An unmanned aerial vehicle based phased array radar radiation characteristic evaluation method

Abstract

This invention discloses a method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV), comprising the following steps: Step 1: acquiring radiation measurement data at various UAV flight paths; Step 2: preprocessing the radiation measurement data; Step 3: generating a deep radiation evaluation feature vector using an improved TimeMixer network; Step 4: quantitatively analyzing the radiation characteristics of the target phased array radar based on the deep radiation evaluation feature vector; Step 5: comprehensively analyzing the overall radiation performance of the target phased array radar using a radiation performance grading mechanism based on Choquet integral; Step 6: performing anomaly localization and radiation field distribution reconstruction for the target phased array radar; and Step 7: generating a performance evaluation report for the target phased array radar. This invention improves the evaluation accuracy of the radiation characteristics of phased array radar through an improved TimeMixer network and a radiation performance grading mechanism based on Choquet integral.

Description

Technical Field

[0001] This invention relates to the field of radar radiation characteristic assessment technology, and in particular to a method for assessing the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV). Background Technology

[0002] With the widespread application of phased array radar in early warning detection, target tracking, airspace surveillance, and intelligent equipment, high-precision evaluation and anomaly diagnosis technologies for the radiation characteristics of phased array radars have received increasing attention. Existing radar radiation performance testing methods mainly rely on fixed test fields, turntable measuring devices, or manually set measurement points for radiation pattern measurement and index analysis. However, these methods generally suffer from the following problems in practical applications: The collected radiation measurement data is susceptible to disturbances from UAV flight, platform attitude changes, environmental noise, and multipath effects, resulting in large fluctuations in radiation intensity signals, low signal-to-noise ratios, and poor stability and reliability of key indicators such as main lobe width, side lobe level, and beam pointing deviation under complex testing conditions. Multi-source measurement data includes information such as radiation intensity, spatial location, attitude angle, and time stamps. These data often exhibit differences in sampling frequency, recording accuracy, and time reference. Existing synchronization and registration methods struggle to achieve precise correspondence between spatial location and radiation response, leading to insufficient consistency of multi-track point measurement data across time and spatial coordinates. For nonlinear, non-stationary, and strongly coupled radiation response data generated under dynamic UAV sampling conditions, traditional pattern fitting methods, statistical analysis methods, and conventional neural network methods struggle to simultaneously characterize temporal variation features, spatial distribution features, and anomalous disturbance features, affecting the accuracy of quantitative evaluation of phased array radar radiation characteristics, performance classification, anomaly localization, and radiation field reconstruction.

[0003] Therefore, how to provide a method for evaluating the radiation characteristics of phased array radar based on unmanned aerial vehicles is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0004] One objective of this invention is to propose a method for evaluating the radiation characteristics of a phased array radar based on unmanned aerial vehicles (UAVs). This invention fully utilizes an improved TimeMixer network and a performance grading analysis method based on Choquet integrals, and describes in detail the quantitative evaluation of radiation characteristics, performance level determination, anomaly location, and radiation field distribution reconstruction for a target phased array radar. It has the advantages of high automation, high evaluation accuracy, and strong anomaly identification capability.

[0005] A method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to an embodiment of the present invention includes the following steps: Step 1: Obtain radiation measurement data at each flight point of the UAV; Step 2: Perform data preprocessing on the radiation measurement data to obtain a standard radiation measurement dataset; Step 3: The standard radiometric measurement dataset is processed through an improved TimeMixer network to perform spatiotemporal multi-scale feature extraction, radiometric decomposition, and cross-scale directional hybrid modeling to obtain a deep radiometric assessment feature vector; the improved TimeMixer network includes an input encoding layer, a spatiotemporal multi-scale feature module, a radiometric decomposition and hybridization module, and a radiometric assessment mapping module; Step 4: Based on the deep radiation assessment feature vector, perform quantitative analysis on the radiation characteristics of the target phased array radar to obtain the quantitative analysis results of the radiation characteristics; Step 5: Based on the quantitative analysis results of radiation characteristics, a radiation performance grading mechanism based on Choquet integral is adopted to comprehensively analyze the overall radiation performance of the target phased array radar and obtain the radiation performance level. Step 6: Based on the quantitative analysis results of radiation characteristics, radiation performance level and standard radiation measurement dataset, perform anomaly localization and radiation field distribution reconstruction on the target phased array radar, obtain the anomaly radiation marker grid and generate the radiation field distribution reconstruction results; Step 7: Based on the results of the quantitative analysis of radiation characteristics, radiation performance level, anomalous radiation marker grid, and radiation field distribution reconstruction, generate a performance evaluation report for the target phased array radar.

[0006] Optionally, the radiation measurement data includes radiation intensity, spatial position vector, attitude angle vector, and time stamp obtained according to the sampling time point.

[0007] Optionally, the data preprocessing includes performing time axis alignment, spatial registration, noise suppression, outlier removal, missing data compensation, amplitude normalization, and temporal feature encoding on the radiometric measurement data, specifically including: The radiometric measurement data is aligned along the time axis according to the time stamp information to obtain the aligned radiometric measurement dataset; the aligned radiometric measurement dataset consists of the aligned radiation intensity sequence, the aligned spatial position vector sequence, the aligned attitude angle vector sequence, and the aligned time stamp sequence; The alignment spatial position vector sequence and the alignment attitude angle vector sequence are transformed from the UAV carrier coordinate system to the target radar coordinate system to obtain the registration spatial position vector sequence and the registration attitude angle vector sequence. The aligned radiance intensity sequence is subjected to noise suppression using a Gaussian filtering method to obtain a denoised radiance intensity sequence. The denoised radiation intensity sequence is identified and removed by the Z-score method. For the removed abnormal sampling points, the missing radiation intensity is compensated by the linear interpolation method to obtain the compensated radiation intensity sequence. The amplitude of the compensated radiation intensity sequence is normalized by the min-max normalization method to obtain the normalized radiation intensity sequence; The aligned time stamp sequence is encoded using sine and cosine coding methods to obtain a time stamp vector sequence; The normalized radiation intensity sequence, the registration spatial position vector sequence, the registration attitude angle vector sequence, and the time stamp vector sequence are correlated to obtain a standard radiation measurement dataset.

[0008] Optionally, step three specifically includes: In the input encoding layer, the standard radiometric measurement data at each sampling time point are further vectorized and linearly transformed to generate an initial fused feature vector sequence; The spatiotemporal multi-scale feature module includes a temporal multi-scale branch and a spatial multi-scale branch; In the time-multiscale branch, a timescale set is defined, which includes K timescales, each timescale corresponding to a sampling window length; At each time scale, the initial fused feature vector sequence is downsampled in segments according to the sampling window length to obtain the downsampled time feature vector sequence; The downsampled time feature sequence is linearly mapped within the scale to obtain the time-scale feature vector sequence. In the spatial multi-scale branch, based on the registration spatial location vector sequence, the spatial distance between the registration spatial location vectors at any two sampling time points is calculated using the L2 norm. Define a set of spatial scales, which includes K spatial scales, each spatial scale corresponding to a neighborhood radius; At each spatial scale, for each sampling time point, the initial fused feature vector corresponding to the sampling time point whose spatial distance is less than or equal to the neighborhood radius is used as the spatial neighborhood set of the current sampling time point; Within the spatial neighborhood set, the initial fused feature vector sequence is averaged and aggregated within the neighborhood to obtain the spatial feature vector at the current sampling time point, and the spatial feature vectors at all sampling time points are combined to form a spatial feature vector sequence. The spatial feature vector sequence is linearly mapped within the scale to obtain the spatial scale feature vector sequence. The temporal feature vector sequence at the k-th time scale is concatenated with the spatial feature vector sequence at the k-th spatial scale to obtain the spatiotemporal joint feature vector sequence; where k ranges from 1 to K. The radiation decomposition and mixing module performs component decomposition and cross-scale directional mixing on the spatiotemporal joint feature vector sequence of K scales to generate a previously decomposable and mixed feature vector sequence. In the radiation assessment mapping module, the previously decomposable mixed feature vector sequence is averaged in the scale dimension to obtain the scale-pooled feature vector sequence. Define a set of evaluation metrics, and for each evaluation metric, map the scaled pooling feature vector sequence to the metric feature vector sequence through a linear transformation; The indicator feature vector sequence is averaged and pooled over time to obtain the indicator average feature vector; the indicator average feature vector of each evaluation indicator is concatenated to obtain the deep radiation evaluation feature vector.

