Radar model sorting method and device and storage medium
By using a three-level cascaded multi-task neural network structure, radar signals are sorted step by step, solving the problem of accurate sorting in complex and dynamic radar signal environments, and achieving efficient and accurate radar model sorting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUILIN CHANGHAI DEV
- Filing Date
- 2025-12-23
- Publication Date
- 2026-05-12
AI Technical Summary
Existing technologies struggle to accurately distinguish different types of radar signals in complex, dense, and dynamically changing radar signal environments, especially in environments with high signal parameter similarity, dynamic signal changes, and dense signals, where traditional methods are ineffective.
A three-level cascaded multi-task neural network structure, including PNet, JNet and ONet, is adopted to sort radar signals step by step through matrix analysis, sorting analysis, batch analysis and screening analysis. The complex nonlinear relationships and temporal characteristics of radar signals are automatically learned by deep learning models, reducing the reliance on human experience and prior knowledge.
It improves the accuracy and robustness of radar signal sorting, reduces incorrect sorting and omissions, adapts to more diverse radar signal characteristics, and achieves near real-time sorting processing speed.
Smart Images
Figure CN122017737A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of radar sorting technology, specifically to a radar model sorting method, device, and storage medium. Background Technology
[0002] In the fields of electronic warfare and radar reconnaissance, radar signal sorting is a crucial step. Because the modern battlefield environment contains a large number of different types and functions of radar systems, which may operate simultaneously or alternately, emitting pulse signals with similar parameters such as frequency, pulse width, and pulse repetition interval (PRI), this dense and complex signal environment presents a significant challenge to radar signal sorting.
[0003] Traditional radar signal sorting methods primarily rely on rule-based and template matching techniques, such as Time Difference of Arrival (TDOA), pulse repetition interval analysis (e.g., fixed PRI, parametric PRI, PRI jitter), and clustering algorithms (e.g., K-means, DBSCAN). While these methods can handle simple signal environments to some extent, they often perform poorly in the following situations: High similarity of signal parameters: Signals emitted by different radars may be very similar in terms of frequency, pulse width, pulse repetition interval, etc., making it difficult to distinguish them using traditional methods.
[0004] Dynamic changes in signal: Radars may employ anti-jamming techniques such as frequency hopping, pulse width jitter, and PRI jitter, causing signal parameters to change rapidly over time, rendering traditional static template-based methods ineffective.
[0005] Dense signal environment: A large number of signals arrive at the same time, which makes the time relationship unclear and makes it difficult to accurately calculate TDOA or perform pulse repetition interval analysis. Summary of the Invention
[0006] The technical problem to be solved by the present invention is to provide a radar model sorting method, device and storage medium to address the shortcomings of the prior art.
[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: A radar model sorting method, comprising the following steps: Import the radar signal PDW data sequence to be sorted, perform matrix analysis on the radar signal PDW data sequence, and obtain the radar signal PDW matrix to be processed. A sorting analysis model is constructed, and the PDW matrix of the radar signal to be processed is sorted and analyzed by the sorting analysis model to obtain multiple original radar signal PDW vectors. A batch analysis model is constructed, and signal batch analysis is performed on each of the original radar signal PDW vectors using the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors. A screening analysis model is constructed, and the original radar signal PDW vectors and batch analysis results are screened and analyzed using the screening analysis model. The analysis results are then used as radar model sorting results.
[0008] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: A radar model sorting device, comprising: The import module is used to import the radar signal PDW data sequence to be sorted; The matrix analysis module is used to perform matrix analysis on the radar signal PDW data sequence to obtain the radar signal PDW matrix to be processed. The sorting and analysis module is used to construct a sorting and analysis model, and to perform sorting and analysis on the radar signal PDW matrix to be processed through the sorting and analysis model to obtain multiple original radar signal PDW vectors. The signal batch analysis module is used to construct a batch analysis model, and to perform signal batch analysis on each of the original radar signal PDW vectors through the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors. The sorting result acquisition module is used to construct a screening analysis model, and to perform screening analysis on all the original radar signal PDW vectors and all the batch analysis results through the screening analysis model, and use the analysis results as the radar model sorting results.
[0009] Based on the radar model sorting method described above, the present invention also provides a radar model sorting system.
[0010] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a radar model sorting system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the radar model sorting method as described above.
[0011] Based on the radar model sorting method described above, the present invention also provides a computer-readable storage medium.
[0012] Another technical solution of the present invention to solve the above-mentioned technical problems is as follows: a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the radar model sorting method as described above.
