Structure tensor anisotropic diffusion interference suppression method guided by time-frequency energy threshold

Abstract

The invention discloses a structure tensor anisotropic diffusion interference suppression method guided by a time-frequency energy threshold, and relates to the technical field of radars, and the method comprises the steps: determining a mixed signal; performing short-time Fourier transform on the mixed signal to obtain a time-frequency diagram; calculating an energy boundary threshold value based on the relationship between the interference-to-signal ratio and the energy of the time-frequency diagram; a core layer is determined from the time-frequency graph based on the energy boundary threshold value, a structure tensor is determined by performing Gaussian smoothing operation on the core layer, and the core layer is an interference core area; using the empirical coefficient and the structure tensor to construct a diffusion tensor used for controlling the diffusion direction and intensity; and carrying out anisotropic diffusion iterative filtering on the time-frequency diagram by using the diffusion tensor to obtain an anti-interference mixed signal, thereby providing an interference suppression method which can realize self-adaptive suppression and maximally maintain target information under the condition that an interference model, threshold experience or a training sample does not need to be preset.

Description

Technical Field

[0001] This invention belongs to the field of radar technology, specifically relating to a time-frequency energy threshold-guided method for suppressing anisotropic diffusion interference of structural tensors. Background Technology

[0002] In modern warfare, mainlobe active jamming (MAJ) has become a core electronic warfare tool for suppressing radar systems. It detects radar signals and generates deception / suppression jamming with waveform coherence and directional coaxiality in real time, causing operational-level harm. (1) Target masking effect: The jammer injects high-power noise (typical JSR>30dB) into the radar main lobe beam, causing receiver saturation. When the jamming bandwidth covers 80% of the radar signal bandwidth, the target detection probability drops significantly. (2) False Target Induction: Dense groups of false targets are generated using intermittent sample-and-forward jamming (ISRJ) based on DRFM (Digital Radio Frequency Storage). For example, in LFM (Linear Frequency Modulation) radar, the pulse width... The sampling forwarding can generate up to A false target, The increased radar pulse width leads to a higher false alarm rate.

[0003] In actual radar countermeasures scenarios, jammers typically perform selective processing and waveform reconstruction on intercepted signals. Moreover, jammers must maintain an energy advantage of at least 10dB over the signal to ensure jamming effectiveness, thus forming a composite scenario of "strong jamming + weak target". Radar systems urgently need to complete jamming suppression under unknown systems, variable power differences, and strict real-time constraints, which poses a severe challenge to traditional algorithms.

[0004] To address the aforementioned challenges, existing traditional methods for interference suppression can be broadly categorized into three types: 1. Non-parametric methods: Non-parametric methods do not rely on specific signal models or parameter settings. They suppress interference by analyzing the statistical characteristics of the signal. These mainly include: Subspace projection method: This method decomposes the received signal into mutually orthogonal signal and noise spaces. First, singular value decomposition is performed on the signal matrix to identify the subspace basis representing the noise components. Then, a projection operator is constructed to project the original signal onto the noise subspace and subtract it, thus preserving the target signal components. This method is widely used in airborne radar radio frequency interference suppression. Time-frequency domain threshold filtering: Based on short-time Fourier transform, the signal is mapped to the time-frequency plane, and the interference area is identified through energy distribution characteristics. A dynamic energy threshold is set, typically 30%-70% of the maximum energy, and a binary mask is generated to remove high-energy areas. Finally, the time-frequency map after masking is inversely transformed to reconstruct the signal. Wavelet multi-scale decomposition: Utilizing the frequency band localization characteristics of wavelet basis functions, the signal is decomposed into components of different scales. By analyzing the anomalous energy in the high-frequency detail coefficients, the frequency band where the interference is located is located. After thresholding or zeroing, the signal is reconstructed using low-frequency approximation coefficients.

[0005] 2. Parameterization Methods: Parameterization methods rely on model assumptions about the signal. Based on these assumptions, parameter information of the signal is extracted, and interference suppression is performed using these parameters. Common parameterization methods include model-based filtering and matched filtering. For intermittent sampling-forwarding interference, a periodic truncation-forwarding model is first established, followed by estimation: key parameters such as pulse width and intermittent sampling time are extracted by detecting envelope dips, and then corresponding interference replicas are generated for coherent cancellation.

[0006] 3. Deep Learning Methods: With the rise of deep learning, interference suppression methods based on deep learning have gradually become a research hotspot. These methods typically train neural network models to learn the characteristics of signals and interference from large amounts of data, and suppress interference through model predictions. The main methods include: Unrolled Deep Networks: mapping iterative optimization algorithms, such as soft thresholding, to neural network layers to achieve hyperparameter adaptive learning; Time-Frequency Graph Segmentation Networks: treating the time-frequency spectrum as an image, using U-Net (Time-Frequency Graph Segmentation Network) to segment interference regions; Generative Adversarial Networks (GANs): constructing a generator-discriminator adversarial framework. The generator learns to map the interfered signal to a clean signal, while the discriminator distinguishes between real and generated signals; the two improve generation quality through alternating adversarial training. After training, the generator can directly process novel interference signals, exhibiting strong adaptability in complex electromagnetic environments.

[0007] Among them, anti-jamming methods based on time-frequency domain analysis have gradually become a research hotspot due to their ability to simultaneously capture the joint time-frequency characteristics of signals, their lack of model dependence, and their good generalization ability. This method performs short-time Fourier transform (STFT) on the mixed radar and jamming signals to obtain their time-frequency domain signals. For the time-frequency map, the OTU (Otsu method) dynamic thresholding method is used for threshold segmentation to obtain the jamming, signal, and noise components. Pixels exceeding the threshold are set to zero, initially achieving suppression of strong interference bands. To further remove the residual energy of the jamming signal after suppression on the time-frequency map, the interference edge notch method is used for interference residue processing. This method can effectively handle different types of jamming signals.

[0008] The existing traditional methods for interference suppression have the following drawbacks: (1) Non-parametric methods: a) Highly dependent on significant differences in interference energy: Separation is only easy when the interference power is much higher than the target-noise floor (above 20dB in strong interference scenarios); when the interference-signal power is close to equilibrium or at a low signal-to-noise ratio, the target features are submerged and the subspace / threshold partitioning fails.