[0009] Optionally, the radiation decomposition and mixing module performs component decomposition and cross-scale directional mixing on the spatiotemporal joint feature vector sequence at K scales to generate a previously decomposable and mixed feature vector sequence, specifically including: The radiation decomposition and mixing module includes a radiation baseline branch, a beam variation branch, an anomaly perturbation branch, a cross-scale mixing branch, and a component fusion branch. In the radiation baseline branch, the spatiotemporal joint feature vector sequence at each scale is locally averaged in the time dimension to obtain the baseline candidate vector sequence at the current scale. The baseline candidate vector sequence is then used to generate the radiation baseline vector sequence through linear mapping. In the beamforming branch, the spatiotemporal joint feature vector at each sampling time point under each scale is subtracted element-wise from the corresponding radiation baseline vector to obtain the beamforming candidate vector; the beamforming candidate vector is generated into a beamforming vector through linear mapping; and the beamforming vectors of all sampling time points are combined into a beamforming vector sequence. In the anomalous perturbation branch, the spatiotemporal joint feature vector at each sampling time point under each scale is subtracted from the corresponding radiation baseline vector and beam change vector to obtain anomalous perturbation candidate vectors. The anomalous perturbation candidate vectors are then used to generate anomalous perturbation vectors through linear mapping. The anomalous perturbation vectors corresponding to all sampling time points are then combined into an anomalous perturbation vector sequence. In the cross-scale directional mixing branch, taking the k-th scale as the reference, the radiation baseline vector sequence of the (k+1)-th scale is upsampled and mapped to the k-th scale, and then weighted and fused with the radiation baseline vector sequence of the k-th scale to obtain the mixed radiation baseline vector sequence of the k-th scale. Using the k-th scale as a reference, the beam change vector sequence of the k-1-th scale is downsampled and mapped to the k-th scale, and then weighted and fused with the beam change vector sequence of the k-th scale to obtain the hybrid beam change vector sequence of the k-th scale. The anomalous perturbation vector sequence at the k-th scale is used to generate an anomalous enhancement weight vector sequence through a linear mapping and a Sigmoid activation function; the anomalous perturbation vector sequence and the anomalous enhancement weight vector sequence are multiplied element-wise to obtain the hybrid anomalous perturbation vector sequence at the k-th scale. In the component fusion branch, the mixed radiation baseline vector sequence, mixed beam change vector sequence, and mixed anomaly perturbation vector sequence at each scale are added element by element to obtain the radiation decomposition mixed feature vector sequence. Arrange the K-scale radiative decomposition hybrid feature vector sequences in scale order to form the past decomposable hybrid feature vector sequence.

[0010] Optionally, the quantitative analysis results of the radiation characteristics consist of the main lobe width, side lobe level, beam pointing deviation, and radiation uniformity, specifically including: The deep radiation assessment feature vector is used to generate azimuth half-power beamwidth estimates and elevation half-power beamwidth estimates by applying two independent linear mappings and Softplus activation. The mean values ​​of the azimuth half-power beamwidth estimate and the elevation half-power beamwidth estimate are calculated to obtain the main lobe width; The deep radiation assessment feature vectors are used to generate the main lobe peak estimate and the maximum side lobe peak estimate through linear mapping, respectively; The sidelobe level is obtained by calculating the difference between the estimated peak value of the main lobe and the estimated peak value of the maximum sidelobe. The deep radiation assessment feature vector is linearly transformed in three-dimensional space to generate a beam pointing estimation vector. The beam pointing estimation vector is vector normalized to obtain the unit beam pointing vector. Based on the theoretical beam pointing azimuth and theoretical beam pointing elevation angles of the target phased array radar, a theoretical unit pointing vector is constructed. Calculate the dot product of the unit beam pointing vector and the theoretical unit pointing vector to obtain the direction cosine value; When the direction cosine value is greater than 1, set the direction cosine value to 1; when the direction cosine value is less than -1, set the direction cosine value to -1; use the inverse cosine value of the direction cosine value as the beam pointing deviation. The number of discrete units is set, and the deep radiation assessment feature vector is mapped to the radiation response estimation vector through a linear transformation based on the number of discrete units; the radiation response estimation vector is then normalized using Softmax to obtain the radiation response probability distribution vector. Calculate the mean and standard deviation of the probability distribution vector of the radiation response to obtain the mean and standard deviation of the radiation response; Set a smoothing coefficient, divide the mean of the radiation response by the sum of the standard deviation of the radiation response and the smoothing coefficient to obtain the radiation dispersion, and calculate the difference between 1 and the radiation dispersion to obtain the radiation uniformity.

[0011] Optionally, step five specifically includes: Set the main lobe width reference value, side lobe level reference value, beam pointing deviation reference value, radiation uniformity reference value, main lobe width tolerance coefficient, side lobe level tolerance coefficient, beam pointing deviation tolerance coefficient, and radiation uniformity tolerance coefficient; The Z-score method was used to standardize the main lobe width, side lobe level, beam pointing deviation, and radiation uniformity to obtain standard main lobe width, standard side lobe level, standard beam pointing deviation, and standard radiation uniformity. The standard main lobe width, standard side lobe level, standard beam pointing deviation, and standard radiation uniformity are respectively negatively exponentially mapped to obtain the main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score. The main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score are arranged in ascending order of value to form an index score vector. Based on the indicator scoring vector, the indicator corresponding to the current ranking position and all subsequent indicators constitute the indicator subset of the current ranking position; Define a fuzzy measure function to map the index subset to real numbers in the interval [0, 1] and obtain the fuzzy measure value; Based on the fuzzy measure function, the index score vector is non-additively aggregated using Choquet integral to obtain the comprehensive radiation score; Set primary, secondary, and tertiary thresholds, ensuring that the primary threshold is greater than the secondary threshold, and the secondary threshold is greater than the tertiary threshold. Determine the radiation performance level based on the comprehensive radiation score. If the comprehensive radiation score is greater than or equal to the Level 1 threshold, the radiation performance level is Level 1; if the comprehensive radiation score is greater than or equal to the Level 2 threshold and less than the Level 1 threshold, the radiation performance level is Level 2; if the comprehensive radiation score is greater than or equal to the Level 3 threshold and less than the Level 2 threshold, the radiation performance level is Level 3; if the comprehensive radiation score is less than the Level 3 threshold, the radiation performance level is Level 4.

