Low-altitude clutter suppression method based on multi-domain combined processing

Abstract

The invention relates to the technical field of radar signal processing, and discloses a low-altitude clutter suppression method based on multi-domain joint processing, comprising the following steps: acquiring an original radar echo signal from a target, and preprocessing to generate a radar echo signal; performing multi-domain feature extraction on the radar echo signal, performing feature fusion on multi-domain features, and outputting a joint judgment result; the joint judgment result comprises a binary label of a distance-Doppler unit, a target or clutter probability and a suppression weight; on the basis of the joint judgment result, clutter suppression is carried out on the specified signal, and an effective echo signal is obtained; and obtaining a final target detection result according to the effective echo signal. According to the technical scheme, complex and changeable low-altitude clutter characteristics can be accurately represented, and a better suppression effect is obtained; the target retention capability is stronger; the potential calculation efficiency can be improved, the calculation burden is reduced while high performance is preserved, and engineering implementation is easier.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and more specifically, to a low-altitude clutter suppression method based on multi-domain joint processing. Background Technology

[0002] Low-altitude detection of low-flying targets has always faced the challenge of low-altitude clutter: the presence of buildings, mountains, and trees creates strong, non-uniformly distributed ground clutter and sea clutter in low-altitude radar echoes. The intensity of these clutters is much higher than that of the echoes from low-altitude flying targets (such as small aircraft, drones, and cruise missiles), severely obscuring the target signal and leading to a decrease in target detection probability and an increase in false alarm probability.

[0003] In existing technologies, time-domain processing, frequency-domain processing, and spatial processing are commonly used to suppress low-altitude clutter. However, time-domain processing is ineffective at suppressing slow-moving or tangentially flying targets (with Doppler frequencies close to zero); it has weak adaptability to non-uniform clutter (such as wind-driven sea waves) and is easily affected by target scintillation; frequency-domain processing is limited by radar system bandwidth and Doppler resolution, and its effectiveness is poor when Doppler ambiguity and clutter spectrum broadening are severe, making it difficult to handle clutter overlapping with the target spectrum; spatial processing has extremely high computational complexity and poor real-time performance; it requires precise array calibration and channel consistency; it lacks robustness to non-stationary clutter environments, and its performance is limited under low-degree-of-freedom arrays. These three methods are usually performed independently or in series in a single domain (time, frequency, and space). In complex low-altitude clutter environments, clutter characteristics often exhibit coupling in multiple domains simultaneously, making single-domain processing difficult to comprehensively and effectively suppress clutter, and easily resulting in the loss of target information or the introduction of new distortions while suppressing clutter.

[0004] Therefore, a technical solution is needed to overcome the limitations of single-domain processing or simple cascade processing in existing low-altitude clutter suppression technologies, so as to not only comprehensively and effectively suppress complex low-altitude clutter, but also retain information on weak low-altitude targets. Summary of the Invention

[0005] To achieve the above objectives, this application provides a low-altitude clutter suppression method based on multi-domain joint processing, comprising the following steps: Acquire the raw radar echo signal from the target, preprocess the raw radar echo signal, and generate a radar echo signal. Multi-domain feature extraction is performed on radar echo signals, including time-domain features, frequency-domain features, and spatial / angular-domain features. Multi-domain features are fused to output a joint decision result; the joint decision result includes the binary label of the range-Doppler cell, the target or clutter probability and / or suppression weight; Based on the joint decision result, clutter suppression is applied to the specified signal to obtain the effective echo signal; based on the effective echo signal, the final target detection result is obtained. The specified signal can be the original radar echo signal or the signal obtained after range-Doppler spectroscopy and beamforming.

[0006] The preprocessing includes analog-to-digital conversion, down-conversion, quadrature demodulation, pulse compression, and channel equalization / calibration operations.

[0007] The time-domain features are the amplitude statistics and / or non-stationarity measures of slow time series, including: amplitude fluctuation statistics, time correlation, and non-equilibrium measures. Frequency domain characteristics include: Doppler spectrum shape, spectral entropy, specific frequency band energy, and harmonic characteristics; among which, the Doppler spectrum shape includes peak value, width, and symmetry; Spatial / angle domain features include spatial spectrum, angle spread, angle of arrival estimate, inter-channel amplitude, and spatial correlation measure.

