 32-dimensional comprehensive encoded feature vector. Then, the feature vectors are normalized and standardized. The 26-dimensional attribute vector includes radar transmitter output power, radar transmitted signal center frequency, radar pulse repetition frequency, radar transmitted pulse duration, radar platform motion direction, antenna maximum radiation direction gain, elevation angle at maximum antenna gain, elevation angle at minimum antenna gain, receiver internal noise degradation, receiver intermediate frequency bandwidth, receiver amplification gain, false alarm probability in noisy background, minimum signal-to-noise ratio threshold for identifiable targets, transmit link loss compensation factor, receive link loss compensation factor, equivalent noise temperature of the receiving system, total transmission loss of the radar system, target radar cross-section, jammer transmit power, jamming signal center frequency, jamming signal bandwidth, atmospheric ambient temperature, precipitation intensity, ionospheric electron density disturbance level, target fluctuation model type, and number of coherent accumulation pulses.

[0095] Figure 3 The vector concatenation process is illustrated, where the directional distance component includes three potential field distance vectors and three jammer distance vectors. The potential field distance vectors, jammer distance vectors, and 26 attribute vectors are sequentially concatenated together to form a 32-dimensional comprehensive encoded feature vector.

[0096] Normalization: (Mapped to [0,1])

[0097] standardization: (Mean 0, Standard Deviation 1)

[0098] in, , , These represent the 32-dimensional comprehensive coding feature vectors. The minimum, maximum, and mean values. The normalized vector, This is the standardized vector.

[0099] S5. Input the model and calculate the detection probability value: The generated 32-dimensional comprehensive encoded feature vector The data is input into a trained 3D UNet network model to calculate the radar detection probability value. The results of data preprocessing and encoding in steps S1-S4 are used as input to the model. The prediction result is obtained through the UNet model. During training, the mean squared error is calculated between the prediction result and the label in S120. The training is iterated continuously to reduce this error loss.

[0100] Specifically, the creation and training process of the 3D UNet network model used in step S5 is as follows:

[0101] S110. Dataset Construction: For multiple different radar samples, repeat steps S1 to S4 to generate a set of encoded feature vectors for all grid points in the 3D grid of each radar sample; summarize the feature vector sets of all radar samples to construct the total dataset; divide the total dataset into a training set and a test set, where the training set is used for model training and the test set is used for model validation; the table shown in step 1 is a sample of radar parameters. Different types of radars have different parameters, corresponding to different radar samples, which are used to input the model for training.

[0102] Specifically, the dataset (100 samples in total) is divided into a training set and a test set. The training set contains 80 radars, and the test set contains 20 radars. The training set is used for parameter training of the 3D UNet model, and the test set is used for model performance verification and generalization ability evaluation. For different radar samples, steps S1 to S4 are repeated to generate a set of encoded feature vectors for all grid points in the 3D grid for each radar sample.

[0103] S120. Baseline Detection Probability Calculation: Based on the radar equation and the Swerling model, calculate the detection probability value of each grid point in the total dataset described in S110 (i.e., each grid point corresponding to each radar sample), as a baseline reference value.

[0104] Specifically, the radar signal-to-interference-plus-noise ratio (SNR) is first calculated using the extended radar equation, as shown in the following formula:

[0105]

[0106] in, Representing the radar power equation, Representing the interference power equation, This represents the environmental degradation equation. Indicates radar transmit power. Indicates radar antenna gain. Indicates the area of ​​troop strength. Indicates the radar transmission frequency. Indicates the pulse repetition frequency. This indicates the distance between the grid points and the target radar. Indicates system losses; Let represent the Boltzmann constant, denoted as 1.38e-23; Indicates ambient temperature. Indicates receiver bandwidth. Represents the noise figure. Indicates interference power. Indicates the bandwidth of the jammer. Indicates the interference frequency. Indicates rainfall rate, This represents the ionospheric disturbance index.

[0107] Then, the detection probability is calculated using Swerling Model I. The formula is as follows:

[0108]

[0109] in, This represents the minimum detectable signal-to-noise ratio.

[0110] Note: The parameters substituted into the formulas above are from the true values ​​of the relevant attributes in S1 and S2, not the preprocessed values ​​in S3. This step uses the results calculated from the true attribute values ​​as the model's labels, while step 3, preprocessing, is the model's input.

