Virtual volume scan data generation method and system for radar data stream

Abstract

The invention relates to the technical field of data processing, and discloses a virtual volume scanning data generation method and system for a radar data stream. The method comprises the following steps: performing multi-dimensional decomposition and parallel preprocessing on an obtained original radar data stream to obtain a multi-channel preprocessed data stream; carrying out adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain a time-varying feature cluster; constructing a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimization scanning mode; performing multi-level spatial interpolation and phase reconstruction on the time-varying feature cluster through a self-optimization scanning mode to obtain a preliminary virtual volume scanning data field; performing target feature enhancement and multi-source environment information fusion on the preliminary virtual volume scanning data field to obtain high-simulation virtual volume scanning data; and performing multi-dimensional consistency verification and intelligent format adaptation on the high-simulation virtual volume scanning data to obtain target virtual volume scanning data. According to the method and the device, the efficiency and the accuracy of generating the virtual volume scanning data for the radar data stream are improved.

Description

Technical Field

[0001] The present application relates to the field of data processing, and in particular to a method and system for generating virtual volume scan data for radar data streams. Background Art

[0002] Existing radar data stream processing technologies mainly rely on traditional signal processing methods and fixed volume scanning patterns. These methods usually use linear filtering, Fourier transform, and statistical analysis techniques to extract and process radar echo information. In terms of virtual volume scanning data generation, existing technologies mostly use simple interpolation algorithms and predefined scanning patterns to simulate the radar scanning process, and simulate environmental interference by adding artificial noise. These methods can achieve certain results when processing static or slowly changing target scenes.

[0003] However, in the face of complex and ever-changing modern radar application scenarios, existing technologies have obvious shortcomings. First, the traditional fixed processing flow is difficult to adapt to the changes in highly dynamic targets and complex electromagnetic environments, resulting in large differences between the generated virtual volume scan data and the actual radar data. Secondly, simple interpolation and noise models cannot accurately reflect the performance characteristics and environmental impacts of real radar systems, reducing the simulation and usability of virtual data. In addition, existing technologies are inefficient in processing massive multi-dimensional radar data streams, making it difficult to meet real-time requirements, and lack an effective verification mechanism for the quality of generated data. Summary of the invention

[0004] The present application provides a method and system for generating virtual volume scan data for radar data streams, which are used to improve the efficiency and accuracy of generating virtual volume scan data for radar data streams.

[0005] In a first aspect, the present application provides a method for generating virtual volume scan data for a radar data stream, the method comprising: performing multi-dimensional decomposition and parallel preprocessing on the acquired original radar data stream to obtain a multi-channel preprocessed data stream; performing adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain a time-varying feature cluster; constructing a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode; performing multi-level spatial interpolation and phase reconstruction on the time-varying feature cluster through the self-optimizing scanning mode to obtain a preliminary virtual volume scan data field; performing target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data; performing multi-dimensional consistency verification and intelligent format adaptation on the highly simulated virtual volume scan data to obtain target virtual volume scan data.

[0006] In combination with the first aspect, in a first implementation manner of the first aspect of the present application, the multi-dimensional decomposition and parallel preprocessing of the acquired original radar data stream to obtain a multi-channel preprocessed data stream includes: performing time domain-frequency domain joint decomposition on the original radar data stream to obtain multi-dimensional decomposition data; performing parallel data stream allocation on the multi-dimensional decomposition data to obtain a parallel processing data stream; performing adaptive filtering on the parallel processing data stream to obtain a denoised data stream; performing dynamic range compression on the denoised data stream to obtain a normalized data stream; performing multi-resolution analysis on the normalized data stream to obtain multi-scale feature data; performing time synchronization calibration on the multi-scale feature data to obtain synchronized feature data; performing time-frequency joint characterization on the synchronized feature data through a wavelet packet transform algorithm to obtain a time-frequency feature map; performing feature enhancement on the time-frequency feature map to obtain an enhanced feature map; performing data stream reorganization on the enhanced feature map to obtain a reconstructed data stream; performing channel allocation and format conversion on the reconstructed data stream to obtain the multi-channel preprocessed data stream.

[0007] In combination with the first aspect, in a second implementation method of the first aspect of the present application, the multi-channel preprocessed data stream is adaptively extracted and dynamically clustered to obtain a time-varying feature cluster, including: performing multi-dimensional feature analysis on the multi-channel preprocessed data stream to obtain an original feature set; performing adaptive threshold segmentation on the original feature set to obtain a segmented feature subset; performing nonlinear dimensionality reduction on the segmented feature subset to obtain a reduced dimensionality feature vector; performing time series correlation analysis on the reduced dimensionality feature vector to obtain a time series feature sequence; dynamically clustering the time series feature sequence through a fuzzy C-means algorithm to obtain an initial feature cluster; performing outlier detection and elimination on the initial feature cluster to obtain an optimized feature cluster; performing time-varying feature extraction on the optimized feature cluster to obtain a time-varying feature matrix; performing feature weight calculation on the time-varying feature matrix to obtain a weighted feature set; performing adaptive feature fusion on the weighted feature set to obtain a fused feature vector; and performing dynamic threshold clustering on the fused feature vector to obtain the time-varying feature cluster.

[0008] In combination with the first aspect, in a third implementation method of the first aspect of the present application, the reconfigurable virtual scanning structure is constructed based on the time-varying feature cluster to obtain a self-optimizing scanning mode, including: performing spatiotemporal characteristic analysis on the time-varying feature cluster to obtain a spatiotemporal characteristic description matrix; performing topological structure mapping on the spatiotemporal characteristic description matrix to obtain an initial virtual scanning topology; dynamically meshing the initial virtual scanning topology to obtain an adaptive scanning grid; performing node importance evaluation on the adaptive scanning grid to obtain a node weight distribution; optimizing and reconstructing the node weight distribution through a graph theory algorithm to obtain an optimized scanning structure; parameterizing the optimized scanning structure to obtain a parameterizable scanning model; dynamically adjusting parameters of the parameterizable scanning model to obtain a dynamic scanning mode; performing simulation verification on the dynamic scanning mode to obtain a simulation performance index; performing multi-objective optimization on the simulation performance index to obtain an optimized scanning parameter set; reconstructing the scanning mode of the optimized scanning parameter set to obtain the self-optimizing scanning mode.

[0009] In combination with the first aspect, in a fourth implementation method of the first aspect of the present application, the time-varying feature clusters are subjected to multi-level spatial interpolation and phase reconstruction through the self-optimizing scanning mode to obtain a preliminary virtual volume scan data field, including: performing spatial distribution analysis on the time-varying feature clusters to obtain a feature space distribution map; performing multi-scale decomposition on the feature space distribution map to obtain a multi-level feature structure; performing adaptive grid generation on the multi-level feature structure to obtain a non-uniform interpolation grid; performing local feature density calculation on the non-uniform interpolation grid to obtain a density weight matrix; performing adaptive interpolation on the density weight matrix through a radial basis function algorithm to obtain a primary interpolation data field; performing boundary condition constraints on the primary interpolation data field to obtain a boundary correction data field; performing local optimization adjustment on the boundary correction data field to obtain a refined interpolation data field; performing phase consistency analysis on the refined interpolation data field to obtain a phase feature map; performing phase reconstruction optimization on the phase feature map to obtain a reconstructed phase field; and fusing the reconstructed phase field with the refined interpolation data field to obtain the preliminary virtual volume scan data field.