[0008] Among them, feature fusion methods include: feature-level fusion, decision-level fusion, model-based fusion, and machine learning fusion; Feature-level fusion refers to concatenating or weighting feature vectors extracted from different domains into a high-dimensional joint feature vector; the high-dimensional joint feature vector can be combined with other methods to obtain joint decision results; Decision-level fusion refers to performing preliminary target / clutter discrimination based on time domain, frequency domain, and spatial domain features or subsets thereof, obtaining multi-dimensional preliminary discrimination results, and then fusing these multi-dimensional preliminary discrimination results to obtain a joint decision result. Model-based fusion refers to using prior statistical models of clutter and targets in different domains to calculate the probability that the joint characteristic of the received signal belongs to the target or clutter. Machine learning fusion refers to inputting a high-dimensional joint feature vector into a trained machine learning model to obtain binary labels or suppression weights for distance-Doppler units.

[0009] Among them, binary labels are symbols used to mark signals as clutter or targets; Target or clutter probability refers to the probability or confidence score of belonging to a target or clutter. The suppression weight is a weight value between 0 and 1, used for subsequent weighting of the original signal or intermediate processing results.

[0010] Furthermore, when the joint decision result is a binary label of the range-Doppler unit, the unit with the binary label "clutter" is directly set to zero or to the noise level, and then a valid echo signal is returned. When the joint decision result is the target probability or clutter probability, the suppression weight is obtained by calculating based on the target probability or clutter probability.

[0011] After obtaining the suppression weights, the effective echo signal is calculated using the following method: Effective echo signal = Original radar echo signal × Suppression weight.

[0012] According to the present invention, the complex and varied low-altitude clutter characteristics (time-varying, frequency-spreading, spatially non-uniform) can be accurately characterized. Compared with single-domain methods, it can more thoroughly suppress clutter, especially for slow targets, spectral spread clutter, and non-uniform clutter, achieving a significantly better suppression effect. It can more finely distinguish weak targets with similar clutter characteristics, reduce target signal loss, improve detection probability, and enhance target retention capability. Furthermore, its sensitivity to changes in clutter environment and array errors is reduced compared to traditional methods such as STAP. It can be selected and optimized according to actual application needs, computing resources, and performance requirements. It can improve potential computational efficiency, reduce computational burden while maintaining high performance, and is easier to implement in engineering. Attached Figure Description

[0013] Figure 1 This is a flowchart illustrating the steps of a low-altitude clutter suppression method based on multi-domain joint processing provided in an embodiment of the present invention. Detailed Implementation

[0014] This invention proposes a multi-domain joint processing framework that utilizes the characteristic information of radar echo signals in multiple dimensions such as time domain, frequency domain, and spatial domain in parallel or deeply coupled manner. Through specific information fusion and joint decision-making mechanisms, it achieves better clutter suppression and target enhancement.

[0015] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.

[0016] Figure 1 A step-by-step diagram of a low-altitude clutter suppression method is provided, as shown in the figure, including the following steps: Step S100: Acquire the raw radar echo signal from the target, preprocess the raw radar echo signal, and generate a radar echo signal. In this step, the echo signals from each channel of the radar front end are received as the raw radar echo signals. The raw radar echo signals are then subjected to preprocessing operations such as analog-to-digital conversion, down-conversion, quadrature demodulation, pulse compression, and channel equalization / calibration to form a uniform and standardized radar echo signal.

[0017] The radar front-end includes a transmitter, an antenna array (with at least airspace processing capabilities), and a receiver (multi-channel), which are used to transmit radar signals and receive echoes from targets and clutter.

[0018] Step S110: Extract multi-domain features from the radar echo signal, the multi-domain features including time domain features, frequency domain features and / or spatial / angular domain features; 1) Time-domain characteristics are the amplitude statistics and / or non-stationarity measures of slow time series, including: amplitude fluctuation statistics, time correlation, and non-equilibrium measures; Time-domain features are obtained by calculating the statistical properties of the signal in the slow time dimension, including obtaining amplitude fluctuation statistics through methods such as mean, variance, and higher-order moments, obtaining time correlation through autocorrelation function, and calculating non-stationarity measures through time variance.

[0019] 2) Frequency domain characteristics include: Doppler spectrum shape, spectral entropy, energy of specific frequency bands, and harmonic characteristics; among which, the Doppler spectrum shape includes peak value, width, and symmetry; Frequency domain features are extracted by performing spectral analysis (such as FFT) on the signal of each distance cell-channel.