[0111] S130 Model Construction and Training: Construct a 3D UNet network model; use the comprehensive encoded feature vector in the training set constructed in S110 as the model input, and use the baseline detection probability value of the corresponding grid point calculated in S120 as the supervision label to train the 3D UNet network model;

[0112] Specifically, the constructed 3D Unet model is as follows: Figure 1 As shown, each box represents a feature vector obtained after performing operations such as concatenation and convolution. The arrows indicate the data flow direction. 3D UNet has one encoding path and one decoding path, each with 4 resolution levels. Each layer in the encoding path contains two 3×3×3 convolutions, each followed by a ReLU regularization layer, and then a 2×2×2 max pooling layer (downsampling) with a stride of 2 in each direction. In the decoding path, each layer contains a 2×2×2 deconvolution layer (upsampling) with a stride of 2, followed by two 3×3×3 convolutions, each followed by a block normalization layer before ReLU to provide the original high-resolution features. The last layer is a 1×1×1 convolutional layer, which can reduce the number of output channels. The final number of output channels is the predicted value.

[0113] The network input is a 32×32×32 pixel set with 32 channels. The output size is 32×32×32×1, where 1 represents the number of channels.

[0114] The 3D UNet network model is trained using the comprehensive encoded feature vectors in the training set constructed by S110 as model inputs and the baseline detection probability values ​​of the corresponding grid points calculated by S120 as supervision labels. The loss function is L2 norm loss (mean squared error).

[0115] S140. Model Validation: The integrated encoded feature vectors from the test set constructed in S110 are input into the 3D UNet network model trained in S130 to obtain the predicted detection probability values. The predicted values ​​are compared with the corresponding baseline detection probability values ​​calculated in S120 to verify the model's performance and generalization ability. If the model output deviates from the radar equation result within a set threshold, the radar detection capability is considered normal. The closer the model output result is to the radar equation result, the better the radar detection capability.

[0116] The model outputs the predicted detection probability value for each grid cell. Based on the predicted detection probability value and its corresponding coordinates, a cloud map is generated and output, which is then displayed on a display device. The cloud map is as follows: Figure 4 As shown.

[0117] This invention also provides an intelligent representation system for radar detection capabilities based on the UNet network model. The system includes a jammer, a data processing device, and a display device, used to intelligently represent the detection capabilities of local radars. The data processing device, acting as the carrier of the UNet model, calculates and displays the output results on the display device, presenting the results as a point cloud map. The intelligent representation uses a deep learning model to replace traditional radar equation calculations, and the model's output closely approximates the calculated results of the radar equations. The UNet model can output tens of thousands of potential field points at once, offering advantages over traditional methods in terms of speed and accuracy.

[0118] The model output is the detection probability value of all grid points. The output is multiple values. The radar's detection capability is compared with the result (label) calculated according to the S120 radar equation. If it is close to the result of the radar equation, it means that the radar model is trained better and the predicted detection probability value is more accurate.

[0119] Technical advantages of the present invention:

[0120] 1. Providing an effective potential field representation method: This invention proposes to use a unified three-dimensional grid coordinate system to represent each point in the radar detection potential field, providing a basic support for the analysis of radar detection capabilities and the fusion calculation of multi-radar detection potential fields in subsequent combat systems.

[0121] 2. Achieve efficient and accurate prediction: The UNet model is proposed to predict the detection probability value of all grid points in the potential field at once. Compared with traditional neural networks or other deep models, the UNet model makes full use of the strong spatial correlation between neighboring points in the potential field, which not only significantly improves the inference speed, but also has a higher prediction accuracy.

[0122] 3. Enhanced Model Interpretability: An innovative data-driven modeling approach is adopted, using the detection probability baseline values ​​calculated by the radar equation and the Swerling model as supervised information to train the deep learning model. This enables the model to approximate the results of traditional physical calculation methods, effectively improving the model's interpretability.

[0123] Any process or method described in the flowcharts of this invention or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, which can be implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device. The computer-readable medium can be any medium containing a program for storage, communication, propagation, or transmission for use by the execution system, apparatus, or device, including read-only memory, magnetic disks, or optical disks.

[0124] In the description of this specification, references to terms such as "embodiment," "example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, those skilled in the art can combine or combine the different embodiments or examples described in this specification and the features therein without causing contradiction.

