Radar signal anti-interference processing method and system based on space-time attention mechanism

Abstract

The invention provides a radar signal anti-interference processing method and system based on a space-time attention mechanism. The method comprises the following steps: mapping an original horizontal polarization channel signal and an original vertical polarization channel signal into an interference suppression space, and generating a three-dimensional feature tensor comprising an amplitude component, a polarization phase difference component and an interference suppression residual component; according to the amplitude component, dividing a radar observation area into multiple layers of target blocks in a spatial dimension, and calculating a space-time association weight between adjacent layers of target blocks through a space-time attention mechanism; and according to the space-time correlation weight, screening out a plurality of target blocks meeting a preset continuous motion condition, and decoding a motion mode of the plurality of target blocks by using a long short-term memory network to generate an anti-interference target trajectory coordinate and a velocity vector. According to the method, the polarization characteristics and the space-time attention mechanism are fused, the anti-interference effect of radar signal processing in a strong interference environment is improved, and then the detection precision of the multi-target motion trail is improved.

Description

Technical Field

[0001] This application relates to the field of radar signal processing technology, and in particular to a radar signal anti-interference processing method and system based on a spatiotemporal attention mechanism. Background Technology

[0002] In radar multi-target tracking scenarios under strong electromagnetic interference, the following three core requirements must be met: first, the ability to resist strong suppression interference to ensure the stability of target signals; second, to avoid confusion of the cross trajectories of multiple moving targets; and third, to maintain tracking accuracy when the target is performing complex maneuvers such as sharp turns or high acceleration changes.

[0003] The current mainstream approach adopts a joint framework of polarization filtering and trajectory prediction. The main steps are as follows: extract the amplitude ratio and phase difference characteristics of the horizontal and vertical polarization channels of the target echo; suppress interference outside the working frequency band through an adaptive filter; finally, use a sliding window to extract the motion segment of the target, input the motion segment into a long short-term memory network, and predict the future trajectory coordinates of the target.

[0004] However, existing solutions have three key drawbacks: First, the interference signal and target polarization characteristics are severely aliased in the two-dimensional plane, and some interference components remain after adaptive filtering, which negatively impacts subsequent trajectory reconstruction. Second, the spatial correlation is rigid, and the use of a fixed sliding window to divide the target trajectory segments results in weak correlation between adjacent motion segments, which can easily cause the tracking trajectory to break when the target performs complex maneuvers. Third, when multiple target trajectories are close to or intersect in space, the existing solutions have limited ability to continuously track a single target and cannot effectively distinguish intersecting trajectories. Summary of the Invention

[0005] This application provides a radar signal anti-jamming processing method and system based on a spatiotemporal attention mechanism to solve the problem that existing technologies cannot cleanly remove interference at the feature level, resulting in interference residues and ultimately causing track trajectory breakage and trajectory confusion.

[0006] Firstly, this application provides a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism, including:

[0007] Acquire the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar;

[0008] The original horizontal polarization channel signal and the original vertical polarization channel signal are mapped into the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components.

[0009] Based on the amplitude components, the radar observation area is divided into multiple target blocks in the spatial dimension, and the spatiotemporal correlation weights between adjacent target blocks are calculated through a spatiotemporal attention mechanism.

[0010] Based on the spatiotemporal correlation weights, multiple target blocks that meet the preset continuous motion conditions are selected, and the motion patterns of the multiple target blocks are decoded using a long short-term memory network to generate anti-interference target trajectory coordinates and velocity vectors.

[0011] Optionally, mapping the original horizontally polarized channel signal and the original vertically polarized channel signal into the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components, and interference suppression residual components includes:

[0012] Calculate the orthogonal phase difference between the original horizontal polarization channel signal and the original vertical polarization channel signal;

[0013] Based on the cosine and sine components of the orthogonal phase difference, an orthogonal basis vector for the interference suppression space is constructed. The orthogonal basis vector includes the principal axis of energy response, the polarization difference axis, and the interference residual separation axis.

[0014] The original horizontal polarization channel signal is projected onto the principal axis of the energy response to generate an amplitude component, and the original vertical polarization channel signal is projected onto the polarization differential axis to generate a polarization phase difference component.

[0015] Based on the amplitude component and the polarization phase difference component, calculate the projection difference between the original horizontal polarization channel signal and the original vertical polarization channel signal on the plane spanned by the principal axis of the energy response and the polarization difference axis.

[0016] The projection difference is projected onto the interference residual separation axis to generate an interference suppression residual component.

[0017] The amplitude component, the polarization phase difference component, and the interference suppression residual component are integrated along the three coordinate axes of the orthogonal basis vector to generate a three-dimensional feature tensor.

[0018] Optionally, projecting the projection difference onto the interference residual separation axis to generate interference suppression residual components includes:

[0019] The projection difference is subjected to amplitude compression to generate a normalized residual amplitude.

[0020] Based on the normalized residual magnitude, a residual density distribution is generated using a distribution function template;

[0021] Discretize the residual density distribution along the interference residual separation axis to generate a residual vector sequence;

[0022] The residual vector sequence is tensor-expanded along the interference residual separation axis to generate interference suppression residual components.

[0023] Optionally, generating the residual density distribution using a distribution function template based on the normalized residual magnitude includes:

[0024] The time-domain waveform of the normalized residual amplitude is scanned to identify the positions of the rising and falling edges of the pulse, and an amplitude fluctuation quantification index is generated.

[0025] Based on the amplitude fluctuation quantification index, a distribution function template is matched from a preset residual distribution function template library;

[0026] Based on the positions of the rising and falling edges of the pulse, statistical moment features are extracted from the normalized residual amplitude. Based on the statistical moment features of the normalized residual amplitude, the shape parameters of the matching distribution function template are calculated.

[0027] Based on the normalized residual magnitude, a residual density distribution is generated using a matching distribution function template with the shape parameters.

[0028] Optionally, generating the residual density distribution based on the normalized residual magnitude using a matching distribution function template with the shape parameters includes:

[0029] According to the radar pulse repetition period, the normalized residual amplitude is divided into multiple pulse window residual sequences;

[0030] Parallel computation is performed on each of the pulse window residual sequences using a matching distribution function template with the shape parameters to generate a window residual density distribution;

[0031] The window residual density distributions of all windows are superimposed along the pulse sequence dimension to generate the pulse cumulative residual distribution;

[0032] The pulse accumulation residual distribution is mapped onto the distance Doppler plane to generate a residual density distribution.

[0033] Optionally, the step of filtering out multiple target blocks that meet preset continuous motion conditions based on the spatiotemporal correlation weights, and decoding the motion patterns of the multiple target blocks using a long short-term memory network to generate anti-interference target trajectory coordinates and velocity vectors includes:

[0034] Motion continuity analysis is performed on the spatiotemporal correlation weights to generate a motion continuity confidence index that characterizes the motion correlation strength of the target block between consecutive observation frames;

[0035] Target blocks that fall within the threshold range of a preset continuous motion condition, based on the motion continuity confidence index, are selected to form a target block set;

[0036] The historical phase sequence of the target block set in the Doppler spectrum is input into a long short-term memory network for motion pattern decoding to generate a target motion vector field.

[0037] The target motion vector field is decomposed into radial velocity components and tangential velocity components. Based on the radial velocity components and tangential velocity components, the trajectory and velocity are jointly calculated to generate the target trajectory coordinates and velocity vector after interference resistance.