[0013] The beneficial effects of this invention are as follows: The radar signal PDW matrix to be processed is obtained through matrix analysis of the radar signal PDW data sequence; the original radar signal PDW vector is obtained through sorting analysis of the radar signal PDW matrix to be processed using a sorting analysis model; batch analysis results are obtained through batch analysis of the original radar signal PDW vector using a batch analysis model; and the original radar signal PDW vector and batch analysis results are screened and analyzed using a filtering analysis model. The analysis results are then used as the radar model sorting results. This effectively reduces erroneous sorting and omissions, improves sorting accuracy, enhances robustness, and increases the speed of sorting processing. It can adapt to more diverse radar signal characteristics, reduces reliance on human experience and prior knowledge, and overcomes the misjudgment problem that may occur with a single model. Attached Figure Description
[0014] Figure 1 This is a flowchart illustrating the radar model sorting method provided in an embodiment of the present invention; Figure 2 This is a network structure diagram of the sorting analysis model of the radar model sorting method provided in the embodiments of the present invention; Figure 3 This is a network structure diagram of the batch analysis model of the radar model sorting method provided in the embodiments of the present invention; Figure 4 This is a network structure diagram of the screening analysis model of the radar model sorting method provided in the embodiments of the present invention; Figure 5 This is a block diagram of a radar model sorting device provided in an embodiment of the present invention. Detailed Implementation
[0015] The principles and features of the present invention are described below with reference to the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.
[0016] Figure 1 This is a flowchart illustrating a radar model sorting method provided in an embodiment of the present invention.
[0017] like Figure 1 As shown, a radar model selection method includes the following steps: S1: Import the radar signal PDW data sequence to be sorted, perform matrix analysis on the radar signal PDW data sequence, and obtain the radar signal PDW matrix to be processed; S2: Construct a sorting analysis model, and use the sorting analysis model to sort and analyze the radar signal PDW matrix to be processed to obtain multiple original radar signal PDW vectors; S3: Construct a batch analysis model, and perform signal batch analysis on each of the original radar signal PDW vectors using the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors; S4: Construct a screening analysis model, and use the screening analysis model to screen and analyze all the original radar signal PDW vectors and all the batch analysis results, and use the analysis results as the radar model sorting results.
[0018] It should be understood that the PDW data sequence to be sorted (i.e., radar signal PDW data sequence) is obtained, which contains parameter information of multiple PDWs.
[0019] Specifically, suppose the radar reconnaissance system collects a batch of PDW data (i.e., radar signal PDW data sequence), which contains 1024 pulses, and each pulse contains three parameters: carrier frequency, pulse width, and time of arrival.
[0020] In the above embodiments, the radar signal PDW matrix to be processed is obtained by matrix analysis of the radar signal PDW data sequence. The original radar signal PDW vector is obtained by sorting the radar signal PDW matrix to be processed through a sorting analysis model. The batch analysis results are obtained by batch analysis of the original radar signal PDW vector through a batch analysis model. The original radar signal PDW vector and the batch analysis results are filtered and analyzed by a screening analysis model. The analysis results are used as the radar model sorting results. This effectively reduces erroneous sorting and omissions, improves the accuracy of sorting, enhances robustness, and increases the speed of sorting processing. It can adapt to more diverse radar signal characteristics, reduces reliance on human experience and prior knowledge, and overcomes the misjudgment problem that may occur with a single model.
[0021] Optionally, as an embodiment of the present invention, the radar signal PDW data sequence includes a first radar signal carrier frequency, a first radar signal pulse width, and a first radar signal arrival time. The process of performing matrix analysis on the radar signal PDW data sequence to obtain the radar signal PDW matrix to be processed includes: The first radar signal carrier frequency, the first radar signal pulse width, and the first radar signal arrival time are vectorized respectively to obtain the first radar signal carrier frequency vector corresponding to the first radar signal carrier frequency, the first radar signal pulse width vector corresponding to the first radar signal pulse width, and the first radar signal arrival time vector corresponding to the first radar signal arrival time. The original radar signal PDW matrix is constructed by the first radar signal carrier frequency vector, the first radar signal pulse width vector, and the first radar signal arrival time vector. Determine whether the number of rows in the original radar signal PDW matrix is the preset number of rows. If yes, use the original radar signal PDW matrix as the radar signal PDW matrix to be processed. If no, fill the original radar signal PDW matrix according to the preset zero vector and use the filling result as the radar signal PDW matrix to be processed.
[0022] Understandably, the parameter information of each PDW (such as carrier frequency, pulse width, and arrival time) (i.e., the first radar signal carrier frequency, the first radar signal pulse width, and the first radar signal arrival time) is organized into a vector (i.e., the first radar signal carrier frequency vector, the first radar signal pulse width vector, and the first radar signal arrival time vector). The entire data sequence is organized into a matrix (i.e., the original radar signal PDW matrix). The number of rows in the matrix is fixed at 1024, representing the number of pulses; the number of columns is 3, corresponding to the three parameters of carrier frequency, pulse width, and arrival time, respectively. If the actual number of pulses is less than 1024, it is padded with zero vectors to form a 1024x3 input matrix (i.e., the radar signal PDW matrix to be processed).