[0020] 3) Spatial characteristics include spatial spectrum, angular spread, angle of arrival estimate, inter-channel amplitude, and spatial correlation measurement characteristics; Spatial features are obtained by calculating array information. For example, the spatial spectrum is obtained by spectral estimation methods such as DBF or Capon, and the spatial correlation matrix features are obtained by eigenvalue distribution.

[0021] Step S120: Perform feature fusion on the multi-domain features and output a joint decision result; The methods for achieving feature fusion include: feature-level fusion, decision-level fusion, model-based fusion, and machine learning fusion. During feature integration, multiple methods may be combined to calculate the joint decision result, or a single method may be used to obtain the joint decision result.

[0022] 1) Feature-level fusion refers to concatenating or weighting feature vectors extracted from different domains (such as time-domain statistics, frequency-domain spectral entropy, and spatial domain DOA) into a high-dimensional joint feature vector; This invention provides an embodiment in which, after obtaining multi-domain features from a signal, feature-level concatenation is performed: The time-domain features are: Frequency domain characteristics: Airspace characteristics: Concatenate them into a high-dimensional joint feature vector , is represented as: (1) in, , For time-domain feature dimensions, For frequency domain feature dimensions, For spatial domain features, the dimension is 1.

[0023] 2) Decision-level fusion refers to: performing preliminary target / clutter discrimination based on time domain, frequency domain, and spatial domain features or subsets thereof to obtain multi-dimensional preliminary discrimination results; the multi-dimensional preliminary discrimination results generally include probability or confidence level, and then using methods such as weighted voting, Dempster-Shafer evidence theory, Bayesian inference, etc., to fuse the multi-dimensional preliminary discrimination results to obtain a joint decision result; According to the embodiments provided by the present invention, the high-dimensional joint feature vector is calculated using a classifier decision function, as follows: (2) in, The probability of the existence of the target. For support vector weights, For sample labels, Radial basis kernel function , This is the SVM kernel width parameter, with a width range of [0.005, 0.05]. This represents the decision bias.

[0024] 3) Model-based fusion refers to: using prior statistical models of clutter and targets in different domains (such as the time-space-frequency joint distribution model of clutter) to calculate the probability that the joint characteristic of the received signal belongs to the target or clutter; 4) Machine learning fusion refers to inputting high-dimensional joint feature vectors into a trained machine learning model to obtain binary labels or suppression weights for distance-Doppler units. Generally, machine learning models include support vector machines, random forests, neural networks (CNN / RNN), etc., and require pre-training with labeled clutter and target sample data.

[0025] The joint decision result includes the binary label of the range-Doppler cell, the target or clutter probability, and the suppression weight; Among them, binary labels are symbols used to mark signals as clutter or targets; Target or clutter probability refers to the probability or confidence score of belonging to a target or clutter. The suppression weight is a weight value between 0 and 1, used for subsequent weighting of the original signal or intermediate processing results.

[0026] In the embodiments provided by the present invention, the probability of the existence of the target is obtained. Then, the suppression weights are calculated using the suppression weight mapping function. , is represented as: (3) Step S130: Based on the joint decision result, clutter suppression is performed on the specified signal to obtain an effective echo signal; and the final target detection result can be obtained based on the effective echo signal; the specified signal can be the original radar echo signal or the signal after range-Doppler spectrum and beamforming.

[0027] Specifically, such as Figure 1 As shown in step S131, when the joint decision result is a binary tag of the range-Doppler unit, the unit with the binary tag "clutter" will be set to zero or set to the noise level, and then the valid echo signal will be returned. As shown in step S132, when the joint decision result is the target or clutter probability, the suppression weight is calculated based on the target or clutter probability; and the effective echo signal is calculated based on the suppression weight, and the calculation method is the same as the calculation method in step S133. As shown in step S133, after obtaining the suppression weight, the effective echo signal is calculated. The calculation method is as follows: Effective echo signal = Original radar echo signal × Suppression weight.

[0028] After step S130, a valid echo signal is obtained, and then necessary post-processing is performed, such as constant false alarm rate detection (CFAR detection) and spot aggregation, and finally the target detection result is output.