[0038] Optionally, the step of integrating the amplitude component, the polarization phase difference component, and the interference suppression residual component along the three coordinate axes of the orthogonal basis vectors to generate a three-dimensional feature tensor includes:

[0039] Alignment operations are performed on the amplitude components along the principal axis of the energy response to generate a first axial characteristic component;

[0040] Alignment operation is performed on the polarization phase difference components along the polarization difference axis to generate a second axial characteristic component;

[0041] A mapping operation is performed on the interference suppression residual component along the interference residual separation axis to generate a third axial characteristic component;

[0042] Perform a three-dimensional tensor synthesis operation on the first axial feature component, the second axial feature component, and the third axial feature component to generate a three-dimensional feature tensor.

[0043] Secondly, this application provides a radar signal anti-jamming processing system based on a spatiotemporal attention mechanism, comprising:

[0044] The acquisition module is used to acquire the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar;

[0045] The mapping module is used to map the original horizontal polarization channel signal and the original vertical polarization channel signal to the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components.

[0046] The partitioning module is used to divide the radar observation area into multiple target blocks in the spatial dimension according to the amplitude component, and to calculate the spatiotemporal correlation weight between adjacent target blocks through a spatiotemporal attention mechanism.

[0047] The filtering module is used to filter out multi-target blocks that meet the preset continuous motion conditions according to the spatiotemporal correlation weights, and use a long short-term memory network to decode the motion mode of the multi-target blocks to generate anti-interference target trajectory coordinates and velocity vectors.

[0048] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are to be invoked and executed by the processing component to implement a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism as described in any of the first aspects.

[0049] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a radar signal anti-interference processing method based on a spatiotemporal attention mechanism as described in any of the first aspects.

[0050] This application provides a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism. The method includes: acquiring the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar; mapping the original horizontal polarization channel signal and the original vertical polarization channel signal to an interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components, and interference suppression residual components; dividing the radar observation area into multiple target blocks in the spatial dimension according to the amplitude components, and calculating the spatiotemporal correlation weight between adjacent target blocks through a spatiotemporal attention mechanism; selecting multiple target blocks that meet preset continuous motion conditions according to the spatiotemporal correlation weights, and decoding the motion mode of the multiple target blocks using a long short-term memory network to generate the anti-jamming target trajectory coordinates and velocity vectors.

[0051] This application simultaneously acquires horizontal and vertical polarization channel signals, preserving the target's full polarization scattering characteristics and providing the original information basis for interference suppression. It maps the dual-polarization signal to the interference suppression space, generating a three-dimensional tensor with fused amplitude, polarization phase difference, and interference suppression residuals, achieving physical decoupling between the interference signal and target features. Based on amplitude components, it hierarchically divides target blocks and adaptively calculates the association weights of adjacent blocks using a spatiotemporal attention mechanism, solving the problem of misassociation caused by the intersection of dense target trajectories. Based on the association weights, it filters continuously moving target blocks and uses a long short-term memory network to decode complex motion patterns, ultimately outputting high-precision trajectory coordinates and velocity vectors resistant to interference. This ultimately achieves robust reconstruction of multi-target trajectories and accurate analysis of motion states under strong interference scenarios.

[0052] Furthermore, this application constructs an interference suppression space with the energy response principal axis, polarization difference axis, and interference residual separation axis as basis vectors by calculating the orthogonal phase difference between the original horizontal and vertical polarization channel signals. The horizontal channel signal is projected onto the energy response axis to generate an amplitude component, and the vertical channel signal is projected onto the polarization difference axis to generate a polarization phase difference component. Based on these two components, the projection difference is calculated, and after amplitude compression, residual density distribution generation, and discretization, it is expanded along the interference residual separation axis to form an interference suppression residual component. Finally, the three components are integrated along the orthogonal basis vector coordinate axes into a three-dimensional feature tensor. The constructed three-dimensional feature tensor enhances anti-interference capability through triple decoupling: the energy response principal axis carries target intensity information, the polarization difference axis separates the target polarization scattering characteristics, and the interference residual separation axis uses residual density distribution transformation to compress the interference signal to an independent dimension. This fundamentally solves the coupling problem between interference and target signals in traditional two-dimensional polarization features, providing a physically interpretable feature basis for subsequent processing.

[0053] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 A flowchart illustrating a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism, provided for embodiments of this application;

[0056] Figure 2 A schematic diagram of a radar signal anti-jamming processing system based on a spatiotemporal attention mechanism is provided for an embodiment of this application;

[0057] Figure 3 This is a schematic diagram of the structure of a computing device provided in an embodiment of this application. Detailed Implementation

[0058] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0059] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 11, 12, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a sequential order, nor do they limit "first" and "second" to different types.

[0060] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0061] To address the problem that existing technologies cannot cleanly remove interference at the feature level, resulting in residual interference and ultimately leading to broken and confused tracking trajectories, this application provides a radar signal anti-interference processing method based on a spatiotemporal attention mechanism. This method employs the following concept: Addressing the pain point of easily broken and miscorrelated multi-target trajectories in strong interference scenarios, this application adopts a phased framework of feature decoupling, dynamic association, and motion decoding. At the polarization signal layer, an interference suppression space is constructed, decoupling the dual-polarization signal into physically meaningful amplitude, phase difference, and interference residual components to form an anti-interference feature base. At the spatiotemporal modeling layer, target blocks are divided based on amplitude hierarchy, and the correlation between adjacent blocks is dynamically learned through a spatiotemporal attention mechanism, overcoming the trajectory fragmentation bottleneck caused by fixed thresholds. At the trajectory generation layer, continuously moving target blocks are selected, and a long short-term memory network is used to decode complex motion patterns, achieving end-to-end reconstruction of anti-interference trajectories.

[0062] Figure 1 A flowchart illustrating a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism, as provided in this application embodiment, is shown below. Figure 1 As shown, the method includes:

[0063] S11. Acquire the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar.

[0064] The original horizontal polarization channel signal refers to the echo signal received by the radar's horizontal polarization antenna, which includes target scattering energy and environmental interference, and is used to extract electromagnetic wave polarization information in the horizontal direction. The original vertical polarization channel signal refers to the echo signal received by the radar's vertical polarization antenna, which forms an orthogonal polarization pair with the horizontal channel, and is used to analyze the target's scattering characteristics of vertically polarized electromagnetic waves.

[0065] In this embodiment, the original horizontal polarization channel signal and the original vertical polarization channel signal are first acquired by the radar receiver. These two signals are composed of electromagnetic wave echo data collected by the horizontal polarization port and the vertical polarization port of the radar antenna, respectively.

[0066] S12. Map the original horizontal polarization channel signal and the original vertical polarization channel signal into the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components.

[0067] The interference suppression space refers to a three-dimensional vector space constructed based on orthogonal phase difference, with its three axes used to separate target intensity, polarization characteristics, and residual interference components, respectively. The amplitude component refers to the projection value of the original horizontal polarization channel signal onto the principal axis of the energy response, reflecting the energy intensity characteristics of the target echo. The polarization phase difference component refers to the projection value of the original vertical polarization channel signal onto the polarization difference axis, characterizing the phase difference between the target's horizontal and vertical polarization echoes. The interference suppression residual component refers to the mapping result of the projection difference on the interference residual separation axis, used to capture residual interference noise after polarization decomposition. The three-dimensional feature tensor is a data structure that integrates the amplitude component, polarization phase difference component, and interference suppression residual component along the three coordinate axes of the orthogonal basis vector, forming the input features for subsequent processing.