[0012] Optionally, step six specifically includes: The abnormal judgment conditions are set as follows: the standard main lobe width is greater than the set main lobe width abnormal threshold, or the standard side lobe level is less than the set side lobe level abnormal threshold, or the standard beam pointing deviation is greater than the set beam pointing deviation abnormal threshold, or the standard radiation uniformity is less than the set radiation uniformity abnormal threshold, or the radiation performance level is level three or four. Based on the results of quantitative analysis of radiation characteristics and radiation performance level, the sampling time points that meet the anomaly judgment conditions are determined, and the set of anomaly sampling points is obtained. Extract the registration spatial location vector of each abnormal sampling time point in the abnormal sampling point set, and aggregate the abnormal sampling points according to the spatial proximity relationship to obtain the abnormal radiation area; Based on the normalized radiation intensity values, registration spatial position vectors, and registration attitude angle vectors at each sampling time point in the standard radiation measurement dataset, a set of discrete spatial radiation response points within the three-dimensional spatial sampling area is constructed. Based on the discrete spatial radiation response point set, the radiation response at unsampled locations within the three-dimensional spatial sampling area is interpolated and meshed to obtain a spatial radiation response mesh covering the three-dimensional spatial sampling area. The anomalous radiation region is mapped onto the spatial radiation response grid to obtain the anomalous radiation distribution region corresponding to the anomalous sampling point, thus forming the anomalous radiation marker grid. Based on the spatial radiation response grid and the anomalous radiation marker grid, coordinate mapping and gridded reconstruction methods are used to generate the directional distribution map, spatial energy distribution map and three-dimensional radiation field distribution map of the target phased array radar in the three-dimensional spatial sampling area, forming the radiation field distribution reconstruction result.

[0013] Optionally, the performance evaluation report includes the target phased array radar's identification information, evaluation time information, three-dimensional spatial sampling area information, radiation characteristic quantitative analysis results, radiation performance level, abnormal radiation area information, radiation field distribution reconstructed image results, and overall performance evaluation conclusions.

[0014] The beneficial effects of this invention are: This invention utilizes UAVs to perform radiation measurements on various track points of a phased array radar. It then performs time axis alignment, spatial registration, noise suppression, outlier removal, missing data compensation, amplitude normalization, and temporal feature encoding on radiation intensity, spatial position vectors, attitude angle vectors, and time markers to construct a standard radiation measurement dataset. This reduces the impact of asynchronous multi-source measurement data, high noise interference, and strong sample discreteness on the evaluation results under dynamic flight sampling conditions. In the feature extraction stage, the standard radiation measurement dataset is input into an improved TimeMixer network. Through the input encoding layer, spatiotemporal multi-scale feature module, radiation decomposition and mixing module, and radiation evaluation mapping module, spatiotemporal multi-scale feature extraction, radiation baseline component extraction, beam change component extraction, and anomaly disturbance component extraction are performed on the radiation data. Cross-scale directional mixing modeling is also employed, enabling the improved TimeMixer to simultaneously characterize the temporal variation characteristics, spatial distribution characteristics, and local anomaly disturbance characteristics of the target phased array radar's radiation response, thereby improving the discriminative ability of deep radiation evaluation feature vectors. In the quantitative analysis stage, the main lobe width, side lobe level, beam pointing deviation, and radiation uniformity are obtained based on the deep radiation assessment feature vector. Furthermore, a radiation performance grading mechanism based on Choquet integrals is employed to non-additively aggregate multiple indicators, fully considering the coupling relationship and comprehensive contribution between each indicator, thus improving the accuracy and reliability of the overall radiation performance level determination. In the post-processing stage, combining the radiation characteristic quantitative analysis results, radiation performance levels, and standard radiation measurement datasets, spatial aggregation of anomalous sampling points and location of anomalous radiation regions are performed. The radiation field distribution within the three-dimensional spatial sampling area is reconstructed, forming an anomalous radiation marker grid, as well as a direction distribution map, a spatial energy distribution map, and a three-dimensional radiation field distribution map. Therefore, this invention can effectively improve the automation level, quantitative analysis accuracy, anomaly identification capability, and engineering application value of phased array radar radiation characteristic assessment. Attached Figure Description

[0015] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of a phased array radar radiation characteristic evaluation method based on UAV proposed in this invention; Figure 2 This is a flowchart of the improved TimeMixer network structure in a phased array radar radiation characteristic evaluation method based on UAV proposed in this invention. Figure 3 This is a flowchart of the radiation performance grading mechanism based on Choquet integral in a phased array radar radiation characteristic evaluation method based on UAV proposed in this invention. Detailed Implementation

[0016] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0017] refer to Figures 1-3 A method for evaluating the radiation characteristics of a phased array radar based on unmanned aerial vehicles (UAVs) includes the following steps: Step 1: Obtain radiation measurement data for each flight point of the UAV; Step 2: Perform data preprocessing on the radiation measurement data to obtain a standard radiation measurement dataset; Step 3: The standard radiometric measurement dataset is processed through an improved TimeMixer network to perform spatiotemporal multi-scale feature extraction, radiometric decomposition, and cross-scale directional hybrid modeling to obtain deep radiometric assessment feature vectors. The improved TimeMixer network includes an input encoding layer, a spatiotemporal multi-scale feature module, a radiometric decomposition and hybridization module, and a radiometric assessment mapping module. Step 4: Based on the deep radiation assessment feature vector, perform quantitative analysis on the radiation characteristics of the target phased array radar to obtain the quantitative analysis results of the radiation characteristics; Step 5: Based on the quantitative analysis results of radiation characteristics, a radiation performance grading mechanism based on Choquet integral is adopted to comprehensively analyze the overall radiation performance of the target phased array radar and obtain the radiation performance level. Step 6: Based on the quantitative analysis results of radiation characteristics, radiation performance level and standard radiation measurement dataset, perform anomaly localization and radiation field distribution reconstruction on the target phased array radar, obtain the anomaly radiation marker grid and generate the radiation field distribution reconstruction results; Step 7: Based on the results of the quantitative analysis of radiation characteristics, radiation performance level, anomalous radiation marker grid, and radiation field distribution reconstruction, generate a performance evaluation report for the target phased array radar.

[0018] In this embodiment, the radiation measurement data includes radiation intensity, spatial position vector, attitude angle vector, and time stamp obtained according to the sampling time point.

[0019] In this embodiment, data preprocessing involves performing time axis alignment, spatial registration, noise suppression, outlier removal, missing data compensation, amplitude normalization, and temporal feature encoding on the radiometric measurement data. Specifically, this includes: The radiometric measurement data is aligned along the time axis according to the time stamp information to obtain the aligned radiometric measurement dataset; the aligned radiometric measurement dataset consists of the aligned radiation intensity sequence, the aligned spatial position vector sequence, the aligned attitude angle vector sequence, and the aligned time stamp sequence; The alignment spatial position vector sequence and the alignment attitude angle vector sequence are transformed from the UAV carrier coordinate system to the target radar coordinate system to obtain the registration spatial position vector sequence and the registration attitude angle vector sequence. The aligned radiance intensity sequence is subjected to noise suppression using a Gaussian filtering method to obtain a denoised radiance intensity sequence. The denoised radiation intensity sequence is identified and removed by the Z-score method. For the removed abnormal sampling points, the missing radiation intensity is compensated by the linear interpolation method to obtain the compensated radiation intensity sequence. The amplitude of the compensated radiation intensity sequence is normalized by the min-max normalization method to obtain the normalized radiation intensity sequence; The aligned time stamp sequence is encoded using sine and cosine coding methods to obtain a time stamp vector sequence; The normalized radiation intensity sequence, the registration spatial position vector sequence, the registration attitude angle vector sequence, and the time stamp vector sequence are correlated to obtain a standard radiation measurement dataset.

