A signal frequency domain noise reduction data processing method and system

Abstract

The application discloses a signal frequency domain noise reduction data processing method and system, and relates to the technical field of data processing. The method comprises the following steps: acquiring power values of a reference cell sequence formed by a current to-be-detected cell and a preset number of reference cells on both sides of the current to-be-detected cell in a radar detection channel; calculating, in parallel, preliminary detection thresholds generated by at least two basic CFAR algorithms based on the power values of the reference cell sequence; extracting a distribution feature vector based on the power values of the reference cell sequence; inputting the distribution feature vector into a preconfigured environment decision maker to obtain a decision factor used for representing a current local environment type; adaptively fusing the preliminary detection thresholds generated by the at least two basic CFAR algorithms based on the decision factor to generate a final detection threshold; and comparing a power value of the current to-be-detected cell with the final detection threshold to output a target detection result. The application effectively improves the detection precision of radar signal frequency domain noise reduction processing.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a signal frequency domain noise reduction data processing method and system. Background Technology

[0002] As radar detection technology continues to expand into complex scenarios, the requirements for the accuracy of target detection and environmental adaptability in the frequency domain processing of pulse Doppler radar signals are increasing. Traditional radar signal frequency domain noise reduction and target detection often use a single basic constant false alarm rate (CFAR) algorithm to determine the detection threshold, which has become a conventional technique for current radar signal processing.

[0003] However, single basic CFAR algorithms have significant limitations in scene adaptation. The unit average CFAR algorithm only has good detection performance in uniform background environments. It is prone to threshold deviation in clutter edge and multi-target interference scenarios. Although the ordered statistical CFAR algorithm can resist some interference, it has insufficient detection sensitivity in uniform backgrounds and cannot meet the detection needs of different local environments. As a result, the robustness of signal frequency domain noise reduction processing and detection accuracy are difficult to guarantee. Summary of the Invention

[0004] This invention provides a signal frequency domain noise reduction data processing method and system, aiming to solve the technical problem of insufficient accuracy in existing signal frequency domain noise reduction detection technology.

[0005] In view of the above problems, the present invention provides a signal frequency domain noise reduction data processing method and system.

[0006] In a first aspect, the present invention provides a signal frequency domain noise reduction data processing method, comprising: Obtain the power value of the reference unit sequence consisting of the currently detected unit and a preset number of reference units on both sides in the radar detection channel; Based on the power values ​​of the reference unit sequence, preliminary detection thresholds generated by at least two basic CFAR algorithms are calculated in parallel. Based on the power values ​​of the reference unit sequence, a distribution feature vector is extracted, wherein the distribution feature vector includes at least local contrast features and peak interference features; The distributed feature vector is input into a pre-configured environmental decision-maker to obtain decision factors that characterize the current local environment type. Based on the decision factors, the preliminary detection thresholds generated by the at least two basic CFAR algorithms are adaptively fused to generate the final detection threshold. The power value of the current unit to be detected is compared with the final detection threshold, and the target detection result is output.

[0007] Secondly, the present invention provides a signal frequency domain noise reduction data processing system, comprising: The power value acquisition module is used to acquire the power value of the reference unit sequence consisting of the current unit to be detected and a preset number of reference units on both sides in the radar detection channel. The CFAR threshold calculation module is used to calculate, in parallel, the preliminary detection thresholds generated by at least two basic CFAR algorithms based on the power values ​​of the reference unit sequence. The feature vector extraction module is used to extract a distribution feature vector based on the power value of the reference unit sequence, wherein the distribution feature vector includes at least local contrast features and peak interference features; The decision factor acquisition module is used to input the distribution feature vector into a pre-configured environment decision-maker to obtain decision factors that characterize the current local environment type; The threshold adaptive fusion module is used to adaptively fuse the preliminary detection thresholds generated by the at least two basic CFAR algorithms based on the decision factors to generate the final detection threshold. The detection result output module is used to compare the power value of the current unit to be detected with the final detection threshold and output the target detection result.

[0008] One or more technical solutions provided in this invention have at least the following technical effects or advantages: This invention provides a signal frequency domain denoising data processing method and system, which realizes the synergistic optimization of radar signal frequency domain denoising and target detection: based on the accurately acquired power value, the limitations of a single algorithm are broken through the parallel operation of multiple CFAR algorithms. The local environment is accurately determined by feature extraction and environmental decision-making, such as local contrast and peak interference. Then, different environmental requirements are dynamically matched through threshold adaptive fusion. Finally, the accurate detection result is output by comparing the power value with the final threshold, which effectively improves the environmental adaptability, target detection accuracy and algorithm robustness of signal frequency domain denoising processing. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a signal frequency domain noise reduction data processing method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a signal frequency domain noise reduction data processing system provided in an embodiment of the present invention; The components represented by each number in the attached diagram are explained below: The module includes a power value acquisition module 11, a CFAR threshold calculation module 12, a feature vector extraction module 13, a decision factor acquisition module 14, a threshold adaptive fusion module 15, and a detection result output module 16. Detailed Implementation

[0010] This invention provides a signal frequency domain noise reduction data processing method and system to address the technical problem of insufficient accuracy in existing signal frequency domain noise reduction detection technologies.

[0011] Example 1, as Figure 1 As shown, the present invention provides a signal frequency domain noise reduction data processing method, the method comprising: S100: Obtain the power value of the reference unit sequence consisting of the current unit to be detected and a preset number of reference units on both sides in the radar detection channel.

[0012] In this embodiment of the invention, the power value of a reference unit sequence consisting of the current target unit and a predetermined number of reference units on both sides in the radar detection channel is obtained. After the radar signal is processed by pulse Doppler, the energy of the target, noise, and clutter will be distributed in different range-Doppler units. Relying solely on the power value of the current target unit, it is impossible to distinguish the target signal from local background noise and clutter. Furthermore, subsequent CFAR threshold calculation and environmental feature extraction both rely on the background power distribution data around the target unit. Therefore, after pulse Doppler processing, it is necessary to select the target unit and its two adjacent units to construct a reference sequence to obtain accurate local background power data, providing a reliable foundation for subsequent processing.

[0013] Step S100 in the method provided in this embodiment of the invention includes: In the radar signal processing chain, after pulse Doppler processing, the power value of the specified range-Doppler unit is obtained as the power value of the current unit to be detected; Centered on the distance-Doppler unit, a predetermined number of adjacent units are selected forward and backward along the distance dimension, and the power values ​​of the adjacent units are arranged in an orderly manner to form the reference unit sequence.

[0014] First, in the radar signal processing chain, after pulse Doppler processing, the power value of a specified range-Doppler cell is obtained as the power value of the current target cell. Pulse Doppler processing refers to the joint signal processing process in which the radar achieves range dimension resolution through pulse compression and performs FFT on data from the same range cell across multiple repetition periods to achieve Doppler dimension resolution. This is a pre-processing step for radar frequency domain target detection. A range-Doppler cell is a two-dimensional unit formed by the radar transmitting pulse signals and receiving echoes. It obtains target range information through range-direction pulse compression processing to form range cells, and performs Doppler FFT processing on data from the same range cell across multiple periods to obtain target velocity information to form Doppler cells. This combined unit is the basic data unit for radar target detection, carrying signal energy at specific range and velocity. In the radar signal processing chain, after completing the entire pulse-Doppler process, including pulse compression and Doppler FFT, the two-dimensional cell corresponding to the target to be detected is selected, and the signal power value of that cell is extracted as the power value of the current target cell.

[0015] For example, a 100×50 range-Doppler two-dimensional matrix is ​​generated after the radar pulse Doppler processing is preset. The range dimension index 1~100 corresponds to different detection distances, and the Doppler dimension index 1~50 corresponds to different motion speeds. The cell with range dimension index 50 and Doppler dimension index 25 is selected as the specified cell to be detected, and the signal power value of the cell is extracted to be 80dB, which is used as the power value of the current cell to be detected.