[0010] In combination with the first aspect, in a fifth implementation method of the first aspect of the present application, the target feature enhancement and multi-source environmental information fusion of the preliminary virtual volume scan data field to obtain high-simulation virtual volume scan data includes: performing target detection and segmentation on the preliminary virtual volume scan data field to obtain a target area data set; performing adaptive contrast enhancement on the target area data set to obtain enhanced target feature data; performing texture refinement processing on the enhanced target feature data to obtain a high-precision target feature field; performing dynamic scattering characteristic simulation on the high-precision target feature field to obtain simulated scattering data; performing environmental noise modeling on the simulated scattering data to obtain noise model parameters; performing correlation analysis on the noise model parameters and geographic information system data to obtain an environmental feature map; performing multi-source information fusion on the environmental feature map to obtain a comprehensive environmental data field; and performing adaptive weighted fusion on the comprehensive environmental data field and the high-precision target feature field to obtain the high-simulation virtual volume scan data.

[0011] In combination with the first aspect, in a sixth implementation method of the first aspect of the present application, the high-simulation virtual volume scan data is subjected to multi-dimensional consistency verification and intelligent format adaptation to obtain target virtual volume scan data, including: performing statistical characteristic analysis on the high-simulation virtual volume scan data to obtain a data statistical feature set; performing consistency comparison between the data statistical feature set and real radar data features to obtain a difference measurement index; performing adaptive threshold judgment on the difference measurement index to obtain a consistency evaluation result; performing multi-dimensional comprehensive evaluation on the consistency evaluation result to obtain a credibility score; performing local optimization adjustment on the data area whose credibility score is lower than a preset threshold to obtain optimized virtual volume scan data; performing data format analysis on the optimized virtual volume scan data to obtain a format feature description; performing matching analysis on the format feature description and the target system format requirements to obtain a format conversion strategy; performing intelligent format conversion on the optimized virtual volume scan data according to the format conversion strategy to obtain the target virtual volume scan data.

[0012] In a second aspect, the present application provides a virtual volume scan data generation system for a radar data stream, the virtual volume scan data generation system for a radar data stream comprising:

[0013] A decomposition module is used to perform multi-dimensional decomposition and parallel preprocessing on the acquired original radar data stream to obtain a multi-channel preprocessed data stream;

[0014] An extraction module, used for performing adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain time-varying feature clusters;

[0015] A construction module, used to construct a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode;

[0016] A reconstruction module, used for performing multi-level spatial interpolation and phase reconstruction on the time-varying feature cluster through the self-optimizing scanning mode to obtain a preliminary virtual volume scanning data field;

[0017] A fusion module, used for performing target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data;

[0018] The adaptation module is used to perform multi-dimensional consistency verification and intelligent format adaptation on the highly simulated virtual body scan data to obtain target virtual body scan data.

[0019] In the technical solution provided by the present application, the efficiency of data processing is greatly improved by multi-dimensional decomposition and parallel preprocessing of the original radar data stream, and high-throughput radar data streams can be processed in real time, laying a solid foundation for the subsequent generation of virtual volume scan data. Secondly, the use of adaptive feature extraction and dynamic clustering technology significantly improves the recognition and characterization capabilities of complex dynamic targets, so that the generated virtual data can more accurately reflect the characteristics of the actual target. The reconfigurable virtual scanning structure constructed based on time-varying feature clusters greatly enhances the adaptability of the system to different radar working modes and improves the flexibility and authenticity of virtual volume scan data. The application of multi-level spatial interpolation and phase reconstruction technology effectively improves the spatial resolution and phase consistency of virtual volume scan data, making the generated data more realistic. The process of target feature enhancement and multi-source environmental information fusion significantly improves the simulation accuracy of virtual data in complex electromagnetic environments, so that the generated data can better simulate the actual radar detection results. Finally, the multi-dimensional consistency verification and intelligent format adaptation mechanism ensure the quality and availability of the generated data, while improving the compatibility with various radar data processing platforms. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0021] Figure 1 A schematic diagram of an embodiment of a method for generating virtual volume scan data for a radar data stream in an embodiment of the present application;

[0022] Figure 2 Schematic diagram of an embodiment of a virtual volume scan data generation system for radar data stream in an embodiment of the present application. DETAILED DESCRIPTION

[0023] Embodiments of the present application provide a method and system for generating virtual volume scan data for radar data stream. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable in appropriate circumstances, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, the process, method, system, product or equipment comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or equipment.

[0024] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of a method for generating virtual volume scan data for radar data stream in an embodiment of the present application includes:

[0025] Step S101, performing multi-dimensional decomposition and parallel preprocessing on the acquired original radar data stream to obtain a multi-channel preprocessed data stream;

[0026] Step S102, performing adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain time-varying feature clusters;

[0027] Step S103, constructing a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode;

[0028] Step S104, performing multi-level spatial interpolation and phase reconstruction on the time-varying feature clusters through a self-optimizing scanning mode to obtain a preliminary virtual volume scanning data field;

[0029] Step S105, performing target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data;

[0030] Step S106: Perform multi-dimensional consistency verification and intelligent format adaptation on the highly simulated virtual body scan data to obtain target virtual body scan data.

[0031] It is understandable that the execution subject of the present application may be a virtual volume scan data generation system for radar data stream, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0032] Specifically, the multi-dimensional decomposition and parallel preprocessing of the acquired raw radar data stream are achieved by applying multi-channel signal processing technology. Specifically, the raw radar data stream is first decomposed into multiple frequency channels, each channel representing a different frequency range. Subsequently, the data of each frequency channel is processed in parallel, including operations such as noise filtering, amplitude normalization, and phase correction. This multi-dimensional decomposition and parallel preprocessing method can effectively improve the efficiency of data processing while retaining the rich information of the original data. Adaptive feature extraction and dynamic clustering of multi-channel preprocessed data streams are achieved by combining time-frequency analysis and machine learning techniques. The adaptive feature extraction process uses methods such as wavelet transform to automatically adjust the extraction parameters according to the time-varying characteristics of the data to capture the dynamic characteristics of the target. Dynamic clustering uses an improved K-means algorithm to classify and track the target in real time based on the extracted features, thereby forming time-varying feature clusters.

[0033] One of the core innovations of the present invention is to construct a reconfigurable virtual scanning structure based on time-varying feature clusters. This process uses graph theory and topology optimization methods to map time-varying feature clusters into a virtual three-dimensional space to form a dynamically adjustable scanning grid. By analyzing the motion characteristics and distribution density of the target, the scanning strategy is automatically optimized to generate a self-optimizing scanning pattern. This method can flexibly adapt to different target scenes and improve the efficiency and accuracy of virtual volume scanning. Multi-level spatial interpolation and phase reconstruction of time-varying feature clusters through self-optimizing scanning patterns are key steps in generating preliminary virtual volume scanning data fields. Multi-level spatial interpolation uses an adaptive interpolation algorithm, such as the Kriging interpolation method, to select the optimal interpolation parameters based on the spatial distribution characteristics of the data. Phase reconstruction uses the principle of phase consistency to ensure that the generated virtual data remains coherent in space and time. This process not only fills the blank areas in the original data, but also ensures the physical rationality of the generated data.

[0034] Target feature enhancement and multi-source environmental information fusion of the preliminary virtual volume scan data field are important links to improve the simulation degree of virtual data. Target feature enhancement uses adaptive contrast enhancement and texture refinement technology to highlight the key features of the target. Multi-source environmental information fusion makes comprehensive use of geographic information system data, meteorological data and electromagnetic environment models to simulate complex radar working environments. This step ensures that the generated virtual data can accurately reflect the complexity of actual radar detection. Finally, multi-dimensional consistency verification and intelligent format adaptation of high-simulation virtual volume scan data are the key to ensuring data quality and availability. Multi-dimensional consistency verification includes statistical characteristic analysis, spatial correlation test and time continuity evaluation, etc., to ensure that the generated virtual data is consistent with the real radar data in all dimensions. Intelligent format adaptation automatically adjusts the data structure and encoding method according to the data format requirements of different radar systems, thereby improving the versatility and compatibility of virtual data.