[0068] In this embodiment, the original horizontal polarization channel signal and the original vertical polarization channel signal are first mapped to the interference suppression space: the orthogonal phase difference between the two signals is calculated, and the cosine and sine components are extracted to construct an orthogonal basis vector containing the principal axis of energy response, the polarization difference axis, and the interference residual separation axis; then the horizontal signal is projected onto the principal axis of energy response to generate the amplitude component, and the vertical signal is projected onto the polarization difference axis to generate the polarization phase difference component; next, the projection difference between the two signals on the energy-polarization plane is calculated based on the amplitude component and the polarization phase difference component; finally, the projection difference is projected onto the interference residual separation axis to generate the interference suppression residual component, and the three components are integrated to form a three-dimensional feature tensor.

[0069] S13. Based on the amplitude component, the radar observation area is divided into multiple target blocks in the spatial dimension, and the spatiotemporal correlation weight between adjacent target blocks is calculated through the spatiotemporal attention mechanism.

[0070] In this context, the spatial dimension refers to the range-azimuth two-dimensional plane of the radar observation area, used as a spatial coordinate system to divide target blocks. The radar observation area refers to the monitored airspace covered by the radar beam, containing the potential spatial locations of all targets to be detected. Multi-layer target blocks refer to gridded units divided in the spatial dimension, each layer corresponding to an azimuth slice within a specific range gate. A target block is the smallest processing unit with similar motion characteristics within a single layer, containing a set of target echoes from a local region. Adjacent layer target blocks refer to two consecutive target block levels in the range dimension, used to analyze the motion continuity of targets across range gates. The spatiotemporal correlation weight refers to the correlation strength value between target blocks calculated through an attention mechanism, reflecting the probability of target position migration between consecutive frames.

[0071] In this embodiment, firstly, the target blocks are divided into multiple layers in the spatial dimension of the radar observation area based on the amplitude component, with each layer of target blocks corresponding to a specific range azimuth unit; secondly, the correlation between adjacent layers of target blocks is calculated through a spatiotemporal attention mechanism: the motion features of the target blocks are extracted between consecutive frames, and the coupling relationship between spatial position and temporal evolution is quantified using an attention weight matrix to generate a spatiotemporal correlation weight that characterizes the strength of target correlation.

[0072] S14. Based on the spatiotemporal correlation weights, select multi-target blocks that meet the preset continuous motion conditions, and use a long short-term memory network to decode the motion patterns of the multi-target blocks to generate the target trajectory coordinates and velocity vectors after interference resistance.

[0073] The preset continuous motion condition is used to determine whether the target is a real moving target rather than random noise or residual interference. A multi-target block refers to a set of target blocks formed through screening, representing the associated echoes of the same target in multiple spatial layers. A Long Short-Term Memory (LSTM) network is a recurrent neural network with forget gates, input gates, and output gates, used to model the long-term dependencies of target motion sequences. It should be noted that the structure of this network is not specifically limited in this embodiment. Motion pattern refers to the velocity change pattern exhibited by the target block set in spatiotemporal evolution, including dynamic characteristics such as acceleration and turning. Target trajectory coordinates refer to the target spatial position sequence after anti-interference processing, represented by three-dimensional spatial coordinates. Velocity vector refers to the instantaneous velocity description of the target motion, including the vector synthesis result of radial and tangential velocity components.

[0074] In this embodiment, firstly, target blocks whose motion continuity confidence reaches the preset continuous motion condition are selected according to the spatiotemporal correlation weight to form a multi-target block set; secondly, the historical phase sequence of the target block in the Doppler spectrum is input into the long short-term memory network, and the target acceleration and turning characteristics are decoded through the gating mechanism; finally, the anti-interference target trajectory coordinates and velocity vector are output, wherein the velocity vector includes radial and tangential components.

[0075] Here is a specific example: An air surveillance radar tracks three high-speed aircraft in a strong electromagnetic interference environment: Aircraft 1 maintains a constant speed and straight flight at an altitude of 10,000 meters; Aircraft 2 performs continuous figure-eight maneuvers at low altitude; Aircraft 3's flight path intersects with the trajectory of a flock of migratory birds. The radar first synchronously acquires the original horizontal polarization channel signal, which contains strong scattered energy from the aircraft's metal components, and simultaneously acquires the original vertical polarization channel signal, which is sensitive to the phase modulation characteristics of the rotor or control surfaces. Then, the dual signals are mapped to the interference suppression space. By calculating the orthogonal phase difference between the two signals (e.g., phase difference Δφ = 32° for Aircraft 2), a three-dimensional orthogonal coordinate system is constructed, consisting of the principal axis of energy response, the polarization difference axis, and the interference residual separation axis. The horizontal signal is projected onto the energy axis to generate an amplitude component, such as 0.92 for Aircraft 1; the vertical signal is projected onto the polarization axis to generate a polarization phase difference component, such as 0.78 for Aircraft 2. The projection difference is calculated based on two components and mapped to the residual axis to generate interference suppression residual components. For example, the residual of aircraft 3 is 0.21, effectively suppressing bird flock interference. Finally, it is integrated into a 128×128×3 three-dimensional feature tensor. Based on the amplitude component, the observation space is divided into 20 range gates, each 5 km apart, and each layer is divided into 8×8 azimuth grids to form multi-layer target blocks. For the target block of aircraft 2 in the adjacent range layer, the spatiotemporal attention mechanism calculates the spatiotemporal correlation between its displacement vectors of 0.3 km and -1.2 km and velocity change of 120 m / s, outputting a spatiotemporal correlation weight of 0.88, reflecting a high confidence correlation; while the correlation weight between aircraft 3 and the bird flock is only 0.12, effectively distinguishing intersecting trajectories. Target blocks that meet the continuous motion condition are selected according to the weights. For example, the motion continuity confidence of aircraft 1 is 0.94, which is greater than the threshold of 0.8, forming a multi-target block set. The Doppler historical phase sequence of Unit 2 (5 consecutive frames) [1.2, 0.8, ..., -0.3] was input into a Long Short-Term Memory (LSTM) network. This network decoded and generated motion vector fields of 650 m / s and -320 m / s. These were then decomposed into a radial velocity component of 580 m / s and a tangential velocity component of 420 m / s. Joint calculations were used to output anti-interference trajectory coordinate sequences such as (12.3, 8.7) and (11.8, 8.1), along with a synthesized velocity vector, thus reducing velocity errors. This scheme improved interference suppression, reduced the number of trajectory breaks in Unit 2 to zero, and lowered the false correlation rate of intersecting trajectories.

[0076] By executing S11 to S14, the embodiments of this application generate a three-dimensional feature tensor through the decoupling mapping of dual-polarized signals in the interference suppression space, thereby improving the separation capability between the target and the interference; the spatiotemporal attention mechanism based on spatially layered target blocks accurately captures the motion correlation of the target; and the combination of long short-term memory network to model the dynamics of continuously moving targets effectively overcomes the trajectory breakage problem under strong interference environment, and realizes stable and reliable target trajectory reconstruction and motion parameter estimation.

[0077] In one possible embodiment, S12, mapping the original horizontal polarization channel signal and the original vertical polarization channel signal into the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components, and interference suppression residual components, including:

[0078] Step 121: Calculate the orthogonal phase difference between the original horizontal polarization channel signal and the original vertical polarization channel signal.