[0009] b) There is no unified standard for threshold selection: energy threshold and singular value threshold need to be determined by experience or offline calibration, which makes it difficult to adapt to different scenarios, resulting in a trade-off between suppression strength and signal fidelity.

[0010] c) Hard threshold resection introduces information gaps: if the threshold is too low, it will weaken the target main lobe; if it is too high, interference will remain, creating a dilemma of "undersuppression or oversuppression".

[0011] (2) Parameterization method: a) Strongly dependent on the interference mechanism model: It is necessary to first accurately establish a mathematical model of "forwarding / sweeping / modulation" and estimate the model parameters (pulse width, modulation rate, etc.). New or variant interference will become ineffective once it deviates from the assumptions.

[0012] b) Long computation time: Establishing interference signal parameter models, estimating parameters, extracting interference in real time, and hardware resource consumption are all limited.

[0013] (3) Deep learning methods: a) Large-scale "interference-free" labeled data is required: It is difficult to collect comprehensive data in real radar confrontation scenarios; the deviation between simulated data and measured distribution leads to limited inter-domain migration.

[0014] b) Poor generalization ability: When faced with unfamiliar interference systems or combinations of interference, the network is prone to over-suppression (target loss) or under-suppression (interference residue); frequent retraining is required.

[0015] Therefore, it has become an important issue how to provide an interference suppression method that can achieve adaptive suppression and preserve target information to the maximum extent without the need for a pre-set interference model, threshold experience, or training samples. Summary of the Invention

[0016] To address the aforementioned problems in the prior art, this invention provides a time-frequency energy threshold-guided method for suppressing anisotropic diffusion interference in structural tensors.

[0017] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for suppressing anisotropic diffusion interference of structure tensors guided by time-frequency energy threshold, the method comprising: Determine the mixed signal; Perform a short-time Fourier transform on the mixed signal to obtain a time-frequency diagram; The energy boundary threshold is calculated based on the relationship between the interference-to-signal ratio and the energy of the time-frequency graph. The core layer is determined from the time-frequency graph based on the energy boundary threshold, and the structure tensor is determined by performing Gaussian smoothing on the core layer; the core layer is the interference core region. A diffusion tensor for controlling the diffusion direction and intensity is constructed using empirical coefficients and the structure tensor. The time-frequency graph is subjected to anisotropic diffusion iterative filtering using the diffusion tensor to obtain an anti-interference mixed signal.

[0018] Optionally, determining the core layer from the time-frequency diagram based on the energy boundary threshold includes: Based on the energy boundary threshold, the time-frequency graph is normalized to obtain the initial core layer mask; Generate a binary mask image of the initial core layer mask; The pixels with a value of 1 in the binary mask image are used to form connected components, and the core layer is generated by filtering the connected components.

[0019] Optionally, a diffusion tensor for controlling the diffusion direction and intensity is constructed using empirical coefficients and the structure tensor, including: The eigenvalues ​​are obtained by performing eigenvalue decomposition on the structure tensor. A diffusion tensor for controlling the diffusion direction and intensity is constructed based on the unit eigenvectors of the eigenvalues ​​and empirical coefficients.

[0020] Optionally, the diffusion tensor includes: ; in, Represents the diffusion tensor; The unit eigenvector representing the eigenvalue; This represents the empirical coefficient.

[0021] Optionally, the time-frequency graph is subjected to anisotropic diffusion iterative filtering using the diffusion tensor to obtain an anti-interference mixed signal, including: The flux divergence is determined based on the diffusion tensor. The flux divergence is used to perform anisotropic diffusion update of the time-frequency graph, and the average gray value of the time-frequency graph after each diffusion update is calculated. The iteration ends when the difference between the average gray value of the time-frequency image obtained in the current iteration and the average gray value of the time-frequency image obtained in the previous iteration is less than a preset threshold, and the anti-interference mixed signal is obtained.

[0022] Optionally, the energy threshold includes: ; in, This represents the energy boundary threshold; Indicates the safety factor; This represents the pixel distribution of the interference signal on the time-frequency graph; This represents the pixel distribution of the target signal on the time-frequency graph; A linear scale representing the signal-to-interference ratio; This represents the proportionality coefficient.

[0023] Optionally, the structure tensor includes: ; in, Represents the structure tensor; Represents the Gaussian kernel function; This represents the rate of change of the grayscale value of the core layer in the horizontal direction; This represents the rate of change of the grayscale value of the core layer in the vertical direction.

[0024] In a second aspect, the present invention provides a time-frequency energy threshold-guided structure tensor anisotropic diffusion interference suppression device, the structure tensor anisotropic diffusion interference suppression device comprising: The first determining module is used to determine the mixed signal; The transformation module is used to perform a short-time Fourier transform on the mixed signal to obtain a time-frequency diagram; The calculation module is used to calculate the energy boundary threshold based on the relationship between the interference-to-information ratio and the energy of the time-frequency graph; The second determining module is used to determine the core layer from the time-frequency diagram based on the energy boundary threshold, and to determine the structure tensor by performing a Gaussian smoothing operation on the core layer; the core layer is the interference core region; A construction module is used to construct a diffusion tensor for controlling the diffusion direction and intensity using empirical coefficients and the structure tensor; The filtering module is used to perform anisotropic diffusion iterative filtering on the time-frequency graph using the diffusion tensor to obtain an anti-interference mixed signal.

[0025] Thirdly, the present invention provides an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a computer program stored in memory, it implements the steps described in any of the time-frequency energy threshold-guided structural tensor anisotropic diffusion interference suppression methods.

[0026] Fourthly, the present invention provides a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, it implements the steps of the method described in any of the time-frequency energy threshold-guided structural tensor anisotropic diffusion interference suppression methods.

[0027] This invention provides a time-frequency energy threshold-guided structure tensor anisotropic diffusion interference suppression method. It calculates an energy boundary threshold based on the relationship between the interference-to-signal ratio and the energy in the time-frequency graph. Based on the energy boundary threshold, a core layer is determined from the time-frequency graph, and a structure tensor is determined by performing Gaussian smoothing on the core layer, where the core layer is the core interference region. A diffusion tensor is constructed using empirical coefficients and the structure tensor to control the diffusion direction and intensity. Anisotropic diffusion iterative filtering is performed on the time-frequency graph using the diffusion tensor, ensuring that energy is fully homogenized along the texture direction and restricted across edge directions. This achieves adaptive suppression without requiring a pre-set interference model, threshold experience, or training samples, while maximizing the preservation of target information, thus providing an interference suppression method with both high suppression depth and strong generalization ability.