[0020] In this embodiment, step three specifically includes: In the input encoding layer, the standard radiometric data at each sampling time point are further concatenated and linearly transformed to generate an initial fused feature vector sequence, specifically: The normalized radiation intensity, registration spatial position vector, and registration attitude angle vector at each sampling time point are concatenated to obtain the measurement attribute vector. The measurement attribute vector and the time stamp vector are linearly transformed to obtain the measurement encoding vector and the time encoding vector, respectively. The measurement encoding vector and the time encoding vector are added element by element, and the initial fused feature vector sequence is formed according to the sampling time point order; The spatiotemporal multiscale feature module includes a temporal multiscale branch and a spatial multiscale branch; In the time-multiscale branch, a timescale set is defined, which includes K timescales, each timescale corresponding to a sampling window length; At each time scale, the initial fused feature vector sequence is downsampled in segments according to the sampling window length to obtain the downsampled time feature vector sequence; The downsampled time feature sequence is linearly mapped within the scale to obtain the time scale feature vector sequence at the current time scale. In the spatial multi-scale branch, based on the registration spatial location vector sequence, the spatial distance between the registration spatial location vectors at any two sampling time points is calculated using the L2 norm. Define a set of spatial scales, which includes K spatial scales, each spatial scale corresponding to a neighborhood radius; At each spatial scale, for each sampling time point, the initial fused feature vector corresponding to the sampling time point whose spatial distance is less than or equal to the neighborhood radius is used as the spatial neighborhood set of the current sampling time point; Within the spatial neighborhood set, the initial fused feature vector sequence is averaged and aggregated within the neighborhood to obtain the spatial feature vector at the current sampling time point, and the spatial feature vectors at all sampling time points are combined to form a spatial feature vector sequence. By performing a linear mapping within the scale of the spatial feature vector sequence, a spatial scale feature vector sequence at the current spatial scale is obtained. The temporal feature vector sequence at the k-th time scale is concatenated with the spatial feature vector sequence at the k-th spatial scale to obtain the spatiotemporal joint feature vector sequence; where k ranges from 1 to K. The radiation decomposition and mixing module performs component decomposition and cross-scale directional mixing on the spatiotemporal joint feature vector sequence of K scales to generate a previously decomposable mixed feature vector sequence. In the radiation assessment mapping module, the previously decomposable mixed feature vector sequence is averaged in the scale dimension to obtain the scale-pooled feature vector sequence. Define a set of evaluation indicators, which includes M evaluation indicators; For each evaluation metric, the scaled pooling feature vector sequence is mapped to the metric feature vector sequence through a linear transformation; The indicator feature vector sequence is averaged and pooled over time to obtain the indicator average feature vector; the indicator average feature vector of each evaluation indicator is concatenated to obtain the deep radiation evaluation feature vector.

[0021] In this embodiment, the radiation decomposition and mixing module performs component decomposition and cross-scale directional mixing on the spatiotemporal joint feature vector sequence of K scales to generate a previously decomposable and mixed feature vector sequence, specifically including: The radiation decomposition and mixing module includes a radiation baseline branch, a beam variation branch, anomaly perturbation branch, a cross-scale mixing branch, and a component fusion branch. In the radiation baseline branch, the spatiotemporal joint feature vector sequence at each scale is locally averaged in the time dimension to obtain the baseline candidate vector sequence at the current scale. The baseline candidate vector sequence is then used to generate the radiation baseline vector sequence through linear mapping. In the beamforming branch, the spatiotemporal joint feature vector at each sampling time point under each scale is subtracted element-wise from the corresponding radiation baseline vector to obtain the beamforming candidate vector; the beamforming candidate vector is generated into a beamforming vector through linear mapping; and the beamforming vectors of all sampling time points are combined into a beamforming vector sequence. In the anomalous perturbation branch, the spatiotemporal joint feature vector at each sampling time point under each scale is subtracted from the corresponding radiation baseline vector and beam change vector to obtain anomalous perturbation candidate vectors. The anomalous perturbation candidate vectors are then used to generate anomalous perturbation vectors through linear mapping. The anomalous perturbation vectors corresponding to all sampling time points are then combined into an anomalous perturbation vector sequence. In the cross-scale directional mixing branch, taking the k-th scale as the reference, the radiation baseline vector sequence of the (k+1)-th scale is upsampled and mapped to the k-th scale, and then weighted and fused with the radiation baseline vector sequence of the k-th scale to obtain the mixed radiation baseline vector sequence of the k-th scale. Using the k-th scale as a reference, the beam change vector sequence of the k-1-th scale is downsampled and mapped to the k-th scale, and then weighted and fused with the beam change vector sequence of the k-th scale to obtain the hybrid beam change vector sequence of the k-th scale. The anomalous perturbation vector sequence at the k-th scale is used to generate an anomalous enhancement weight vector sequence through a linear mapping and a Sigmoid activation function; the anomalous perturbation vector sequence and the anomalous enhancement weight vector sequence are multiplied element-wise to obtain the hybrid anomalous perturbation vector sequence at the k-th scale. In the component fusion branch, the mixed radiation baseline vector sequence, mixed beam change vector sequence, and mixed anomaly perturbation vector sequence at each scale are added element by element to obtain the radiation decomposition mixed feature vector sequence. Arrange the K-scale radiative decomposition hybrid feature vector sequences in scale order to form the past decomposable hybrid feature vector sequence.

[0022] In this invention, the improved TimeMixer network generally inherits the basic technical approach of the original TimeMixer network. Both belong to the time series feature extraction network based on the multi-scale modeling idea, and their core processing flow can be summarized as input representation, multi-scale feature extraction, decomposition and fusion, and task output mapping. Specifically, the original TimeMixer network first encodes the input time series, then models the sequence information at different time scales through a multi-scale feature extraction structure to obtain local change features and global trend features. Subsequently, a decomposable fusion structure is used to decompose and fuse the multi-scale features across scales to enhance the interaction between features at different scales. Finally, the output mapping structure maps the fused deep features to the target task space for prediction or evaluation.

[0023] The difference lies in the fact that the improved TimeMixer network does not directly adopt the structure of the original TimeMixer network for general time series prediction. Instead, it is specifically modified based on the original framework to suit the UAV radiometric measurement scenario. Specifically, at the input end, the improved TimeMixer network encodes the normalized radiation intensity, registration spatial position vector, registration attitude angle vector, and time stamp vector together to form an initial fused feature vector sequence, thereby achieving a unified representation of multi-source measurement data. In the multi-scale feature extraction part, the original TimeMixer network mainly emphasizes temporal scale modeling, while the improved TimeMixer network adds a spatial multi-scale branch. It constructs a spatial neighborhood set through neighborhood radius constraints and concatenates temporal and spatial scale features one by one to form a spatiotemporal joint feature vector sequence, thus expanding the original single-time multi-scale structure into a spatiotemporal multi-scale coupled structure. In the decomposition and mixing part, the original TimeMixer network tends to abstractly model general temporal components, while the improved TimeMixer network further refines this into radiation baseline branches, beam variation branches, and anomalous perturbation branches, and sets up cross-scale directional mixing branches and component fusion branches. This enables the network to perform more targeted component decomposition and fusion of different spatiotemporal features around the phased array radar radiation assessment task. At the output end, the improved TimeMixer network maps the scale-pooled features to multiple assessment index feature spaces through a radiation assessment mapping module, and forms a deep radiation assessment feature vector through time-dimensional pooling and vector concatenation, thereby serving subsequent quantitative analysis tasks of radiation characteristics such as main lobe width, side lobe level, beam pointing deviation, and radiation uniformity.