[0035] For example, in a certain test, the original radar data stream contained Doppler spectrum data at 1000 time points. Through multi-dimensional decomposition, the data was divided into 10 frequency channels, each containing data at 100 time points. After parallel preprocessing, the noise level was reduced by about 6dB. The adaptive feature extraction process identified three dynamic targets and formed time-varying feature clusters. Based on these feature clusters, a 100×100×100 virtual scanning grid was constructed, and the self-optimizing scanning mode reduced the number of scans from the traditional 10,000 to 7,500. Multi-level spatial interpolation and phase reconstruction generated a preliminary virtual volume scan data field with a resolution of 1m×1m×1m. Target feature enhancement and environmental information fusion took into account the local terrain undulations and atmospheric attenuation, so that the generated high-simulation virtual volume scan data had a correlation coefficient of 0.95 with the actual radar data in terms of signal-to-noise ratio and target features. Multi-dimensional consistency verification showed that the generated data was more than 90% consistent with the real data in terms of power spectral density, spatial autocorrelation function and time series characteristics. Finally, through intelligent format adaptation, the generated target virtual volume scan data can be directly used in various types of radar signal processing systems, greatly improving the practicality and versatility of virtual data.

[0036] In the embodiment of the present application, by multi-dimensional decomposition and parallel preprocessing of the original radar data stream, the efficiency of data processing is greatly improved, and high-throughput radar data streams can be processed in real time, laying a solid foundation for the subsequent generation of virtual volume scan data. Secondly, the use of adaptive feature extraction and dynamic clustering technology significantly improves the recognition and characterization capabilities of complex dynamic targets, so that the generated virtual data can more accurately reflect the characteristics of the actual target. The reconfigurable virtual scanning structure constructed based on time-varying feature clusters greatly enhances the adaptability of the system to different radar working modes and improves the flexibility and authenticity of virtual volume scan data. The application of multi-level spatial interpolation and phase reconstruction technology effectively improves the spatial resolution and phase consistency of virtual volume scan data, making the generated data more realistic. The process of target feature enhancement and multi-source environmental information fusion significantly improves the simulation accuracy of virtual data in complex electromagnetic environments, so that the generated data can better simulate the actual radar detection results. Finally, the multi-dimensional consistency verification and intelligent format adaptation mechanism ensure the quality and availability of the generated data, while improving the compatibility with various radar data processing platforms.

[0037] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0038] (1) Perform a joint time-domain and frequency-domain decomposition on the original radar data stream to obtain multi-dimensional decomposition data;

[0039] (2) performing parallel data stream allocation on the multi-dimensional decomposed data to obtain parallel processing data streams;

[0040] (3) Adaptively filter the parallel processed data stream to obtain a noise-reduced data stream;

[0041] (4) performing dynamic range compression on the noise reduction data stream to obtain a normalized data stream;

[0042] (5) performing multi-resolution analysis on the normalized data stream to obtain multi-scale feature data;

[0043] (6) performing time synchronization calibration on the multi-scale feature data to obtain synchronized feature data;

[0044] (7) Performing time-frequency joint characterization on the synchronization feature data by using a wavelet packet transform algorithm to obtain a time-frequency feature map;

[0045] (8) Performing feature enhancement on the time-frequency feature map to obtain an enhanced feature map;

[0046] (9) performing data stream reorganization on the enhanced feature map to obtain a reconstructed data stream;

[0047] (10) Channel allocation and format conversion are performed on the reconstructed data stream to obtain a multi-channel preprocessed data stream.

[0048] Specifically, the original radar data stream is decomposed in the time domain and frequency domain. This process uses the short-time Fourier transform (STFT) technology to convert the time domain signal into a time-frequency representation. STFT obtains the spectrum information that changes with time by Fourier transforming the signal in different time windows, thereby generating multi-dimensional decomposition data. These data contain the energy distribution information of the radar echo at different times and frequencies. The multi-dimensional decomposition data is distributed in parallel data streams. This step uses multi-threading or distributed computing technology to distribute the data to multiple processing units. Each processing unit is responsible for processing data within a specific time or frequency range, thereby forming a parallel processing data stream. This parallel processing method significantly improves the efficiency of data processing.

[0049] Adaptive filtering of the parallel processing data stream is a key step in reducing noise. Adaptive filters dynamically adjust the filter parameters based on the statistical characteristics of the signal, such as using the least mean square error (LMS) algorithm or the recursive least squares (RLS) algorithm. These algorithms minimize the error between the output signal and the desired signal by continuously adjusting the filter coefficients to obtain a de-noised data stream. The de-noised data stream is then subjected to dynamic range compression. This process uses techniques such as logarithmic compression or piecewise linear compression to map the wide dynamic range of the radar signal to a narrower range to obtain a normalized data stream. This not only facilitates subsequent processing, but also preserves the details of weak signals.

[0050] Multi-resolution analysis is performed on the normalized data stream to capture the characteristics of the signal at different scales. Here, the discrete wavelet transform (DWT) technique is used to decompose the signal into sub-signals of different frequency bands. DWT decomposes the signal in multiple layers through a high-pass and low-pass filter group to obtain multi-scale feature data, including approximate coefficients and detail coefficients. Time synchronization calibration is an important step to ensure that the multi-scale feature data are aligned in time. This process uses correlation analysis or phase alignment technology to compensate for the possible time delay between features of different scales, thereby obtaining synchronized feature data. The wavelet packet transform algorithm is an extension of the DWT, which performs a more detailed frequency band division of the signal. In this method, the wavelet packet transform performs a time-frequency joint characterization of the synchronized feature data to generate a time-frequency feature map. This process includes steps such as selecting the best basis function, calculating the transform coefficients, and reconstructing the signal, and finally obtains a high-resolution time-frequency representation.

[0051] The purpose of feature enhancement of the time-frequency feature map is to highlight key information. This step may include techniques such as contrast enhancement, edge sharpening, or nonlinear mapping. For example, the grayscale distribution of the time-frequency feature map is adjusted using histogram equalization technology to enhance the overall contrast of the image, thereby obtaining an enhanced feature map. Data stream reconstruction is the process of converting the enhanced feature map back to a time domain signal. This step reconstructs the information in the time-frequency domain into a time domain signal through an inverse transform operation, such as an inverse wavelet packet transform, to obtain a reconstructed data stream. This reconstruction process needs to take into account the preservation of phase information to ensure the integrity of the signal.

[0052] Finally, the reconstructed data stream is assigned channels and converted in format. Channel assignment allocates the reconstructed data stream to different processing channels according to the specific requirements of the radar system. Format conversion adjusts the data to a format that meets the subsequent processing requirements, such as IEEE 754 floating point format or a specific binary format, and finally obtains a multi-channel pre-processed data stream.

[0053] For example, suppose a pulse Doppler radar system collects 10 seconds of raw data with a sampling rate of 1MHz. First, STFT is used for joint decomposition in the time domain and frequency domain, and a 20ms time window and a 50% overlap rate are selected to obtain 500 frames of time-frequency spectrograms. These data are distributed to 8 parallel processing units. In the adaptive filtering stage, the LMS algorithm is used to iteratively optimize each frequency bin with a step size of 10^-4, reducing the noise level by 6dB on average. Dynamic range compression uses logarithmic compression to compress the original 90dB dynamic range to 60dB. Multi-resolution analysis uses 4-layer DWT decomposition to obtain sub-signals of 5 frequency bands. Time synchronization calibration corrects the maximum time deviation of 200μs between features of different scales by calculating the cross-correlation function. The wavelet packet transform selects the db4 wavelet basis and performs 6-layer decomposition to obtain 64 frequency bands of time-frequency feature maps. Feature enhancement uses adaptive histogram equalization to increase the information entropy of the time-frequency feature map from 5.6 to 6.8. Data stream reorganization and format conversion ultimately generate 8 channels of preprocessed data streams, each channel contains 1250 time sampling points and 64 frequency bins, and the data is stored in 32-bit floating point format. This detailed preprocessing process provides high-quality input data for subsequent virtual volume scan data generation.