[0028] The present invention will now be described in further detail with reference to the accompanying drawings. Attached Figure Description

[0029] Figure 1 This is a flowchart illustrating a time-frequency energy threshold-guided method for suppressing anisotropic diffusion interference in structural tensors, as provided in an embodiment of the present invention. Figure 2 This is a diagram comparing the effects of the two methods on suppressing false target interference. Figure 3 This is a diagram comparing the interference suppression effects of two intermittent sampling methods; Figure 4 This is a schematic diagram comparing the suppression effects of comb-like interference from two different methods. Figure 5 This is a schematic diagram comparing the signal-to-interference ratio improvement effect of two methods after intermittent sampling interference suppression under different interference-to-input ratios; Figure 6 This is a schematic diagram comparing the signal-to-interference ratio improvement effect of two methods after suppressing comb-spectrum interference at different interference-to-input ratios; Figure 7 This is a diagram comparing the signal-to-interference ratio improvement effect of two methods after suppressing false target interference under different interference-to-input ratios; Figure 8 This is a schematic diagram of a time-frequency graph energy threshold-guided structure tensor anisotropic diffusion interference suppression device provided in an embodiment of the present invention; Figure 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0030] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0031] To address the shortcomings of existing interference suppression methods, such as reliance on strong interference energy differences, model prior dependence, massive labeled data requirements, and poor real-time performance and generalization, this invention provides a time-frequency energy threshold-guided method for suppressing anisotropic diffusion interference in structure tensors. (See also...) Figure 1 , Figure 1 This is a flowchart illustrating a time-frequency energy threshold-guided method for suppressing anisotropic diffusion interference in structure tensors, as provided in an embodiment of the present invention. The method specifically includes the following steps: Step S101: Determine the mixed signal.

[0032] In this embodiment of the invention, the target signal For radar to receive The target is in a short time. The signal reflected back: ; in, Indicates the time window function; Indicates time delay; , Indicates the distance to the target. Represents the speed of light; Represents the imaginary unit; Indicates the Doppler frequency; , Indicates radial velocity; Indicates the carrier frequency.

[0033] In this embodiment of the invention, the interference types of the interference signal include suppression interference and deception interference, specifically including comb spectrum interference signal, intermittent sampling interference signal and dense false target interference signal.

[0034] The mechanism of comb-like spectral interference is equivalent to in Narrowband or band-limited noise signals are superimposed on discrete frequency points, exhibiting an equally spaced grid pattern in the frequency domain. This pattern has a wide coverage area and is easily coupled with radar sidelobes. The time-domain expression of the comb-spectrum interference signal is: ; in, This indicates a comb-like interference signal; Indicates the amplitude of the comb spectrum interference signal; Indicates the carrier frequency of the comb spectrum interference signal; This represents the frequency variation value of the comb-spectrum interference signal; .

[0035] Intermittent sampling interference is initiated by the jammer transmitting waveforms to the radar. The time-domain expression of the intermittently sampled interference signal is obtained by slicing and sampling, and then forwarding it periodically or pseudo-randomly: ; in, This indicates intermittent sampling interference signals; , Indicates the number of intermittent sampling interference samples; , Indicates the number of intermittent sampling interference sub-pulses; Indicates the intermittent sampling interference pulse width; This indicates the sampling period for intermittent sampling interference.

[0036] Dense decoy jamming involves the jammer replicating the radar's transmitted waveform in real time. Multiple sets of time-delay-Doppler shifts are then superimposed to create a high-density false echo in the range-velocity plane, resulting in the dense false target interference signal. The time-domain expression for this signal is: ; in, This indicates interference signals from densely packed decoys; , Indicates the number of false targets; Indicates the first The magnitude of interference from false targets; Indicates the first The time delay position of a fake target relative to the real target; Indicates the first The delay amount of a dummy target; Indicates the first Doppler frequency shift of a dummy target.

[0037] Based on this, the mixed signal generated from the target signal, noise signal, and interference signal is: ; in, Indicates a mixed signal; Indicates the target echo signal; This refers to the collective term for all the aforementioned interference signals, including comb-spectrum interference signals, intermittent sampling interference signals, and dense false target interference signals; This indicates a noise signal.

[0038] After generating the interference signal, a series of interference data corresponding to the selected pulse in the echo is extracted. Gaussian white noise is mixed into this series of interference signals, and the interference-to-noise ratio is set to 20dB. By adding noise, a more realistic mixed signal is obtained.

[0039] Step S102: Perform a short-time Fourier transform on the mixed signal to obtain a time-frequency diagram.

[0040] First, the principle of obtaining the time-frequency graph will be explained: The Hamming window function is defined as follows: ; in, Indicates the Hamming window function; Indicates the local sampling index label of the Hamming window function; , Indicates the length of the window.

[0041] Based on the Hamming window function, for each time index 𝑟, i.e., the position the window has slid across, a signal segment is extracted: ; in, The index number represents a local sampling segment of the signal, i.e., a mixed signal truncated by a window function. The signal segment; Indicates the mixed signal at time point The value at; For each windowed signal segment, a Fourier transform (DFT) is performed to obtain the local spectrum of the signal at time index t. Then, the window is slid to the next time index, and the whole process is repeated to obtain a complete image of the signal frequency changing over time, i.e., the frequency spectrum.

[0042] For the aforementioned mixed signal, which is a non-stationary signal, the Short-Time Fourier Transform (STFT) is used to perform time-frequency transformation in this embodiment of the invention: ; in, Represents a time-frequency graph; , This indicates the total number of frequency coordinate points; The reliability of the time-frequency plot will be verified below, proving that the total energy calculated from the time-frequency plot and the total energy of the mixed signal are in a fixed proportional relationship.

[0043] First, we will prove the energy conservation of the time-frequency power spectrum: According to Parseval's theorem, the total energy or average power of a signal is exactly the same whether calculated in the time domain or the frequency domain. The sum of squares or the integral of squares in the time domain is equal to the sum of squares or the integral of squares in the frequency domain; therefore, for a signal of length... discrete signals That is, mixed signal, here yes The discrete sequence obtained after sampling satisfies the following conditions: the time-domain average power and the frequency-domain average power. ; in, This represents the N-point DFT.