[0023] In the above embodiments, matrix analysis is performed on the radar signal PDW data sequence to obtain the radar signal PDW matrix to be processed, which effectively reduces erroneous sorting and omissions, improves sorting accuracy, enhances robustness, and increases the speed of sorting processing.
[0024] Optionally, as an embodiment of the present invention, such as Figure 2 As shown, the sorting analysis model includes a first long short-term memory network and a second long short-term memory network. The process of sorting and analyzing the radar signal PDW matrix to be processed through the sorting analysis model to obtain multiple original radar signal PDW vectors includes: The radar signal PDW matrix to be processed is captured by the first long short-term memory network to obtain the radar signal PDW hidden state sequence. The second long short-term memory network is used to extract features from the radar signal PDW sequence to be processed, resulting in multiple original radar signal PDW vectors.
[0025] Understandably, the 1024x3 input matrix (i.e., the radar signal PDW matrix to be processed) is input into the PNet network (i.e., the first long short-term memory network and the second long short-term memory network). The PNet network structure is a two-layer long short-term memory network (LSTM) (i.e., the first long short-term memory network and the second long short-term memory network). The first LSTM (i.e., the first long short-term memory network) receives the input sequence (i.e., the radar signal PDW matrix to be processed), captures the dependence of pulse parameters in the time dimension, and outputs the hidden state sequence (i.e., the radar signal PDW hidden state sequence). The second LSTM (i.e., the second long short-term memory network) receives the output of the first LSTM (i.e., the radar signal PDW hidden state sequence), further extracts high-level features, and generates a 1024-dimensional output vector (i.e., the original radar signal PDW vector). The output dimension of PNet (i.e., the first long short-term memory network and the second long short-term memory network) is 1024-dimensional, and each dimension corresponds to an input PDW. Its value represents the label of the potential radar signal to which the PDW is assigned (i.e., the preliminary sorting result). The role of PNet is to generate all possible sorting results and provide a basis for subsequent judgment.
[0026] Specifically, the data of these 1024 PDWs (dimension 1024×3) (i.e., the radar signal PDW matrix to be processed) is input into the PNet model (i.e., the first long short-term memory network and the second long short-term memory network). After the 1024×3 input of the PNet (i.e., the radar signal PDW matrix to be processed) is processed by two layers of LSTM (i.e., the first long short-term memory network and the second long short-term memory network), a 1024-dimensional vector (i.e., the original radar signal PDW vector) is output. Assuming the output vector is [l1, l2, ..., l1024], each li is a value representing the sorting result label index of the potential radar signal to which the i-th PDW is assigned (e.g., 1, 2, 3... represent different radars).
[0027] In the above embodiments, the sorting analysis model is used to sort and analyze the PDW matrix of the radar signal to be processed, and multiple original radar signal PDW vectors are obtained. This can capture the dependence of pulse parameters in the time dimension and further extract high-level features, providing a basis for subsequent judgment.
[0028] Optionally, as an embodiment of the present invention, such as Figure 3 As shown, the batch analysis model includes a third long short-term memory network and a first fully connected layer. The process of performing batch analysis on each of the original radar signal PDW vectors using the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors includes: Each of the original radar signal PDW vectors is combined with any of the remaining original radar signal PDW vectors to obtain multiple pairs of original radar signal PDW vectors corresponding to each of the original radar signal PDW vectors. Extract from each of the original radar signal PDW vector pairs multiple second radar signal carrier frequencies and multiple second radar signal pulse widths corresponding to each of the original radar signal PDW vector pairs respectively; The carrier frequencies of the multiple second radar signals corresponding to each original radar signal PDW vector pair and the pulse widths of the multiple second radar signals corresponding to each original radar signal PDW vector pair are concatenated to obtain the concatenated radar signal PDW matrix corresponding to each original radar signal PDW vector pair. Determine whether the number of rows in the spliced radar signal PDW matrix is a preset number of rows. If yes, use the spliced radar signal PDW matrix as the processed radar signal PDW matrix. If no, fill the spliced radar signal PDW matrix according to a preset zero vector and use the filling result as the processed radar signal PDW matrix, thereby obtaining the processed radar signal PDW matrix corresponding to each original radar signal PDW vector pair. The third long short-term memory network is used to extract features from each of the processed radar signal PDW matrices to obtain the first feature-extracted radar signal PDW matrix corresponding to each of the original radar signal PDW vector pairs. The first fully connected layer maps each of the radar signal PDW matrices after the first feature extraction to obtain the first mapped radar signal PDW matrix corresponding to each of the original radar signal PDW vector pairs. The first confidence level corresponding to each original radar signal PDW vector pair is extracted from each of the first mapped radar signal PDW matrices; Determine whether the first confidence level is greater than or equal to a preset first confidence level threshold. If yes, use the first radar source batch result as the batch analysis result; otherwise, use the second radar source batch result as the batch analysis result, thereby obtaining the batch analysis result corresponding to each of the original radar signal PDW vectors.
