A radar operating mode recognition method and system based on a drone cluster

Abstract

The application belongs to the field of electronic countermeasures, and specifically discloses a radar working mode recognition method and system based on a UAV cluster, which comprises the following steps: dividing space into small spaces in units of wave positions, extracting PDW flow information from the UAV reconnaissance pulse signals in each small space after PDW extraction; inputting the PDW flow information into a Bi-LSTM neural network for first feature extraction, and into a ResNet18 network for second feature extraction; combining the first feature and the second feature to obtain the radar working mode recognized by each UAV through an MLP network; and fusing the radar working modes recognized by all the UAVs through a DS theory to obtain the working mode of a multifunction phased array radar. The application overcomes the blind area and vulnerability problem of single-machine single-point reconnaissance, and significantly improves the recognition rate, anti-interference resilience and decision confidence of the UAV in identifying the radar working mode under severe conditions such as strong interference and partial node failure.

Description

Technical Field

[0001] This application belongs to the field of electronic warfare, and more specifically, relates to a radar operating mode recognition method and system based on unmanned aerial vehicle (UAV) swarms. Background Technology

[0002] Radar operating mode recognition, as a core component of electronic reconnaissance, directly impacts the effectiveness of battlefield situational awareness and threat assessment. However, with the development of modern radar technology towards complex modes such as waveform agility and adaptive scanning, traditional single-platform reconnaissance-based recognition methods have revealed numerous limitations: First, the limited perspective makes single-point reconnaissance susceptible to terrain obstruction and beam scanning, leading to missed signals; second, the system faces the risk of "single-point failure," as interference or destruction of the main reconnaissance platform paralyzes the entire reconnaissance system; third, the "single source" causes a sharp drop in reliability when facing low signal-to-noise ratio signals or deceptive interference, resulting in high rates of misjudgment and missed judgment. Although existing research has proposed various recognition algorithms based on pulse descriptor word (PDW) statistical features, intra-pulse analysis, and even deep learning models, their essence remains within the single-point observation paradigm, making them ill-suited to the complex and intense modern electromagnetic spectrum warfare environment. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the purpose of this application is to provide a radar operating mode recognition method and system based on UAV swarms, aiming to solve the problem that traditional single-platform reconnaissance-based radar operating mode recognition methods are difficult to adapt to complex and intense electromagnetic spectrum confrontation environments.

[0004] The first aspect of this application relates to a radar operating mode recognition method based on a swarm of unmanned aerial vehicles (UAVs), wherein the UAV swarm surrounds a multi-functional phased array radar at different wave positions, and includes the following steps: Step S1: Divide the space into several small spaces in units of wave positions, and extract several PDW stream information from the UAV reconnaissance pulse signal in each small space by performing PDW extraction. Step S2: Input the PDW stream information into the Bi-LSTM neural network for temporal first feature extraction, and then input it into the ResNet18 network for spatial second feature extraction. After merging the first and second features, pass them through the MLP network to obtain the radar operating mode identified by each UAV. Then, fuse all the radar operating modes identified by all UAVs through DS theory to obtain the operating mode of the multi-functional phased array radar.

[0005] More preferably, the training methods for Bi-LSTM and ResNet18 networks are as follows: A dataset for recognizing the operating modes of a multi-functional phased array radar was established, which includes: the operating modes of the multi-functional phased array radar and their corresponding operating mode parameters, as well as the pulse signals of UAV swarm reconnaissance. The pulse signals from UAV swarm reconnaissance were used as inputs to the Bi-LSTM neural network and the ResNet18 network, and the operating modes of the multi-function phased array radar and their corresponding operating mode parameters were used as outputs to train the Bi-LSTM and ResNet18 networks.

[0006] More preferably, the phased array radar has the following operating modes: search mode, tracking mode, and composite mode; the search mode includes velocity search mode and search-while-ranging mode; the tracking mode includes single-target tracking mode; and the composite mode includes scan-while-tracking mode and search-plus-tracking mode.

[0007] More preferably, the method for acquiring PDW stream information is as follows: based on the transmitted pulse of the multi-function phased array radar, the UAV reconnaissance pulse signal in each small space is acquired, and PDW stream information is obtained after PDW extraction. Each PDW stream information includes arrival time, signal amplitude, pulse width, carrier frequency and intra-pulse modulation.