[0044] A signal segment of length L is extracted at time position r and multiplied point-by-point with a window function to obtain a locally windowed signal. A sliding window is then introduced. Afterwards, each frame also satisfies Parseval. Let the result of multiplying the window of time index 𝑟 with the signal be: ; in, This represents the signal segment corresponding to the time index r, i.e., the locally windowed signal; Represents the time index of a discrete signal The value at; right Doing NFFT and point DFT According to Parseval's theorem, we can obtain: ; Therefore, the difference between the temporal energy of this frame and the energy of that frame is only a fixed coefficient. The energy of each pixel in a whole frame corresponds one-to-one with the discrete signal. This is the energy gain of the Hamming window, which will be treated as a constant coefficient thereafter and will not affect the proportional derivation.

[0045] Based on the discrete Moyal formula, i.e., the generalization of STFT-Parseval, we obtain: ; After time-frequency energy normalization, the power average across the entire time-frequency plane differs from the original time-domain average power by only a constant. .

[0046] This proves that there is a one-to-one correspondence and a fixed proportional relationship between the energy values ​​of pixels in the time-frequency graph and the energy of the mixed signal. Therefore, energy analysis on the time-frequency graph is completely reliable.

[0047] Step S103: Calculate the energy boundary threshold based on the relationship between the interference-to-signal ratio and the energy of the time-frequency graph.

[0048] In this embodiment of the invention, the expression for the interference-to-signal ratio is defined as: ; in, Indicates the dry-to-sound ratio; This represents the average power of the interference signal; This indicates the peak power of the target echo signal after pulse compression; This represents the average power of the target echo signal; This represents the scaling factor, which is the ratio of the peak power to the average power of the target echo signal, and is approximately equal to the duty cycle. Convert the dry-to-sound ratio to a linear scale: ; in, A linear scale representing the dry-to-information ratio.

[0049] Analyze the energy distribution characteristics in the time-frequency graph: (1) Upper bound of single pixel energy: The energy of a single pixel in any time-frequency graph can never exceed the total energy of the entire graph. Only when the interference energy is 100% concentrated in a single pixel does that pixel's energy equal the total energy. This sets a theoretical upper limit for pixel energy: ; in, This represents the value of a single pixel in a time-frequency graph; This indicates the time coordinate index corresponding to a single pixel in the time-frequency graph; This indicates the frequency coordinate index corresponding to a single pixel in the time-frequency graph; Indicates a time index; This represents the frequency coordinate points in the time-frequency graph; This represents the energy normalization constant introduced by Parseval's theorem, the value of which depends on N and the energy of the window function. In the derivation, it is considered as a fixed proportionality coefficient; .

[0050] If and only if The equality holds, meaning that a single pixel is "conserved" only when all energy is extremely concentrated at that pixel.

[0051] Energy spans multiple pixel lower bounds: Define set Ω as the region of interfering pixels on the time-frequency map, and its number is... A lower bound estimate is set for pixel energy: ; in, Indicates the number of interfering pixels; Because pixel energy is non-negative: ; in, This represents the average energy value in the time-frequency graph.

[0052] Averaged per pixel: ; in, This represents the average energy of each pixel in the interference portion of the time-frequency graph.

[0053] Therefore, for any pixel: ; This illustrates the ratio between pixel power and actual energy: ∝1 / 𝐾.

[0054] Assume that the set of pixels in the interference region in the time-frequency diagram is The target region pixel set is : ; in, This represents the energy level of all pixels on the time-frequency graph; This represents the energy that belongs to the target signal portion; This represents the energy that belongs to the interference signal portion; Represents the target main lobe pixel set; Representing the set of pixels with interference peaks, we can obtain the following by simultaneously solving the equations: ; in, This represents the peak energy of a single pixel in the time-frequency graph of the target signal. This represents the pixel distribution of the target signal on the time-frequency graph; This represents the average energy of a single pixel in the time-frequency graph of the target signal; This represents the peak energy of a single pixel in the time-frequency graph of the interference signal; This represents the pixel distribution of the interference signal on the time-frequency graph; This represents the average energy of a single pixel in the time-frequency graph of the interference signal. Therefore, the ratio of the maximum pixel energy value of the interference signal to the maximum pixel energy value of the target signal on the time-frequency map can be determined as follows: ; in, This represents the maximum pixel energy value of the interference signal on the time-frequency graph; This represents the maximum pixel energy value of the target signal on the time-frequency graph; Indicates the proportionality coefficient; A linear scale representing the signal-to-interference ratio; In the simulation, the average power of the simulated interference signal and the scaling factor are... Approximately equal to the signal duty cycle, the energy of a single pixel is related not only to the original signal power but also to its distribution within the pixel. The threshold selection in the time-frequency plot is based on the ratio of the interference to the pixel energy value with the highest signal strength.

[0055] Based on this, in this embodiment of the invention, the process of determining the energy boundary threshold is as follows: ; in, This represents the time-frequency distribution of the mixed signal after normalization. This represents the amplitude of the mixed signal on the time-frequency graph; After normalizing the power spectrum, if the pixel corresponding to the interference signal is assigned to the global maximum value, then: ; To simultaneously select interfering targets and exclude targets in any JSR scenario, a safety factor is introduced. The safety factor depends on the impact of noise; if there is no noise, then... =1, defining the energy boundary threshold: ; in, This indicates the energy boundary threshold.

[0056] Step S104: Determine the core layer from the time-frequency map based on the energy boundary threshold, and determine the structure tensor by performing Gaussian smoothing operation on the core layer; the core layer is the interference core region.

[0057] In this embodiment of the invention, the core layer is determined from the time-frequency graph based on an energy boundary threshold, including: Based on the energy boundary threshold, the initial core layer mask is obtained by normalizing the time-frequency graph. Generate a binary mask image of the initial core layer mask; The pixels with a value of 1 in the binary mask image are used to form connected components, and the core layer is generated by filtering the connected components.

[0058] In this embodiment of the invention, the regions with abnormally high energy that are connected in a continuous pattern in the time-frequency graph represent concentrated areas of strong interference. After normalizing the time-frequency graph, the value of each pixel in the graph is between 0 and 1: ; in, Indicates the initial core layer mask; After calculation, for example, in the case of an interference-to-signal ratio greater than 10dB and the presence of noise, A value of 0.5 satisfies the core layer threshold. The purpose of this is to perform preliminary screening. Pixels with energy values ​​exceeding the energy threshold are highly likely to be interference. The coarse screening aims to quickly focus on high-energy regions.