[0016] Secondly, taking the range-Doppler cell as the center, a predetermined number of adjacent cells are selected forward and backward along the range dimension. The power values ​​of these adjacent cells are then arranged in an orderly manner to form the reference cell sequence. The reference cell sequence refers to the set of adjacent cells selected along the range dimension, centered on the current cell to be detected. It is used to characterize the signal power distribution pattern of the local background around the cell to be detected. The range dimension refers to the dimension in the radar signal that characterizes the straight-line distance between the target and the radar. It is the core dimension of the range-Doppler two-dimensional cell, and selecting cells along this dimension ensures the spatial continuity of the background data. Taking the range-Doppler cell to be detected as the center, a predetermined number of adjacent cells are selected forward and backward along the range dimension. The power values ​​of these adjacent cells are then arranged in an orderly manner according to the ascending order of the range dimension index to form the reference cell sequence.

[0017] For example, five adjacent cells are selected forward and backward along the distance dimension. Centered on the cell at distance dimension 50 and Doppler dimension 25, five cells at distance dimensions 45-49 and Doppler dimension 25 are selected forward, with power values ​​of 12dB, 15dB, 13dB, 14dB, and 16dB respectively. Five cells at distance dimensions 51-55 and Doppler dimension 25 are selected backward, with power values ​​of 17dB, 15dB, 14dB, 13dB, and 12dB respectively. The power values ​​of the above 10 cells are arranged in order according to the distance dimension index to obtain the reference cell sequence: [12dB, 15dB, 13dB, 14dB, 16dB, 17dB, 15dB, 14dB, 13dB, 12dB].

[0018] In this embodiment of the invention, by accurately collecting power data of the unit to be detected and the surrounding local background, the spatial continuity and data authenticity of the reference unit sequence are ensured. This provides standardized basic data for the subsequent parallel calculation of the initial threshold and extraction of environmental distribution feature vectors in the multi-basic CFAR algorithm, and avoids deviations in subsequent threshold calculation and environmental decision-making due to distortion in the selection of background data.

[0019] S200: Based on the power values ​​of the reference unit sequence, calculate the preliminary detection thresholds generated by at least two basic CFAR algorithms in parallel.

[0020] In this embodiment of the invention, preliminary detection thresholds generated by at least two basic CFAR algorithms are calculated in parallel based on the power values ​​of the reference cell sequence. A single CFAR algorithm can only adapt to specific radar detection environments. The cell-averaged CFAR algorithm has high detection accuracy against a uniform background, but is prone to threshold drift at clutter edges and in multi-target interference scenarios. The ordered statistical CFAR algorithm has strong anti-interference capabilities, but lacks sensitivity in a uniform background. To balance the advantages of different algorithms and provide multiple sets of basic thresholds for subsequent adaptive threshold fusion, preliminary detection thresholds for at least two basic CFAR algorithms need to be calculated in parallel based on the power values ​​of the reference cell sequence, ensuring reliable basic detection thresholds support different scenarios.

[0021] Step S200 in the method provided in this embodiment of the invention includes: Establish a basic CFAR algorithm set, which includes at least the cell average CFAR algorithm and the ordered statistical CFAR algorithm; For the reference unit sequence, each algorithm in the basic CFAR algorithm set is invoked synchronously; Among them, the unit average CFAR algorithm calculates the arithmetic mean of the power values ​​of all units in the reference unit sequence, multiplies it by a preset scaling factor, and generates the first preliminary detection threshold. Meanwhile, the ordered statistical CFAR algorithm sorts the power values ​​of the reference unit sequence, selects the power value at a specified position, multiplies it by a preset scaling factor, and generates a second preliminary detection threshold. The first preliminary detection threshold and the second preliminary detection threshold are output as preliminary detection thresholds to be fused.

[0022] First, a basic CFAR algorithm set is established, which includes at least the cell-average CFAR algorithm and the ordered statistical CFAR algorithm. The basic CFAR algorithm set refers to a combination of various classic constant false alarm rate (CFAR) detection algorithms, used to simultaneously generate basic detection thresholds with different logics, providing candidate thresholds for subsequent adaptive fusion. The cell-average CFAR algorithm (CA-CFAR) is a classic CFAR algorithm that determines the background power by calculating the arithmetic mean of the power values ​​of the reference cell sequence, and then combines it with a scaling factor to generate the detection threshold; it is suitable for uniform noise clutter backgrounds. The ordered statistical CFAR algorithm (OS-CFAR) is a CFAR algorithm that sorts the power values ​​of the reference cell sequence, selects a specified position value to represent the background power, and then combines it with a scaling factor to generate the detection threshold; it has stronger anti-clutter and multi-target interference capabilities. The basic CFAR algorithm set is constructed, ensuring that it includes at least two basic algorithms: the cell-average CFAR algorithm and the ordered statistical CFAR algorithm.

[0023] Secondly, for the reference cell sequence, each algorithm in the basic CFAR algorithm set is synchronously invoked. Synchronous invocation means that within the same data processing cycle, all algorithms in the basic CFAR algorithm set are started simultaneously, processing the same reference cell sequence in parallel, without any specific execution order. Using the reference cell sequence obtained in S100 as unified input data, the cell averaging CFAR algorithm and the ordered statistical CFAR algorithm are started synchronously, and the threshold calculation process is executed independently for each. For example, using the reference cell sequence [12dB, 15dB, 13dB, 14dB, 16dB, 17dB, 15dB, 14dB, 13dB, 12dB] as input, the cell averaging CFAR algorithm and the ordered statistical CFAR algorithm in the set are synchronously invoked, and threshold calculations are performed in parallel.

[0024] The cell-averaged CFAR algorithm calculates the arithmetic mean of the power values ​​of all cells in the reference cell sequence, multiplies it by a preset scaling factor, and generates a first preliminary detection threshold. The arithmetic mean is the sum of all linear power values ​​in the reference cell sequence divided by the number of cells, used to characterize the average power level of the uniform background. The scaling factor is a preset constant coefficient used to scale the average background power to the detection threshold, ensuring a stable false alarm rate. The first preliminary detection threshold refers to the candidate detection threshold calculated by the cell-averaged CFAR algorithm, providing the first set of threshold data for subsequent fusion. The arithmetic mean of the power values ​​of all cells in the reference cell sequence is calculated, and the average is multiplied by the preset scaling factor; the result is the first preliminary detection threshold.

[0025] For example, the power values ​​of the reference cell sequence are logarithmically scaled in dB, specifically [12, 15, 13, 14, 16, 17, 15, 14, 13, 12] dB. Before performing the arithmetic mean, each power value is converted to a linear scale: The linear power sequence (in mW) is obtained as: [15.85, 31.62, 19.95, 25.12, 39.81, 50.12, 31.62, 25.12, 19.95, 15.85]. The linear mean is calculated as: (15.85 + 3162 + 19.95 + 25.12 + 39.81 + 5012 + 3162 + 25.12 + 19.95 + 15.85) / 10 = 27.50 mW. Converting back to logarithmic scale: 10 × log 10 (27.50)≈14.39dB, multiplied by the preset scaling factor 1.5, the first preliminary detection threshold is: 14.39×1.5≈21.59dB.

[0026] Simultaneously, the ordered statistical CFAR algorithm sorts the power values ​​of the reference unit sequence, selects the power value at a specified position, multiplies it by a preset scaling factor, and generates a second preliminary detection threshold. Ordered statistical sorting refers to rearranging the power values ​​of the reference unit sequence from smallest to largest to obtain an ordered power sequence. The specified position is a preset fixed sorting position; selecting the power value at this position represents the background power and can avoid abnormally high power interference. The second preliminary detection threshold refers to the candidate detection threshold calculated by the ordered statistical CFAR algorithm, providing a second set of threshold data for subsequent fusion. The power values ​​of the reference unit sequence are sorted from smallest to largest, the power value at a preset specified position is selected, and this value is multiplied by a preset scaling factor to obtain the second preliminary detection threshold.

[0027] For example, the reference unit sequence [12,15,13,14,16,17,15,14,13,12] is sorted from smallest to largest to obtain the ordered sequence: [12,12,13,13,14,14,15,15,16,17]; the preset specified position is the 8th position, corresponding to a power value of 15dB; the scaling factor is still 1.5, and the second preliminary detection threshold = 15 × 1.5 = 22.5dB.