[0008] More preferably, the operation of the Bi-LSTM neural network is as follows:

[0009]

[0010]

[0011]

[0012]

[0013] in, , and These are the forget gate, input gate, and output gate. All three gates are controlled by the current input. Output at the previous moment Controlled The unit consists of a forget gate that controls the historical state. Enter the current state Opening degree, input gate control and Enter Opening degree, output gate control Enter the current output The degree of opening.

[0014] More preferably, the ResNet network's basic architecture is a residual block, in which the input propagates forward faster through cross-layer data paths. The residual block initially contains two identical 3... One convolutional layer, followed by a batch normalization layer and a ReLU activation function; then, by using a cross-layer data path, skipping two convolution operations, the input is directly added before the final ReLU activation function.

[0015] The second aspect of this application relates to a radar operating mode recognition system based on a drone swarm, comprising: a PDW stream information acquisition module and an operating mode recognition module; The PDW stream information acquisition module is used to divide the space into several small spaces in units of wave positions, and to extract several PDW stream information from the UAV reconnaissance pulse signals in each small space after PDW extraction. The working mode recognition module is used to input PDW stream information into Bi-LSTM neural network for temporal first feature extraction, and then input it into ResNet18 network for spatial second feature extraction. The first and second features are merged and then passed through MLP network to obtain the radar working mode of each UAV. The radar working modes of all UAVs are fused through DS theory to obtain the working mode of multi-function phased array radar.

[0016] More preferably, the phased array radar has the following operating modes: search mode, tracking mode, and composite mode; the search mode includes velocity search mode and search-while-ranging mode; the tracking mode includes single-target tracking mode; and the composite mode includes scan-while-tracking mode and search-plus-tracking mode.

[0017] More preferably, the radar operating mode recognition system further includes a network training module, which includes a training dataset acquisition unit and a training unit; The training dataset acquisition unit is used to establish a multi-functional phased array radar operating mode recognition dataset, which includes: the operating modes of the multi-functional phased array radar and their corresponding operating mode parameters, as well as UAV swarm reconnaissance pulse signals. The training unit is used to train the Bi-LSTM and ResNet18 networks by using the UAV swarm reconnaissance pulse signals as inputs and the operating modes of the multi-function phased array radar and their corresponding operating mode parameters as outputs.

[0018] More preferably, the method for acquiring PDW stream information is as follows: based on the transmitted pulse of the multi-function phased array radar, the UAV reconnaissance pulse signal in each small space is acquired, and PDW stream information is obtained after PDW extraction. Each PDW stream information includes arrival time, signal amplitude, pulse width, carrier frequency and intra-pulse modulation.

[0019] More preferably, the operation of the Bi-LSTM neural network is as follows:

[0020]

[0021]

[0022]

[0023]

[0024] in, , and These are the forget gate, input gate, and output gate. All three gates are controlled by the current input. Output at the previous moment Controlled The unit consists of a forget gate that controls the historical state. Enter the current state Opening degree, input gate control and Enter Opening degree, output gate control Enter the current output The degree of opening.

[0025] More preferably, the basic architecture of ResNet is a residual block, in which the input propagates forward faster through cross-layer data paths. The residual block initially contains two identical 3... One convolutional layer, followed by a batch normalization layer and a ReLU activation function; then, by using a cross-layer data path, skipping two convolution operations, the input is directly added before the final ReLU activation function.

[0026] Thirdly, this application provides an electronic device, comprising: at least one memory for storing a program; and at least one processor for executing the program stored in the memory, wherein when the program stored in the memory is executed, the processor is configured to execute the method described in the first aspect or any possible implementation thereof.

[0027] Fourthly, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0028] Fifthly, this application provides a computer program product that, when run on a processor, causes the processor to perform the method described in the first aspect or any possible implementation thereof.

[0029] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0030] Overall, the technical solutions conceived in this application have the following beneficial effects compared with the prior art: This application provides a radar operating mode recognition method and system based on UAV swarms. By deploying multiple UAVs to form a distributed reconnaissance network, radar signals are synchronously acquired from different spatial orientations. Each UAV node first uses a hybrid model combining Long Short-Term Memory (LSTM) and ResNet to intelligently identify its local PDW sequence, fully leveraging the advantages of LSTM in processing sequence data and ResNet in extracting deep features. Then, the recognition results and their confidence levels output by each node are uploaded as evidence to a fusion center for composite decision-making using DS evidence theory. The innovation of this architecture lies in integrating edge computing theory into collaborative reconnaissance. It not only forms a "global view" through multi-view observation, fundamentally overcoming the blind spots and vulnerabilities of single-point reconnaissance, but also utilizes the powerful processing capabilities of DS theory for uncertainty and conflicting evidence, achieving interaction and cross-validation of multi-source information. This significantly improves the recognition rate, anti-interference resilience, and decision confidence of UAV swarms under harsh conditions such as strong interference and partial node failure, providing a valuable exploration for the future development of distributed electronic reconnaissance technology. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of a drone swarm provided in an embodiment of this application.