[0059] A binary mask is generated based on the energy boundary threshold to transform the grayscale image represented by the continuous core layer mask into a clear black-and-white mask represented by the binary mask, thus achieving preliminary region segmentation. The white areas, i.e., areas with a value of 1, are "candidate interference areas," and the black areas, i.e., areas with a value of 0, are the target signal areas. The binary mask is calculated as follows: ; in, This represents a binary mask. If the value is greater than the energy threshold, it is recorded as 1; otherwise, it is recorded as 0. Represents the step function; In a binary mask, all pixels with a value of 1 form a connected component. : ; Not all connected components are valid interference. Those connected components with very small areas are likely scattered noise points and need to be filtered out, retaining only those with large areas. Not less than The components are used to obtain a new core candidate set. : ; ; in, Indicates the core layer; This represents the minimum area threshold used to filter out small, potentially noisy, connected components.

[0060] In this embodiment of the invention, the original image of the core layer gradient field for: ; in, This represents the rate of change of the grayscale values ​​of the original image in the horizontal direction; This represents the rate of change of the grayscale values ​​of the original image in the vertical direction; Represents the original image; Represents the coordinates of pixels in the original image; the gradient field reflects the direction and magnitude of the fastest brightness change at each pixel in the original image, and is the most basic information for perceiving image edges and textures; superscript This represents the transpose of a matrix.

[0061] In this embodiment of the invention, the structure tensor is constructed by performing Gaussian smoothing on the original image: ; in, This represents the image after Gaussian smoothing. Represents the original image; This represents the Gaussian kernel function, which is the core of the Gaussian smoothing operation. It is used to perform a weighted average on the original image to achieve smoothing, i.e., noise reduction and blurring. The standard deviation of the Gaussian kernel determines the degree of Gaussian smoothing. The larger the value, the more obvious the smoothing effect, and the greater the degree to which the original image is blurred; Indicates the convolution operation; Under Gaussian smoothing, the structure tensor Defined as: ; in, Represents the structure tensor; Represents the Gaussian kernel function; This represents the rate of change of the grayscale value of the core layer in the horizontal direction; This represents the rate of change of the grayscale value of the core layer in the vertical direction; Write the structure tensor in component form : ; in, , , These are all components of the structure tensor. .

[0062] In this embodiment of the invention, unlike traditional methods that rely solely on energy thresholds or filtering templates, the structure tensor is constructed using a time-frequency gradient field, which can accurately extract the local principal direction and intensity of change. The principal direction extracted by the structure tensor not only encodes the texture direction of the interference stripes but also provides clear guiding coordinates for the diffusion direction.

[0063] Step S105: Construct a diffusion tensor to control the diffusion direction and intensity using empirical coefficients and structural tensors.

[0064] In this embodiment of the invention, a diffusion tensor for controlling the diffusion direction and intensity is constructed using empirical coefficients and a structure tensor, including: Eigenvalues ​​are obtained by performing eigenvalue decomposition on the structure tensor; A diffusion tensor for controlling the direction and intensity of diffusion is constructed based on unit eigenvectors of eigenvalues ​​and empirical coefficients.

[0065] For structure tensor Performing eigenvalue decomposition yields: ; in, express Two eigenvalues, of which It is the largest eigenvalue, corresponding to the edge normal direction. The smallest eigenvalue corresponds to the direction of the edge tangent. The largest eigenvalue represents the intensity of the most drastic gradient change in a local region, while the smallest eigenvalue represents the intensity of the most gradual gradient change in a local region.

[0066] In an embodiment of the present invention, The corresponding unit eigenvector is: ; in, ; The corresponding unit eigenvector is: ; Then, based on and The diffusion tensor is constructed as follows: First, empirical coefficients are designed, including the diffusion coefficient in the structural direction. and the diffusion coefficient in the vertical direction : ; ; in, Indicates the preset minimum diffusion rate; Indicates edge sensitivity; Used to represent coherence, i.e. the intensity of anisotropy, indicating that the stronger the directionality of a local region, such as a clear edge; a value close to 0 indicates that the local region is isotropic, such as a flat region or noise.

[0067] By performing eigenvalue decomposition on the structure tensor, unit eigenvectors of the edge normal direction and tangent direction are obtained. and their corresponding eigenvalues The degree of local anisotropy can be calculated. This metric controls the diffusion coefficients in each direction of the diffusion tensor, ensuring rapid diffusion in the texture direction and limited diffusion across texture directions, thereby achieving structural protection and homogenization of disturbance energy.

[0068] Unlike deep learning methods that rely on the generalization ability of training data, this invention directly generates the corresponding tensor field through structural tensor analysis, giving the diffusion behavior directional guidance in each iteration. This "image structure"-based guidance mechanism can significantly reduce the risk of erroneous diffusion covering the target echo.

[0069] Then, based on the unit eigenvectors of the eigenvalues ​​and the designed empirical coefficients, a diffusion tensor is constructed, let: ; in, Unit eigenvectors representing eigenvalues; Represents the empirical coefficient; This represents the operation of constructing a diagonal matrix; Constructed diffusion tensor for: ; Among them, superscript This represents the matrix transpose operation; Diffusion Tensor It preserves the local orientation-dominated diffusion characteristics at each point, and the tensor field maintains the same size as the original time-frequency image, adapting to subsequent image diffusion operations. Compared to methods based on fixed templates or orientation-independent diffusion kernels, it supports constructing an independent diffusion tensor for each pixel, enhancing the adaptability of this invention in non-uniform interference backgrounds. Diffusion Tensor Provides a physically constrained direction-aware tensor field for iterative diffusion, improving convergence accuracy.

[0070] make: ; in, express Eigenvalue components in the horizontal direction; express Eigenvalue components in the vertical direction; express Eigenvalue components in the horizontal direction; express Eigenvalue components in the vertical direction; Therefore, the component matrix of the diffusion tensor can be expressed as: ; in, Indicates the first component; Indicates the second component; Indicates the third component; In this embodiment of the invention, within a selected scale, the time-frequency gradient field is calculated and a structure tensor is constructed, simultaneously encoding the direction and magnitude of local amplitude changes; then, the structure tensor is subjected to eigenvalue decomposition to calculate its eigenvalues ​​and eigenvectors; and the dynamic diffusion coefficients of the structural direction and vertical direction are introduced to calculate the final diffusion flux expression, finally obtaining a tensor field of the same size as the original image, providing a basis for subsequent diffusion iterations.