[0028] Finally, the first preliminary detection threshold and the second preliminary detection threshold are output as the preliminary detection thresholds to be fused. For example, the preliminary detection thresholds to be fused are: the first preliminary detection threshold is 21.59 dB, and the second preliminary detection threshold is 22.5 dB.

[0029] In this embodiment of the invention, by constructing a set of multiple algorithms for parallel computation, two sets of preliminary detection thresholds corresponding to unit average CFAR and ordered statistical CFAR are generated simultaneously. This not only retains the detection advantage of unit average CFAR in uniform backgrounds, but also has the anti-interference characteristics of ordered statistical CFAR. This provides diverse and highly adaptable candidate thresholds for subsequent environment-based adaptive threshold fusion, solving the problem of insufficient scene adaptability of a single algorithm.

[0030] S300: Extract a distribution feature vector based on the power values ​​of the reference unit sequence, wherein the distribution feature vector includes at least local contrast features and peak interference features.

[0031] In this embodiment of the invention, a distribution feature vector is extracted based on the power values ​​of the reference unit sequence. This distribution feature vector includes at least local contrast features and peak interference features. The local environment type of a radar signal can be intuitively characterized by the power distribution features of the reference unit sequence. A single power value cannot distinguish between power abrupt changes at clutter edges and peak interference in multi-target scenarios, and the environment decision-maker needs to rely on quantified features to determine the environment type. Therefore, it is necessary to extract local contrast features and peak interference features from the reference unit sequence and construct a distribution feature vector to provide accurate quantitative basis for subsequent environment decision-making.

[0032] Step S300 in the method provided in this embodiment of the invention includes: The reference unit sequence is divided into a forward reference window and a backward reference window; Calculate the average power value of all cells in the forward reference window as the forward reference window average, and calculate the average power value of all cells in the backward reference window as the backward reference window average. The absolute value of the difference between the mean of the forward reference window and the mean of the backward reference window is calculated as the local contrast feature; Calculate the sliding median sequence of the reference unit sequence; The number of units in the reference unit sequence whose power values ​​exceed a certain proportion of the sliding median at their corresponding positions is counted as the peak interference feature.

[0033] First, the reference cell sequence is divided into a forward reference window and a backward reference window. In CFAR processing, the forward and backward reference windows are subsets of reference cells located on either side of the cell to be detected, used to estimate the local background power. The power difference between the two windows can be used to determine whether clutter edges exist. Using the cell to be detected corresponding to the reference cell sequence as the boundary, the reference cells on one side of the cell to be detected in the sequence are divided into a forward reference window, and the reference cells on the other side are divided into a backward reference window. The number of cells contained in the two windows is a preset value. For example, if the reference cell sequence is [12dB,15dB,13dB,14dB,16dB,17dB,15dB,14dB,13dB,12dB], using the cell to be detected as the boundary, the first 5 cells are the forward reference window: [12,15,13,14,16], and the last 5 cells are the backward reference window: [17,15,14,13,12].

[0034] Next, the mean power value of all cells within the forward reference window is calculated as the forward reference window mean, and the mean power value of all cells within the backward reference window is calculated as the backward reference window mean. The forward reference window mean is the arithmetic mean of the power values ​​of all reference cells within the forward reference window, used to characterize the background power level in front of the cell to be detected. The backward reference window mean is the arithmetic mean of the power values ​​of all reference cells within the backward reference window, used to characterize the background power level behind the cell to be detected. The power values ​​in the forward and backward reference windows are summed respectively, and then divided by the number of cells in the corresponding window to obtain the forward and backward reference window mean. For example, the total power of the forward reference window is: 12+15+13+14+16=70, mean=70 / 5=14dB; the total power of the backward reference window is: 17+15+14+13+12=71, mean=71 / 5=14.2dB.

[0035] Next, the absolute value of the difference between the mean of the forward reference window and the mean of the backward reference window is calculated as the local contrast feature. The local contrast feature, the absolute value of the difference between the mean of the forward reference window and the mean of the backward reference window, is used to quantify the degree of difference in background power on both sides of the detected unit; a larger difference indicates a higher probability of clutter edges. The difference between the mean of the forward reference window and the mean of the backward reference window is calculated, and the absolute value of this difference is taken as the local contrast feature. For example, local contrast feature = |14 - 14.2| = 0.2 dB.

[0036] Then, the sliding median sequence of the reference unit sequence is calculated. The sliding median sequence is a new sequence formed by calculating the median of the power values ​​within w units before and after each position in the reference unit sequence and arranging them in order; it is used to smooth random noise and serves as a local background benchmark for judging whether the unit power is an outlier peak. When the sliding window length is odd, such as w=3, each position in the reference unit sequence is traversed: non-first and last positions: the power values ​​of the 1 unit before, the current position, and the 1 unit after the current position (a total of 3 units) are taken, and the median is calculated; first position: only the power values ​​of the current position and the 1 unit after the current position (a total of 2 units) are taken, and the median is calculated; last position: only the power values ​​of the current position and the 1 unit before the current position (a total of 2 units) are taken, and the median is calculated; the medians of all positions are arranged in the original sequence order to form the sliding median sequence.

[0037] For example, the reference cell sequence is: [12,15,13,14,16,17,15,14,13,12], a total of 10 cells, with indices 1 to 10. The sliding window length is set to w=3. The calculation is performed bit by bit: Index 1: only indices 1 and 2 [12,15] are taken, and the median is (12+15) / 2=13.5; Index 2: indices 1, 2, and 3 are taken, [12,15,13], and sorted [12,13,15], with the median being 13; the other index positions are calculated in the same way, and the final sliding median sequence is: [13.5,13,14,14,16,16,15,14,13,12.5].

[0038] Finally, the number of units in the reference unit sequence whose power values ​​exceed a certain percentage of the sliding median at their corresponding positions is counted, and this number is used as the peak interference feature. The peak interference feature refers to the number of units in the reference unit sequence whose power values ​​exceed a certain percentage of the sliding median at their corresponding positions. It is used to quantify the density of multi-target spike interference in a local scene; a higher number indicates more severe multi-target interference. A preset power exceedance percentage is used. Each position in the reference unit sequence is traversed, and it is determined whether the unit power value exceeds the preset percentage of the sliding median at the corresponding position. The total number of units meeting the condition is counted, which is the peak interference feature. For example, if the preset power exceedance percentage is 10%, i.e., unit power > sliding median × 1.1, it is determined to be a peak interference unit. A position-by-position comparison shows that only the power of the second position in the sequence (15dB) is greater than 13 × 1.1 = 14.3dB, resulting in a final peak interference feature of 1, indicating one peak interference unit. The exceedance percentage can be preset according to parameters specific to the actual radar scenario: a higher percentage results in a stricter judgment and a smaller peak interference feature; a lower percentage results in a more lenient judgment and makes it easier to detect peak interference.

[0039] In this embodiment of the invention, local contrast features are obtained by dividing the reference window and calculating the mean difference, and peak interference features are obtained by statistical analysis of the sliding median sequence. The power distribution differences and interference levels of the reference unit sequence are accurately quantified. The resulting distribution feature vector can intuitively characterize the clutter edge and multi-target interference characteristics, providing a reliable quantitative input for the subsequent environmental decision-maker and ensuring the accuracy of environmental type determination.

[0040] S400: Input the distribution feature vector into the pre-configured environment decision-maker to obtain decision factors that characterize the current local environment type.

[0041] In this embodiment of the invention, the distributed feature vector is input to a pre-configured environment decision-maker to obtain decision factors characterizing the current local environment type. The local contrast features and peak interference features extracted in the preceding steps are only quantified values ​​and cannot directly guide the CFAR threshold fusion strategy. Different radar local environments require different CFAR algorithm weights / selection strategies. Therefore, a multi-level condition lookup table based on prior rules, i.e., the environment decision-maker, needs to be constructed to associate feature values ​​with environment types and output decision factors characterizing the environment type, providing clear strategy guidance for subsequent threshold adaptive fusion.