[0032] Figure 2 This is the intent of the BiLSTM-ResNet18-DS algorithm provided in the embodiments of this application.

[0033] Figure 3 This is an RNN network structure diagram provided in the embodiments of this application.

[0034] Figure 4 This is an LSTM network structure diagram provided in the embodiments of this application.

[0035] Figure 5 This is a schematic diagram of the Bi-LSTM operation process provided in the embodiments of this application.

[0036] Figure 6 This is a residual block structure diagram provided in the embodiments of this application.

[0037] Figure 7 This is a ResNet18 network structure diagram provided in the embodiments of this application.

[0038] Figure 8(a) is a schematic diagram showing the relationship between the number of learning rounds and the training set loss provided in the embodiments of this application.

[0039] Figure 8(b) is a schematic diagram showing the relationship between the number of learning rounds and the accuracy of the training set provided in the embodiments of this application.

[0040] Figure 8(c) is a schematic diagram showing the relationship between the number of learning rounds and the accuracy of the validation set provided in the embodiments of this application.

[0041] Figure 9 This is a test result confusion matrix diagram provided in the embodiments of this application. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0043] In this application, the term "and / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A existing alone, A and B existing simultaneously, and B existing alone. In this application, the symbol " / " indicates that the related objects are in an "or" relationship, for example, A / B means A or B.

[0044] In this application, the terms “first” and “second” are used to distinguish different objects, rather than to describe a specific order of objects.

[0045] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0046] In the description of the embodiments in this application, unless otherwise stated, "multiple" means two or more.

[0047] This application provides a radar operating mode recognition method and system based on UAV swarms. By deploying multiple UAVs to form a distributed reconnaissance network, radar signals are synchronously acquired from different spatial orientations. Each UAV node first uses a hybrid model combining Long Short-Term Memory (LSTM) and ResNet to intelligently identify its local PDW sequence, fully leveraging the advantages of LSTM in processing sequence data and ResNet in extracting deep features. Then, the recognition results and their confidence levels output by each node are uploaded as evidence to a fusion center for composite decision-making using DS evidence theory. The innovation of this architecture lies in integrating edge computing theory into collaborative reconnaissance. It not only forms a "global view" through multi-view observation, fundamentally overcoming the blind spots and vulnerabilities of single-point reconnaissance, but also utilizes the powerful processing capabilities of DS theory for uncertainty and conflicting evidence, achieving interaction and cross-validation of multi-source information. This significantly improves the recognition rate, anti-interference resilience, and decision confidence of UAV swarms under harsh conditions such as strong interference and partial node failure, providing a valuable exploration for the future development of distributed electronic reconnaissance technology.

[0048] The embodiments of this application are described below with reference to the accompanying drawings.

[0049] This application provides a radar operating mode recognition method based on UAV swarms, specifically including the following steps: Step S1: Establish a dataset for recognizing the operating modes of a multi-functional phased array radar. Step S1.1: Introduction to the typical operating modes of a multi-functional phased array radar: This application considers the operating modes of a multi-functional phased array radar. In a multi-functional phased array radar, the formation of synthetic beams is one of its core characteristics, allowing for flexible configuration according to different tactical intentions and application scenarios. Based on this characteristic, the operating modes of a phased array radar can be divided into three main categories: search mode, tracking mode, and composite mode. The search mode includes velocity search mode and search-while-ranging mode; the tracking mode includes single-target tracking; and the composite mode includes scan-while-tracking mode and search-plus-tracking mode. Step S1.1.1: Speed ​​Search Mode Velocity Search (VS) is a long-range search and detection mode primarily used to search for high-speed moving targets within a specific airspace. This mode is commonly used in the early stages of target detection. Its signal typically employs a high Pulse Repetition Frequency (PRF) because a high PRF not only generates more pulses per unit time, making the radar system's velocity measurement more accurate, but also provides a clutter-free region in the frequency domain, allowing detection at greater distances than with clutter. From a reconnaissance and defense perspective, while it provides target azimuth and velocity information, it does not provide range information, making it unsuitable for direct use with precision-guided weapon systems; its threat level is relatively low. VS mode typically uses a line-by-line scanning method, with an azimuth search range of: , , .