[0054] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0055] (1) Perform multi-dimensional feature analysis on the multi-channel preprocessed data stream to obtain the original feature set;

[0056] (2) Adaptively threshold segment the original feature set to obtain a segmentation feature subset;

[0057] (3) Perform nonlinear dimensionality reduction on the segmentation feature subset to obtain a reduced dimensionality feature vector;

[0058] (4) Perform time series correlation analysis on the reduced dimension feature vector to obtain a time series feature sequence;

[0059] (5) Dynamically cluster the time series feature sequences using the fuzzy C-means algorithm to obtain the initial feature clusters;

[0060] (6) Detect and remove outliers on the initial feature cluster to obtain the optimized feature cluster;

[0061] (7) performing time-varying feature extraction on the optimized feature cluster to obtain a time-varying feature matrix;

[0062] (8) Calculating feature weights on the time-varying feature matrix to obtain a weighted feature set;

[0063] (9) Adaptively fuse the weighted feature set to obtain a fused feature vector;

[0064] (10) Perform dynamic threshold clustering on the fused feature vector to obtain time-varying feature clusters.

[0065] Specifically, a multi-dimensional feature analysis is performed on the multi-channel preprocessed data stream, which involves a comprehensive analysis of the time domain, frequency domain and statistical characteristics of the data. Specifically, short-time Fourier transform is used to extract time-frequency features, and high-order statistical analysis is used to capture the non-Gaussian characteristics of the signal. At the same time, statistical features such as entropy and correlation coefficient are calculated to obtain the original feature set. The original feature set is adaptively thresholded. This step uses the OTSU algorithm, which automatically determines the optimal threshold by maximizing the inter-class variance. The OTSU algorithm iteratively calculates the inter-class variance under different thresholds, and selects the threshold that maximizes the inter-class variance as the segmentation point to obtain the segmentation feature subset. This method can effectively divide the feature space into multiple regions with significant differences. Nonlinear dimensionality reduction is performed on the segmentation feature subset to reduce the redundancy of the data and extract essential features. The t-SNE (t-distributed stochastic neighbor embedding) algorithm is used here. The algorithm first calculates the conditional probability distribution of the data points in the high-dimensional space, and then reconstructs this distribution in the low-dimensional space. The dimensionality reduction is achieved by minimizing the KL divergence between the two distributions, and finally the reduced-dimensional feature vector is obtained.

[0066] Time series correlation analysis is the study of the correlation of reduced-dimensional feature vectors in the time dimension. This process uses autocorrelation functions and cross-correlation functions to calculate the correlation coefficients under different time delays, thereby revealing the periodicity and time dependence of the features and obtaining a time series feature sequence. The fuzzy C-means algorithm is a soft clustering method used to dynamically cluster time series feature sequences. The algorithm iteratively optimizes the objective function and assigns a membership vector to each data point, indicating the degree to which it belongs to each cluster. The specific steps include initializing the cluster center, calculating the distance from the data point to the cluster center, updating the membership matrix and cluster center until convergence, and finally obtaining the initial feature cluster.

[0067] Outlier detection and removal of the initial feature cluster is to improve the reliability of the clustering results. The local outlier factor (LOF) algorithm is used here, which detects outliers by comparing the local density of an object with its neighboring objects. The LOF value of each data point is calculated, and the points significantly higher than the threshold are marked as abnormal and removed to obtain the optimized feature cluster. Time-varying feature extraction is to extract feature information that changes over time from the optimized feature cluster. This process uses sliding window technology to calculate statistics and spectral features in different time windows, and track the changing trends of these features, and finally form a time-varying feature matrix.

[0068] The feature weight calculation aims to evaluate the importance of different features. The information gain-based method is used to calculate the contribution of each feature to the target classification. Specifically, the importance of the feature is quantified by calculating the change in system entropy before and after the feature is introduced, thereby obtaining a weighted feature set. Adaptive feature fusion is the process of integrating features from different sources and properties. Here, the adaptive weighted summation method is used to dynamically adjust the weights according to the reliability and information content of the features to achieve the optimal combination of features and obtain a fused feature vector.

[0069] Perform dynamic threshold clustering on the fused feature vector. This step uses the DBSCAN (density-based spatial clustering) algorithm, which can find clusters of any shape and is insensitive to noisy data. DBSCAN defines clusters by setting the neighborhood radius and the minimum number of points, dynamically adjusts these parameters to adapt to the changing characteristics of the data, and finally obtains time-varying feature clusters.

[0070] For example, suppose a phased array radar system collects 100 frames of multi-channel preprocessed data streams, each frame contains 1024 time sampling points and 32 frequency channels. Multidimensional feature analysis extracts 20 time domain features (such as mean, variance, skewness, etc.) and 30 frequency domain features (such as power spectrum density, center frequency, etc.) of each frame of data to form a 50-dimensional original feature set. Adaptive threshold segmentation uses the OTSU algorithm to divide the feature space into 3 regions to obtain the segmentation feature subset. Nonlinear dimensionality reduction uses the t-SNE algorithm to reduce the 50-dimensional features to 3 dimensions and obtain the reduced dimensionality feature vector. Time series correlation analysis calculates the autocorrelation function of 0-10 frame delays to obtain an 11-dimensional time series feature sequence. The fuzzy C-means algorithm sets the number of clusters to 5, performs 10 iterative clustering on the time series feature sequence, and obtains the initial feature cluster. Outlier detection uses the LOF algorithm, sets k=5 to calculate the local outlier factor, removes data points with LOF values ​​greater than 2, and obtains the optimized feature cluster. Time-varying feature extraction uses a sliding window of 20 frames, extracting 10 statistical features per frame to form a 100x10 time-varying feature matrix. Feature weight calculation is based on information gain, assigning weights to 10 features to obtain a weighted feature set. Adaptive feature fusion uses an exponential weighted average method to fuse features from different time windows to obtain a 10-dimensional fused feature vector. Finally, the DBSCAN algorithm sets the neighborhood radius to 0.5 and the minimum number of points to 4, dynamically clusters the fused feature vector, and finally obtains a time-varying feature cluster.

[0071] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0072] (1) Performing spatiotemporal characteristic analysis on the time-varying feature clusters to obtain a spatiotemporal characteristic description matrix;

[0073] (2) performing topological structure mapping on the spatiotemporal characteristic description matrix to obtain an initial virtual scanning topology;

[0074] (3) Dynamically meshing the initial virtual scanning topology to obtain an adaptive scanning mesh;

[0075] (4) Evaluate the node importance of the adaptive scanning grid and obtain the node weight distribution;

[0076] (5) Optimizing and reconstructing the node weight distribution through the graph theory algorithm to obtain the optimized scanning structure;

[0077] (6) performing parameterized modeling on the optimized scanning structure to obtain a parameterizable scanning model;

[0078] (7) dynamically adjusting parameters of the parameterizable scanning model to obtain a dynamic scanning mode;

[0079] (8) Perform simulation verification on the dynamic scanning mode to obtain simulation performance indicators;

[0080] (9) Perform multi-objective optimization on the simulation performance indicators to obtain an optimized scanning parameter set;

[0081] (10) The scanning pattern of the optimized scanning parameter set is reconstructed to obtain a self-optimizing scanning pattern.