[0042] Step S400 in the method provided in this embodiment of the invention includes: The environmental decision-maker is a multi-level condition lookup table based on prior rules.

[0043] Specifically, the distributed feature vector is input to a pre-configured environmental decision-maker to obtain decision factors characterizing the current local environment type, including: Establish a lookup table containing several decision condition levels, where each decision condition level corresponds to a threshold comparison of a distribution feature; The local contrast features are compared with a preset first threshold value: if the local contrast features are greater than the first threshold value, the current environment is determined to be a clutter edge scene, and the lookup table directly outputs the selection instruction for the ordered statistical CFAR algorithm as the decision factor; or, the preset weight vectors for the cell average CFAR algorithm and the ordered statistical CFAR algorithm are output as the decision factor. If the local contrast feature is less than or equal to the first threshold value, then proceed to the second level of judgment; At the second level, the peak interference feature is compared with a preset second threshold value: if the peak interference feature is greater than the second threshold value, the current environment is determined to be a multi-target interference scenario, and the lookup table outputs the decision factor that prioritizes the use of the ordered statistical CFAR algorithm. If the peak interference feature is less than or equal to the second threshold value, the current environment is determined to be a uniform background scene, and the lookup table outputs the decision factor that preferentially adopts the cell average CFAR algorithm. The lookup table for the judgment condition hierarchy is obtained based on statistical analysis of a large amount of scenario data, and the preset first threshold value and second threshold value are determined through offline optimization.

[0044] First, a lookup table containing several decision condition levels is established, with each decision condition level corresponding to a threshold comparison of a distribution feature. This lookup table is obtained based on statistical analysis of a large amount of scene data. The environment decision-maker is a multi-level condition lookup table based on prior rules. Its function is to map the quantized values ​​of the distribution feature vector to the environment type and output the corresponding decision factor. The multi-level condition lookup table has a structure containing multiple layers of feature threshold judgment logic, with each layer corresponding to a threshold comparison of a distribution feature, progressively narrowing the range of environment type determination. Prior rules refer to judgment logic summarized based on knowledge from the radar signal processing domain and a large amount of scene data; for example, a high local contrast feature corresponds to a clutter edge scene.

[0045] For example, the constructed multi-level condition lookup table adopts two-layer judgment logic. The first level compares the local contrast features with the first threshold value to determine whether it is a clutter edge scene. If it does not meet the requirements, it enters the second level. The second level compares the peak interference features with the second threshold value to distinguish between multi-target interference scenes and uniform background scenes. Different scenes correspond to different types of decision factors.

[0046] The method further includes inputting the distributed feature vector into a pre-configured environmental decision-maker to obtain decision factors characterizing the current local environment type, and also includes: Collect measured or high-fidelity simulated radar data that include at least three typical scenarios: uniform background, clutter edge, and multi-target interference, and extract the corresponding distribution feature vectors. Domain experts, combining their signal processing knowledge, label the measured or high-fidelity simulated radar data samples with real-world environment type tags, which serve as a labeled sample set. Based on the labeled sample set, with the optimization objective of maximizing the distinguishability of different types of samples in the feature space, a preset optimization algorithm is used to iteratively search for the optimal combination of the first threshold value corresponding to the local contrast feature and the second threshold value corresponding to the peak interference feature.

[0047] First, measured or high-fidelity simulated radar data is collected, encompassing at least three typical scenarios: uniform background, clutter edge, and multi-target interference. The corresponding distribution feature vectors are then extracted. A uniform background scenario refers to a conventional detection scenario where noise and clutter power distribution within the radar detection area is stable, without power abrupt changes or multi-target spike interference. A clutter edge scenario refers to a scenario where the radar's local background power undergoes drastic changes, such as the boundary between land and sea, or areas with sudden changes in clutter density. A multi-target interference scenario refers to a scenario where multiple target power spikes exist in the reference cell sequence, interfering with the normal estimation of background power. High-fidelity simulated radar data refers to simulated echo data generated through a professional radar simulation platform, exhibiting characteristics highly consistent with measured data, and can replace measured data to reduce acquisition costs.

[0048] Specifically, collect measured radar data covering at least three typical scenarios: uniform background, clutter edge, and multi-target interference, or generate corresponding high-fidelity simulated radar data; for each set of radar data, calculate local contrast features and peak interference features according to the S300 feature extraction process to form a distribution feature vector corresponding to each set of data.

[0049] For example, 1000 sets of radar data samples were collected, including 300 sets of uniform background scenes, 400 sets of clutter edge scenes, and 300 sets of multi-target interference scenes. For all samples, operations such as reference window division, mean difference calculation, and moving median sequence calculation were performed to extract the local contrast features and peak interference features corresponding to each set of samples, resulting in 1000 sets of standardized distribution feature vectors.

[0050] Secondly, domain experts, combining their signal processing knowledge, annotate the measured or high-fidelity simulated radar data samples with realistic environment type labels, forming an annotated sample set. Environment type labels, annotated by domain experts, are classification identifiers used to characterize the actual detection scenarios corresponding to the radar data, categorized into three types: uniform background, clutter edges, and multi-target interference. The annotated sample set is a standardized set of samples formed by associating radar data, distribution feature vectors, and realistic environment type labels one-to-one; it forms the core data foundation for threshold optimization. Domain experts with expertise in radar signal processing, combining data power distribution characteristics and actual detection scenarios, annotate each set of radar data samples with corresponding realistic environment type labels; integrating and associating the original data, distribution feature vectors, and environment labels forms an annotated sample set that can be used for optimization calculations.

[0051] For example, domain experts verify and label 1,000 sets of data samples one by one, labeling 300 sets of uniform background data with uniform background labels, 400 sets of clutter edge data with clutter edge labels, and 300 sets of multi-target interference data with multi-target interference labels; and matching the sample data, feature vectors and labels one by one to complete the construction of the labeled sample set.

[0052] Furthermore, based on the labeled sample set, with the optimization objective of maximizing the discriminative power of different types of samples in the feature space, a preset optimization algorithm is used to iteratively search for the optimal combination of a first threshold value corresponding to the local contrast feature and a second threshold value corresponding to the peak interference feature. Feature space discriminative power refers to the degree to which samples of different environment types are separated and not confused in the two-dimensional space composed of local contrast and peak interference features; the higher the discriminative power, the more accurate the environment determination. The preset optimization algorithm refers to an iterative algorithm used to automatically search for the optimal threshold, such as grid search or genetic algorithm, which can traverse threshold combinations and calculate the optimization effect. The optimal threshold combination refers to the pairing of the first and second threshold values ​​that maximizes the discriminative power of samples from different scenarios and maximizes the accuracy of the environment decision-maker.

[0053] Specifically, based on the labeled sample set, the first threshold value of the local contrast feature and the second threshold value of the peak interference feature are used as parameters to be optimized. With the goal of maximizing the distinguishability of different types of samples in the feature space, different threshold combinations are traversed through a preset optimization algorithm, and the environmental judgment accuracy of each combination is calculated iteratively. The threshold combination with the highest judgment accuracy and the greatest sample distinguishability is selected as the final optimal threshold.

[0054] For example, grid search is selected as the preset optimization algorithm to traverse all combinations of local contrast feature thresholds from 0.1dB to 1.0dB and peak interference feature thresholds from 0.5 to 1.5. After iterative calculation, when the first threshold value is 0.5dB and the second threshold value is 0.8, the feature space discrimination of the three types of scene samples is the largest and the environment judgment accuracy is the highest. Therefore, this set of values ​​is determined to be the optimal threshold combination.