[0079] Among them, the orthogonal phase difference refers to the arctangent value obtained by dividing the real part of the phase difference between the horizontal and vertical polarized channel signals by the imaginary part, which reflects the difference in phase response of the target scatterer to orthogonally polarized electromagnetic waves.

[0080] In the embodiments of this application, the ratio of the real part to the imaginary part of the phase difference between the original horizontal polarization channel signal and the original vertical polarization channel signal is first calculated to obtain the orthogonal phase difference characterizing the orthogonal characteristics of the dual polarization signal.

[0081] Step 122: Based on the cosine and sine components of the orthogonal phase difference, construct the orthogonal basis vectors of the interference suppression space. The orthogonal basis vectors include the principal axis of energy response, the polarization difference axis, and the interference residual separation axis.

[0082] In this context, the cosine component refers to the projection coefficient of the orthogonal phase difference along the real axis, used to construct the unit vector basis of the energy response principal axis. The sine component refers to the projection coefficient of the orthogonal phase difference along the imaginary axis, serving as the mathematical basis for constructing the polarization difference axis. The orthogonal basis vectors are a set of mutually perpendicular unit vectors formed by the energy response principal axis, the polarization difference axis, and the interference residual separation axis, used to define the coordinate system of the interference suppression space. The energy response principal axis is a coordinate axis with the cosine component as the direction reference, used to extract the energy intensity characteristics of the target echo. The polarization difference axis is a coordinate axis with the sine component as the direction reference, used to separate the phase difference characteristics of horizontal and vertical polarized signals. The interference residual separation axis is a coordinate axis generated by the cross product of the energy response principal axis and the polarization difference axis, used to capture the residual interference components after orthogonal projection.

[0083] In this embodiment, the cosine and sine components of the orthogonal phase difference are first extracted as basis elements; then, the cosine components are used to define the principal axis of the energy response, and the sine components are used to define the polarization difference axis; finally, the interference residual separation axis perpendicular to the first two axes is generated by the vector cross product, and a complete three-dimensional orthogonal basis vector is constructed.

[0084] Step 123: Project the original horizontal polarization channel signal onto the principal axis of the energy response to generate the amplitude component, and project the original vertical polarization channel signal onto the polarization difference axis to generate the polarization phase difference component.

[0085] In this embodiment, the original horizontal polarization channel signal is first multiplied by the unit vector of the energy response principal axis to generate an amplitude component reflecting the target echo intensity; then the original vertical polarization channel signal is projected onto the polarization difference axis unit vector to generate a polarization phase difference component characterizing the polarization difference.

[0086] Step 124: Based on the amplitude component and the polarization phase difference component, calculate the projection difference between the original horizontal polarization channel signal and the original vertical polarization channel signal on the plane spanned by the principal axis of the energy response and the polarization difference axis.

[0087] Among them, the projection difference refers to the magnitude of the projection vector of the dual-polarized signal on the energy-polarization plane, which reflects the degree of deviation between the target echo and the ideal polarization model.

[0088] In the embodiments of this application, firstly, the vector coordinates of the amplitude component and the polarization phase difference component are used; secondly, the Euclidean distance of the projection vectors of the horizontal and vertical polarized signals on the plane formed by the principal axis of the energy response and the polarization difference axis is calculated to generate a projection difference quantity characterizing the signal difference.

[0089] Step 125: Project the projection difference onto the interference residual separation axis to generate the interference suppression residual component.

[0090] In the embodiments of this application, the projection difference is first multiplied by the interference residual to separate the axial unit vector; then, the amplitude fluctuation is suppressed by absolute value limiting; finally, the interference suppression residual component is generated by logarithmic transformation.

[0091] Step 126: Integrate the amplitude component, polarization phase difference component, and interference suppression residual component along the three coordinate axes of the orthogonal basis vector to generate a three-dimensional feature tensor.

[0092] The coordinate axes refer to the three directional dimensions of the orthogonal basis vectors, including the principal axis of energy response, the polarization difference axis, and the interference residual separation axis.

[0093] In the embodiments of this application, firstly, the amplitude components are arranged along the principal axis of the energy response to form the first dimension; secondly, the polarization phase difference components are arranged along the polarization difference axis to form the second dimension; finally, the interference suppression residual components are extended along the interference residual separation axis to form the third dimension, thus synthesizing a three-dimensional feature tensor.

[0094] Here is a specific example: When a certain air surveillance radar processes a low-altitude maneuvering aircraft, it first calculates the orthogonal phase difference between the real part (0.65, imaginary part 0.12) of the horizontally polarized signal and the real part (-0.21, imaginary part 0.73) of the vertically polarized signal. Through the product of the real and imaginary parts and the calculation using the arctangent function, a phase difference of 122.8 degrees is obtained, reflecting the modulation effect of the aircraft's control surface deflection on the electromagnetic wave scattering characteristics. Subsequently, a three-dimensional orthogonal coordinate system is constructed based on this phase difference: the cosine component -0.54 is used to define the principal axis of the energy response [-0.54, 0, 0], and the sine component 0.84 is used to define the polarization difference axis [0, 0.84, 0]. The interference residual separation axis [0, 0, -1] is generated through vector cross product. For a target flying at a constant speed at high altitude, the horizontal polarization signal [0.92, 0.15] is projected onto the principal axis of the energy response using a dot product to obtain an amplitude component of 0.50 representing the target intensity. Simultaneously, the vertical polarization signal [-0.08, 0.42] is projected onto the polarization difference axis to generate a phase difference component of 0.35 reflecting the polarization characteristics. For a target intersecting with a flock of birds' trajectories, the coordinates of its projection points (-0.21, 0) and (0, 0.38) on the energy-polarization plane are calculated, and the projection difference of 0.43 is obtained using the Euclidean distance formula. This difference is projected onto the residual separation axis, and after absolute value limiting and logarithmic transformation, an interference suppression residual component of 0.15 is finally generated. All feature components are integrated along the three coordinate axes into a 128×128×3 three-dimensional feature tensor: at the target location (60,75), the energy layer is assigned a value of 0.50, the polarization layer is assigned a value of 0.35, and the residual layer is assigned a value of 0.00; while the residual layer for intersecting targets is increased to 0.15, effectively separating bird flock interference features. This processing takes only 2 milliseconds with GPU acceleration, achieving physical interpretability of feature decoupling, namely, the energy layer highlights the reflection of the metallic fuselage, the polarization layer captures the dynamic modulation characteristics of the control surfaces, and the residual layer isolates natural interference, establishing a highly discriminative feature basis for subsequent target tracking.

[0095] By executing steps 121 to 126, this embodiment of the application achieves signal decoupling by constructing a three-dimensional orthogonal space: the energy axis preserves the essential features of the target, the polarization axis extracts discriminative differences, and the residual axis separates noise interference; the finally generated three-dimensional feature tensor fuses multi-dimensional information to provide a highly discriminative input basis for subsequent processing.

[0096] In one possible embodiment, step 125, projecting the projection difference onto the interference residual separation axis to generate an interference suppression residual component, includes:

[0097] Step a1: Perform amplitude compression on the projection difference to generate normalized residual amplitude.

[0098] Amplitude compression refers to the process of applying amplitude limits and nonlinear transformations to the projected difference, including threshold truncation to eliminate extreme value interference and logarithmic scaling to enhance the visibility of weak signals. Normalized residual amplitude refers to the standardized residual signal output after amplitude compression, whose numerical range is constrained within a unit interval, used to unify the measurement benchmark for interference of different intensities.