[0050] Step 1.1.2: Search-while-measuring mode Range While Search (RWS) mode is the most common operating mode for airborne fire control radars. It is primarily used to search for targets within a specific airspace, while also providing a rough estimate of target range, azimuth, and altitude, thus mitigating the ranging ambiguity problem in VS mode to some extent. This mode has two sub-modes: High Range While Search (HRWS) mode and Medium Range While Search (MRWS) mode. Search (MRWS) mode; Compared to VS mode, HRWS mode is also used for long-range search and detection, but HRWS mode has a lower PRF, sacrificing velocity range and accuracy, but improving ranging range and accuracy. However, the ranging accuracy of this mode is still not high, and its application is relatively limited. Compared to HRWS mode, MRWS mode uses a medium PRF, improving ranging range and accuracy, but introducing velocity ambiguity. This is usually solved by changing the PRF group. However, changing the PRF group reduces the number of pulses that can be coherently processed for one wavelet within the same dwell time, resulting in a decrease in coherent gain and a decrease in radar detection range. Therefore, RWS mode is usually used in place of HRWS and MRWS modes to compensate for the detection capability of this mode for long-range targets. From a reconnaissance and defense perspective, RWS mode can only provide rough information on the target's range, velocity, and azimuth, and is not suitable for direct use in precision-guided weapon systems. Its threat level is relatively low. RWS mode is similar to VS mode in its scanning method, also using a line-by-line scanning method. The azimuth search range is: , , ; Step 1.1.3: Scan-and-track mode Track While Search (TWS) mode is a working mode that tracks targets by introducing track processing during the search process. When the radar searches and detects a specific airspace, it also needs to automatically determine whether the detected target is a new target, and then perform track correlation on the target data to achieve target tracking. The working parameters of TWS mode are similar to those of RWS mode, but the difference is that TWS mode needs to perform track processing on the range, velocity, azimuth, and other information of the detected target to predict the target's position at the next moment for tracking. Because TWS mode requires a higher data rate to improve tracking accuracy, it needs to transmit fewer pulses per wave position to speed up the scan speed and thus increase the data rate. From a reconnaissance and defense perspective, TWS mode can perform a certain degree of coarse tracking, but the tracking accuracy is too low, and its threat level is moderate. TWS mode is similar to RWS mode in its scanning method, which is also a line-by-line scanning method, and the azimuth search range is: , , ; Step 1.1.4: Search and Tracking Mode Track and Search (TAS) mode utilizes the flexible scanning capabilities of array antennas and the principle of time division to provide different data rates for search and tracking based on mission requirements, alternating between the two. In TWS mode, search and tracking are coupled, while in TAS mode, they are independent. It prioritizes data rates for tracking each key target, then uses idle time to scan and search the remaining airspace. Therefore, compared to TWS mode, TAS mode offers more efficient resource management and scheduling, significantly improving tracking stability and accuracy. From a reconnaissance and defense perspective, TAS mode can track targets stably and accurately, indicating a higher level of threat. Step 1.1.5: Single Target Tracking Mode Single Target Tracking (STT) mode is a classic operating mode for tracking targets. In STT mode, the multi-functional radar allocates all resources to track a single target, providing real-time, accurate, and continuous information on the speed, range, and azimuth of a single target within a specific airspace. Therefore, STT mode has an extremely high data rate and extremely high tracking accuracy, enabling "surgical strikes" against targets within a specific airspace, thus posing a very high threat to the reconnaissance side. Therefore, the parameters of the multi-functional phased array radar operating modes are shown in Table 1 above. Table 1