[0029] Specifically, iterate through all the preliminary sorting results (i.e., the original radar signal PDW vectors) generated by PNet. For any two different preliminary sorting results (denoted as sorting result A and sorting result B) (i.e., the original radar signal PDW vectors), extract the frequency (i.e., the second radar signal carrier frequency) and pulse width (i.e., the second radar signal pulse width) of their corresponding PDWs, and form two 2048-dimensional vectors respectively (the frequency parameter (i.e., the second radar signal carrier frequency) and the pulse width parameter (i.e., the second radar signal pulse width) are concatenated, with the first 1024 dimensions being the frequency parameter and the last 1024 dimensions being the pulse width parameter). Concatenate these two vectors column by column to form a 2048x2 matrix (i.e., the concatenated radar signal PDW matrix). If the number of PDWs contained in sorting result A or sorting result B (i.e., the original radar signal PDW vector) is less than 1024, it is padded with zero vectors. This process is repeated to generate all possible combinations of sorting results and their corresponding 2048x2 matrices (i.e., the processed radar signal PDW matrices). Each 2048x2 matrix is fed into the second-level network JNet (i.e., the third long short-term memory network and the first fully connected layer). The JNet network structure consists of a 1-layer LSTM (i.e., the third long short-term memory network) connected to a 1-layer fully connected layer (i.e., the first fully connected layer). The first LSTM layer receives the 2048x2 matrix (i.e., the processed radar signal PDW matrix), and the second LSTM layer is connected to the output of the first LSTM layer to further extract the two sorting results. Finally, the fully connected layer receives the output of the second LSTM layer and generates a 2D output vector (i.e., the first mapped radar signal PDW matrix). The 2D output vector of the JNet network is used to determine whether sorting result A and sorting result B belong to the same radar signal batch. A threshold is set (i.e., the preset first confidence threshold). If the first element of the output vector (i.e., the first confidence) is greater than the threshold (i.e., the preset first confidence threshold), then sorting result A and sorting result B are determined to belong to the same batch (i.e., the first radar source batch result); otherwise, they are determined not to belong to the same batch (i.e., the second radar source batch result). JNet traverses and judges all possible tag pairs and outputs the same batch judgment result (i.e., the batch analysis result).
[0030] Understandably, PNet outputs 6 sorting result labels (i.e., the original radar signal PDW vectors). Now, it is necessary to determine which labels correspond to PDWs belonging to the same radar. JNet needs to process all possible label pairs (i.e., original radar signal PDW vector pairs). For example, comparing sorting result labels 1 and 2, extracting the frequency (i.e., the carrier frequency of the second radar signal) and pulse width parameters (i.e., the pulse width of the second radar signal) and concatenating them into a 2048×2 input (i.e., the concatenated radar signal PDW matrix), and feeding it into JNet (i.e., the third long short-term memory network and the first fully connected layer). JNet outputs a 2D vector (i.e., the first mapped radar signal PDW matrix), such as [0.85, 0.15], indicating that sorting result labels 1 and 2 have an 85% probability of belonging to the same batch. Similarly, JNet will process all combinations (1, 3), (1, 4), ..., (5, 6), etc. (a total of 6*5 / 2 pairs). Based on the output of JNet, a threshold (such as 0.7) is set (i.e., the preset first confidence threshold), and label pairs with a confidence level (i.e., the first confidence level) higher than the threshold (i.e., the preset first confidence threshold) are judged as the same batch.
[0031] In the above embodiments, the batch analysis model is used to perform signal batch analysis on each original radar signal PDW vector to obtain batch analysis results, which improves the speed of sorting and processing, can adapt to more diverse radar signal characteristics, reduces the reliance on human experience and prior knowledge, and overcomes the misjudgment problem that may occur with a single model.
[0032] Optionally, as an embodiment of the present invention, such as Figure 4 As shown, the screening analysis model includes multiple convolutional layers and a second fully connected layer. The process of screening and analyzing all the original radar signal PDW vectors and all the batch analysis results through the screening analysis model, and using the analysis results as the radar model selection results, includes: The carrier frequency of the third radar signal corresponding to each of the original radar signal PDW vectors, the pulse width of the third radar signal corresponding to each of the original radar signal PDW vectors, and the arrival time of the second radar signal corresponding to each of the original radar signal PDW vectors are extracted respectively. The radar signal PDW matrix corresponding to each of the original radar signal PDW vectors is constructed by the carrier frequency of the third radar signal corresponding to each of the original radar signal PDW vectors, the pulse width of the third radar signal corresponding to each of the original radar signal PDW vectors, and the arrival time of the second radar signal corresponding to each of the original radar signal PDW vectors. Determine whether the number of rows in the radar signal PDW matrix to be judged is a preset number of rows. If yes, use the radar signal PDW matrix to be judged as the judged radar signal PDW matrix. If no, fill the radar signal PDW matrix to be judged according to the preset zero vector and use the filling result as the judged radar signal PDW matrix, thereby obtaining the judged radar signal PDW matrix corresponding to each of the original radar signal PDW vectors. By performing feature extraction on each of the judged radar signal PDW matrices through multiple convolutional layers, a second feature-extracted radar signal PDW matrix corresponding to each of the original radar signal PDW vectors is obtained. The second fully connected layer is used to map each of the radar signal PDW matrices after the second feature extraction to obtain the second mapped radar signal PDW matrix corresponding to each of the original radar signal PDW vectors. The second confidence level corresponding to the PDW vector of each original radar signal is extracted from each of the second mapped radar signal PDW matrices; If the second confidence level is greater than or equal to the preset second confidence level threshold, the batch analysis result is used as the screening analysis result, thereby obtaining the screening analysis result corresponding to each of the original radar signal PDW vectors, and all the screening analysis results are used as the radar model sorting result.