[0055] Based on this, the distribution feature vector is input into a pre-configured environment decision-maker to obtain decision factors characterizing the current local environment type: First, the local contrast features are compared with a preset first threshold value. If the local contrast features are greater than the first threshold value, the current environment is determined to be a clutter edge scene, and the lookup table directly outputs a selection instruction for the ordered statistical CFAR algorithm as a decision factor; alternatively, a preset weight vector for the cell-average CFAR algorithm and the ordered statistical CFAR algorithm is output as a decision factor. A clutter edge scene refers to a scene where the radar's local background power distribution changes abruptly. Examples include the land-sea boundary and clutter density abrupt changes, where the local contrast features are significantly higher than those of a uniform background. The local contrast features extracted by S300 are compared with the preset first threshold value. If the local contrast features > the first threshold value, the scene is determined to be a clutter edge scene, and the lookup table outputs a selection instruction for the ordered statistical CFAR algorithm.

[0056] Secondly, if the local contrast feature is less than or equal to the first threshold value, the process proceeds to the second level of judgment. If the local contrast feature is less than or equal to the first threshold value: no specific scene is determined, and the process proceeds to the second level of judgment. For example, if the local contrast feature is 0.2dB and the first threshold value is 0.5dB, the comparison result is 0.2dB ≤ 0.5dB, which does not meet the clutter edge scene determination condition, and the process proceeds to the second level of judgment.

[0057] At the second level, the peak interference feature is compared with a preset second threshold value. If the peak interference feature is greater than the second threshold value, the current environment is determined to be a multi-target interference scenario, and the lookup table outputs the decision factor that prioritizes the ordered statistical CFAR algorithm. A multi-target interference scenario refers to a scenario where there are multiple target spikes in the reference unit sequence. In this scenario, the peak interference feature will be significantly higher than that of a uniform background scenario. The peak interference feature extracted in the previous step is compared with the preset second threshold value. If the peak interference feature is greater than the second threshold value, the current environment is determined to be a multi-target interference scenario. For example, if the current peak interference feature is 1 and the preset second threshold value is 0.8, the comparison shows that 1 is greater than 0.8, therefore the current environment is determined to be a multi-target interference scenario.

[0058] Subsequently, if the peak interference feature is less than or equal to the second threshold, the current environment is determined to be a uniform background scene, and the lookup table outputs the decision factor that prioritizes the cell-average CFAR algorithm. A uniform background scene refers to a scene where the radar local background power distribution is stable, without obvious power abrupt changes or spike interference. In this scene, both the local contrast feature and the peak interference feature are at a low level. The peak interference feature extracted in the previous step is compared with the preset second threshold. If the peak interference feature is less than or equal to the second threshold, the current environment is determined to be a uniform background scene. The decision factor is a quantitative result or instruction-type result used to characterize the current local environment type, and is divided into two categories: selection instructions pointing to a specific CFAR algorithm, and weight vectors corresponding to the two basic CFAR algorithms.

[0059] Specifically, based on the environment type determination, corresponding decision factors are output. For clutter edge scenarios, the output is a selection instruction pointing to the ordered statistical CFAR algorithm or a weight vector emphasizing the ordered statistical CFAR algorithm; for multi-target interference scenarios, the output is a weight vector with higher weights in the ordered statistical CFAR algorithm; and for uniform background scenarios, the output is a weight vector with higher weights in the cell average CFAR algorithm. For example, in the current environment of multi-target interference, the lookup table outputs a weight vector [0.3, 0.7] as a decision factor, where 0.3 is the weight of the cell average CFAR algorithm and 0.7 is the weight of the ordered statistical CFAR algorithm.

[0060] In this embodiment of the invention, a multi-level condition lookup table is constructed as an environment decision maker. Combined with offline optimized threshold values, accurate determination of local environment types is achieved. The quantified distribution feature vector is transformed into a decision factor that can directly guide threshold fusion. This not only ensures the regularity and interpretability of environment determination, but also improves the adaptability of threshold values ​​through offline optimization. It provides a precise strategy basis for subsequent threshold adaptive fusion and solves the problem that feature values ​​cannot directly guide algorithm selection.

[0061] S500: Based on the decision factor, the preliminary detection thresholds generated by the at least two basic CFAR algorithms are adaptively fused to generate the final detection threshold.

[0062] In this embodiment of the invention, based on the decision factor, the preliminary detection thresholds generated by the at least two basic CFAR algorithms are adaptively fused to generate the final detection threshold. The preceding steps have generated preliminary detection thresholds for two basic algorithms: Cell Average CFAR and Ordered Statistical CFAR, and the environment decision maker outputs a decision factor adapted to the current local environment. Different radar scenarios have different requirements for the adaptability of the detection thresholds; a single preliminary threshold cannot cover the detection accuracy of all scenarios. Therefore, it is necessary to adaptively fuse the preliminary thresholds based on the decision factor to ensure that the final detection threshold accurately matches the current environmental characteristics, guaranteeing the accuracy and robustness of target detection.

[0063] Step S500 in the method provided in this embodiment of the invention includes: When the decision factor is an identifier pointing to a specific basic CFAR algorithm, the preliminary detection threshold corresponding to the specific basic CFAR algorithm is directly selected as the final detection threshold. When the decision factor is a weight vector, each preliminary detection threshold is multiplied by its corresponding weight and then summed to obtain the final detection threshold after weighted fusion. Specifically, for clutter edge and multi-target interference scenarios, the ordered statistical CFAR algorithm is given higher weight; for uniform background scenarios, the cell average CFAR algorithm is given higher weight.

[0064] First, determine the type of decision factors. Decision factors are divided into two categories: identifiers pointing to a specific basic CFAR algorithm (instruction type), directly specifying the threshold for using a particular algorithm; and weight vectors (quantization type), assigning fusion weights to different CFAR algorithms. Read the decision factors output by the S400 and identify their data type: if it is a preset algorithm selection instruction / identifier, such as a character or instruction code, it is determined to be an identifier pointing to a specific CFAR algorithm; if it is a numerical vector and the sum of its elements is 1, it is determined to be a weight vector. For example, if the decision factor is SELECT_OS_CFAR, it is determined to be an identifier pointing to a specific CFAR algorithm; if the decision factor is [0.3, 0.7], it is determined to be a weight vector.

[0065] Secondly, when the decision factor is an identifier pointing to a specific basic CFAR algorithm, the preliminary detection threshold corresponding to that specific basic CFAR algorithm is directly selected as the final detection threshold. An algorithm identifier refers to an instruction-type identifier used to directly specify the use of a threshold from a certain basic CFAR algorithm, such as SELECT_CA_CFAR pointing to cell average CFAR or SELECT_OS_CFAR pointing to ordered statistical CFAR. If the decision factor is an identifier pointing to a specific CFAR algorithm, the preliminary detection threshold generated by the basic CFAR algorithm corresponding to that identifier is directly extracted and used as the final detection threshold.

[0066] For example, S200 has calculated the first preliminary detection threshold of the cell average CFAR as 21.59dB and the second preliminary detection threshold of the ordered statistical CFAR as 22.5dB; Scenario 1 (clutter edge): the decision factor is SELECT_OS_CFAR (an identifier pointing to the ordered statistical CFAR), and 22.5dB is directly selected as the final detection threshold; Scenario 2 (uniform background): if the decision factor is SELECT_CA_CFAR (an identifier pointing to the cell average CFAR), 21.59dB is directly selected as the final detection threshold.

[0067] Secondly, when the decision factor is a weight vector, each preliminary detection threshold is multiplied by its corresponding weight and then summed to obtain the final detection threshold after weighted fusion. The weight vector refers to the fusion weight values ​​assigned to the two basic CFAR algorithms; the first value corresponds to the weight of the unit average CFAR algorithm, and the second value corresponds to the weight of the ordered statistical CFAR algorithm. Weighted fusion refers to the fusion method of multiplying the preliminary detection thresholds of each basic CFAR algorithm by their corresponding weights and then summing the results to obtain the final threshold. If the decision factor is a weight vector, the weight values ​​of the corresponding unit average CFAR and ordered statistical CFAR are first extracted from the vector. Then, the preliminary detection thresholds of the two algorithms are multiplied by their corresponding weights, and finally, the product results are summed to obtain the final detection threshold after weighted fusion. Specifically, ordered statistical CFAR is given a higher weight in clutter edge / multi-target interference scenarios, while unit average CFAR is given a higher weight in uniform background scenarios.