[0099] In this embodiment, the projection difference is first compressed by a preset threshold upper limit to cut off abnormally high amplitude points. Then, a logarithmic function is used to nonlinearly scale the truncated amplitude. Finally, the normalized residual amplitude is generated by dividing by the reference energy value.

[0100] Step a2: Generate the residual density distribution using the distribution function template based on the normalized residual amplitude.

[0101] The distribution function template refers to a mathematical model in a pre-defined library of probability density functions, constructed based on the Weibull or Rayleigh distribution, used to fit the statistical characteristics of the residual amplitude. The residual density distribution refers to the result of probabilistic modeling of the normalized residual amplitude using the distribution function template, reflecting the statistical distribution law of the residual signal.

[0102] In this embodiment, the pulse characteristics of the time-domain waveform of the normalized residual amplitude are first identified by scanning. Then, the amplitude fluctuation quantization index is calculated based on the position of the rising and falling edges of the pulse. Subsequently, the distribution function template is matched from the preset template library. Then, the statistical moment characteristics of the residual amplitude are extracted to calculate the template shape parameters. Finally, the residual density distribution is generated using the parameterized distribution function template.

[0103] Step a3: Discretize the residual density distribution along the interference residual separation axis to generate a residual vector sequence.

[0104] Discretization refers to the process of transforming a continuous residual density distribution into a finite number of discrete points along the interference residual separation axis, achieving continuous signal vectorization through equally spaced sampling. The residual vector sequence is the set of probability density values ​​arranged in axial coordinate order after discretization, forming a one-dimensional structured residual representation.

[0105] In this embodiment, firstly, equally spaced discrete coordinate points are set along the axis of the interference residual separation. Secondly, the probability density value of the residual density distribution at each coordinate point is calculated. Then, the probability density values ​​are arranged in coordinate order to generate a residual vector sequence.

[0106] Step a4: Perform tensor expansion on the residual vector sequence along the interference residual separation axis to generate interference suppression residual components.

[0107] Tensor expansion refers to the operation of copying the residual vector sequence along the axis of the interference residual separation, and achieving three-dimensional integration with other components through the Kronecker product.

[0108] In the embodiments of this application, the residual vector sequence is first copied and extended along the interference residual separation axis, then tensor multiplication is performed with the dimensions of the amplitude component and the polarization phase difference component, and finally the interference suppression residual component of the three-dimensional structure is generated.

[0109] Here is a specific example: First, the airborne radar system performs amplitude compression on the projection difference of a low-altitude UAV target: setting an upper threshold to truncate sudden interference, and then dividing the normalized residual amplitude by the mean of the background noise after logarithmic scaling. Next, the pulse waveform characteristics of this residual amplitude are analyzed, a Weibull distribution template is matched, and shape parameters are calculated to generate a residual density probability model. Then, discrete coordinate points are set along the interference residual separation axis to sample probability densities, forming a residual vector sequence. Finally, this sequence is extended along the residual axis to three-dimensional space and synthesized with the energy and polarization dimensions to form the final interference suppression residual component.

[0110] By executing steps a1 to a4, the embodiments of this application suppress abnormal interference pulses through amplitude compression, enhance the anti-fluctuation capability by fitting the residual characteristics using a statistical distribution model, achieve structured transformation of continuous signals through discretization, and finally ensure the integrity of three-dimensional features through tensor expansion, thus providing robust protection for interference suppression.

[0111] In one possible embodiment, step a2, generating a residual density distribution using a distribution function template based on the normalized residual magnitude, includes:

[0112] Step a21: Scan the time-domain waveform of the normalized residual amplitude, identify the rising and falling edges of the pulse, and generate an amplitude fluctuation quantification index.

[0113] In this context, the time-domain waveform refers to the curve shape of the normalized residual amplitude changing over time, reflecting the amplitude fluctuation characteristics of the interference signal. The identified pulse rising and falling edge positions directly determine the calculation benchmark for the amplitude fluctuation quantification index. First, the effective range of the pulse is defined using the rising edge position as the starting point and the falling edge position as the ending point. Second, the ratio of the maximum change in the normalized residual amplitude within this range to the time span is calculated to generate the amplitude change rate reflecting the steepness of the interference pulse. Finally, this change rate is multiplied by the pulse width normalization coefficient to output the amplitude fluctuation quantification index characterizing the intensity of the interference impact. The pulse rising edge refers to the transition segment in the time-domain waveform where the residual amplitude transitions from a low level to a high level, used to locate the start time of the interference pulse. The falling edge refers to the transition segment in the time-domain waveform where the residual amplitude transitions from a high level to a low level, used to locate the end time of the interference pulse. The position refers to the coordinate points of the pulse rising and falling edges on the time axis, marked by the sampling point sequence number. The amplitude fluctuation quantification index is a scalar value calculated based on the rate of change of amplitude between the rising and falling edges, used to measure the steepness of the interference pulse.

[0114] In this embodiment, the time-domain waveform of the normalized residual amplitude is first scanned, and the signal slope abrupt change point is detected by differential operation; secondly, the rising edge position of the pulse is identified as the turning point of the waveform from low amplitude to high amplitude, and the falling edge position is identified as the turning point of high amplitude to low amplitude; finally, the amplitude change rate between the rising edge and the falling edge is calculated to generate an amplitude fluctuation quantification index.

[0115] Step a22: Match a distribution function template from the preset residual distribution function template library based on the amplitude fluctuation quantification index.

[0116] The pre-set residual distribution function template library refers to a pre-stored collection of statistical models, including probability density function prototypes such as Rayleigh and Weibull distributions. The physical meaning of a distribution function template is a mathematical abstract model of the statistical characteristics of interference residuals. The template content includes probability density function forms defining the probability distribution law of residual amplitudes, such as Weibull or Rayleigh distributions; adjustable shape and scale parameters controlling the distribution shape; and mapping rules between quantitative indicators of amplitude fluctuations and distribution types. The core difference between different templates lies in their fundamentally different distribution types and physical meanings: the distribution function is suitable for modeling uniform background noise generated by a large number of weak scatterers, characterized by a single scale parameter and no shape parameter; while the Weibull distribution is suitable for sudden fluctuations dominated by a few strong scatterers, sensitively reflecting the steepness of the pulse through its shape parameter. The two correspond to different physical mechanisms of interference generation.

[0117] In this embodiment, the numerical range of the amplitude fluctuation quantification index is first determined; then, a matching distribution type is selected from a preset residual distribution function template library. When the index is below a threshold, the Rayleigh distribution template is matched, and when it is above the threshold, the Weibull distribution template is matched; finally, the selected distribution function template is output.

[0118] Step a23: Extract statistical moment features from the normalized residual amplitude based on the positions of the rising and falling edges of the pulse. Calculate the shape parameters of the matching distribution function template based on the statistical moment features of the normalized residual amplitude.

[0119] Among these, statistical moment characteristics refer to the numerical features describing the statistical properties of residual amplitudes, including the mean reflecting the central location and the variance reflecting the degree of dispersion. Shape parameters are key variables that control the shape of the distribution function curve; for example, the shape coefficient in the Weibull distribution determines the skewness of the distribution.

[0120] In this embodiment, firstly, the effective segment of the normalized residual amplitude is extracted based on the positions of the rising and falling edges of the pulse; secondly, the statistical moment features of the segment are extracted, including the mean of the first-order original moment and the variance of the second-order central moment; finally, the statistical moment features are input into the parameter calculator of the distribution function template, and the shape parameters are output.