[0071] Step S106: Use the diffusion tensor to perform anisotropic diffusion iterative filtering on the time-frequency graph to obtain the anti-interference mixed signal.

[0072] In this embodiment of the invention, based on the diffusion tensor, the continuous form of anisotropic diffusion is: ; Wherein, the left side of the equal sign is the rate of change of gray level over time, and the right side is the net effect of energy inflow and outflow, i.e., divergence; It represents the gray value of the original image. It is the brightness information of the image at a certain location. In the continuous form of anisotropic diffusion, it describes the change of gray value over time. This represents the gradient operator, which is used to calculate the gradient of the image grayscale function, reflecting information such as the direction and rate of change of the function value.

[0073] The time-frequency graph is divided into grids according to the following rules: ,in The width of the grid. The length of the grid and the spatial step size are given by the given information. .

[0074] Then, the central difference method is used to calculate the rate of change of brightness for each grid in the horizontal and vertical directions: ; in, Indicates the rate of change of brightness in the horizontal direction; Indicates the rate of change of brightness in the vertical direction; express Brightness of the adjacent point on the right; express Brightness of the left neighboring point; Indicates the first During iteration, position The value at; Indicates the number of iterations.

[0075] In this embodiment of the invention, anisotropic diffusion iterative filtering of the time-frequency graph is performed using a diffusion tensor to obtain an anti-interference mixed signal, including: The flux divergence is determined based on the diffusion tensor; Anisotropic diffusion updates of the time-frequency plot are performed using flux divergence, and the average gray value of the time-frequency plot after each diffusion update is calculated. The iteration ends when the difference between the average gray value of the time-frequency image obtained in the current iteration and the average gray value of the time-frequency image obtained in the previous iteration is less than a preset threshold, and the anti-interference mixed signal is obtained.

[0076] In this embodiment of the invention, the diffusion flux is obtained based on the diffusion tensor as follows: ; in, This represents the diffusion flux in the horizontal direction; This represents the diffusion flux in the vertical direction; Calculate flux divergence based on diffusion flux: ; in, Indicates flux divergence; Next, flux divergence is used to perform diffusion updates on the time-frequency map, updating the grayscale values ​​at each location in the time-frequency map in each iteration: ; in, Indicates the first The image obtained from the next iteration; Indicates the first The image obtained from the next iteration; This parameter controls the flow rate and determines the speed of image diffusion updates. The speed at which image grayscale values ​​are updated.

[0077] In this embodiment of the invention, the average gray value of the image obtained through iteration is calculated to measure the overall gray level of the image during the iteration process, thereby determining whether convergence has occurred. The first The average gray value of the image obtained in the second iteration and the The average gray value of the image obtained in the second iteration If the difference between the two is less than a preset threshold, then... If the energy converges and stabilizes, and the image no longer changes significantly, the iteration ends; otherwise, the iteration continues.

[0078] Specifically, the convergence detection calculation method is as follows: ; If the above equation holds true, then the energy convergence is considered stable, and the iteration is terminated; otherwise, the iteration continues.

[0079] After the iteration ends, we get the first... The time-frequency domain representation of the image obtained in the next iteration is: Convert it into a time-domain waveform to obtain the interference-resistant mixed signal: ; in, This represents the mixed signal after interference suppression; Angular frequency is a physical quantity that describes the speed of vibration or oscillation of an object. In Fourier analysis, including the inverse short-time Fourier transform, it is used to characterize the frequency characteristics of a signal in the frequency domain by analyzing the frequency domain signals at different angular frequencies ω.

[0080] In each iteration, the divergence is calculated based on the fixed diffusion tensor, using a step size. To ensure numerical stability and monitor the image mean in real time, when the change in the average value between two consecutive iterations is small enough to be negligible, it is automatically determined that "convergence has been achieved" and subsequent iterations are stopped immediately. In this embodiment of the invention, an energy boundary threshold is calculated based on the relationship between the interference-to-signal ratio and the energy in the time-frequency graph. The core layer is determined from the time-frequency graph based on the energy boundary threshold, and a structure tensor is determined by performing Gaussian smoothing on the core layer, where the core layer is the core interference region. A diffusion tensor is constructed using empirical coefficients and the structure tensor to control the diffusion direction and intensity. Anisotropic diffusion iterative filtering is performed on the time-frequency graph using the diffusion tensor, ensuring that the energy is fully homogenized along the texture direction and diffused in a restricted manner across the edge direction. This achieves adaptive suppression without the need for a pre-set interference model, threshold experience, or training samples, and maximizes the preservation of target information, thus providing an interference suppression method with both high suppression depth and strong generalization ability.

[0081] The simulation experiment of the time-frequency energy threshold guided structure tensor anisotropic diffusion interference suppression method provided by the embodiments of the present invention is as follows: Traditional time-frequency domain dynamic threshold filtering algorithms often rely on pre-set fixed thresholds or significant differences in interference energy. While effective in simple interference environments with minimal parameter fluctuations, fixed thresholds struggle to balance suppression strength and signal fidelity when facing complex interference with diverse spectral patterns and constantly changing power and bandwidth, easily leading to undersuppression or oversuppression. To address this, this invention introduces a time-frequency map energy threshold-guided method. By calculating and deriving the relationship between time-domain power, frequency power, time-frequency domain power, and pixel amplitude in the time-frequency domain, the thresholds for interference and target signals in the time-frequency map under a given interference-to-signal ratio (ISR) are obtained. The time-frequency plane is divided into a core layer (mainly containing interference peaks) and a background layer (containing targets, noise, and residual interference). Table 1 compares the interference suppression metrics of traditional methods, OTU dynamic thresholding and notch filtering algorithms, and the method of this invention. It can be seen that the proposed algorithm significantly improves the ISR.

[0082] Table 1 Comparison of interference suppression indices for different methods

[0083] To address the shortcomings of traditional methods, such as heavy reliance on interference mechanism models and long computation time, and the need for large-scale data annotation in deep learning methods, this method uses structural tensors to perceive the direction of interference strips, efficiently diffuses along the strips, and strictly suppresses interference across strips, ensuring accurate interference localization. Since the image energy distribution varies greatly under different JSNR, signal bandwidth, or interference patterns, a global mean adaptive early stopping diffusion iteration mechanism is adopted during the iteration process. This mechanism can automatically perceive this difference, ensuring that it stops after a few iterations under weak interference conditions, while iterating several more times before convergence under strong interference conditions. This ensures both the suppression effect and significantly saves computing power and time.