[0033] Specifically, based on the batch judgment results of JNet, signals belonging to the same batch are merged to form several candidate sorting results (i.e., batch analysis results). Each candidate result contains a set of PDWs (PDW matrices) judged to be in the same batch and their labels. Each candidate sorting result (i.e., the radar signal PDW matrix after judgment) is input into the third-level network ONet (i.e., multiple convolutional layers and a second fully connected layer). ONet (i.e., multiple convolutional layers and a second fully connected layer) adopts a neural network structure of 4 convolutional layers followed by 1 fully connected layer (i.e., the second fully connected layer). The dimension of the input data is 1024×3, and the number of rows is fixed at 1024, representing the number of pulses. If the actual number of pulses is less than 1024, The zero vector is used for padding; the number of columns is 3, corresponding to the three parameters of carrier frequency, pulse width, and time of arrival. The output dimension of ONet is 2-dimensional (i.e., the radar signal PDW matrix after the second mapping), which represents the confidence or probability of whether the candidate sorting result is the correct final sorting result. ONet verifies all candidate sorting results and selects the sorting results judged as correct as the final radar signal sorting output (i.e., the screening analysis result). All candidate sorting results judged as correct by ONet are selected. These candidate sorting results are the final radar signal sorting results (i.e., the screening analysis result). Each correct candidate sorting result represents an independent radar signal source.
[0034] It should be understood that, based on the judgment results of JNet (i.e., batch analysis results), PDWs with the same sorting result label are merged. For example, if 1, 2, and 5 are determined to be in the same batch, the corresponding PDWs (PDWs of sorting results labeled 1, 2, and 5 in the original input data) are merged into a single candidate sorting result. The three parameters corresponding to the candidate sorting result—carrier frequency (i.e., the carrier frequency of the third radar signal), pulse width (i.e., the pulse width of the third radar signal), and arrival time (i.e., the arrival time of the second radar signal)—are used to form a vector representing the batch of signals (i.e., the PDW matrix of the radar signals to be judged). The dimension is adjusted to 1024×3 and input to ONet (i.e., multiple convolutional layers and a second fully connected layer). ONet outputs a 2D vector (i.e., the second mapped radar signal PDW matrix), such as [0.95, 0.05], indicating that this candidate sorting result has a 95% probability of being correct. Otherwise, the candidate sorting result is discarded. This operation is performed on all candidate sorting results. Finally, all candidate sorting results correctly judged by ONet (i.e., the screening analysis results) are selected as the final radar signal sorting results.
[0035] In the above embodiments, the screening analysis model is used to screen and analyze all original radar signal PDW vectors and all batch analysis results, and the analysis results are used as radar model sorting results. This improves the speed of sorting and processing, can adapt to more diverse radar signal characteristics, reduces reliance on human experience and prior knowledge, and overcomes the misjudgment problem that may occur with a single model.
[0036] Alternatively, as another embodiment of the present invention, the present invention belongs to the field of UAV reconnaissance technology, specifically relating to a radar model selection method based on channelization processing and a two-level deep learning network, which is particularly suitable for fast and accurate radar model selection based on IQ data collected by a high sampling rate broadband receiver when UAV computing resources are limited.
[0037] Optionally, as another embodiment of the present invention, addressing the problems of high computational complexity and long processing time resulting from directly processing high-sampling-rate raw IQ data in the prior art, making real-time identification difficult under the limited resources of UAVs, and the difficulty of effectively identifying broadband interference with a single sub-channel, the present invention aims to provide a radar model selection method based on channelized data. This method aims to reduce computational complexity and improve processing speed under the condition of limited UAV computing resources, enabling more comprehensive and accurate identification of various communication interference types (single-tone interference, multi-tone interference, linear frequency modulation, BPSK, QPSK, comb spectrum interference, narrowband noise interference, and broadband noise interference, etc.), to meet the actual needs of UAV reconnaissance and countermeasures.