[0125] While embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and alterations to the above embodiments within the scope of the present invention.

Claims

1. A method for intelligently representing radar detection capabilities based on a UNet network model, characterized in that, The method includes the following steps: S1. Obtain the detection capability-related attributes of the radar to be analyzed: the attributes include radar attributes, force attributes, environmental attributes, and configuration item attributes; S2. Construct a three-dimensional grid coordinate system for the potential field: Construct a three-dimensional grid coordinate system covering the detection range of the radar to be analyzed; calculate the distance components of each grid point in the three-dimensional grid coordinate system relative to the radar and the jammer on the X, Y, and Z coordinate axes; S3. Data Preprocessing: Preprocess the attribute data obtained in step S1 and the directional distance component data calculated in step S2; S4. Feature Encoding: The normalized and standardized distance data of each grid point calculated in step S2 are spliced ​​and fused with the radar, troop, environment and configuration attribute data preprocessed in step S3 to form a comprehensive encoded feature vector for each grid point. S5. Input the model and calculate the detection probability value: The generated 32-dimensional comprehensive encoded feature vector The input is fed into the trained 3D UNet network model to calculate the radar detection probability value; Step S2 is implemented as follows: First, a standardized 3D mesh coordinate system is constructed, including: S2.1 Coordinate System Construction: Using the radar's geographical location as the origin of the coordinate system, longitude, latitude, and altitude are used... , , () indicates; construct a three-dimensional Cartesian grid coordinate system covering its detection range; S2.2 Mesh Vertex Normalization and Alignment: Calculate the initial geographic coordinates of the bottom left vertex of the mesh; S2.3 Grid point coordinate generation: Based on the standardized vertex coordinates and preset scale factor, generate the geometric coordinates of all 32×32×32 grid points in the 3D grid; then calculate the directional distance components; S2.4 For radar: Set the radar's position coordinates in the three-dimensional grid coordinate system as ( , , ); Calculate each grid point ( , , The three-dimensional directional range components relative to the radar position; S2.5 For the jammer: Set the position coordinates of the jammer in the three-dimensional grid coordinate system as ( , , ); Calculate each grid point ( , , The three-dimensional directional distance components relative to the jammer's position.

2. The intelligent representation method for radar detection capability based on the UNet network model according to claim 1, characterized in that, The radar attributes include radar transmit power, radar antenna gain, radar transmit frequency, pulse repetition frequency, receiver bandwidth, and noise figure.

3. The intelligent representation method for radar detection capability based on the UNet network model according to claim 2, characterized in that, Force attributes include force cross-sectional area, jamming power, jammer bandwidth, and jamming frequency.

4. The intelligent representation method for radar detection capability based on the UNet network model according to claim 3, characterized in that, Environmental attributes include ambient temperature, rainfall rate, and ionospheric disturbance index.

5. The intelligent representation method for radar detection capability based on the UNet network model according to claim 1, characterized in that, Step S3 includes the following steps: S3.1 Set normalization factor: Set the corresponding normalization factor for each attribute; S3.2 Calculate the normalization ratio: Calculate the ratio of the true value of each attribute to its corresponding normalization factor: The resulting ratio is used as the encoded vector value of this attribute and is then used as input for subsequent models.

6. The intelligent representation method for radar detection capability based on the UNet network model according to claim 1, characterized in that, In step S4, the directional distance component and attribute data of each grid point are first concatenated and fused to generate a 32-dimensional comprehensive encoded feature vector. Then, the feature vectors are normalized and standardized.

7. The intelligent representation method for radar detection capability based on the UNet network model according to claim 6, characterized in that, The specific formulas for normalization and standardization are as follows: Normalization: Mapped to [0,1] standardization: Mean 0, Standard deviation 1 in, , , These represent the 32-dimensional comprehensive coding feature vectors. The minimum, maximum, and mean values; The normalized vector, This is the standardized vector.

8. The intelligent representation method for radar detection capability based on the UNet network model according to claim 1, characterized in that, In step S5, the results of data preprocessing and encoding in steps S1-S4 are used as input to the model. The prediction results are obtained through the UNet model, and the error loss is reduced through iterative training.

9. A radar detection capability intelligent characterization system based on the UNet network model, characterized in that, The system is used to implement the intelligent representation method for radar detection capability based on the UNet network model according to any one of claims 1-8.

Images