[0084] Figure 2 This is a diagram comparing the false target interference suppression effects of the two methods. Figure 2 (a) in the diagram is the time-frequency diagram of the echo affected by false target interference. Figure 2 (b) in the diagram is the time-frequency diagram of the echo affected by false target interference. Figure 2 (c) in the figure is the time-frequency diagram of edge echoes after removing false target interference using the OTU dynamic threshold and notch filtering algorithm. Figure 2 (d) in the diagram is a schematic diagram of the final false target interference suppression of the present invention. Figure 2 In the diagram (e), the pulse compression plot is obtained after suppressing false target interference using the OTU dynamic threshold and notch filter algorithm. Figure 2 In Figure (f), the time-domain signal compression plot after false target interference suppression according to the present invention is shown. The original signal-to-interference ratio (SIR) is -25 dB. After suppression of false target interference using the OTU dynamic threshold and notch filter algorithm, the SIR is 13.80 dB, representing an improvement of 38.80 dB. The SIR after suppression of false target interference according to the present invention is 21.35 dB, representing an improvement of 46.35 dB.

[0085] Figure 3 This is a diagram comparing the interference suppression effects of two intermittent sampling methods. Figure 3 (a) in the diagram is the time-frequency diagram of the echo affected by intermittent sampling interference. Figure 3 (b) in the diagram is the time-frequency diagram of the echo affected by intermittent sampling interference. Figure 3 (c) in the figure is the time-frequency diagram of edge echoes removed by OTU dynamic threshold and notch filtering algorithm to remove intermittent sampling interference. Figure 3 (d) in the diagram is a schematic diagram of the final intermittent sampling interference suppression of the present invention. Figure 3 In the diagram (e), the pulse compression plot is obtained after suppressing intermittent sampling interference using the OTU dynamic threshold and notch filter algorithm. Figure 3 In Figure (f), the signal pulse compression plot in the time domain after intermittent sampling interference suppression according to the present invention is shown. The original signal-to-interference ratio (SIR) is -25 dB. After intermittent sampling interference suppression using the OTU dynamic threshold and notch filter algorithm, the SIR is 12.19 dB, representing an improvement of 37.19 dB. The SIR after intermittent sampling interference suppression according to the present invention is 19.71 dB, representing an improvement of 44.71 dB.

[0086] Figure 4 This is a schematic diagram comparing the suppression effects of two methods on comb-like interference. Figure 4 (a) in the diagram is the time-frequency diagram of the echo affected by comb-spectrum interference. Figure 4 (b) in the diagram is the time-frequency diagram of the echo affected by comb-spectrum interference. Figure 4 (c) in the figure is the time-frequency diagram of edge echoes removed by OTU dynamic thresholding and notch filtering algorithm for comb spectrum interference. Figure 4 (d) in the diagram is a schematic diagram of the final comb-like spectrum interference suppression of this invention. Figure 4 In the diagram, (e) represents the pulse compression plot after OTU dynamic thresholding and notch filtering algorithm comb spectrum interference suppression. Figure 4 In Figure (f), the time-domain signal compression plot after the inter-comb spectrum interference suppression of this invention is shown. The original signal-to-interference ratio (SIR) is -25 dB. After SIR suppression using the OTU dynamic threshold and notch filter algorithm, the SIR is 9.85 dB, representing an improvement of 34.85 dB. The SIR after SIR suppression using the inter-comb spectrum interference of this invention is 13.58 dB, representing an improvement of 38.58 dB.

[0087] Figure 5 This is a diagram comparing the signal-to-interference ratio (SIR) improvement effect of two methods after intermittent sampling interference suppression at different SIR ratios. Figure 6 This diagram illustrates the comparison of the signal-to-interference ratio (SIR) improvement effect of two methods after suppressing comb-spectral interference at different SIR ratios. Figure 7 This diagram illustrates the comparison of the signal-to-interference ratio improvement effect of two methods after suppressing false target interference under different interference-to-input ratios. By showing the interference suppression effect of the two methods under slice interference, comb spectrum interference, false target interference, and different interference-to-input ratios, it can be seen that the interference suppression advantage of this method is significant.

[0088] Based on the same inventive concept, embodiments of the present invention also provide a time-frequency energy threshold-guided structure tensor anisotropic diffusion interference suppression device, see [link to previous document]. Figure 8 , Figure 8 This is a schematic diagram of a time-frequency energy threshold-guided structure tensor anisotropic diffusion interference suppression device provided in an embodiment of the present invention. The structure tensor anisotropic diffusion interference suppression device includes: The first determining module 801 is used to determine the mixed signal; Transformation module 802 is used to perform short-time Fourier transform on the mixed signal to obtain a time-frequency diagram; Calculation module 803 is used to calculate the energy boundary threshold based on the relationship between the interference-to-signal ratio and the energy of the time-frequency graph; The second determining module 804 is used to determine the core layer from the time-frequency diagram based on the energy boundary threshold, and to determine the structure tensor by performing a Gaussian smoothing operation on the core layer; the core layer is an interference core region; Module 805 is used to construct a diffusion tensor for controlling the diffusion direction and intensity using empirical coefficients and the structure tensor; The filtering module 806 is used to perform anisotropic diffusion iterative filtering on the time-frequency graph using the diffusion tensor to obtain an anti-interference mixed signal.

[0089] In this embodiment of the invention, an energy boundary threshold is calculated based on the relationship between the interference-to-signal ratio and the energy in the time-frequency graph. The core layer is determined from the time-frequency graph based on the energy boundary threshold, and a structure tensor is determined by performing Gaussian smoothing on the core layer, where the core layer is the core interference region. A diffusion tensor is constructed using empirical coefficients and the structure tensor to control the diffusion direction and intensity. Anisotropic diffusion iterative filtering is performed on the time-frequency graph using the diffusion tensor, ensuring that the energy is fully homogenized along the texture direction and diffused in a restricted manner across the edge direction. This achieves adaptive suppression without the need for a pre-set interference model, threshold experience, or training samples, and maximizes the preservation of target information, thus providing an interference suppression method with both high suppression depth and strong generalization ability.