[0121] Step a24: Based on the normalized residual magnitude, generate the residual density distribution using a matching distribution function template with shape parameters.

[0122] In this embodiment, the normalized residual amplitude is first used as the input independent variable; then, a matching distribution function template with shape parameters is called; finally, the probability density value corresponding to each amplitude point is calculated to generate the residual density distribution function.

[0123] Here is a specific example: First, the shipborne radar system scans the normalized residual amplitude waveform of ship targets in a sea clutter environment, identifies the pulse rise and fall edges, and calculates the amplitude fluctuation quantification index. Next, based on this index, a Weibull distribution function template is matched from a template library. Then, within the interval defined by the pulse edges, the mean and variance statistical moments of the residual amplitude are extracted, and the shape parameters of the template are calculated. Finally, the normalized residual amplitude is input into the Weibull distribution template with shape parameters to generate the residual density probability distribution function.

[0124] By executing steps a21 to a24, this embodiment of the application dynamically matches the optimal statistical model with waveform features, accurately extracts residual characteristics using pulse edge information, and combines statistical moments to parameterize and adjust the distribution shape, thereby achieving adaptive modeling of complex interference environments and improving the physical rationality of residual density distribution.

[0125] In one possible embodiment, step a24, generating a residual density distribution based on the normalized residual magnitude using a matching distribution function template with shape parameters, includes:

[0126] Step b1: Divide the normalized residual amplitude into multiple pulse window residual sequences according to the radar pulse repetition period.

[0127] The radar pulse repetition period refers to the fixed time interval between two adjacent pulses emitted by the radar, which determines the maximum detection range and serves as the basis for signal segmentation. The pulse window residual sequence refers to the normalized residual amplitude data segment within a single pulse period, containing residual information for a specific range gate.

[0128] In the embodiments of this application, the pulse repetition period parameter of the radar system is first obtained, and then the continuous normalized residual amplitude is divided into multiple equal-length segments using the period length as a time window. Finally, each segment forms a pulse window residual sequence.

[0129] Step b2: Perform parallel computation on each pulse window residual sequence using a matching distribution function template with shape parameters to generate a window residual density distribution.

[0130] Parallel computing refers to the technique of using multi-core processors to perform the same operation on multiple data windows simultaneously, accelerating the calculation of distribution functions through parallel processing of graphics processing unit (GPU) threads. The window residual density distribution is the probability density function of the residual amplitude within a single pulse window, reflecting local statistical characteristics.

[0131] In this embodiment, a matching distribution function template with shape parameters is first loaded, then the probability density is calculated independently for each pulse window residual sequence on the graphics processor, and finally a corresponding window residual density distribution function is generated for each window.

[0132] Step b3: Overlay the window residual density distributions of all windows along the pulse sequence dimension to generate the pulse cumulative residual distribution.

[0133] Here, the pulse sequence dimension refers to the pulse window index axis arranged in chronological order, used to organize the stacking direction of multi-window data. The pulse cumulative residual distribution refers to the comprehensive probability model formed by superimposing all window density distributions along the pulse sequence dimension.

[0134] In the embodiments of this application, all window residual density distributions are first aligned along the pulse time series dimension, then the density values ​​of the corresponding distance cells are added point by point, and finally a pulse cumulative residual distribution characterizing the overall statistical properties is generated.

[0135] Step b4: Map the pulse accumulation residual distribution onto the distance Doppler plane to generate the residual density distribution.

[0136] The range-Doppler plane refers to a two-dimensional coordinate system with target range as the horizontal axis and Doppler frequency shift as the vertical axis, used for spatial frequency domain signal representation. The relationship between the range-Doppler plane and orthogonal basis vectors lies in the fact that the range-Doppler plane integrates the three-dimensional information of the orthogonal basis vectors through coordinate transformation. The range dimension corresponds to the target echo intensity characteristics along the principal axis of the energy response, while the Doppler dimension incorporates the phase dynamic change characteristics along the polarization difference axis. The noise statistical characteristics captured by the interference residual separation axis in the pulse accumulation residual distribution are encoded as the amplitude intensity at each coordinate point on the range-Doppler plane, thus uniformly mapping the energy, polarization, and residual characteristics of the orthogonal spatial decomposition to the classical radar signal representation domain.

[0137] In the embodiments of this application, a distance-Doppler plane coordinate system is first established, then the distance dimension of the pulse cumulative residual distribution is mapped to the horizontal axis of the plane, and the Doppler frequency is mapped to the vertical axis. Finally, a two-dimensional residual density distribution is generated by bilinear interpolation.

[0138] Here is a specific example: First, the airborne radar system divides the normalized residual amplitude of a cruise missile target into multiple pulse window residual sequences according to the pulse repetition period. Next, Weibull distribution calculations are performed in parallel on a graphics processor to generate the residual density distribution for each window. Then, the density functions of all windows are superimposed along the pulse sequence dimension to form the pulse cumulative residual distribution. Finally, this distribution is mapped to the range-Doppler plane to generate a spatialized residual density distribution that includes range-velocity interference features.

[0139] By executing steps b1 to b4, the embodiments of this application achieve temporal localization analysis of interference through pulse segmentation, improve processing efficiency through parallel computing, enhance statistical robustness by superimposing pulse dimensions, and finally map to the distance to the Doppler plane to form a spatially resolvable interference distribution characterization.

[0140] In one possible embodiment, S14, based on spatiotemporal correlation weights, select multi-target blocks that meet preset continuous motion conditions, and use a long short-term memory network to decode the motion patterns of the multi-target blocks to generate anti-interference target trajectory coordinates and velocity vectors, including:

[0141] Step 141: Perform motion continuity analysis on the spatiotemporal correlation weights to generate a motion continuity confidence index that characterizes the motion correlation strength of the target block between consecutive observation frames.

[0142] Motion continuity analysis refers to analyzing the state transition probability of a target block between consecutive observation frames using a Markov chain model, reflecting the physical rationality of the target's trajectory. Consecutive observation frames refer to a sequence of signal snapshots collected by radar at fixed time intervals, forming the time reference for target motion analysis. Motion correlation strength refers to parameters that quantify the cross-frame positional correlation of the target block, including displacement vector similarity and acceleration continuity. The motion continuity confidence index is a normalized scalar value ranging from 0 to 1; a larger value indicates that the target's trajectory conforms more closely to physical laws.

[0143] In this embodiment, motion continuity analysis is first performed on the spatiotemporal correlation weights: the displacement change of the target block between consecutive observation frames is extracted, the displacement direction consistency coefficient and velocity change smoothness are calculated, and then the two indicators are fused to generate a motion continuity confidence index characterizing the motion correlation strength.

[0144] Step 142: Select target blocks whose motion continuity confidence index is within the threshold range of the preset continuous motion conditions to form a target block set.

[0145] The threshold range refers to the preset confidence level range, with the lower limit excluding random noise interference and the upper limit filtering out maneuvering targets with sudden changes in direction. The target block set refers to the spatially correlated unit group formed through screening, representing the echo set of the same target in multiple range gates.

[0146] In this embodiment, firstly, a threshold range upper and lower limits for a preset continuous motion condition are set; secondly, the motion continuity confidence index is compared with the range; then, target blocks whose index values ​​are within the range are selected; and finally, they are aggregated into a target block set containing multi-spatial-layer associated targets.

[0147] Step 143: Input the historical phase sequence of the target block set in the Doppler spectrum into the long short-term memory network for motion pattern decoding to generate the target motion vector field.