[0024] Through the aforementioned improvements, the enhanced TimeMixer network is better suited to the specific application scenario of evaluating the radiation characteristics of phased array radar under dynamic sampling conditions of UAVs. First, by jointly encoding radiation intensity, spatial location, attitude angle, and temporal information, the network gains a more complete input description capability, mitigating the information loss problem caused by single temporal inputs. Second, by introducing spatial multi-scale branches and constructing spatiotemporal joint features, the network can not only extract the time-varying patterns of radiation response but also characterize the correlation between different spatial sampling points, thereby improving its ability to represent radar directional distribution and spatial energy changes. Third, by refining the decomposed hybrid structure into three dedicated branches—radiation baseline, beam variation, and anomalous disturbance—the network can capture the overall radiation profile, beam dynamic changes, and local anomalous disturbances respectively, making the feature representation closer to the radiation mechanism of phased array radar. Finally, through feature mapping and aggregation oriented towards evaluation indicators, a more direct correspondence is established between the deep radiation evaluation feature vector output by the network and subsequent quantitative analysis of radiation characteristics, thus improving the accuracy and stability of main lobe width estimation, side lobe level analysis, beam deviation identification, anomaly localization, and radiation field distribution reconstruction.

[0025] In this embodiment, the quantitative analysis results of radiation characteristics consist of main lobe width, side lobe level, beam pointing deviation, and radiation uniformity, specifically including: The deep radiation assessment feature vector is used to generate azimuth half-power beamwidth estimates and elevation half-power beamwidth estimates by applying two independent linear mappings and Softplus activation. The mean values ​​of the azimuth half-power beamwidth estimate and the elevation half-power beamwidth estimate are calculated to obtain the main lobe width; The deep radiation assessment feature vectors are used to generate the main lobe peak estimate and the maximum side lobe peak estimate through linear mapping, respectively; The sidelobe level is obtained by calculating the difference between the estimated peak value of the main lobe and the estimated peak value of the maximum sidelobe. The deep radiation assessment feature vector is linearly transformed in three-dimensional space to generate a beam pointing estimation vector. The beam pointing estimation vector is vector normalized to obtain the unit beam pointing vector. Based on the theoretical beam pointing azimuth and theoretical beam pointing elevation angles of the target phased array radar, a theoretical unit pointing vector is constructed. Calculate the dot product of the unit beam pointing vector and the theoretical unit pointing vector to obtain the direction cosine value; When the direction cosine value is greater than 1, set the direction cosine value to 1; when the direction cosine value is less than -1, set the direction cosine value to -1; use the inverse cosine value of the direction cosine value as the beam pointing deviation. The number of discrete units is set, and the deep radiation assessment feature vector is mapped to the radiation response estimation vector through a linear transformation based on the number of discrete units; the radiation response estimation vector is then normalized using Softmax to obtain the radiation response probability distribution vector. Calculate the mean and standard deviation of the probability distribution vector of the radiation response to obtain the mean and standard deviation of the radiation response; Set a smoothing coefficient, divide the mean of the radiation response by the sum of the standard deviation of the radiation response and the smoothing coefficient to obtain the radiation dispersion, and calculate the difference between 1 and the radiation dispersion to obtain the radiation uniformity.

[0026] In this embodiment, step five specifically includes: Set the main lobe width reference value, side lobe level reference value, beam pointing deviation reference value, radiation uniformity reference value, main lobe width tolerance coefficient, side lobe level tolerance coefficient, beam pointing deviation tolerance coefficient, and radiation uniformity tolerance coefficient; The Z-score method is used to standardize the main lobe width, side lobe level, beam pointing deviation, and radiation uniformity to obtain standard main lobe width, standard side lobe level, standard beam pointing deviation, and standard radiation uniformity. For example, the standard main lobe width obtained by the Z-score method is obtained by calculating the absolute difference between the main lobe width and the reference value of the main lobe width, and dividing the absolute difference by the main lobe width tolerance coefficient to obtain the standard main lobe width. The standard main lobe width, standard side lobe level, standard beam pointing deviation, and standard radiation uniformity are respectively negatively exponentially mapped to obtain the main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score. The main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score are arranged in ascending order of value to form an index score vector. Based on the index scoring vector, the index at the current sorting position and all subsequent indices constitute the index subset at the current sorting position. For example, if the main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score are 0.42, 0.78, 0.35, and 0.61 respectively, then after sorting from smallest to largest, the index scoring vector can be written as [0.35, 0.42, 0.61, 0.78], and the corresponding sorted indices are beam pointing deviation index, main lobe width index, radiation uniformity index, and side lobe level index. At this point, the first subset of indicators is formed by the indicator corresponding to the first ranking position and all subsequent indicators, namely {beam pointing deviation indicator, main lobe width indicator, radiation uniformity indicator, and side lobe level indicator}; the second subset of indicators is formed by the indicator corresponding to the second ranking position and all subsequent indicators, namely {main lobe width indicator, radiation uniformity indicator, and side lobe level indicator}; the third subset of indicators is formed by the indicator corresponding to the third ranking position and all subsequent indicators, namely {radiation uniformity indicator, and side lobe level indicator}; and the fourth subset of indicators is formed by the indicator corresponding to the fourth ranking position, namely {side lobe level indicator}. Then, in the Choquet integral calculation, the fuzzy measure values ​​corresponding to these four subsets of indicators are called respectively, multiplied by the difference between adjacent ranking scores, and accumulated to obtain the comprehensive radiation score.

[0027] A fuzzy measure function is defined to map the subset of indicators to real numbers in the interval [0, 1] to obtain fuzzy measure values. The fuzzy measure values ​​are used to characterize the contribution of the corresponding subset of indicators to the comprehensive radiation score. The fuzzy measure value of the empty set is set to 0, and the fuzzy measure value of the whole set is set to 1. If any first subset of indicators is contained in any second subset of indicators, then the fuzzy measure value of the first subset of indicators is less than or equal to the fuzzy measure value of the second subset of indicators. Based on the fuzzy measure function, the index score vector is non-additively aggregated using Choquet integral to obtain the comprehensive radiation score. For example, if the main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score are 0.42, 0.78, 0.35, and 0.61 respectively, sorting them in ascending order yields the index score vector [0.35, 0.42, 0.61, 0.78]. The corresponding sorted indices are beam pointing deviation index, main lobe width index, radiation uniformity index, and side lobe level index. This constructs four index subsets: {beam pointing deviation index, main lobe width index, radiation uniformity index, side lobe level index}, {main lobe width index, radiation uniformity index, side lobe level index}, {radiation uniformity index, side lobe level index}, and {side lobe level index}. If the fuzzy measure values ​​corresponding to the above index subsets are 1.0... Given scores of 0, 0.82, 0.58, and 0.30, the initial value 0 is subtracted from the first sorted score of 0.35 to obtain 0.35, which is then multiplied by the fuzzy measure value of the first subset of indicators, 1.00, resulting in 0.35. Similarly, the first sorted score of 0.35 is subtracted from the second sorted score of 0.42 to obtain 0.07, which is then multiplied by the fuzzy measure value of the second subset of indicators, 0.82, resulting in 0.0574. Finally, the score is subtracted from the initial value of 0.30 from the third sorted score of 0.61. The second score of 0.42 yields 0.19, which is multiplied by the fuzzy measure value of the third indicator subset of 0.58 to get 0.1102. The fourth score after sorting is 0.78, which is minus the third score after sorting of 0.61 to get 0.17. This is then multiplied by the fuzzy measure value of the fourth indicator subset of 0.30 to get 0.051. Finally, 0.35, 0.0574, 0.1102 and 0.051 are added together to obtain the comprehensive radiation score of 0.5686.