[0051] Step S1.2: Use the multi-functional phased array radar dataset received by UAV swarm reconnaissance Imagine a swarm of drones on a sphere, surrounding a multi-functional phased array radar at different wavefronts. This means different drones are positioned at different wavefronts of the multi-functional phased array radar. A schematic diagram of the drone swarm is shown below. Figure 1 As shown; Because the drone swarm surrounds the multi-functional phased array radar in a spherical shape, meaning that each drone is equidistant from the multi-functional phased array radar, the amplitude ratio of the radar pulse signals detected by different drones is positively correlated only with the radar antenna gain ratio at the different drone positions, without considering atmospheric attenuation loss. Assuming that wave positions are arranged throughout the entire space, the following is obtained: If the signal amplitude detected at any position within the same wave position is approximately equal to the signal amplitude detected at the center of the wave position, then the entire space is divided into units of wave positions. Each small space contains the same signal detected within it. The multi-functional phased array radar transmits pulses as shown in Table 1. The UAV within each small space detects the pulse signals and extracts the PDW (Pulse Description Word) to obtain the PDW stream. Each PDW mainly includes arrival time, signal amplitude, pulse width, and carrier frequency. The signal amplitude in the PDW is replaced by the radar antenna gain. Because the receiver is affected by noise, a threshold value can be set based on factors such as the distance between the UAV and the radar and the receiver noise level. When the signal amplitude is less than the threshold value At that time, it is assumed that the signal was not received, that is, its PDW information was not received; Based on the above standards, a working mode recognition dataset was established, with the PDW stream dataset size set to [size to be specified]. Based on the different wave positions of the UAV, the different operating modes of the radar, and the different parameters under the radar operating modes, multiple representative operating modes were simulated. PDW stream information (time of arrival, signal amplitude, pulse width, and carrier frequency); Step S2: Build the BiLSTM-ResNet18-DS model The BiLSTM-ResNet18-DS algorithm, for a single UAV, uses Bi-LSTM for temporal feature extraction from the input PDW stream data and ResNet18 for spatial feature extraction. The extracted features are then merged and passed through an MLP network to obtain the radar operating mode classification result. Results from multiple UAVs are fused at the result level using DS theory. The specific structure is as follows: Figure 2 As shown; Step S2.1: Construct the BiLSTM network model Long Short-Term Memory (LSTM) networks are a special type of Recurrent Neural Network (RNN) that effectively addresses long-term dependencies. All RNNs are chains of repetitive neural network modules. Their processing layers are remarkably simple, typically a single tanh layer, using the current input and the output from the previous time step to obtain the current output. Compared to traditional neural networks, with simple modifications, LSTM can utilize information learned in the previous time step for learning in the current time step. The specific structure is as follows... Figure 3 As shown; However, simple RNN network structures often suffer from gradient vanishing and gradient explosion during training, causing network parameters to fail to converge for extended periods. In 1997, Schmidhuber invented the Long Short-Term Memory (LSTM) network, which partially solved this problem. Its structure is as follows: Figure 4 As shown.

[0052] By simulating the characteristics of human memory, it introduces the concept of gates into traditional RNN networks to control the opening degree of information flow in network loops, achieving the purpose of filtering key local information and synthesizing features over long periods; the operation definition of the LSTM neural network is as follows:

[0053]

[0054]

[0055]

[0056]

[0057] In the definition of LSTM operation, , and These are the forget gate, input gate, and output gate. All three gates are controlled by the current input. Output at the previous moment Controlled The unit consists of a forget gate that controls the historical state. Enter the current state Opening degree, input gate control and Enter Opening degree, output gate control Enter the current output Compared to simple RNNs, LSTMs not only add a gating structure, which improves the network's trainability and sequence feature synthesis capabilities, but also inherit the characteristic of simple RNNs to adapt to variable-length inputs. This makes LSTM networks more suitable for synthesizing information from local radiation source signal features extracted by convolution operations. Bi-LSTM (Bi-directional Long Short-Term Memory) is composed of forward LSTM and backward LSTM. The operation process of Bi-LSTM is as follows: Figure 5 As shown.

[0058] Step S2.2: Build the ResNet18 network model When neural networks become too deep, they can experience vanishing or exploding gradients. While regularization methods can mitigate this, further increases in network depth lead to model degradation. Therefore, deep residual networks are used to address these issues. The core of this network structure is the residual block. This residual structure allows deep neural network models to have up to 100 or even 1000 layers.

[0059] The basic architecture of ResNet is residual blocks. In residual blocks, the input can propagate forward faster through cross-layer data paths. A residual block initially contains two identical 3... One convolutional layer, followed by a batch normalization layer and a ReLU activation function; then, using a cross-layer data path, these two convolutional operations are skipped, and the input is directly appended before the final ReLU activation function; this design requires the outputs of the two convolutional layers to have the same shape as the input, so that they can be added; if you want to change the number of channels, an additional 1 is needed. A convolutional layer is used to transform the input into the required shape before performing addition operations, specifically as follows: Figure 6 As shown; a ResNet network can be formed by stacking residual blocks. This application adopts an 18-layer ResNet network structure, and its network structure diagram is shown below. Figure 7 As shown; Step S2.3: Fusion of DS Evidence Theory Dempster-Shafer (DS) evidence theory, as a reasoning theory, can handle uncertain information without requiring any prior knowledge and is widely used in the field of target recognition through multi-sensor fusion. DS evidence theory is based on a finite, non-empty, and mutually exclusive set, which is called the recognition frame. ; by recognition framework The set consisting of all subsets is called The power set, denoted as ; Definition 1: Let To identify the frame, the function satisfy:

[0060]

[0061] Then it is called For identification framework The basic probability assignment function on; , Called The basic probability assignment (BPA) represents the probability distribution of a proposition. confidence level As evidence The focal length; Definition 2: Let To identify the frame, the function To identify the basic probability assignment function on the framework. Then it is called by Defined functions For identification framework Trust function on, Indicates the proposition The degree of genuine trust is... The lower bound estimate of the level of trust; by Defined function for The likelihood function on Indicates the proposition The degree of trust that is not false is a measure of trust in... The upper limit of the level of trust, Then it means Trust interval; Definition 3: Let the recognition frame be defined. Multiple pieces of evidence The basic probability distribution is , Represent The corresponding focal element can be fused according to the DS synthesis rules. After one piece of evidence, regarding the proposition Basic probability distribution:

[0062] in, , representing the conflict coefficient between pieces of evidence. This is called the normalization factor; when When, it indicates that multiple pieces of evidence are completely consistent; when When, it indicates that the evidence is completely conflicting; when When the evidence is partially compatible, it indicates that the evidence is compatible with each other. The DS evidence theory synthesis rules can reflect the effect of evidence combination. For multiple incomplete conflicting pieces of evidence, the orthogonal sum of their basic credibility distributions can be calculated using the above rules to obtain a new basic credibility distribution. Analysis shows that the synthesis rules satisfy the commutative and associative laws. Therefore, in the process of evidence synthesis, the order of evidence does not need to be considered, and the above formula can be used directly. The fusion result is calculated.

[0063] Step S3: Simulation Analysis Simulation conditions: The graphics card used is NVIDIA GeForce RTX 3090, CUDA version 12.1 is used, and PyTorch version 2.1.2 is used to build the network algorithm.

[0064] Step S3.1: Single-machine network identification simulation and analysis Learning process: BiLSTM-ResNet, ResNet18, LSTM and other algorithms are used for learning. The relationship between the number of learning rounds and the training set loss is shown in Figure 8(a), the relationship between the number of learning rounds and the training set accuracy is shown in Figure 8(b), and the relationship between the number of learning rounds and the validation set accuracy is shown in Figure 8(c). During the learning process, as shown in Figures 8(a) and 8(b), all three methods reached their limits on the training set, achieving an accuracy of approximately 100%. Figure 8(c) shows that the LSTM method, which extracts features temporally, exhibits better stability in validation set accuracy. The ResNet18 method, which extracts features spatially, shows poorer stability in validation set accuracy but can randomly learn better results. The BiLSTM-ResNet method combines the advantages of both methods, extracting features both temporally and spatially, achieving the highest validation set accuracy with good stability. After learning with BiLSTM-ResNet, the validation set accuracy improved by 2% compared to ResNet18 and Bi-LSTM, reaching 99.67%.

[0065] Testing process: The test results of the BiLSTM-ResNet trained model on the test set data are as follows: Figure 9 As shown, the BiLSTM-ResNet algorithm has a small amount of obfuscation for VS and HRWS, and a small amount of obfuscation for MRWS and TWS.

[0066] Step S3.2: Multi-machine fusion recognition simulation and analysis The simulation conditions are consistent with those for single-machine network recognition. The relationship between the number of UAVs and the accuracy under the BiLSTM-ResNet algorithm is shown in Table 2.

[0067] Table 2

[0068] The table showing the relationship between the fusion accuracy of each algorithm and the number of drones indicates that the fusion effect improves the accuracy of recognition for a single drone.

[0069] The radar operating mode recognition system based on UAV swarm provided in this application is described below. The radar operating mode recognition system based on UAV swarm described below can be referred to in correspondence with the radar operating mode recognition method based on UAV swarm described above.

[0070] The second aspect of this application relates to a radar operating mode recognition system based on a drone swarm, comprising: a PDW stream information acquisition module and an operating mode recognition module; The PDW stream information acquisition module is used to divide the space into several small spaces in units of wave positions, and to extract several PDW stream information from the UAV reconnaissance pulse signals in each small space after PDW extraction. The working mode recognition module is used to input PDW stream information into Bi-LSTM neural network for temporal first feature extraction, and then input it into ResNet18 network for spatial second feature extraction. The first and second features are merged and then passed through MLP network to obtain the radar working mode of each UAV. The radar working modes of all UAVs are fused through DS theory to obtain the working mode of multi-function phased array radar.