[0148] The Doppler spectrum refers to the frequency-amplitude distribution generated by the Fourier transform of the target block echo signal, reflecting radial motion characteristics. The historical phase sequence refers to the sequence of the main lobe phase of the target block in the Doppler spectrum over time, carrying target acceleration information. The target motion vector field refers to the structured data output by the Long Short-Term Memory network, containing the velocity vector distribution of spatial location.

[0149] In this embodiment, the historical phase sequence in the Doppler spectrum is first extracted from the target block set, then input into a long short-term memory network, outliers are filtered through a forget gate, the state is updated by the input gate, the instantaneous velocity vector is generated by the output gate, and finally the target motion vector field is synthesized.

[0150] Step 144: Decompose the target motion vector field into radial velocity components and tangential velocity components. Based on the radial velocity components and tangential velocity components, perform joint trajectory and velocity calculation to generate the target trajectory coordinates and velocity vector after interference resistance.

[0151] The radial velocity component refers to the projection of the target motion vector onto the radar beam direction, used to calculate the target range change rate. The tangential velocity component refers to the planar component of the target motion vector perpendicular to the beam direction, determining the target azimuth change. Joint trajectory and velocity calculation refers to the Kalman filtering process that simultaneously optimizes the position coordinates and velocity vectors, ensuring the physical consistency of the motion parameters.

[0152] In this embodiment, the target motion vector field is first decomposed into a radial velocity component in the radar line of sight and a tangential velocity component in the vertical direction. Then, the range change is calculated based on the radial component integration, and the azimuth change is calculated based on the tangential component. Finally, the target trajectory coordinates and the synthesized velocity vector after anti-jamming are fused together.

[0153] Here is a specific example: First, motion continuity analysis is performed on the spatiotemporal correlation weights of a low-altitude UAV target block to generate a motion continuity confidence index. Next, target blocks with index values ​​between 0.8 and 0.95 are selected to form a set. Then, the historical phase sequence of this set in the Doppler spectrum is input into a long short-term memory network for decoding to generate the target motion vector field. Finally, the vector field is decomposed into radial and tangential velocity components, and the anti-jamming three-dimensional track coordinates and velocity vector are output through joint calculation.

[0154] By executing steps 141 to 144, this embodiment of the application eliminates non-existent targets through motion continuity analysis, accurately models complex maneuvers using long short-term memory networks, and achieves trajectory reconstruction by combining radial and tangential velocity decomposition, effectively overcoming the trajectory breakage problem under strong interference.

[0155] In one possible embodiment, step 126, integrating the amplitude component, polarization phase difference component, and interference suppression residual component along the three coordinate axes of the orthogonal basis vectors to generate a three-dimensional feature tensor, includes:

[0156] Step c1: Perform an alignment operation on the amplitude component along the principal axis of the energy response to generate the first axial characteristic component.

[0157] The alignment operation refers to the mathematical processing of transforming feature data to the reference axis, including coordinate rotation to eliminate directional deviations and scale normalization to eliminate dimensional differences. The first axial feature component refers to the normalized data layer generated after the amplitude components are aligned along the principal axes of the energy response, reflecting the spatial distribution characteristics of the target echo intensity.

[0158] In this embodiment, the original coordinate values ​​of the amplitude component along the principal axis of the energy response are first obtained, then the coordinate transformation is performed based on the unit vector of this axis, then the scale difference between distance units is eliminated by least squares fitting, and finally the first axial feature component with uniform scale is generated.

[0159] Step c2: Perform an alignment operation on the polarization phase difference component along the polarization difference axis to generate the second axial characteristic component.

[0160] The second axial characteristic component refers to the phase difference distribution layer formed after the polarization phase difference components are aligned along the polarization difference axis, which characterizes the continuous change of polarization characteristics in the observation area.

[0161] In this embodiment, the distribution data of the polarization phase difference component along the polarization difference axis is first read, then the cosine value of the angle between the component and the axial reference direction is calculated, then the data is aligned to the positive axial direction by rotation transformation, and finally a second axial feature component with the same direction is generated.

[0162] Step c3: Perform a mapping operation on the interference suppression residual component along the interference residual separation axis to generate a third axial characteristic component.

[0163] The mapping operation refers to the process of converting a continuous probability distribution into discrete axial coordinates, achieving data structuring through domain compression and range quantization. The third axial feature component refers to the discretized residual layer generated by the mapping operation from the interference suppression residual component, carrying spatial distribution information of noise statistical characteristics.

[0164] In this embodiment, the probability density function of the interference suppression residual component along the interference residual separation axis is first determined, then the function value is normalized to a preset dynamic range, then converted into discrete coordinate values ​​through linear mapping, and finally the dimension-matched third axial feature component is generated.

[0165] Step c4: Perform a three-dimensional tensor synthesis operation on the first axial feature component, the second axial feature component, and the third axial feature component to generate a three-dimensional feature tensor.

[0166] Among them, the three-dimensional tensor synthesis operation refers to the process of constructing a high-order tensor by orthogonally relating the three axial feature components, and achieving dimensional fusion through the Kronecker product.

[0167] In this embodiment, the first axial feature component is first used as the X-axis data layer in three-dimensional space, the second axial feature component is used as the Y-axis data layer, the third axial feature component is used as the Z-axis data layer, and finally the three-dimensional feature tensor is synthesized by Kronecker product operation.

[0168] Here is a specific example: First, an energy response principal axis alignment operation is performed on the amplitude component of a certain aircraft target to generate the first axial feature component. Next, the polarization phase difference component is rotated and aligned along the polarization difference axis to generate the second axial feature component. Then, the interference suppression residual component is mapped to discrete coordinate values ​​along the residual separation axis to generate the third axial feature component. Finally, the three feature layers are fused through a 3D tensor synthesis operation to generate a 3D feature tensor for target recognition.

[0169] By executing steps c1 to c4, the embodiments of this application achieve spatial consistency representation of multi-source features through axial alignment, transform continuous statistics into computable data structures through mapping transformation, and finally construct a three-dimensional information carrier compatible with energy, polarization and noise features through tensor synthesis, providing structured input for deep learning models.

[0170] Figure 2 A schematic diagram of a radar signal anti-jamming processing system based on a spatiotemporal attention mechanism is provided for an embodiment of this application, as shown below. Figure 2 As shown, the system includes:

[0171] The acquisition module 21 is used to acquire the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar.

[0172] The mapping module 22 is used to map the original horizontal polarization channel signal and the original vertical polarization channel signal to the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components.

[0173] The partitioning module 23 is used to divide the radar observation area into multiple target blocks in the spatial dimension according to the amplitude component, and calculate the spatiotemporal correlation weight between adjacent target blocks through a spatiotemporal attention mechanism.

[0174] The filtering module 24 is used to filter out multi-target blocks that meet the preset continuous motion conditions based on the spatiotemporal correlation weights, and use a long short-term memory network to decode the motion mode of the multi-target blocks to generate the target trajectory coordinates and velocity vectors after anti-interference.

[0175] Figure 2 The aforementioned radar signal anti-jamming processing system based on a spatiotemporal attention mechanism can perform... Figure 1 The implementation principle and technical effects of the radar signal anti-jamming processing method based on spatiotemporal attention mechanism described in the illustrated embodiment will not be repeated here. The specific operation methods of each module and unit in the radar signal anti-jamming processing system based on spatiotemporal attention mechanism in the above embodiments have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0176] In one possible design, Figure 2 The radar signal anti-jamming processing system based on a spatiotemporal attention mechanism shown in the embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32.