[0028] Set primary, secondary, and tertiary thresholds, ensuring that the primary threshold is greater than the secondary threshold, and the secondary threshold is greater than the tertiary threshold. Determine the radiation performance level based on the comprehensive radiation score. If the comprehensive radiation score is greater than or equal to the Level 1 threshold, the radiation performance level is Level 1; if the comprehensive radiation score is greater than or equal to the Level 2 threshold and less than the Level 1 threshold, the radiation performance level is Level 2; if the comprehensive radiation score is greater than or equal to the Level 3 threshold and less than the Level 2 threshold, the radiation performance level is Level 3; if the comprehensive radiation score is less than the Level 3 threshold, the radiation performance level is Level 4.

[0029] In this embodiment, step six specifically includes: The abnormal judgment conditions are set as follows: the standard main lobe width is greater than the set main lobe width abnormal threshold, or the standard side lobe level is less than the set side lobe level abnormal threshold, or the standard beam pointing deviation is greater than the set beam pointing deviation abnormal threshold, or the standard radiation uniformity is less than the set radiation uniformity abnormal threshold, or the radiation performance level is level three or four. Based on the quantitative analysis results of radiation characteristics and radiation performance level, the sampling time points that meet the anomaly judgment conditions are determined, and the set of anomaly sampling points is obtained. Extract the registration spatial location vector of each abnormal sampling time point in the abnormal sampling point set, and aggregate the abnormal sampling points according to the spatial proximity relationship to obtain the abnormal radiation area; Based on the normalized radiation intensity values, registration spatial position vectors, and registration attitude angle vectors at each sampling time point in the standard radiation measurement dataset, a set of discrete spatial radiation response points within the three-dimensional spatial sampling area is constructed. Based on the discrete spatial radiation response point set, the radiation response at unsampled locations within the three-dimensional spatial sampling area is interpolated and meshed to obtain a spatial radiation response mesh covering the three-dimensional spatial sampling area. The anomalous radiation region is mapped onto the spatial radiation response grid to obtain the anomalous radiation distribution region corresponding to the anomalous sampling point, thus forming the anomalous radiation marker grid. Based on the spatial radiation response grid and the anomalous radiation marker grid, coordinate mapping and gridded reconstruction methods are used to generate the directional distribution map, spatial energy distribution map and three-dimensional radiation field distribution map of the target phased array radar in the three-dimensional spatial sampling area, forming the radiation field distribution reconstruction result.

[0030] In this embodiment, the performance evaluation report includes the target phased array radar's identification information, evaluation time information, three-dimensional spatial sampling area information, radiation characteristic quantitative analysis results, radiation performance level, abnormal radiation area information, radiation field distribution reconstructed image results, and overall performance evaluation conclusions.

[0031] Example 1: To verify the feasibility of this invention in practice, the method of this invention was applied to the external radiation characteristics test and health status assessment of a certain type of vehicle-mounted phased array radar. This type of phased array radar operates in the X-band and is mainly used for medium- and short-range target detection and beam tracking. After long-term operation, it may experience problems such as performance degradation of local T / R components on the array surface, beam pointing offset, and sidelobe lift.

[0032] In the implementation of this invention, the test site was a semi-open electromagnetic test area, with a test range of approximately 800 meters long, 600 meters wide, and 200 meters high. The UAV platform used was a six-rotor industrial-grade flight platform with a maximum endurance of 38 minutes, equipped with a broadband radiation measurement antenna, an inertial navigation unit, a differential positioning module, and a high-precision timing module. To ensure sufficient coverage of the target radar's main lobe, sidelobes, and edge radiation areas, a spherical three-dimensional spatial sampling area was first constructed based on the theoretical beam pointing range of the target radar. The sampling radius was set to 120 to 260 meters, the azimuth coverage angle range was -70° to 70°, and the elevation coverage angle range was -25° to 25°. Within this spatial range, considering the UAV's endurance and safe flight constraints, six orbital paths and four elevation-crossing paths were planned, totaling 1680 path points with an average point spacing of approximately 4.6 meters. Each track point carries spatial location, attitude constraints, and time stamp information to ensure a one-to-one correspondence between subsequent radiometric measurement data and spatial location information.

[0033] After the test began, the UAV performed flight measurement tasks according to the predetermined trajectory. During the flight, it continuously received radiation signals emitted by the target phased array radar and simultaneously recorded the radiation intensity, spatial position vector, attitude angle vector, and time stamp at each sampling moment. A total of 1680 sets of raw sampling records were obtained in one complete test, of which 1634 sets were valid radiation sampling records. The remaining records contained a small number of attitude fluctuations and signal losses caused by sudden gusts of wind. After the data acquisition was completed, the raw radiation measurement data were preprocessed to form a standard radiation measurement dataset.

[0034] In the feature extraction stage, the standard radiometric measurement dataset is input into the improved TimeMixer network. The network input encodes and fuses the normalized radiation intensity, registration spatial position vector, registration attitude angle vector, and time stamp vector to obtain an initial fused feature vector sequence. The spatiotemporal multi-scale feature module simultaneously establishes temporal and spatial multi-scale branches. The temporal scale is set to four levels, with sampling window lengths corresponding to 8, 16, 32, and 64 points, respectively; the spatial scale is set to four levels, with neighborhood radii of 3 meters, 6 meters, 12 meters, and 20 meters, respectively. In the radiometric decomposition and mixing module, the spatiotemporal joint features are decomposed into radiation baseline components, beam variation components, and anomalous perturbation components. Cross-scale directional mixing then enhances the ability to identify main lobe shift, sidelobe lift, and local anomalous energy regions. The radiation assessment mapping module outputs a deep radiation assessment feature vector and quantifies the main lobe width, sidelobe level, beam pointing deviation, and radiation uniformity.

[0035] To demonstrate the beneficial effects of this invention, three comparative schemes are provided in this embodiment. Comparative scheme one uses a traditional sparse ground measurement point method, employing manual point placement and conventional interpolation estimation to complete the radiation pattern evaluation. Comparative scheme two uses UAV sampling combined with traditional statistical analysis methods, performing only mean, variance, and local fitting analysis, without using a deep network. Comparative scheme three uses UAV sampling combined with the original TimeMixer network. The comparison results are shown in Table 1.

[0036] Table 1. Comparison of different schemes in the evaluation of the radiation characteristics of phased array radar.

[0037] As shown in Table 1, the present invention exhibits significant advantages in all comparative indicators. Compared with Comparative Scheme 1, the main lobe width estimation error of the present invention is reduced from 1.84° to 0.39°, the side lobe level estimation error is reduced from 3.27dB to 0.82dB, and the beam pointing deviation estimation error is reduced from 1.42° to 0.24°, indicating that the present invention can more accurately extract and characterize the radiation characteristics of the target phased array radar. The radiation uniformity estimation error is reduced from 0.118 to 0.026, the anomaly region identification accuracy is improved from 72.6% to 96.7%, and the average deviation of anomaly positioning is reduced from 9.8m to 1.8m, indicating that the present invention has higher reliability and accuracy in anomaly radiation identification and spatial positioning.