[0071] More preferably, the phased array radar has the following operating modes: search mode, tracking mode, and composite mode; the search mode includes velocity search mode and search-while-ranging mode; the tracking mode includes single-target tracking mode; and the composite mode includes scan-while-tracking mode and search-plus-tracking mode.

[0072] More preferably, the radar operating mode recognition system further includes a network training module, which includes a training dataset acquisition unit and a training unit; The training dataset acquisition unit is used to establish a multi-functional phased array radar operating mode recognition dataset, which includes: the operating modes of the multi-functional phased array radar and their corresponding operating mode parameters, as well as UAV swarm reconnaissance pulse signals. The training unit is used to train the Bi-LSTM and ResNet18 networks by using the UAV swarm reconnaissance pulse signals as inputs and the operating modes of the multi-function phased array radar and their corresponding operating mode parameters as outputs.

[0073] More preferably, the method for acquiring PDW stream information is as follows: based on the transmitted pulse of the multi-function phased array radar, the UAV reconnaissance pulse signal in each small space is acquired, and PDW stream information is obtained after PDW extraction. Each PDW stream information includes arrival time, signal amplitude, pulse width, carrier frequency and intra-pulse modulation.

[0074] More preferably, the operation of the Bi-LSTM neural network is as follows:

[0075]

[0076]

[0077]

[0078]

[0079] in, , and These are the forget gate, input gate, and output gate. All three gates are controlled by the current input. Output at the previous moment Controlled The unit consists of a forget gate that controls the historical state. Enter the current state Opening degree, input gate control and Enter Opening degree, output gate control Enter the current output The degree of opening.

[0080] More preferably, the basic architecture of ResNet is a residual block, in which the input propagates forward faster through cross-layer data paths. The residual block initially contains two identical 3... One convolutional layer, followed by a batch normalization layer and a ReLU activation function; then, by using a cross-layer data path, skipping two convolution operations, the input is directly added before the final ReLU activation function.

[0081] It is understood that the detailed functional implementation of each of the above units / modules can be found in the description in the aforementioned method embodiments, and will not be repeated here.

[0082] It should be understood that the above-described device is used to execute the methods in the above embodiments. The implementation principle and technical effect of the corresponding program modules in the device are similar to those described in the above methods. The working process of the device can be referred to the corresponding process in the above methods, and will not be repeated here.

[0083] Based on the methods in the above embodiments, this application provides an electronic device / image signal generator / network device / transmitter / terminal / base station / industrial control computer. This electronic device / image signal generator / network device / transmitter / terminal / base station / industrial control computer may include: a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The processor can call logical instructions in the memory to execute the methods in the above embodiments.

[0084] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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 described in the various embodiments of this application.

[0085] Based on the methods in the above embodiments, this application provides a computer-readable storage medium storing a computer program that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0086] Based on the methods in the above embodiments, this application provides a computer program product that, when run on a processor, causes the processor to execute the methods in the above embodiments.

[0087] It is understood that the processor in the embodiments of this application can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. A general-purpose processor can be a microprocessor or any conventional processor.

[0088] The method steps in this application embodiment can be implemented in hardware or by a processor executing software instructions. The software instructions can consist of corresponding software modules, which can be stored in random access memory (RAM), flash memory, read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, CD-ROMs, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor, enabling the processor to read information from and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can reside in an ASIC.

[0089] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted through the computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0090] It is understood that the various numerical designations used in the embodiments of this application are merely for the convenience of description and are not intended to limit the scope of the embodiments of this application.

[0091] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A radar operating mode recognition method based on unmanned aerial vehicle (UAV) swarms, wherein, A swarm of unmanned aerial vehicles (UAVs) surrounds a multi-functional phased array radar at different wave positions, characterized by the following steps: Step S1: Divide the space into several small spaces in units of wave positions, and extract several PDW stream information from the UAV reconnaissance pulse signal in each small space by performing PDW extraction. Step S2: Input the PDW stream information into the Bi-LSTM neural network for temporal first feature extraction, and then input it into the ResNet18 network for spatial second feature extraction. After merging the first and second features, pass them through the MLP network to obtain the radar operating mode identified by each UAV. Then, fuse all the radar operating modes identified by all UAVs through DS theory to obtain the operating mode of the multi-functional phased array radar.