[0177] The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0178] The processing component 32 is used to perform the following process: acquiring the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar; mapping the original horizontal polarization channel signal and the original vertical polarization channel signal to the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components; dividing the radar observation area into multiple target blocks in the spatial dimension according to the amplitude components, and calculating the spatiotemporal correlation weight between adjacent target blocks through a spatiotemporal attention mechanism; selecting multiple target blocks that meet the preset continuous motion conditions according to the spatiotemporal correlation weights, and decoding the motion mode of the multiple target blocks using a long short-term memory network to generate the anti-interference target trajectory coordinates and velocity vectors.

[0179] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0180] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Random Access Memory (RAM), Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read Only Memory (PROM), Read Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0181] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0182] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0183] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0184] The computing device can be a physical device or an elastic computing host provided by a cloud computing platform. In this case, the computing device can refer to a cloud server, and the aforementioned processing components, storage components, etc., can be basic server resources rented or purchased from the cloud computing platform.

[0185] This application also provides a computer storage medium storing a computer program, which, when executed by a computer, can perform the above-described functions. Figure 1 The embodiment shown is a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism.

[0186] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0187] The device embodiments described above are merely illustrative. 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 modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0188] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A radar signal anti-jamming processing method based on a spatiotemporal attention mechanism, characterized in that, include: Acquire the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar; The original horizontal polarization channel signal and the original vertical polarization channel signal are mapped into the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components. Based on the amplitude components, the radar observation area is divided into multiple target blocks in the spatial dimension, and the spatiotemporal correlation weights between adjacent target blocks are calculated through a spatiotemporal attention mechanism. Based on the spatiotemporal correlation weights, multiple target blocks that meet the preset continuous motion conditions are selected, and the motion patterns of the multiple target blocks are decoded using a long short-term memory network to generate anti-interference target trajectory coordinates and velocity vectors.

2. The method according to claim 1, characterized in that, The process of mapping the original horizontally polarized channel signal and the original vertically polarized channel signal into the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components, and interference suppression residual components includes: Calculate the orthogonal phase difference between the original horizontal polarization channel signal and the original vertical polarization channel signal; Based on the cosine and sine components of the orthogonal phase difference, an orthogonal basis vector for the interference suppression space is constructed. The orthogonal basis vector includes the principal axis of energy response, the polarization difference axis, and the interference residual separation axis. The original horizontal polarization channel signal is projected onto the principal axis of the energy response to generate an amplitude component, and the original vertical polarization channel signal is projected onto the polarization differential axis to generate a polarization phase difference component. Based on the amplitude component and the polarization phase difference component, calculate the projection difference between the original horizontal polarization channel signal and the original vertical polarization channel signal on the plane spanned by the principal axis of the energy response and the polarization difference axis. The projection difference is projected onto the interference residual separation axis to generate an interference suppression residual component. The amplitude component, the polarization phase difference component, and the interference suppression residual component are integrated along the three coordinate axes of the orthogonal basis vector to generate a three-dimensional feature tensor.

3. The method according to claim 2, characterized in that, The step of projecting the projection difference onto the interference residual separation axis to generate interference suppression residual components includes: The projection difference is subjected to amplitude compression to generate a normalized residual amplitude. Based on the normalized residual magnitude, a residual density distribution is generated using a distribution function template; Discretize the residual density distribution along the interference residual separation axis to generate a residual vector sequence; The residual vector sequence is tensor-expanded along the interference residual separation axis to generate interference suppression residual components.

4. The method according to claim 3, characterized in that, The step of generating a residual density distribution using a distribution function template based on the normalized residual magnitude includes: The time-domain waveform of the normalized residual amplitude is scanned to identify the positions of the rising and falling edges of the pulse, and an amplitude fluctuation quantification index is generated. Based on the amplitude fluctuation quantification index, a distribution function template is matched from a preset residual distribution function template library; Based on the positions of the rising and falling edges of the pulse, statistical moment features are extracted from the normalized residual amplitude. Based on the statistical moment features of the normalized residual amplitude, the shape parameters of the matching distribution function template are calculated. Based on the normalized residual magnitude, a residual density distribution is generated using a matching distribution function template with the shape parameters.

5. The method according to claim 4, characterized in that, The step of generating a residual density distribution based on the normalized residual amplitude using a matching distribution function template with the shape parameters includes: According to the radar pulse repetition period, the normalized residual amplitude is divided into multiple pulse window residual sequences; Parallel computation operations are performed on each of the pulse window residual sequences using a matching distribution function template with the shape parameters to generate a window residual density distribution; The window residual density distributions of all windows are superimposed along the pulse sequence dimension to generate the pulse cumulative residual distribution; The pulse accumulation residual distribution is mapped onto the distance Doppler plane to generate a residual density distribution.

6. The method according to claim 1, characterized in that, The step of selecting multiple target blocks that meet preset continuous motion conditions based on the spatiotemporal correlation weights, and decoding the motion patterns of the multiple target blocks using a long short-term memory network to generate anti-interference target trajectory coordinates and velocity vectors includes: Motion continuity analysis is performed on the spatiotemporal correlation weights to generate a motion continuity confidence index that characterizes the motion correlation strength of the target block between consecutive observation frames; Target blocks that fall within the threshold range of a preset continuous motion condition, based on the motion continuity confidence index, are selected to form a target block set; The historical phase sequence of the target block set in the Doppler spectrum is input into a long short-term memory network for motion pattern decoding to generate a target motion vector field. The target motion vector field is decomposed into radial velocity components and tangential velocity components. Based on the radial velocity components and tangential velocity components, the trajectory and velocity are jointly calculated to generate the target trajectory coordinates and velocity vector after interference resistance.

7. The method according to claim 2, characterized in that, The step of integrating the amplitude component, the polarization phase difference component, and the interference suppression residual component along the three coordinate axes of the orthogonal basis vectors to generate a three-dimensional feature tensor includes: Alignment operations are performed on the amplitude components along the principal axis of the energy response to generate a first axial characteristic component; Alignment operation is performed on the polarization phase difference components along the polarization difference axis to generate a second axial characteristic component; A mapping operation is performed on the interference suppression residual component along the interference residual separation axis to generate a third axial characteristic component; Perform a three-dimensional tensor synthesis operation on the first axial feature component, the second axial feature component, and the third axial feature component to generate a three-dimensional feature tensor.

8. A radar signal anti-jamming processing system based on a spatiotemporal attention mechanism, characterized in that, include: The acquisition module is used to acquire the original horizontal polarization channel signal and the original vertical polarization channel signal of the radar; The mapping module is used to map the original horizontal polarization channel signal and the original vertical polarization channel signal to the interference suppression space to generate a three-dimensional feature tensor containing amplitude components, polarization phase difference components and interference suppression residual components. The partitioning module is used to divide the radar observation area into multiple target blocks in the spatial dimension according to the amplitude component, and to calculate the spatiotemporal correlation weight between adjacent target blocks through a spatiotemporal attention mechanism. The filtering module is used to filter out multi-target blocks that meet the preset continuous motion conditions according to the spatiotemporal correlation weights, and use a long short-term memory network to decode the motion mode of the multi-target blocks to generate anti-interference target trajectory coordinates and velocity vectors.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a radar signal anti-jamming processing method based on a spatiotemporal attention mechanism as described in any one of claims 1 to 7.

Images