[0082] Specifically, the time-varying feature clusters are subjected to spatiotemporal characteristic analysis. This process involves calculating the statistical characteristics of the feature clusters in the time and space dimensions, including the mean, variance, correlation coefficient, etc., to form a spatiotemporal characteristic description matrix. The matrix contains the distribution information and change trends of the feature clusters at different spatiotemporal locations. Topological structure mapping is performed on the spatiotemporal characteristic description matrix. This step uses the multidimensional scaling (MDS) algorithm to map the high-dimensional spatiotemporal characteristic description matrix to a low-dimensional space, maintaining the relative distance relationship between data points. The MDS algorithm achieves dimensionality reduction by minimizing the stress function of the distance between points in the high-dimensional space and the low-dimensional space, and obtains the initial virtual scanning topology.

[0083] The purpose of dynamic meshing of the initial virtual scan topology is to create an adaptable scan structure. An adaptive quadtree algorithm is used here to dynamically adjust the mesh density according to the complexity of the topology. The algorithm starts with a coarse mesh, subdivides in feature-dense areas, and keeps it coarse in feature-sparse areas, ultimately resulting in an adaptive scan mesh. Node importance assessment quantifies the contribution of each node in the adaptive scan mesh. This process uses the PageRank algorithm, which treats the scan mesh as a graph structure and calculates the PageRank value of each node. PageRank iteratively calculates the importance of nodes, taking into account the connectivity of nodes and the importance of neighboring nodes, and ultimately results in a node weight distribution.

[0084] The graph theory algorithm is used to optimize and reconstruct the node weight distribution. The spectral clustering algorithm is used here, which clusters based on the eigenvalues ​​and eigenvectors of the Laplacian matrix of the graph. By calculating the similarity matrix between nodes, constructing the Laplacian matrix, and then performing eigendecomposition on it, clustering using the first k eigenvectors, the optimized scanning structure is obtained. Parametric modeling is to convert the optimized scanning structure into a controllable mathematical model. This step uses nonlinear least squares fitting to represent the geometric characteristics and topological relationships of the scanning structure as a series of parametric equations. By minimizing the sum of squares of the error between the actual structure and the parametric model, a parameterizable scanning model is obtained.

[0085] Dynamic parameter adjustment is to optimize the parameterizable scanning model in real time according to the characteristics of real-time radar data stream. Here, the gradient descent method is used to iteratively update the parameter value by calculating the partial derivative of the objective function with respect to each parameter, so that the model can better adapt to the current scanning requirements and obtain the dynamic scanning mode. Simulation verification is to evaluate the performance of the dynamic scanning mode. This process uses the Monte Carlo simulation method to generate a large number of random scenes, simulate the radar scanning process, calculate performance indicators such as coverage, resolution, and energy efficiency, and form simulation performance indicators.

[0086] Multi-objective optimization is to find a balance between multiple performance indicators. Here, the NSGA-II (Non-dominated Sorting Genetic Algorithm II) algorithm is used. This algorithm generates a series of Pareto optimal solutions in the target space through genetic operations and non-dominated sorting, thereby obtaining an optimized scanning parameter set. Finally, the scanning pattern of the optimized scanning parameter set is reconstructed to generate the final self-optimizing scanning pattern. This step comprehensively considers the optimization parameters and the original scanning structure, and uses interpolation and smoothing techniques to construct a continuous and smooth scanning trajectory.

[0087] For example, suppose a phased array radar system needs to generate virtual volume scan data. First, the spatiotemporal characteristics of the data containing 100 time-varying feature clusters are analyzed to obtain a 100x50 spatiotemporal characteristic description matrix. The MDS algorithm is used to reduce the 50-dimensional characteristics to 3 dimensions to form the initial virtual scan topology. The adaptive quadtree algorithm is used to mesh the topology to generate an adaptive scan grid containing 1024 nodes. The PageRank algorithm is iterated 10 times to calculate the importance score of each node and form the node weight distribution. The spectral clustering algorithm uses the first 5 eigenvectors to divide the nodes into 10 categories to obtain the optimized scan structure. The nonlinear least squares fitting method uses 20 parameters to describe the scan structure and construct a parameterizable scan model. The gradient descent method uses a learning rate of 0.01 and iterates 100 times to adjust the parameters to obtain a dynamic scan mode. The Monte Carlo method generates 1000 random scenarios for simulation and calculates 5 key performance indicators. The NSGA-II algorithm sets the population size to 100 and evolves for 50 generations to obtain 20 sets of optimized scan parameters. Finally, through cubic spline interpolation, these 20 sets of parameters are reconstructed into a continuous self-optimizing scanning pattern, which contains 200 scanning points and covers the range of ±60° azimuth and ±30° elevation. This process provides a highly optimized and adaptive scanning strategy for virtual volume scan data generation.

[0088] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0089] (1) Perform spatial distribution analysis on the time-varying feature clusters to obtain a feature spatial distribution map;

[0090] (2) Perform multi-scale decomposition on the feature space distribution map to obtain a multi-level feature structure;

[0091] (3) performing adaptive grid generation on the multi-level feature structure to obtain a non-uniform interpolation grid;

[0092] (4) Calculate the local feature density of the non-uniform interpolation grid to obtain a density weight matrix;

[0093] (5) Adaptively interpolating the density weight matrix through a radial basis function algorithm to obtain a primary interpolation data field;

[0094] (6) performing boundary condition constraints on the primary interpolation data field to obtain a boundary correction data field;

[0095] (7) Locally optimizing and adjusting the boundary correction data field to obtain a refined interpolation data field;

[0096] (8) Performing phase consistency analysis on the refined interpolation data field to obtain a phase feature map;

[0097] (9) performing phase reconstruction optimization on the phase characteristic map to obtain a reconstructed phase field;

[0098] (10) The reconstructed phase field and the refined interpolation data field are fused to obtain a preliminary virtual volume scan data field.

[0099] Specifically, the spatial distribution of time-varying feature clusters is analyzed. This process uses the kernel density estimation (KDE) algorithm to calculate the probability density distribution of features in space. KDE generates a feature space distribution map by placing a kernel function (such as a Gaussian kernel) around each data point and then superimposing all kernel functions to obtain a continuous density estimate. The feature space distribution map is decomposed at multiple scales. Wavelet transform technology is used here, specifically discrete wavelet transform (DWT). DWT decomposes the signal into approximate coefficients and detail coefficients through high-pass and low-pass filters, and recursively applies them to the approximate coefficients to form a multi-level decomposition structure. This step generates a multi-level feature structure containing information at different scales.

[0100] Adaptive mesh generation is performed on multi-level feature structures to create a non-uniform interpolation mesh that adapts to the data distribution. A quadtree algorithm (for 2D data) or an octree algorithm (for 3D data) is used here. The algorithm starts with a coarse mesh and recursively subdivides the mesh based on the local feature density, generating a denser mesh in feature-rich areas while maintaining a larger mesh size in feature-sparse areas. The local feature density is calculated on the non-uniform interpolation mesh, and this step uses the k-nearest neighbor (k-NN) algorithm. For each mesh node, the average distance to its k nearest neighbors is calculated as the inverse of the local density. This process generates a density weight matrix that reflects the local feature distribution.

[0101] The density weight matrix is ​​adaptively interpolated by the radial basis function (RBF) algorithm. RBF interpolation is a powerful multivariate interpolation method that uses a linear combination of radially symmetric functions centered on the data points to approximate the interpolation function. In this method, Gaussian RBF is selected as the basis function, and its width parameter is adaptively adjusted according to the local density to obtain the primary interpolation data field. The boundary condition constraints are enforced on the primary interpolation data field to ensure that the interpolation results are consistent with the actual physical boundary conditions. This step uses the Lagrange multiplier method to introduce the boundary conditions as constraints into the interpolation problem, and obtains the boundary correction data field that satisfies the boundary conditions by solving the constrained optimization problem.