2. The radar operating mode recognition method according to claim 1, characterized in that, The training methods for Bi-LSTM and ResNet18 networks are as follows: A dataset for recognizing the operating modes of a multi-functional phased array radar was established, which includes: the operating modes of the multi-functional phased array radar and their corresponding operating mode parameters, as well as the pulse signals of UAV swarm reconnaissance. The pulse signals from UAV swarm reconnaissance were used as inputs to the Bi-LSTM neural network and the ResNet18 network, and the operating modes of the multi-function phased array radar and their corresponding operating mode parameters were used as outputs to train the Bi-LSTM and ResNet18 networks.

3. The radar operating mode recognition method according to claim 2, characterized in that, The operating modes of a phased array radar include search mode, track mode, and composite mode; the operating modes of the search mode include velocity search mode and search-while-ranging mode; the operating modes of the track mode include single-target tracking mode; and the operating modes of the composite mode include scan-while-tracking mode and search-plus-tracking mode.

4. The radar operating mode recognition method according to claim 2, characterized in that, The method for acquiring PDW stream information is as follows: based on the transmitted pulse of the multi-function phased array radar, the UAV reconnaissance pulse signal in each small space is acquired, and PDW stream information is obtained after PDW extraction. Each PDW stream information includes arrival time, signal amplitude, pulse width, carrier frequency and intra-pulse modulation.

5. The radar operating mode recognition method according to claim 1, characterized in that, The operations of the Bi-LSTM neural network are as follows: in, , and These are the forget gate, input gate, and output gate. All three gates are controlled by the current input. Output at the previous moment Controlled The unit consists of a forget gate that controls the historical state. Enter the current state Opening degree, input gate control and Enter Opening degree, output gate control Enter the current output The degree of opening.

6. The radar operating mode recognition method according to claim 1 or 5, characterized in that, The basic architecture of ResNet is residual blocks. In residual blocks, the input propagates forward more quickly through cross-layer data paths. The residual block initially contains two identical 3... One convolutional layer, followed by a batch normalization layer and a ReLU activation function; then, by using a cross-layer data path, skipping two convolution operations, the input is directly added before the final ReLU activation function.

7. A radar operating mode recognition system based on unmanned aerial vehicle (UAV) swarms, characterized in that, include: PDW stream information acquisition module and working mode recognition module; The PDW stream information acquisition module is used to divide the space into several small spaces in units of wave positions, and to extract several PDW stream information from the UAV reconnaissance pulse signals in each small space after PDW extraction. The working mode recognition module is used to input PDW stream information into Bi-LSTM neural network for temporal first feature extraction, and then input it into ResNet18 network for spatial second feature extraction. The first and second features are merged and then passed through MLP network to obtain the radar working mode of each UAV. The radar working modes of all UAVs are fused through DS theory to obtain the working mode of multi-functional phased array radar.

8. The radar operating mode recognition system according to claim 7, characterized in that, The method for acquiring PDW stream information is as follows: based on the transmitted pulse of the multi-function phased array radar, the UAV reconnaissance pulse signal in each small space is acquired, and PDW stream information is obtained after PDW extraction. Each PDW stream information includes arrival time, signal amplitude, pulse width, carrier frequency and intra-pulse modulation.

9. The radar operating mode recognition system according to claim 7, characterized in that, The radar operating mode recognition system also includes a network training module, which comprises a training dataset acquisition unit and a training unit. The training dataset acquisition unit is used to establish a multi-functional phased array radar operating mode recognition dataset, which includes: the operating modes of the multi-functional phased array radar and their corresponding operating mode parameters, as well as UAV swarm reconnaissance pulse signals. The training unit is used to train the Bi-LSTM and ResNet18 networks by using the UAV swarm reconnaissance pulse signals as inputs and the operating modes of the multi-function phased array radar and their corresponding operating mode parameters as outputs.

10. The radar operating mode recognition system according to any one of claims 7 to 9, characterized in that, The operations of the Bi-LSTM neural network are as follows: in, , and These are the forget gate, input gate, and output gate. All three gates are controlled by the current input. Output at the previous moment Controlled The unit consists of a forget gate that controls the historical state. Enter the current state Opening degree, input gate control and Enter Opening degree, output gate control Enter the current output The degree of opening; The basic architecture of ResNet is residual blocks. In residual blocks, the input propagates forward more quickly through cross-layer data paths. The residual block initially contains two identical 3... One convolutional layer, followed by a batch normalization layer and a ReLU activation function; then, by using a cross-layer data path, skipping two convolution operations, the input is directly added before the final ReLU activation function.

Images