[0038] Alternatively, as another embodiment of the present invention, the present invention can more comprehensively analyze radar signal characteristics and significantly improve the accuracy, completeness and robustness of radar signal sorting through multi-stage collaborative processing and verification, and is especially suitable for complex, dense and dynamically changing radar signal environments.
[0039] Alternatively, as another embodiment of the present invention, compared with the prior art, the present invention has the following significant advantages: 1. High sorting accuracy: Through a three-level cascaded multi-task network structure, PNet is responsible for the initial sorting in breadth, JNet is responsible for batch judgment, and ONet is responsible for the final result verification. The progressive process effectively reduces incorrect sorting and omissions, and improves the accuracy of sorting.
[0040] 2. Strong robustness: Deep learning models can automatically learn complex nonlinear relationships and temporal characteristics in PDW data, and have better adaptability to small changes in signal parameters, frequency hopping, pulse width jitter, etc., and perform better in complex and dense signal environments.
[0041] 3. Multi-stage verification mechanism: JNet's batch judgment and ONet's result verification provide dual protection, ensuring the reliability of the final sorting results and overcoming the misjudgment problem that may occur with a single model.
[0042] 4. High level of intelligence: Compared with traditional rule-based methods, this invention utilizes the self-learning ability of neural networks, which can adapt to more diverse radar signal characteristics and reduce reliance on human experience and prior knowledge.
[0043] 5. Potential high efficiency: Although the model structure is relatively complex, by optimizing the network design and training, it is expected to achieve near real-time sorting speed while ensuring accuracy.
[0044] Alternatively, as another embodiment of the present invention, the protection points of the present invention are as follows: Multi-task cascaded structure: The core innovation lies in the use of three cascaded multi-task neural networks (PNet, JNet, ONet), forming a well-defined and progressively advancing processing flow. This structured design itself is an important safeguard.
[0045] PNet's pre-generation capability: The first-stage PNet utilizes an LSTM network to generate all possible sorting results from the PDW data. This solves the problem of traditional methods potentially missing combinations, providing a rich foundation for subsequent accurate screening. Its ability to generate all possible results is key.
[0046] JNet's batch decision-making: The second-level JNet focuses on determining which candidate sorting results generated by PNet are "in the same batch," i.e., from the same radar pulse sequence. This specialized batch decision-making task design is the key difference between this invention and general signal processing methods.
[0047] Final verification by ONet: The third-level ONet is responsible for the final verification and optimization of the batch of results selected by JNet, ensuring the accuracy of the sorting results. This three-stage stepwise verification mechanism is the technical means by which this invention improves accuracy.
[0048] Overall synergy: Although the three modules each undertake different tasks, they work together to achieve a complete process from coarse-grained generation of sorting candidates to fine-grained judgment and confirmation through cascading. This overall collaborative approach and the resulting performance improvement are the main protection points.
[0049] In short, the protection point of this invention is: a method for radar signal sorting using a three-level cascaded multi-task neural network structure, especially the unique collaborative processing flow of PNet generating all possible sorting results, JNet performing batch judgment, and ONet performing final confirmation, and the resulting high accuracy and high coverage sorting effect.
[0050] Figure 5 This is a module block diagram of a radar model sorting device provided in an embodiment of the present invention.
[0051] Alternatively, as another embodiment of the present invention, such as Figure 5 As shown, a radar model sorting device includes: The import module is used to import the radar signal PDW data sequence to be sorted; The matrix analysis module is used to perform matrix analysis on the radar signal PDW data sequence to obtain the radar signal PDW matrix to be processed. The sorting and analysis module is used to construct a sorting and analysis model, and to perform sorting and analysis on the radar signal PDW matrix to be processed through the sorting and analysis model to obtain multiple original radar signal PDW vectors. The signal batch analysis module is used to construct a batch analysis model, and to perform signal batch analysis on each of the original radar signal PDW vectors through the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors. The sorting result acquisition module is used to construct a screening analysis model, and to perform screening analysis on all the original radar signal PDW vectors and all the batch analysis results through the screening analysis model, and use the analysis results as the radar model sorting results.
[0052] Optionally, as an embodiment of the present invention, the radar signal PDW data sequence includes a first radar signal carrier frequency, a first radar signal pulse width, and a first radar signal arrival time, and the matrix analysis module is specifically used for: The first radar signal carrier frequency, the first radar signal pulse width, and the first radar signal arrival time are vectorized respectively to obtain the first radar signal carrier frequency vector corresponding to the first radar signal carrier frequency, the first radar signal pulse width vector corresponding to the first radar signal pulse width, and the first radar signal arrival time vector corresponding to the first radar signal arrival time. The original radar signal PDW matrix is constructed by the first radar signal carrier frequency vector, the first radar signal pulse width vector, and the first radar signal arrival time vector. Determine whether the number of rows in the original radar signal PDW matrix is the preset number of rows. If yes, use the original radar signal PDW matrix as the radar signal PDW matrix to be processed. If no, fill the original radar signal PDW matrix according to the preset zero vector and use the filling result as the radar signal PDW matrix to be processed.