[0102] The boundary correction data field is locally optimized and adjusted, and the moving least squares method (MLS) is used here. MLS uses local weighted least squares fitting to improve the interpolation results by defining a compactly supported weight function around each evaluation point. This process produces a more refined and smoother refined interpolation data field. The refined interpolation data field is subjected to phase consistency analysis, and this step uses local phase correlation analysis technology. This technology calculates the phase spectrum within the local window and analyzes the phase relationship between adjacent regions, thereby obtaining a phase feature map reflecting phase continuity and consistency.

[0103] The phase reconstruction optimization is performed on the phase feature map, and an iterative phase recovery algorithm is used here. The algorithm iteratively optimizes the phase distribution by alternately applying amplitude and phase constraints, and finally obtains a reconstructed phase field that satisfies the physical constraints. Finally, the reconstructed phase field is fused with the refined interpolation data field, and this step uses a weighted average method. According to the credibility of the phase reconstruction, weights are assigned to the amplitude and phase information, and the final preliminary virtual volume scan data field is obtained through weighted combination.

[0104] For example, suppose a 3D phased array radar system needs to generate virtual volume scan data. First, the data containing 1000 time-varying feature clusters is spatially analyzed, and KDE is performed using a Gaussian kernel (bandwidth of 0.1) to obtain a 100x100x100 3D feature space distribution map. This distribution map is decomposed by 3 layers of DWT, and the Daubechies-4 wavelet is used to obtain a multi-level feature structure. The octree algorithm is used to generate adaptive grids with a density threshold of 0.01, and finally generates about 50,000 non-uniformly distributed grid nodes. The k-NN algorithm (k=10) is used to calculate the local density of each node to form a density weight matrix. RBF interpolation uses a Gaussian basis function, and its width parameter is dynamically adjusted between 0.05 and 0.5 to generate a primary interpolation data field. The boundary condition constraints take into account the physical scanning range of the radar (azimuth ±60°, elevation ±30°), and the Lagrange multiplier method is used for correction. MLS optimization uses cubic polynomial basis functions and exponential weight functions (support radius is 0.1) to obtain a refined interpolation data field. Phase consistency analysis uses a 16x16x16 local window to calculate the phase correlation between adjacent windows. The iterative phase recovery algorithm is repeated 50 times with an error threshold set to 0.1 to reconstruct the phase field. Finally, the reconstructed phase field is fused with the refined interpolation data field using the phase credibility as a weight (range 0.5-1) to obtain a preliminary virtual volume scan data field of 200x200x100, covering the complete radar scan range.

[0105] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0106] (1) Target detection and segmentation are performed on the preliminary virtual volume scan data field to obtain a target area data set;

[0107] (2) Adaptively enhance the contrast of the target area data set to obtain enhanced target feature data;

[0108] (3) Performing texture refinement processing on the enhanced target feature data to obtain a high-precision target feature field;

[0109] (4) Simulating the dynamic scattering characteristics of the high-precision target characteristic field to obtain simulated scattering data;

[0110] (5) Modeling environmental noise for simulated scattering data to obtain noise model parameters;

[0111] (6) Correlation analysis between noise model parameters and geographic information system data to obtain environmental feature mapping;

[0112] (7) Perform multi-source information fusion on environmental feature mapping to obtain a comprehensive environmental data field;

[0113] (8) Adaptively weighted fusion of the comprehensive environmental data field and the high-precision target feature field is performed to obtain highly simulated virtual volume scanning data.

[0114] Specifically, target detection and segmentation are performed on the preliminary virtual volume scan data field. This process uses the constant false alarm rate detection (CFAR) algorithm combined with morphological processing. The CFAR algorithm identifies potential targets through dynamic threshold detection, while morphological processing is used to remove noise and connect target areas, and finally obtain the target area data set. The target area data set is adaptively contrast enhanced. Here, histogram equalization technology is used, but the adaptive local histogram equalization (CLAHE) algorithm is adopted. CLAHE performs histogram equalization in a small window and limits the amplitude of contrast enhancement to avoid excessive noise enhancement, thereby obtaining enhanced target feature data.

[0115] Texture refinement is performed on the enhanced target feature data. This step uses the Gaussian-Laplacian pyramid algorithm. This algorithm enhances texture details at different scales by constructing a multi-resolution representation of the image. Specifically, the image is first Gaussian blurred and downsampled, then the difference between adjacent layers is calculated, and finally a high-precision target feature field is obtained through weighted synthesis. The dynamic scattering characteristic simulation of the high-precision target feature field is to simulate the radar scattering characteristics of the target at different observation angles. A hybrid technology combining the physical optics (PO) method and the geometric optics (GO) method is used here. The PO method is used to simulate the scattering of large-scale structures, while the GO method is used to process small-scale structures and multiple reflections to obtain simulated scattering data.

[0116] Environmental noise modeling is the process of adding real-world noise to simulated scattering data. Here, the generalized Gaussian distribution (GGD) model is used to describe complex environmental noise. The GGD model can flexibly represent various types of noise distributions by adjusting shape parameters, scale parameters, and location parameters to obtain noise model parameters. The noise model parameters are correlated with geographic information system (GIS) data, and this step uses spatial interpolation and statistical correlation analysis methods. First, GIS data (such as terrain, vegetation cover, etc.) are aligned with the radar scanning area, and then the correlation coefficient between the noise parameters and geographic features is calculated, and finally the environmental feature map is obtained.

[0117] Multi-source information fusion is the process of integrating environmental feature mapping with other relevant information (such as meteorological data, electromagnetic environment data, etc.). The Dempster-Shafer evidence theory framework is adopted here, which allows the combination of uncertain information from different sources. By defining the basic probability assignment function of each information source and using the Dempster combination rule, the comprehensive environmental data field is obtained. Finally, the comprehensive environmental data field is adaptively weighted fused with the high-precision target feature field. This step uses an adaptive weighting method based on the signal-to-noise ratio (SNR). According to the local SNR estimate, the weights of the target features and environmental features are dynamically adjusted. While retaining the target information, the environmental impact is reasonably introduced, and finally the highly simulated virtual volume scan data is obtained.

[0118] For example, suppose an X-band phased array radar system needs to generate virtual volume scan data covering an area of ​​10 km × 10 km. First, a two-dimensional CFAR algorithm (guard unit number 16, reference unit number 32, false alarm rate 10) is used. -6) Target detection was performed on the 512×512 preliminary virtual volume scan data field, and denoising was performed with 3×3 pixel morphological opening operations to obtain a data set containing 5 target areas. The CLAHE algorithm used an 8×8 local window and limited the contrast enhancement factor to 0.02 to enhance the target area. The Gaussian-Laplacian pyramid algorithm constructed a 4-layer pyramid and used a Gaussian kernel with a standard deviation of 1 for blurring to enhance texture details. The dynamic scattering characteristic simulation considered the geometric shape of the target (such as cubes and cylinders), used the PO method to calculate the primary scattering, and the GO method to process the secondary reflection to generate a scattering database with 360° azimuth and -10° to +30° elevation. The environmental noise modeling used the GGD model, whose shape parameter varied from 1.5 to 2.5 to simulate the noise characteristics from Rayleigh distribution to Gaussian distribution. The GIS data included a 20m resolution digital elevation model and a 1:50000 land use map, which were aligned with the radar data using the Kriging interpolation method. The Dempster-Shafer framework integrates five information sources, including terrain, vegetation, meteorology (temperature, humidity, wind speed), and electromagnetic interference, with the credibility of each source ranging from 0.6 to 0.9. Finally, based on the estimated local SNR (ranging from 0dB to 30dB), the sigmoid function is used to dynamically adjust the weights of target and environmental features to generate the final 1024×1024×256 (distance×azimuth×altitude) high-simulation virtual volume scan data. This process generates virtual radar data with rich details and real environmental effects, providing high-quality simulation scenarios for radar system design and algorithm verification.