[0068] For example, the initial threshold for average CFAR of cells is 21.59 dB, and the initial threshold for ordered statistical CFAR is 22.5 dB; Scenario 1 (multi-target interference): the decision factor is a weight vector [0.3, 0.7], where 0.3 is the weight of average CFAR of cells and 0.7 is the weight of ordered statistical CFAR; the final detection threshold is 21.59 × 0.3 + 22.5 × 0.7 = 22.227 dB; Scenario 2 (uniform background): the decision factor is a weight vector [0.8, 0.2]; the final detection threshold is 21.59 × 0.8 + 22.5 × 0.2 = 21.772 dB; Scenario 3 (clutter edge): the decision factor is a weight vector [0.1, 0.9]; the final detection threshold is 21.59 × 0.1 + 22.5 × 0.9 = 22.409 dB.

[0069] In this embodiment of the invention, by distinguishing the types of decision factors, the final detection threshold is generated using either direct selection or weighted fusion, achieving adaptive matching between the threshold and the local radar environment: in clutter edge / multi-target interference scenarios, ordered statistical CFAR thresholds or high weights are preferentially used to avoid threshold deviations caused by interference; in uniform background scenarios, cell average CFAR thresholds or high weights are preferentially used to ensure detection sensitivity. This adaptive fusion strategy solves the problem of poor scenario adaptability of a single CFAR threshold, improving the accuracy and scenario adaptability of the final detection threshold.

[0070] S600: Compare the power value of the current unit to be detected with the final detection threshold, and output the target detection result.

[0071] In this embodiment of the invention, the power value of the current unit to be detected is compared with the final detection threshold, and the target detection result is output. The preceding steps have generated a final detection threshold adapted to the current local environment, but simply calculating the threshold cannot directly output usable radar target information; it is necessary to complete the single-unit target decision by comparing the power value of the unit to be detected with the final threshold, and then convert the discrete decision results into continuous target points through adjacent unit aggregation processing, finally outputting structured target information to meet the standardization requirements of target points and attribute information for subsequent radar trajectory tracking.

[0072] Step S600 in the method provided in this embodiment of the invention includes: If the power value of the current unit to be detected is greater than or equal to the final detection threshold, it is determined that there is a radar target in the current unit to be detected, and a decision signal indicating that there is a target is output. If the power value of the current unit to be detected is less than the final detection threshold, it is determined that there is no radar target in the current unit to be detected, and a decision signal of no target is output. Perform the above decision on each unit to be detected to generate a one-dimensional preliminary detection result vector; The preliminary detection result vector is subjected to adjacent unit aggregation processing to merge the spatially consecutive target decisions into a single target point trace, and the center position and comprehensive signal-to-noise ratio of the target point trace are calculated as attribute information. The generated target points and attribute information are output to the subsequent radar data processing module for track tracking.

[0073] First, if the power value of the current unit to be detected is greater than or equal to the final detection threshold, it is determined that there is a radar target in the current unit to be detected, and a decision signal indicating that there is a target is output.

[0074] Furthermore, if the power value of the current unit to be detected is less than the final detection threshold, it is determined that there is no radar target for the current unit to be detected, and a decision signal indicating no target is output.

[0075] The decision signal is a binary result characterizing whether a radar target exists in a single detection unit, and is divided into a target-present decision signal and a target-absent decision signal. The current detection unit power value refers to the power value of the specified range-Doppler unit extracted in S100. The current detection unit power value obtained in S100 is compared with the final detection threshold generated in S500: if the power value ≥ the final detection threshold: it is determined that a radar target exists in the unit, and a target-present decision signal is output; if the power value < the final detection threshold: it is determined that no radar target exists in the unit, and a target-absent decision signal is output.

[0076] For example, the current power value of the unit to be detected is 80dB; the final detection threshold is 22.409dB; the comparison result is 80dB≥22.409dB, so it is determined that there is a radar target in the unit, and a target detection signal is output.

[0077] Then, the above decision is performed on each target unit to generate a one-dimensional preliminary detection result vector. The preliminary detection result vector is a one-dimensional vector formed by arranging the decision signals of each target unit in an ordered manner according to the spatial order of the range-Doppler units. The vector dimension is consistent with the number of target units, and the elements are quantized values ​​indicating whether a target is present or absent, such as 1 / 0. All target units in the radar detection channel are traversed, and the target decision operation is performed on each unit one by one. Then, the decision signals of all units are arranged sequentially according to the spatial order of the units, such as in the range dimension from near to far index order, ultimately forming a one-dimensional preliminary detection result vector.

[0078] For example, the radar detection channel is set to contain 10 detectable units in the range dimension, with indices from 1 to 10, and the Doppler dimension of all units is 25. Target decision is performed on each of the 10 units sequentially. Among them, the power values ​​of the 8 units with indices 1 to 4 and 7 to 10 are all less than the final detection threshold of 22.409 dB, and the decision result is no target, represented by a logic value of 0. The power values ​​of the 2 units with indices 5 and 6 are both greater than or equal to the final detection threshold, and the decision result is a target, represented by a logic value of 1. All decision signals are arranged in order of unit indices from 1 to 10 to generate a one-dimensional preliminary detection result vector: [0,0,0,0,1,1,0,0,0,0].

[0079] Based on this, the preliminary detection result vector undergoes adjacent cell aggregation processing, merging spatially consecutive target-determined units into a single target point trace. The center position and overall signal-to-noise ratio (SNR) of this target point trace are then calculated as attribute information. Adjacent cell aggregation processing refers to merging spatially consecutive target-determined units in the preliminary detection result vector into a complete target point trace, avoiding multi-target misjudgments caused by discrete decisions and restoring the actual spatial morphology of the target. A target point trace is a continuous set of target units formed after aggregation, containing attributes such as center position and overall SNR, and is the basic representation unit of radar targets. The center position refers to the geometric center index of the aggregated unit segment, used to accurately represent the spatial position of the target. The overall SNR is the average SNR of all units within the aggregated unit segment, calculated by subtracting the mean of the reference unit sequence from the power value of the unit to be detected.

[0080] Specifically, the process first traverses the initial detection result vector to identify all spatially continuous target decision unit segments; then, all units in the same continuous segment are merged to form a complete target point trace; next, the center position of the target point trace is calculated, which is the arithmetic mean of the continuous unit indices; finally, the overall signal-to-noise ratio of the target point trace is calculated by first calculating the signal-to-noise ratio of each unit in the aggregated unit segment, then summing the signal-to-noise ratios of all units and dividing by the number of units to obtain the overall signal-to-noise ratio.

[0081] For example, the initial detection result vector [0,0,0,0,1,1,0,0,0,0] can be traversed to identify a unique continuous segment of target decision units, namely the two units with indices 5 and 6. These two units (5 and 6) are merged to generate a target point. The center position of this point is calculated by taking the arithmetic mean of the two unit indices, i.e., (5+6) / 2=5.5. Combined with the Doppler dimension 25, the center position of the target point is determined to be distance dimension 5.5 and Doppler dimension 25. The measured power values ​​of units with indices 5 and 6 are then supplemented. The power value of index 5 is 80dB, the power value of index 6 is 78dB, and the average value of the reference unit sequence is 14.1dB. The signal-to-noise ratios (SNRs) of the two units are calculated as follows: SNR of index 5 = 80 - 14.1 = 65.9dB, SNR of index 6 = 78 - 14.1 = 63.9dB. The overall SNR is then calculated as (65.9 + 63.9) / 2 = 64.9dB. The final generated target point attribute information is as follows: center position: distance dimension 5.5, Doppler dimension 25; overall SNR 64.9dB.

[0082] Finally, the generated target point traces and attribute information are output to the subsequent radar data processing module for track tracking. The radar data processing module is a core module in the radar system responsible for track tracking, target identification, and threat assessment; it requires structured target point trace information as input to function properly. The target point trace's center position, overall signal-to-noise ratio, and other attribute information are encapsulated into a standardized data format recognizable by the subsequent radar data processing module. This encapsulated structured information is then output to the subsequent radar data processing modules, such as track tracking, via the radar's internal data bus, providing support for subsequent processing.