[0053] Optionally, as an embodiment of the present invention, the sorting analysis model includes a first long short-term memory network and a second long short-term memory network, and the sorting analysis module is specifically used for: The radar signal PDW matrix to be processed is captured by the first long short-term memory network to obtain the radar signal PDW hidden state sequence. The second long short-term memory network is used to extract features from the radar signal PDW sequence to be processed, resulting in multiple original radar signal PDW vectors.
[0054] Optionally, another embodiment of the present invention provides a radar model sorting system, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the radar model sorting method as described above. This system can be a computer or similar system.
[0055] Optionally, another embodiment of the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the radar model sorting method as described above.
[0056] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.
[0057] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described apparatus and unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.
[0058] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed.
[0059] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiments of the present invention, depending on actual needs.
[0060] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.
[0061] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0062] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A radar model selection method, characterized in that, Includes the following steps: Import the radar signal PDW data sequence to be sorted, perform matrix analysis on the radar signal PDW data sequence, and obtain the radar signal PDW matrix to be processed. A sorting analysis model is constructed, and the PDW matrix of the radar signal to be processed is sorted and analyzed by the sorting analysis model to obtain multiple original radar signal PDW vectors. A batch analysis model is constructed, and signal batch analysis is performed on each of the original radar signal PDW vectors using the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors. A screening analysis model is constructed, and the original radar signal PDW vectors and batch analysis results are screened and analyzed using the screening analysis model. The analysis results are then used as radar model sorting results.
2. The radar model sorting method according to claim 1, characterized in that, The radar signal PDW data sequence includes a first radar signal carrier frequency, a first radar signal pulse width, and a first radar signal arrival time. The process of performing matrix analysis on the radar signal PDW data sequence to obtain the radar signal PDW matrix to be processed includes: The first radar signal carrier frequency, the first radar signal pulse width, and the first radar signal arrival time are vectorized respectively to obtain the first radar signal carrier frequency vector corresponding to the first radar signal carrier frequency, the first radar signal pulse width vector corresponding to the first radar signal pulse width, and the first radar signal arrival time vector corresponding to the first radar signal arrival time. The original radar signal PDW matrix is constructed by the first radar signal carrier frequency vector, the first radar signal pulse width vector, and the first radar signal arrival time vector. Determine whether the number of rows in the original radar signal PDW matrix is the preset number of rows. If yes, use the original radar signal PDW matrix as the radar signal PDW matrix to be processed. If no, fill the original radar signal PDW matrix according to the preset zero vector and use the filling result as the radar signal PDW matrix to be processed.
3. The radar model sorting method according to claim 1, characterized in that, The sorting analysis model includes a first long short-term memory network and a second long short-term memory network. The process of sorting and analyzing the radar signal PDW matrix to be processed through the sorting analysis model to obtain multiple original radar signal PDW vectors includes: The radar signal PDW matrix to be processed is captured by the first long short-term memory network to obtain the radar signal PDW hidden state sequence. The second long short-term memory network is used to extract features from the radar signal PDW sequence to be processed, resulting in multiple original radar signal PDW vectors.
4. The radar model sorting method according to claim 1, characterized in that, The batch analysis model includes a third long short-term memory network and a first fully connected layer. The process of performing batch analysis on each of the original radar signal PDW vectors using the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors includes: Each of the original radar signal PDW vectors is combined with any of the remaining original radar signal PDW vectors to obtain multiple pairs of original radar signal PDW vectors corresponding to each of the original radar signal PDW vectors. Extract from each of the original radar signal PDW vector pairs multiple second radar signal carrier frequencies and multiple second radar signal pulse widths corresponding to each of the original radar signal PDW vector pairs respectively; The carrier frequencies of the multiple second radar signals corresponding to each original radar signal PDW vector pair and the pulse widths of the multiple second radar signals corresponding to each original radar signal PDW vector pair are concatenated to obtain the concatenated radar signal PDW matrix corresponding to each original radar signal PDW vector pair. Determine whether the number of rows in the spliced radar signal PDW matrix is a preset number of rows. If yes, use the spliced radar signal PDW matrix as the processed radar signal PDW matrix. If no, fill the spliced radar signal PDW matrix according to a preset zero vector and use the filling result as the processed radar signal PDW matrix, thereby obtaining the processed radar signal PDW matrix corresponding to each original radar signal PDW vector pair. The third long short-term memory network is used to extract features from each of the processed radar signal PDW matrices to obtain the first feature-extracted radar signal PDW matrix corresponding to each of the original radar signal PDW vector pairs. The first fully connected layer maps each of the radar signal PDW matrices after the first feature extraction to obtain the first mapped radar signal PDW matrix corresponding to each of the original radar signal PDW vector pairs. The first confidence level corresponding to each original radar signal PDW vector pair is extracted from each of the first mapped radar signal PDW matrices; Determine whether the first confidence level is greater than or equal to a preset first confidence level threshold. If yes, use the first radar source batch result as the batch analysis result; otherwise, use the second radar source batch result as the batch analysis result, thereby obtaining the batch analysis result corresponding to each of the original radar signal PDW vectors.