[0119] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0120] (1) Analyze the statistical characteristics of the highly simulated virtual body scan data to obtain a data statistical feature set;

[0121] (2) Compare the consistency of the data statistical feature set with the real radar data features to obtain the difference measurement index;

[0122] (3) Perform adaptive threshold judgment on the difference measurement index to obtain the consistency evaluation result;

[0123] (4) Conduct a comprehensive multi-dimensional evaluation of the consistency assessment results to obtain a credibility score;

[0124] (5) Locally optimizing and adjusting the data area whose credibility score is lower than a preset threshold to obtain optimized virtual volume scan data;

[0125] (6) performing data format analysis on the optimized virtual volume scan data to obtain a format feature description;

[0126] (7) Matching and analyzing the format feature description with the target system format requirements to obtain a format conversion strategy;

[0127] (8) According to the format conversion strategy, the optimized virtual volume scan data is intelligently converted into a format to obtain the target virtual volume scan data.

[0128] Specifically, the statistical characteristics of the highly simulated virtual volume scan data are analyzed. This process involves calculating a variety of statistics, including mean, variance, skewness, kurtosis, entropy, etc., as well as the estimation of power spectral density (PSD) and autocorrelation function (ACF). PSD is calculated by the Welch method, using Hanning window and 50% overlap, while ACF is estimated by fast Fourier transform (FFT) method. These statistical features together constitute the data statistical feature set, which fully describes the statistical characteristics of the virtual data. The consistency of the data statistical feature set is compared with the real radar data features. Two measurement methods, Kullback-Leibler (KL) divergence and Bhattacharyya distance, are used here. KL divergence measures the difference between two probability distributions, while Bhattacharyya distance measures the degree of overlap between two probability distributions. By calculating these two distances between virtual data and real data on each statistical feature, a set of difference measurement indicators is obtained.

[0129] The OTSU algorithm is used to perform adaptive threshold judgment on the difference metric. The OTSU algorithm automatically determines the optimal threshold by maximizing the inter-class variance and divides the difference metric into two categories: "acceptable" and "need to be optimized". This process automatically adapts to the distribution characteristics of different features and obtains objective consistency evaluation results. The consistency evaluation results are evaluated in multiple dimensions using a fuzzy comprehensive evaluation method. This method first establishes an evaluation factor set and a comment set, and then determines the weight vector and membership matrix through expert scoring or machine learning methods. Through fuzzy synthesis operations, the final credibility score is obtained, which comprehensively reflects the simulation quality of virtual data in multiple dimensions.

[0130] Local optimization adjustment is performed on the data area with a credibility score lower than the preset threshold. Here, an iterative optimization method based on gradient descent is used. First, the low-scoring area is identified, and then the simulation quality of the local data is gradually improved by fine-tuning the data generation parameters of the area (such as noise model parameters, scattering characteristic parameters, etc.), and finally the optimized virtual volume scan data is obtained.

[0131] Perform data format analysis on the optimized virtual volume scan data. This step involves detailed analysis of data structure, encoding method, metadata, etc. By parsing the data header, checking the data type, analyzing the data organization method, etc., a format feature description containing detailed information on the data format is generated.

[0132] The format feature description is matched with the target system format requirements. Here, a rule-based expert system method is adopted to predefine a series of format conversion rules, and then the optimal conversion path from the current format to the target format is determined through rule matching and reasoning to obtain the format conversion strategy.

[0133] Finally, according to the format conversion strategy, the optimized virtual volume scan data is converted into an intelligent format. This process may involve operations such as data rearrangement, type conversion, compression / decompression, and metadata reconstruction. By executing a series of conversion operations defined in the conversion strategy, the target virtual volume scan data that meets the requirements of the target system is finally obtained.

[0134] For example, assume that an X-band phased array radar system generates 1024×1024×256 (range×azimuth×altitude) high-simulation virtual volume scan data. First, the statistical features are calculated, including mean (-45dB), standard deviation (8dB), skewness (0.2), kurtosis (3.1) and entropy (6.8). PSD is calculated using 1024-point FFT and Hanning window with 50% overlap, and ACF is calculated to 100 delay points. Compared with the real radar data, KL divergence (average value 0.05) and Bhattacharyya distance (average value 0.03) are calculated. The OTSU algorithm sets the KL divergence threshold to 0.08 and the Bhattacharyya distance threshold to 0.05. The fuzzy comprehensive evaluation uses 5 evaluation factors (amplitude statistics, phase statistics, time correlation, spectrum characteristics, spatial distribution), 3 comment levels (excellent, good, poor), and the weights are determined by the AHP method. The obtained credibility score is 0.85 (full score 1). For local areas with scores lower than 0.7, the gradient descent method is used for optimization, the learning rate is set to 0.01, and 50 iterations are performed. Data format analysis shows that the current format is a 32-bit floating point number, big endian, with 64 bytes of header information. The target system requires a 16-bit fixed point number, little endian, and 128 bytes of header information. Based on this information, a format conversion strategy is generated that includes data type conversion, byte order adjustment, and header information reconstruction. Finally, the format conversion is performed to obtain a 2GB target virtual volume scan data file that meets the requirements of the target system. This process ensures that the generated virtual radar data is not only highly consistent with the real data in terms of statistical characteristics, but also can be seamlessly integrated into the target radar system, providing reliable data support for radar performance evaluation and algorithm verification.

[0135] The above describes the method for generating virtual volume scan data for radar data stream in the embodiment of the present application. The following describes the system for generating virtual volume scan data for radar data stream in the embodiment of the present application. Figure 2 In one embodiment of the present application, a virtual volume scan data generation system for radar data stream includes:

[0136] A decomposition module 201 is used to perform multi-dimensional decomposition and parallel preprocessing on the acquired original radar data stream to obtain a multi-channel preprocessed data stream;

[0137] An extraction module 202, configured to perform adaptive feature extraction and dynamic clustering on the multi-channel pre-processed data stream to obtain a time-varying feature cluster;

[0138] A construction module 203 is used to construct a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode;

[0139] A reconstruction module 204 is used to perform multi-level spatial interpolation and phase reconstruction on the time-varying feature clusters through the self-optimizing scanning mode to obtain a preliminary virtual volume scanning data field;

[0140] A fusion module 205 is used to perform target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data;

[0141] The adaptation module 206 is used to perform multi-dimensional consistency verification and intelligent format adaptation on the highly simulated virtual body scan data to obtain target virtual body scan data.

[0142] Through the collaboration of the above components, the efficiency of data processing is greatly improved by multi-dimensional decomposition and parallel preprocessing of the original radar data stream, which can process high-throughput radar data streams in real time, laying a solid foundation for the subsequent generation of virtual volume scan data. Secondly, the use of adaptive feature extraction and dynamic clustering technology significantly improves the recognition and characterization capabilities of complex dynamic targets, so that the generated virtual data can more accurately reflect the characteristics of actual targets. The reconfigurable virtual scanning structure constructed based on time-varying feature clusters greatly enhances the adaptability of the system to different radar working modes and improves the flexibility and authenticity of virtual volume scan data. The application of multi-level spatial interpolation and phase reconstruction technology effectively improves the spatial resolution and phase consistency of virtual volume scan data, making the generated data more realistic. The process of target feature enhancement and multi-source environmental information fusion significantly improves the simulation accuracy of virtual data in complex electromagnetic environments, so that the generated data can better simulate the actual radar detection results. Finally, multi-dimensional consistency verification and intelligent format adaptation mechanism ensure the quality and availability of generated data, while improving compatibility with various radar data processing platforms.