[0029] The low-altitude clutter suppression method based on multi-domain joint processing provided by this invention comprehensively utilizes multi-dimensional information to more accurately characterize the complex and ever-changing low-altitude clutter characteristics (time-varying, frequency-spreading, spatial non-uniformity). Compared with single-domain methods, it can suppress clutter more thoroughly, especially for slow targets, spectral spread clutter, and non-uniform clutter, achieving a significantly better suppression effect. Through joint decision-making, it can more finely distinguish weak targets with similar clutter characteristics, reduce target signal loss, improve detection probability, and enhance target retention capability. Furthermore, its sensitivity to changes in the clutter environment and array errors is relatively lower than that of traditional methods such as STAP, making the signal suppression more robust. The fusion strategy can be selected and optimized according to actual application needs, computing resources, and performance requirements, thus offering good flexibility. Compared with ultra-complex algorithms such as full-dimensional STAP, this invention can improve potential computational efficiency by selecting effective features and fusion strategies, reducing the computational burden while maintaining high performance, making it easier to implement in engineering.

[0030] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A low-altitude clutter suppression method based on multi-domain joint processing, characterized in that, The method comprises the following steps: Obtaining original radar echo signals from a target, preprocessing the original radar echo signals to generate radar echo signals; Extracting multi-domain features from the radar echo signals, wherein the multi-domain features include time-domain features, frequency-domain features and space / angle-domain features; Fusing the multi-domain features to output a joint decision result, wherein the joint decision result includes a binary label of a range-Doppler cell, a target or clutter probability and a suppression weight; Based on the joint decision result, performing clutter suppression on a specified signal to obtain an effective echo signal, and obtaining a final target detection result according to the effective echo signal.

2. The low-sky-wave suppression method according to claim 1, characterized in that, The preprocessing includes analog-to-digital conversion, frequency down-conversion, quadrature demodulation, pulse compression and channel equalization / calibration operations.

3. The low-sky-wave suppression method according to claim 1, characterized in that, The time-domain features are amplitude statistical features and / or non-stationarity measurement features of a slow time sequence, including amplitude fluctuation statistics, time correlation and non-equilibrium measurement; The frequency-domain features include Doppler spectrum shape, spectrum entropy, specific frequency band energy and harmonic features, wherein the Doppler spectrum shape includes peak value, width and symmetry; The space / angle-domain features include spatial spectrum, angle spread, angle of arrival estimation value, inter-channel amplitude and spatial correlation measurement features.

4. The low-sky-wave suppression method according to claim 1, characterized by, The feature fusion method includes feature-level fusion, decision-level fusion, model-based fusion and machine learning fusion; The feature-level fusion refers to splicing or weighted combination of feature vectors extracted from different domains into a high-dimensional joint feature vector; the high-dimensional joint feature vector can be combined with other methods to obtain a joint decision result; The decision-level fusion refers to performing preliminary target / clutter discrimination based on time-domain, frequency-domain and space-domain features or subsets thereof to obtain a multi-dimensional preliminary discrimination result, and fusing the multi-dimensional preliminary discrimination result to obtain a joint decision result; The model-based fusion refers to calculating the probability of a joint feature quantity of a received signal belonging to a target or a clutter by using prior statistical models of the clutter and the target in different domains; The machine learning fusion refers to inputting a high-dimensional joint feature vector into a trained machine learning model to obtain a binary label of a range-Doppler cell or a suppression weight.

5. The low-sky-wave suppression method according to claim 1, characterized by, The binary label is a mark for labeling a signal as a clutter or a target; The target or clutter probability refers to a probability or confidence score of belonging to a target or a clutter; The suppression weight is a weight value between 0 and 1, which is used for subsequent weighting of original signals or intermediate processing results.

6. The low-sky-wave suppression method according to claim 1, characterized in that, When the joint decision result is a binary label of a range-Doppler cell, the cells with the binary label of "clutter" are set to zero or a noise level, and then the effective echo signal is returned; When the joint decision result is a target or clutter probability, the target or clutter probability is used to calculate a suppression weight.

7. The low-sky-wave suppression method according to claim 6, characterized in that, After obtaining the suppression weight, the effective echo signal is calculated by: Effective echo signal = original radar echo signal × suppression weight.

8. The low-sky-wave suppression method according to claim 1, characterized by, The specified signal supports original radar echo signals and signals after range-Doppler spectrum and beamforming.

Images