[0083] For example, the target point attribute information generated above is encapsulated into a standardized data format. The encapsulated content includes the target point number, center position, overall signal-to-noise ratio, and number of aggregation units. Specifically, it can be represented as: target point number TR001, center position: range dimension 5.5, Doppler dimension 25, overall signal-to-noise ratio 64.9dB, number of aggregation units 2. This standardized data is sent to the track tracking module through the radar's internal data bus for subsequent continuous updating, tracking, and identification of target tracks.

[0084] In this embodiment of the invention, by comparing unit thresholds, generating result vectors, aggregating adjacent units, and outputting structured data, the accuracy of target determination for a single unit to be detected is ensured, while effectively solving the problem of multi-target misjudgment caused by discrete decision-making. The generated structured target point traces and attribute information are fully compatible with the input requirements of the radar's subsequent track tracking module, realizing a closed loop from threshold calculation to output of usable target information, thereby improving the practicality and engineering value of radar target detection results.

[0085] Through the specific implementation methods described above, the embodiments of the present invention achieve the following technical effects: This invention provides a signal frequency domain denoising data processing method and system. Through a collaborative process involving reference cell sequence power acquisition, parallel calculation of preliminary thresholds using multi-basic CFAR, extraction of distribution feature vectors, generation of environmental decision factors, adaptive fusion of detection thresholds, and output of target detection results, the system can accurately determine the scene type based on the power distribution characteristics of the radar local environment, dynamically adapt and optimize the detection thresholds. It retains the detection sensitivity of the cell-averaged CFAR algorithm in uniform backgrounds while fully leveraging the anti-interference advantages of the ordered statistical CFAR algorithm in clutter edges and multi-target interference scenarios. This effectively solves the problems of poor scene adaptability and easy distortion of detection thresholds in traditional single CFAR algorithms, improving the robustness of radar signal frequency domain denoising processing and target detection accuracy. At the same time, by aggregating adjacent cells to form standardized target points, it can directly provide reliable data support for subsequent track tracking modules, improving the overall target detection performance of radar in complex detection environments.

[0086] Example 2, as Figure 2 As shown, the present invention provides a signal frequency domain noise reduction data processing system, the system comprising: The power value acquisition module 11 is used to acquire the power value of the reference unit sequence consisting of the current unit to be detected and a preset number of reference units on both sides in the radar detection channel. The CFAR threshold calculation module 12 is used to calculate, in parallel, the preliminary detection thresholds generated by at least two basic CFAR algorithms based on the power values ​​of the reference unit sequence. The feature vector extraction module 13 is used to extract a distribution feature vector based on the power value of the reference unit sequence, wherein the distribution feature vector includes at least local contrast features and peak interference features. The decision factor acquisition module 14 is used to input the distribution feature vector into a pre-configured environment decision-maker to obtain decision factors that characterize the current local environment type. The threshold adaptive fusion module 15 is used to adaptively fuse the preliminary detection thresholds generated by the at least two basic CFAR algorithms based on the decision factor to generate the final detection threshold. The detection result output module 16 is used to compare the power value of the current unit to be detected with the final detection threshold and output the target detection result.

[0087] In one embodiment, the power value acquisition module 11 is further configured to: In the radar signal processing chain, after pulse Doppler processing, the power value of the specified range-Doppler unit is obtained as the power value of the current unit to be detected; Centered on the distance-Doppler unit, a predetermined number of adjacent units are selected forward and backward along the distance dimension, and the power values ​​of the adjacent units are arranged in an orderly manner to form the reference unit sequence.

[0088] In one embodiment, the CFAR threshold calculation module 12 is further configured to: Establish a basic CFAR algorithm set, which includes at least the cell average CFAR algorithm and the ordered statistical CFAR algorithm; For the reference unit sequence, each algorithm in the basic CFAR algorithm set is invoked synchronously; Among them, the unit average CFAR algorithm calculates the arithmetic mean of the power values ​​of all units in the reference unit sequence, multiplies it by a preset scaling factor, and generates the first preliminary detection threshold. Meanwhile, the ordered statistical CFAR algorithm sorts the power values ​​of the reference unit sequence, selects the power value at a specified position, multiplies it by a preset scaling factor, and generates a second preliminary detection threshold. The first preliminary detection threshold and the second preliminary detection threshold are output as preliminary detection thresholds to be fused.

[0089] In one embodiment, the feature vector extraction module 13 is further configured to: The reference unit sequence is divided into a forward reference window and a backward reference window; Calculate the average power value of all cells in the forward reference window as the forward reference window average, and calculate the average power value of all cells in the backward reference window as the backward reference window average. The absolute value of the difference between the mean of the forward reference window and the mean of the backward reference window is calculated as the local contrast feature; Calculate the sliding median sequence of the reference unit sequence; The number of units in the reference unit sequence whose power values ​​exceed a certain proportion of the sliding median at their corresponding positions is counted as the peak interference feature.

[0090] In one embodiment, the decision factor acquisition module 14 is further configured to: The environmental decision-maker is a multi-level condition lookup table based on prior rules.

[0091] Specifically, the distributed feature vector is input to a pre-configured environmental decision-maker to obtain decision factors characterizing the current local environment type, including: Establish a lookup table containing several decision condition levels, where each decision condition level corresponds to a threshold comparison of a distribution feature; The local contrast features are compared with a preset first threshold value: if the local contrast features are greater than the first threshold value, the current environment is determined to be a clutter edge scene, and the lookup table directly outputs the selection instruction for the ordered statistical CFAR algorithm as the decision factor; or, the preset weight vectors for the cell average CFAR algorithm and the ordered statistical CFAR algorithm are output as the decision factor. If the local contrast feature is less than or equal to the first threshold value, then proceed to the second level of judgment; At the second level, the peak interference feature is compared with a preset second threshold value: if the peak interference feature is greater than the second threshold value, the current environment is determined to be a multi-target interference scenario, and the lookup table outputs the decision factor that prioritizes the use of the ordered statistical CFAR algorithm. If the peak interference feature is less than or equal to the second threshold value, the current environment is determined to be a uniform background scene, and the lookup table outputs the decision factor that preferentially adopts the cell average CFAR algorithm. The lookup table for the judgment condition hierarchy is obtained based on statistical analysis of a large amount of scenario data, and the preset first threshold value and second threshold value are determined through offline optimization.

[0092] The method further includes inputting the distributed feature vector into a pre-configured environmental decision-maker to obtain decision factors characterizing the current local environment type, and also includes: Collect measured or high-fidelity simulated radar data that include at least three typical scenarios: uniform background, clutter edge, and multi-target interference, and extract the corresponding distribution feature vectors. Domain experts, combining their signal processing knowledge, label the measured or high-fidelity simulated radar data samples with real-world environment type tags, which serve as a labeled sample set. Based on the labeled sample set, with the optimization objective of maximizing the distinguishability of different types of samples in the feature space, a preset optimization algorithm is used to iteratively search for the optimal combination of the first threshold value corresponding to the local contrast feature and the second threshold value corresponding to the peak interference feature.

[0093] In one embodiment, the threshold adaptive fusion module 15 is further configured to: When the decision factor is an identifier pointing to a specific basic CFAR algorithm, the preliminary detection threshold corresponding to the specific basic CFAR algorithm is directly selected as the final detection threshold. When the decision factor is a weight vector, each preliminary detection threshold is multiplied by its corresponding weight and then summed to obtain the final detection threshold after weighted fusion. Specifically, for clutter edge and multi-target interference scenarios, the ordered statistical CFAR algorithm is given higher weight; for uniform background scenarios, the cell average CFAR algorithm is given higher weight.