[0143] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0144] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for generating virtual volume scan data for radar data stream, characterized in that: The virtual volume scan data generation method for radar data stream includes: Perform multi-dimensional decomposition and parallel preprocessing on the acquired original radar data stream to obtain a multi-channel preprocessed data stream; Performing adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain time-varying feature clusters; Building a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode; Performing multi-level spatial interpolation and phase reconstruction on the time-varying feature clusters through the self-optimizing scanning mode to obtain a preliminary virtual volume scanning data field; Performing target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data; The highly simulated virtual body scan data is subjected to multi-dimensional consistency verification and intelligent format adaptation to obtain target virtual body scan data.

2. The method for generating virtual volume scan data for radar data stream according to claim 1, characterized in that: The multi-dimensional decomposition and parallel preprocessing of the acquired original radar data stream to obtain a multi-channel preprocessed data stream includes: Performing a time-domain-frequency-domain joint decomposition on the original radar data stream to obtain multi-dimensional decomposition data; Performing parallel data stream allocation on the multi-dimensional decomposed data to obtain parallel processing data streams; Adaptively filtering the parallel processed data stream to obtain a noise-reduced data stream; Performing dynamic range compression on the noise reduction data stream to obtain a normalized data stream; Performing multi-resolution analysis on the normalized data stream to obtain multi-scale feature data; Performing time synchronization calibration on the multi-scale feature data to obtain synchronized feature data; Performing time-frequency joint characterization on the synchronization feature data by using a wavelet packet transform algorithm to obtain a time-frequency feature graph; Performing feature enhancement on the time-frequency feature map to obtain an enhanced feature map; Performing data stream reorganization on the enhanced feature map to obtain a reconstructed data stream; Channel allocation and format conversion are performed on the reconstructed data stream to obtain the multi-channel preprocessed data stream.

3. The method for generating virtual volume scan data for radar data stream according to claim 1, characterized in that: The step of performing adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain a time-varying feature cluster includes: Performing multi-dimensional feature analysis on the multi-channel preprocessed data stream to obtain an original feature set; Performing adaptive threshold segmentation on the original feature set to obtain a segmentation feature subset; Performing nonlinear dimensionality reduction on the segmentation feature subset to obtain a reduced dimensionality feature vector; Performing time series correlation analysis on the dimension-reduced feature vector to obtain a time series feature sequence; Dynamically clustering the time series feature sequence by using a fuzzy C-means algorithm to obtain an initial feature cluster; Performing outlier detection and elimination on the initial feature cluster to obtain an optimized feature cluster; Performing time-varying feature extraction on the optimized feature cluster to obtain a time-varying feature matrix; Calculating feature weights on the time-varying feature matrix to obtain a weighted feature set; Performing adaptive feature fusion on the weighted feature set to obtain a fused feature vector; Dynamic threshold clustering is performed on the fused feature vector to obtain the time-varying feature cluster.

4. The method for generating virtual volume scan data for radar data stream according to claim 1, characterized in that: The step of constructing a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode includes: Performing spatiotemporal characteristic analysis on the time-varying feature cluster to obtain a spatiotemporal characteristic description matrix; Performing topological structure mapping on the spatiotemporal characteristic description matrix to obtain an initial virtual scanning topology; Dynamically meshing the initial virtual scanning topology to obtain an adaptive scanning mesh; Performing node importance evaluation on the adaptive scanning grid to obtain node weight distribution; The node weight distribution is optimized and reconstructed by using a graph theory algorithm to obtain an optimized scanning structure; Performing parameterized modeling on the optimized scanning structure to obtain a parameterizable scanning model; Dynamically adjusting parameters of the parameterizable scanning model to obtain a dynamic scanning mode; Performing simulation verification on the dynamic scanning mode to obtain simulation performance indicators; Performing multi-objective optimization on the simulation performance index to obtain an optimized scanning parameter set; The optimized scanning parameter set is subjected to scanning pattern reconstruction to obtain the self-optimizing scanning pattern.

5. The method for generating virtual volume scan data for radar data stream according to claim 1, characterized in that: The method of performing multi-level spatial interpolation and phase reconstruction on the time-varying feature clusters through the self-optimizing scanning mode to obtain a preliminary virtual volume scanning data field includes: Performing spatial distribution analysis on the time-varying feature clusters to obtain a feature spatial distribution map; Performing multi-scale decomposition on the feature space distribution map to obtain a multi-level feature structure; Performing adaptive grid generation on the multi-level characteristic structure to obtain a non-uniform interpolation grid; Performing local feature density calculation on the non-uniform interpolation grid to obtain a density weight matrix; Adaptively interpolating the density weight matrix using a radial basis function algorithm to obtain a primary interpolation data field; Executing boundary condition constraints on the primary interpolation data field to obtain a boundary correction data field; Performing local optimization adjustment on the boundary correction data field to obtain a refined interpolation data field; Performing phase consistency analysis on the refined interpolation data field to obtain a phase characteristic map; Performing phase reconstruction optimization on the phase characteristic map to obtain a reconstructed phase field; The reconstructed phase field and the refined interpolation data field are fused to obtain the preliminary virtual volume scan data field.

6. The method for generating virtual volume scan data for radar data stream according to claim 1, characterized in that: The step of performing target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data includes: Performing target detection and segmentation on the preliminary virtual volume scan data field to obtain a target area data set; Adaptively enhancing the contrast of the target area data set to obtain enhanced target feature data; Performing texture refinement processing on the enhanced target feature data to obtain a high-precision target feature field; Performing dynamic scattering characteristic simulation on the high-precision target characteristic field to obtain simulated scattering data; Performing environmental noise modeling on the simulated scattering data to obtain noise model parameters; Performing correlation analysis on the noise model parameters and geographic information system data to obtain an environmental feature map; Performing multi-source information fusion on the environmental feature map to obtain a comprehensive environmental data field; The comprehensive environment data field and the high-precision target feature field are adaptively weighted fused to obtain the high-simulation virtual volume scan data.

7. The method for generating virtual volume scan data for radar data stream according to claim 1, characterized in that: The performing multi-dimensional consistency verification and intelligent format adaptation on the highly simulated virtual body scan data to obtain target virtual body scan data includes: Performing statistical characteristic analysis on the highly simulated virtual body scan data to obtain a data statistical feature set; Performing consistency comparison between the data statistical feature set and the real radar data features to obtain a difference measurement index; Performing adaptive threshold judgment on the difference measurement index to obtain a consistency evaluation result; Conduct a multi-dimensional comprehensive evaluation on the consistency assessment results to obtain a credibility score; Performing local optimization adjustment on the data area whose credibility score is lower than a preset threshold to obtain optimized virtual body scan data; Performing data format analysis on the optimized virtual volume scan data to obtain a format feature description; Performing matching analysis on the format feature description and the target system format requirement to obtain a format conversion strategy; According to the format conversion strategy, the optimized virtual volume scan data is intelligently format converted to obtain the target virtual volume scan data.

8. A virtual volume scan data generation system for radar data stream, characterized in that: The virtual volume scan data generation system for radar data stream comprises: A decomposition module is used to perform multi-dimensional decomposition and parallel preprocessing on the acquired original radar data stream to obtain a multi-channel preprocessed data stream; An extraction module, used for performing adaptive feature extraction and dynamic clustering on the multi-channel preprocessed data stream to obtain time-varying feature clusters; A construction module, used to construct a reconfigurable virtual scanning structure based on the time-varying feature cluster to obtain a self-optimizing scanning mode; A reconstruction module, used for performing multi-level spatial interpolation and phase reconstruction on the time-varying feature cluster through the self-optimizing scanning mode to obtain a preliminary virtual volume scanning data field; A fusion module, used for performing target feature enhancement and multi-source environmental information fusion on the preliminary virtual volume scan data field to obtain highly simulated virtual volume scan data; The adaptation module is used to perform multi-dimensional consistency verification and intelligent format adaptation on the highly simulated virtual body scan data to obtain target virtual body scan data.

Images