[0038] Furthermore, compared with Comparative Scheme 2 and Comparative Scheme 3, this invention also demonstrates significant advantages in terms of root mean square error (RMSE) of radiation field reconstruction, consistency rate of comprehensive performance classification, and effective sample utilization. Specifically, the RMSE of radiation field reconstruction is reduced to 0.034, indicating that this invention can more accurately recover the radiation field distribution of the target phased array radar within the three-dimensional spatial sampling area. The consistency rate of comprehensive performance classification reaches 98.3%, demonstrating that the radiation performance classification mechanism based on Choquet integrals in this invention can more effectively reflect the combined influence of multiple radiation indicators. The effective sample utilization rate reaches 96.1%, while the total time for a single test is shortened to 82 minutes, indicating that this invention not only improves data utilization efficiency but also takes into account testing efficiency.

[0039] Therefore, this invention can effectively solve the problems of large noise interference in radiation measurement data, difficulty in accurately aligning multi-source data, insufficient feature extraction capability, low anomaly location accuracy, and large radiation field reconstruction error in the prior art. It has the advantages of high evaluation accuracy, strong anomaly identification capability, good reconstruction effect, reliable classification results, and high efficiency in engineering applications.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV), characterized in that, Includes the following steps: Step 1: Obtain radiation measurement data for each flight point of the UAV; Step 2: Perform data preprocessing on the radiation measurement data to obtain a standard radiation measurement dataset; Step 3: The standard radiometric measurement dataset is processed through an improved TimeMixer network to perform spatiotemporal multi-scale feature extraction, radiometric decomposition, and cross-scale directional hybrid modeling to obtain a deep radiometric assessment feature vector; the improved TimeMixer network includes an input encoding layer, a spatiotemporal multi-scale feature module, a radiometric decomposition and hybridization module, and a radiometric assessment mapping module; Step 4: Based on the deep radiation assessment feature vector, perform quantitative analysis on the radiation characteristics of the target phased array radar to obtain the quantitative analysis results of the radiation characteristics; Step 5: Based on the quantitative analysis results of radiation characteristics, a radiation performance grading mechanism based on Choquet integral is adopted to comprehensively analyze the overall radiation performance of the target phased array radar and obtain the radiation performance level. Step 6: Based on the quantitative analysis results of radiation characteristics, radiation performance level and standard radiation measurement dataset, perform anomaly localization and radiation field distribution reconstruction on the target phased array radar, obtain the anomaly radiation marker grid and generate the radiation field distribution reconstruction results; Step 7: Based on the results of the quantitative analysis of radiation characteristics, radiation performance level, anomalous radiation marker grid, and radiation field distribution reconstruction, generate a performance evaluation report for the target phased array radar.

2. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The radiation measurement data includes radiation intensity, spatial position vector, attitude angle vector, and time stamp obtained according to the sampling time point.

3. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The data preprocessing involves performing time axis alignment, spatial registration, noise suppression, outlier removal, missing data compensation, amplitude normalization, and temporal feature encoding on the radiometric measurement data. Specifically, this includes: The radiometric measurement data is aligned along the time axis according to the time stamp information to obtain the aligned radiometric measurement dataset; the aligned radiometric measurement dataset consists of the aligned radiation intensity sequence, the aligned spatial position vector sequence, the aligned attitude angle vector sequence, and the aligned time stamp sequence; The alignment spatial position vector sequence and the alignment attitude angle vector sequence are transformed from the UAV carrier coordinate system to the target radar coordinate system to obtain the registration spatial position vector sequence and the registration attitude angle vector sequence. The aligned radiance intensity sequence is subjected to noise suppression using a Gaussian filtering method to obtain a denoised radiance intensity sequence. The denoised radiation intensity sequence is identified and removed by the Z-score method. For the removed abnormal sampling points, the missing radiation intensity is compensated by the linear interpolation method to obtain the compensated radiation intensity sequence. The amplitude of the compensated radiation intensity sequence is normalized by the min-max normalization method to obtain the normalized radiation intensity sequence; The aligned time stamp sequence is encoded using sine and cosine coding methods to obtain a time stamp vector sequence; The normalized radiation intensity sequence, the registration spatial position vector sequence, the registration attitude angle vector sequence, and the time stamp vector sequence are correlated to obtain a standard radiation measurement dataset.

4. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, Step three specifically includes: In the input encoding layer, the standard radiometric measurement data at each sampling time point are further vectorized and linearly transformed to generate an initial fused feature vector sequence; The spatiotemporal multi-scale feature module includes a temporal multi-scale branch and a spatial multi-scale branch; In the time-multiscale branch, a timescale set is defined, which includes K timescales, each timescale corresponding to a sampling window length; At each time scale, the initial fused feature vector sequence is downsampled in segments according to the sampling window length to obtain the downsampled time feature vector sequence; The downsampled time feature sequence is linearly mapped within the scale to obtain the time-scale feature vector sequence. In the spatial multi-scale branch, based on the registration spatial location vector sequence, the spatial distance between the registration spatial location vectors at any two sampling time points is calculated using the L2 norm. Define a set of spatial scales, which includes K spatial scales, each spatial scale corresponding to a neighborhood radius; At each spatial scale, for each sampling time point, the initial fused feature vector corresponding to the sampling time point whose spatial distance is less than or equal to the neighborhood radius is used as the spatial neighborhood set of the current sampling time point; Within the spatial neighborhood set, the initial fused feature vector sequence is averaged and aggregated within the neighborhood to obtain the spatial feature vector at the current sampling time point, and the spatial feature vectors at all sampling time points are combined to form a spatial feature vector sequence. The spatial feature vector sequence is linearly mapped within the scale to obtain the spatial scale feature vector sequence. The temporal feature vector sequence at the k-th time scale is concatenated with the spatial feature vector sequence at the k-th spatial scale to obtain the spatiotemporal joint feature vector sequence; where k ranges from 1 to K. The radiation decomposition and mixing module performs component decomposition and cross-scale directional mixing on the spatiotemporal joint feature vector sequence of K scales to generate a previously decomposable and mixed feature vector sequence. In the radiation assessment mapping module, the previously decomposable mixed feature vector sequence is averaged in the scale dimension to obtain the scale-pooled feature vector sequence. Define a set of evaluation metrics, and for each evaluation metric, map the scaled pooling feature vector sequence to the metric feature vector sequence through a linear transformation; The indicator feature vector sequence is averaged and pooled over time to obtain the indicator average feature vector; the indicator average feature vector of each evaluation indicator is concatenated to obtain the deep radiation evaluation feature vector.

5. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 4, characterized in that, The radiation decomposition and mixing module performs component decomposition and cross-scale directional mixing on the spatiotemporal joint feature vector sequence at K scales to generate a previously decomposable and mixed feature vector sequence, specifically including: The radiation decomposition and mixing module includes a radiation baseline branch, a beam variation branch, an anomaly perturbation branch, a cross-scale mixing branch, and a component fusion branch. In the radiation baseline branch, the spatiotemporal joint feature vector sequence at each scale is locally averaged in the time dimension to obtain the baseline candidate vector sequence at the current scale. The baseline candidate vector sequence is then used to generate the radiation baseline vector sequence through linear mapping. In the beamforming branch, the spatiotemporal joint feature vector at each sampling time point under each scale is subtracted element-wise from the corresponding radiation baseline vector to obtain the beamforming candidate vector; the beamforming candidate vector is generated into a beamforming vector through linear mapping; and the beamforming vectors of all sampling time points are combined into a beamforming vector sequence. In the anomalous perturbation branch, the spatiotemporal joint feature vector at each sampling time point under each scale is subtracted from the corresponding radiation baseline vector and beam change vector to obtain anomalous perturbation candidate vectors. The anomalous perturbation candidate vectors are then used to generate anomalous perturbation vectors through linear mapping. The anomalous perturbation vectors corresponding to all sampling time points are then combined into an anomalous perturbation vector sequence. In the cross-scale directional mixing branch, taking the k-th scale as the reference, the radiation baseline vector sequence of the (k+1)-th scale is upsampled and mapped to the k-th scale, and then weighted and fused with the radiation baseline vector sequence of the k-th scale to obtain the mixed radiation baseline vector sequence of the k-th scale. Using the k-th scale as a reference, the beam change vector sequence of the k-1-th scale is downsampled and mapped to the k-th scale, and then weighted and fused with the beam change vector sequence of the k-th scale to obtain the hybrid beam change vector sequence of the k-th scale. The anomalous perturbation vector sequence at the k-th scale is used to generate an anomalous enhancement weight vector sequence through a linear mapping and a Sigmoid activation function; the anomalous perturbation vector sequence and the anomalous enhancement weight vector sequence are multiplied element-wise to obtain the hybrid anomalous perturbation vector sequence at the k-th scale. In the component fusion branch, the mixed radiation baseline vector sequence, mixed beam change vector sequence, and mixed anomaly perturbation vector sequence at each scale are added element by element to obtain the radiation decomposition mixed feature vector sequence. Arrange the K-scale radiative decomposition hybrid feature vector sequences in scale order to form the past decomposable hybrid feature vector sequence.

6. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The quantitative analysis results of the radiation characteristics consist of the main lobe width, side lobe level, beam pointing deviation, and radiation uniformity, specifically including: The deep radiation assessment feature vector is used to generate azimuth half-power beamwidth estimates and elevation half-power beamwidth estimates by applying two independent linear mappings and Softplus activation. The mean values ​​of the azimuth half-power beamwidth estimate and the elevation half-power beamwidth estimate are calculated to obtain the main lobe width; The deep radiation assessment feature vectors are used to generate the main lobe peak estimate and the maximum side lobe peak estimate through linear mapping, respectively; The sidelobe level is obtained by calculating the difference between the estimated peak value of the main lobe and the estimated peak value of the maximum sidelobe. The deep radiation assessment feature vector is linearly transformed in three-dimensional space to generate a beam pointing estimation vector. The beam pointing estimation vector is vector normalized to obtain the unit beam pointing vector. Based on the theoretical beam pointing azimuth and theoretical beam pointing elevation angles of the target phased array radar, a theoretical unit pointing vector is constructed. Calculate the dot product of the unit beam pointing vector and the theoretical unit pointing vector to obtain the direction cosine value; When the direction cosine value is greater than 1, set the direction cosine value to 1; when the direction cosine value is less than -1, set the direction cosine value to -1; use the inverse cosine value of the direction cosine value as the beam pointing deviation. The number of discrete units is set, and the deep radiation assessment feature vector is mapped to the radiation response estimation vector through a linear transformation based on the number of discrete units; the radiation response estimation vector is then normalized using Softmax to obtain the radiation response probability distribution vector. Calculate the mean and standard deviation of the probability distribution vector of the radiation response to obtain the mean and standard deviation of the radiation response; Set a smoothing coefficient, divide the mean of the radiation response by the sum of the standard deviation of the radiation response and the smoothing coefficient to obtain the radiation dispersion, and calculate the difference between 1 and the radiation dispersion to obtain the radiation uniformity.

7. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, Step five specifically includes: Set the main lobe width reference value, side lobe level reference value, beam pointing deviation reference value, radiation uniformity reference value, main lobe width tolerance coefficient, side lobe level tolerance coefficient, beam pointing deviation tolerance coefficient, and radiation uniformity tolerance coefficient; The Z-score method was used to standardize the main lobe width, side lobe level, beam pointing deviation, and radiation uniformity to obtain standard main lobe width, standard side lobe level, standard beam pointing deviation, and standard radiation uniformity. The standard main lobe width, standard side lobe level, standard beam pointing deviation, and standard radiation uniformity are respectively negatively exponentially mapped to obtain the main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score. The main lobe width score, side lobe level score, beam pointing deviation score, and radiation uniformity score are arranged in ascending order of value to form an index score vector. Based on the indicator scoring vector, the indicator corresponding to the current ranking position and all subsequent indicators constitute the indicator subset of the current ranking position; Define a fuzzy measure function to map the index subset to real numbers in the interval [0, 1] and obtain the fuzzy measure value; Based on the fuzzy measure function, the index score vector is non-additively aggregated using Choquet integral to obtain the comprehensive radiation score; Set primary, secondary, and tertiary thresholds, ensuring that the primary threshold is greater than the secondary threshold, and the secondary threshold is greater than the tertiary threshold. Determine the radiation performance level based on the comprehensive radiation score. If the comprehensive radiation score is greater than or equal to the Level 1 threshold, the radiation performance level is Level 1; if the comprehensive radiation score is greater than or equal to the Level 2 threshold and less than the Level 1 threshold, the radiation performance level is Level 2; if the comprehensive radiation score is greater than or equal to the Level 3 threshold and less than the Level 2 threshold, the radiation performance level is Level 3; if the comprehensive radiation score is less than the Level 3 threshold, the radiation performance level is Level 4.

8. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, Step six specifically includes: The abnormal judgment conditions are set as follows: the standard main lobe width is greater than the set main lobe width abnormal threshold, or the standard side lobe level is less than the set side lobe level abnormal threshold, or the standard beam pointing deviation is greater than the set beam pointing deviation abnormal threshold, or the standard radiation uniformity is less than the set radiation uniformity abnormal threshold, or the radiation performance level is level three or four. Based on the results of quantitative analysis of radiation characteristics and radiation performance level, the sampling time points that meet the anomaly judgment conditions are determined, and the set of anomaly sampling points is obtained. Extract the registration spatial location vector of each abnormal sampling time point in the abnormal sampling point set, and aggregate the abnormal sampling points according to the spatial proximity relationship to obtain the abnormal radiation area; Based on the normalized radiation intensity values, registration spatial position vectors, and registration attitude angle vectors at each sampling time point in the standard radiation measurement dataset, a set of discrete spatial radiation response points within the three-dimensional spatial sampling area is constructed. Based on the discrete spatial radiation response point set, the radiation response at unsampled locations within the three-dimensional spatial sampling area is interpolated and meshed to obtain a spatial radiation response mesh covering the three-dimensional spatial sampling area. The anomalous radiation region is mapped onto the spatial radiation response grid to obtain the anomalous radiation distribution region corresponding to the anomalous sampling point, thus forming the anomalous radiation marker grid. Based on the spatial radiation response grid and the anomalous radiation marker grid, coordinate mapping and gridded reconstruction methods are used to generate the directional distribution map, spatial energy distribution map and three-dimensional radiation field distribution map of the target phased array radar in the three-dimensional spatial sampling area, forming the radiation field distribution reconstruction result.

9. The method for evaluating the radiation characteristics of a phased array radar based on an unmanned aerial vehicle (UAV) according to claim 1, characterized in that, The performance evaluation report includes the target phased array radar's identification information, evaluation time information, three-dimensional spatial sampling area information, radiation characteristic quantitative analysis results, radiation performance level, abnormal radiation area information, radiation field distribution reconstruction image results, and overall performance evaluation conclusions.

Images