[0094] In one embodiment, the detection result output module 16 is further configured to: If the power value of the current unit to be detected is greater than or equal to the final detection threshold, it is determined that there is a radar target in the current unit to be detected, and a decision signal indicating that there is a target is output. If the power value of the current unit to be detected is less than the final detection threshold, it is determined that there is no radar target in the current unit to be detected, and a decision signal of no target is output. Perform the above decision on each unit to be detected to generate a one-dimensional preliminary detection result vector; The preliminary detection result vector is subjected to adjacent unit aggregation processing to merge the spatially consecutive target decisions into a single target point trace, and the center position and comprehensive signal-to-noise ratio of the target point trace are calculated as attribute information. The generated target points and attribute information are output to the subsequent radar data processing module for track tracking.

[0095] It should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A signal frequency domain noise reduction data processing method, characterized in that, The method includes: Obtain the power value of the reference unit sequence consisting of the currently detected unit and a preset number of reference units on both sides in the radar detection channel; Based on the power values ​​of the reference unit sequence, preliminary detection thresholds generated by at least two basic CFAR algorithms are calculated in parallel. Based on the power values ​​of the reference unit sequence, a distribution feature vector is extracted, wherein the distribution feature vector includes at least local contrast features and peak interference features, including: The reference unit sequence is divided into a forward reference window and a backward reference window; Calculate the average power value of all cells in the forward reference window as the forward reference window average, and calculate the average power value of all cells in the backward reference window as the backward reference window average. The absolute value of the difference between the mean of the forward reference window and the mean of the backward reference window is calculated as the local contrast feature; Calculate the sliding median sequence of the reference unit sequence; The number of units in the reference unit sequence whose power values ​​exceed a certain proportion of the sliding median at their corresponding positions is counted as the peak interference feature; The distributed feature vector is input into a pre-configured environmental decision-maker to obtain decision factors characterizing the current local environment type, including: Establish a lookup table containing several decision condition levels, where each decision condition level corresponds to a threshold comparison of a distribution feature; The local contrast features are compared with a preset first threshold value: if the local contrast features are greater than the first threshold value, the current environment is determined to be a clutter edge scene, and the lookup table directly outputs the selection instruction for the ordered statistical CFAR algorithm as the decision factor; or, the preset weight vectors for the cell average CFAR algorithm and the ordered statistical CFAR algorithm are output as the decision factor. If the local contrast feature is less than or equal to the first threshold value, then proceed to the second level of judgment; At the second level, the peak interference feature is compared with a preset second threshold value: if the peak interference feature is greater than the second threshold value, the current environment is determined to be a multi-target interference scenario, and the lookup table outputs the decision factor that prioritizes the use of the ordered statistical CFAR algorithm. If the peak interference feature is less than or equal to the second threshold value, the current environment is determined to be a uniform background scene, and the lookup table outputs the decision factor that preferentially adopts the cell average CFAR algorithm. The lookup table for the judgment condition hierarchy is obtained based on statistical analysis of a large amount of scenario data, and the preset first threshold value and second threshold value are determined through offline optimization. Based on the decision factors, the preliminary detection thresholds generated by the at least two basic CFAR algorithms are adaptively fused to generate the final detection threshold. The power value of the current unit to be detected is compared with the final detection threshold, and the target detection result is output.

2. The signal frequency domain noise reduction data processing method according to claim 1, characterized in that, The power value of the reference element sequence formed by the currently detected element and a preset number of reference elements on both sides in the radar detection channel is obtained, including: In the radar signal processing chain, after pulse Doppler processing, the power value of the specified range-Doppler unit is obtained as the power value of the current unit to be detected; Centered on the distance-Doppler unit, a predetermined number of adjacent units are selected forward and backward along the distance dimension, and the power values ​​of the adjacent units are arranged in an orderly manner to form the reference unit sequence.

3. The signal frequency domain noise reduction data processing method according to claim 1, characterized in that, Based on the power values ​​of the reference cell sequence, preliminary detection thresholds generated by at least two basic CFAR algorithms are calculated in parallel, including: Establish a basic CFAR algorithm set, which includes at least the cell average CFAR algorithm and the ordered statistical CFAR algorithm; For the reference unit sequence, each algorithm in the basic CFAR algorithm set is invoked synchronously; Among them, the unit average CFAR algorithm calculates the arithmetic mean of the power values ​​of all units in the reference unit sequence, multiplies it by a preset scaling factor, and generates the first preliminary detection threshold. Meanwhile, the ordered statistical CFAR algorithm sorts the power values ​​of the reference unit sequence, selects the power value at a specified position, multiplies it by a preset scaling factor, and generates a second preliminary detection threshold. The first preliminary detection threshold and the second preliminary detection threshold are output as preliminary detection thresholds to be fused.

4. The signal frequency domain noise reduction data processing method according to claim 1, characterized in that, The distributed feature vector is input into a pre-configured environment decision-maker to obtain decision factors for characterizing the current local environment type, including: the environment decision-maker is a multi-level condition lookup table based on prior rules.

5. The signal frequency domain noise reduction data processing method according to claim 1, characterized in that, The distribution feature vector is input into a pre-configured environment decision-maker to obtain decision factors characterizing the current local environment type, and the process further includes: Collect measured or high-fidelity simulated radar data that include at least three typical scenarios: uniform background, clutter edge, and multi-target interference, and extract the corresponding distribution feature vectors. Domain experts, combining their signal processing knowledge, label the measured or high-fidelity simulated radar data samples with real-world environment type tags, which serve as a labeled sample set. Based on the labeled sample set, with the optimization objective of maximizing the distinguishability of different types of samples in the feature space, a preset optimization algorithm is used to iteratively search for the optimal combination of the first threshold value corresponding to the local contrast feature and the second threshold value corresponding to the peak interference feature.

6. The signal frequency domain noise reduction data processing method according to claim 1, characterized in that, Based on the decision factors, the preliminary detection thresholds generated by the basic CFAR algorithm are adaptively fused to generate the final detection thresholds, including: When the decision factor is an identifier pointing to a specific basic CFAR algorithm, the preliminary detection threshold corresponding to the specific basic CFAR algorithm is directly selected as the final detection threshold; When the decision factor is a weight vector, each preliminary detection threshold is multiplied by its corresponding weight and then summed to obtain the final detection threshold after weighted fusion. Specifically, for clutter edge and multi-target interference scenarios, the ordered statistical CFAR algorithm is given higher weight; for uniform background scenarios, the cell average CFAR algorithm is given higher weight.

7. The signal frequency domain noise reduction data processing method according to claim 1, characterized in that, The power value of the current unit to be detected is compared with the final detection threshold, and the target detection result is output, including: If the power value of the current unit to be detected is greater than or equal to the final detection threshold, it is determined that there is a radar target in the current unit to be detected, and a decision signal indicating that there is a target is output. If the power value of the current unit to be detected is less than the final detection threshold, it is determined that there is no radar target for the current unit to be detected, and a decision signal of no target is output. Perform the above decision for each unit to be detected to generate a one-dimensional preliminary detection result vector; The preliminary detection result vector is subjected to adjacent unit aggregation processing to merge the spatially consecutive target decisions into a single target point trace, and the center position and comprehensive signal-to-noise ratio of the target point trace are calculated as attribute information. The generated target points and attribute information are output to the subsequent radar data processing module for track tracking.

8. A signal frequency domain noise reduction data processing system, characterized in that, The system is used to implement the signal frequency domain noise reduction data processing method according to any one of claims 1-7, the system comprising: The power value acquisition module is used to acquire the power value of the reference unit sequence consisting of the current unit to be detected and a preset number of reference units on both sides in the radar detection channel. The CFAR threshold calculation module is used to calculate, in parallel, the preliminary detection thresholds generated by at least two basic CFAR algorithms based on the power values ​​of the reference unit sequence. The feature vector extraction module is used to extract a distribution feature vector based on the power value of the reference unit sequence, wherein the distribution feature vector includes at least local contrast features and peak interference features; The decision factor acquisition module is used to input the distribution feature vector into a pre-configured environment decision-maker to obtain decision factors that characterize the current local environment type; The threshold adaptive fusion module is used to adaptively fuse the preliminary detection thresholds generated by the at least two basic CFAR algorithms based on the decision factors to generate the final detection threshold. The detection result output module is used to compare the power value of the current unit to be detected with the final detection threshold and output the target detection result.

Images