[0090] This invention also provides an electronic device, such as... Figure 9 As shown, it includes a processor 901, a communication interface 902, a memory 903, and a communication bus 904, wherein the processor 901, the communication interface 902, and the memory 903 communicate with each other through the communication bus 904. Memory 903 is used to store computer programs; When the processor 901 executes the program stored in the memory 603, it implements the method steps of any of the above-mentioned time-frequency energy threshold-guided structural tensor anisotropic diffusion interference suppression methods.

[0091] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of representation, only one thick line is used in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0092] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0093] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0094] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0095] The present invention also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the method steps of any of the above-described time-frequency energy threshold-guided structural tensor anisotropic diffusion interference suppression methods.

[0096] Optionally, the computer-readable storage medium may be non-volatile memory (NVM), such as at least one disk storage device.

[0097] Optionally, the aforementioned computer-readable storage medium may also be at least one storage device located remotely from the aforementioned processor.

[0098] In another embodiment of the present invention, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to execute the steps of the method described in any of the time-frequency energy threshold guided structural tensor anisotropic diffusion interference suppression methods.

[0099] It should be noted that the terms "first," "second," etc., are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention.

[0100] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0101] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0102] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0103] For the embodiments of the device / electronic device / storage medium, since they are basically similar to the method embodiments, the description is relatively simple, and relevant parts can be referred to in the description of the method embodiments.

[0104] It should be noted that the device, electronic device, and storage medium in the embodiments of the present invention are respectively the device, electronic device, and storage medium for suppressing structural tensor anisotropic diffusion interference guided by the above-mentioned time-frequency energy threshold. Therefore, all embodiments of the above-mentioned structural tensor anisotropic diffusion interference suppression method guided by the above-mentioned time-frequency energy threshold are applicable to the device, electronic device, and storage medium, and can achieve the same or similar beneficial effects.

[0105] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A method for suppressing anisotropic diffusion interference of structure tensors guided by time-frequency energy threshold, characterized in that, The method for suppressing anisotropic diffusion interference of the structural tensor includes: Determine the mixed signal; Perform a short-time Fourier transform on the mixed signal to obtain a time-frequency diagram; The energy boundary threshold is calculated based on the relationship between the interference-to-signal ratio and the energy of the time-frequency graph. The core layer is determined from the time-frequency graph based on the energy boundary threshold, and the structure tensor is determined by performing Gaussian smoothing on the core layer; the core layer is the interference core region. A diffusion tensor for controlling the diffusion direction and intensity is constructed using empirical coefficients and the structure tensor. The time-frequency graph is subjected to anisotropic diffusion iterative filtering using the diffusion tensor to obtain an anti-interference mixed signal.

2. The method for suppressing anisotropic diffusion interference of structural tensors according to claim 1, characterized in that, The core layer is determined from the time-frequency diagram based on the energy boundary threshold, including: Based on the energy boundary threshold, the time-frequency graph is normalized to obtain the initial core layer mask; Generate a binary mask image of the initial core layer mask; The pixels with a value of 1 in the binary mask image are used to form connected components, and the core layer is generated by filtering the connected components.

3. The method for suppressing anisotropic diffusion interference of structural tensors according to claim 1, characterized in that, A diffusion tensor for controlling the diffusion direction and intensity is constructed using empirical coefficients and the structure tensor, including: The eigenvalues ​​are obtained by performing eigenvalue decomposition on the structure tensor. A diffusion tensor for controlling the diffusion direction and intensity is constructed based on the unit eigenvectors of the eigenvalues ​​and empirical coefficients.

4. The method for suppressing anisotropic diffusion interference of structural tensors according to claim 3, characterized in that, The diffusion tensor includes: ； in, Represents the diffusion tensor; The unit eigenvector representing the eigenvalue; This represents the empirical coefficient.

5. The method for suppressing anisotropic diffusion interference of structural tensors according to claim 1, characterized in that, Anisotropic diffusion iterative filtering is performed on the time-frequency graph using the diffusion tensor to obtain an anti-interference mixed signal, including: The flux divergence is determined based on the diffusion tensor; The flux divergence is used to perform anisotropic diffusion update of the time-frequency graph, and the average gray value of the time-frequency graph after each diffusion update is calculated. The iteration ends when the difference between the average gray value of the time-frequency image obtained in the current iteration and the average gray value of the time-frequency image obtained in the previous iteration is less than a preset threshold, and the anti-interference mixed signal is obtained.

6. The method for suppressing anisotropic diffusion interference of structural tensors according to claim 1, characterized in that, The energy threshold includes: ； in, This represents the energy boundary threshold; Indicates the safety factor; This represents the pixel distribution of the interference signal on the time-frequency graph; This represents the pixel distribution of the target signal on the time-frequency graph; A linear scale representing the signal-to-interference ratio; This represents the proportionality coefficient.

7. The method for suppressing anisotropic diffusion interference of structural tensors according to claim 1, characterized in that, The structure tensor includes: ； in, Represents the structure tensor; Represents the Gaussian kernel function; This represents the rate of change of the grayscale value of the core layer in the horizontal direction; This represents the rate of change of the grayscale value of the core layer in the vertical direction.

8. A time-frequency energy threshold-guided structure tensor anisotropic diffusion interference suppression device, characterized in that, The structure tensor anisotropic diffusion interference suppression device includes: The first determining module is used to determine the mixed signal; The transformation module is used to perform a short-time Fourier transform on the mixed signal to obtain a time-frequency diagram; The calculation module is used to calculate the energy boundary threshold based on the relationship between the interference-to-information ratio and the energy of the time-frequency graph; The second determining module is used to determine the core layer from the time-frequency diagram based on the energy boundary threshold, and to determine the structure tensor by performing a Gaussian smoothing operation on the core layer; the core layer is the interference core region; A construction module is used to construct a diffusion tensor for controlling the diffusion direction and intensity using empirical coefficients and the structure tensor; The filtering module is used to perform anisotropic diffusion iterative filtering on the time-frequency graph using the diffusion tensor to obtain an anti-interference mixed signal.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; When a processor executes a computer program stored in memory, it implements the structural tensor anisotropic diffusion interference suppression method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the structural tensor anisotropic diffusion interference suppression method according to any one of claims 1-7.

Images