5. The radar model sorting method according to claim 1, characterized in that, The screening analysis model includes multiple convolutional layers and a second fully connected layer. The process of using the screening analysis model to perform screening analysis on all the original radar signal PDW vectors and all the batch analysis results, and using the analysis results as the radar model selection results, includes: The carrier frequency of the third radar signal corresponding to each of the original radar signal PDW vectors, the pulse width of the third radar signal corresponding to each of the original radar signal PDW vectors, and the arrival time of the second radar signal corresponding to each of the original radar signal PDW vectors are extracted respectively. The radar signal PDW matrix corresponding to each of the original radar signal PDW vectors is constructed by the carrier frequency of the third radar signal corresponding to each of the original radar signal PDW vectors, the pulse width of the third radar signal corresponding to each of the original radar signal PDW vectors, and the arrival time of the second radar signal corresponding to each of the original radar signal PDW vectors. Determine whether the number of rows in the radar signal PDW matrix to be judged is a preset number of rows. If yes, use the radar signal PDW matrix to be judged as the judged radar signal PDW matrix. If no, fill the radar signal PDW matrix to be judged according to the preset zero vector and use the filling result as the judged radar signal PDW matrix, thereby obtaining the judged radar signal PDW matrix corresponding to each of the original radar signal PDW vectors. By performing feature extraction on each of the judged radar signal PDW matrices through multiple convolutional layers, a second feature-extracted radar signal PDW matrix corresponding to each of the original radar signal PDW vectors is obtained. The second fully connected layer is used to map each of the radar signal PDW matrices after the second feature extraction to obtain the second mapped radar signal PDW matrix corresponding to each of the original radar signal PDW vectors. The second confidence level corresponding to the PDW vector of each original radar signal is extracted from each of the second mapped radar signal PDW matrices; If the second confidence level is greater than or equal to the preset second confidence level threshold, the batch analysis result is used as the screening analysis result, thereby obtaining the screening analysis result corresponding to each of the original radar signal PDW vectors, and all the screening analysis results are used as the radar model sorting result.
6. A radar model sorting device, characterized in that, include: The import module is used to import the radar signal PDW data sequence to be sorted; The matrix analysis module is used to perform matrix analysis on the radar signal PDW data sequence to obtain the radar signal PDW matrix to be processed. The sorting and analysis module is used to construct a sorting and analysis model, and to perform sorting and analysis on the radar signal PDW matrix to be processed through the sorting and analysis model to obtain multiple original radar signal PDW vectors. The signal batch analysis module is used to construct a batch analysis model, and to perform signal batch analysis on each of the original radar signal PDW vectors through the batch analysis model to obtain the batch analysis results corresponding to each of the original radar signal PDW vectors. The sorting result acquisition module is used to construct a screening analysis model, and to perform screening analysis on all the original radar signal PDW vectors and all the batch analysis results through the screening analysis model, and use the analysis results as the radar model sorting results.
7. The radar model sorting device according to claim 6, characterized in that, The radar signal PDW data sequence includes the first radar signal carrier frequency, the first radar signal pulse width, and the first radar signal arrival time. The matrix analysis module is specifically used for: The first radar signal carrier frequency, the first radar signal pulse width, and the first radar signal arrival time are vectorized respectively to obtain the first radar signal carrier frequency vector corresponding to the first radar signal carrier frequency, the first radar signal pulse width vector corresponding to the first radar signal pulse width, and the first radar signal arrival time vector corresponding to the first radar signal arrival time. The original radar signal PDW matrix is constructed by the first radar signal carrier frequency vector, the first radar signal pulse width vector, and the first radar signal arrival time vector. Determine whether the number of rows in the original radar signal PDW matrix is the preset number of rows. If yes, use the original radar signal PDW matrix as the radar signal PDW matrix to be processed. If no, fill the original radar signal PDW matrix according to the preset zero vector and use the filling result as the radar signal PDW matrix to be processed.
8. The radar model sorting device according to claim 6, characterized in that, The sorting analysis model includes a first long short-term memory network and a second long short-term memory network. The sorting analysis module is specifically used for: The radar signal PDW matrix to be processed is captured by the first long short-term memory network to obtain the radar signal PDW hidden state sequence. The second long short-term memory network is used to extract features from the radar signal PDW sequence to be processed, resulting in multiple original radar signal PDW vectors.
9. A radar model sorting device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the radar model sorting method as described in any one of claims 1 to 5.
10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the radar model sorting method as described in any one of claims 1 to 5.