Radar detection method and system based on multi-threshold filtering and space-time cross validation

Abstract

The invention provides a radar detection method and system based on multi-threshold filtering and space-time cross validation, is applied to the field of radar data processing, and is used for performing data preprocessing on millimeter wave radar original data to generate preprocessed data, the preprocessed data is composed of an invalid region marking result, an abnormal value filtering result and a background estimation and elimination result; processing the noise statistical characteristics and the space-time correlation information of the scanning lines to generate adaptive filtering parameters and space-time cross validation thresholds; processing the preprocessed data to generate target filtering data; processing the target filtering data and the adaptive filtering parameters to generate a target detection result, the target detection result being used for representing target distribution of different signal intensities; and processing the target detection result, the space-time cross validation threshold and the scanning line space-time correlation information based on adaptive multi-threshold filtering and a space-time cross validation model to generate a final target detection and verification result.

Description

Technical Field

[0001] The present application relates to the field of radar data processing, and in particular to a radar detection method and system based on multi-threshold filtering and spatiotemporal cross-validation. Background Art

[0002] In the ocean, wave fluctuations and complex meteorological conditions (such as rainfall and fog) generate significant noise. This noise is superimposed on the radar echo signal, resulting in a low signal-to-noise ratio (SNR) in the raw data acquired by millimeter-wave radar systems. The background noise has a mean of approximately 20, and a standard deviation of approximately 8. This causes valid target signals to be buried in the noise, making accurate identification difficult. The amplitude and distribution of valid target signals vary significantly, ranging from 50 to 200, and target signals also attenuate at different distances. Furthermore, there are blind spots at close range and attenuation zones at long distances, further complicating target detection.

[0003] Traditional fixed threshold processing methods use a uniform threshold to process radar data, making them inadequate for the complex and changing ocean environment. In practical applications, this approach can either over-filter the data, filtering out smaller, useful target signals as noise and resulting in loss of useful signals; or fail to effectively suppress noise, retaining excessive noise that interferes with target detection, significantly reducing its accuracy.

[0004] Existing systems lack the ability to collaboratively analyze data from adjacent scan lines and fail to fully utilize the temporal and spatial correlations of radar data. This results in insufficient anti-interference capabilities, particularly for weak target detection, making it difficult to accurately extract weak target signals from noise. Consequently, traditional methods and existing systems exhibit significant deficiencies in processing millimeter-wave radar data in marine environments, failing to meet the demands for high signal-to-noise ratios and accurate target detection in practical applications. Summary of the Invention

[0005] The purpose of the present invention is to solve the above problems and provide a radar detection method and system based on multi-threshold filtering and spatiotemporal cross-validation.

[0006] The technical solution of the present application is implemented as follows: The present application proposes a radar detection method and system based on multi-threshold filtering and spatiotemporal cross-validation. The method first obtains the millimeter-wave radar raw data, noise statistical characteristics and scan line spatiotemporal correlation information, and then pre-processes the raw data, and obtains the pre-processed data by marking invalid areas, filtering outliers, and estimating and eliminating the background. Then, adaptive filtering parameters and spatiotemporal cross-validation thresholds are generated based on the noise statistical characteristics and spatiotemporal correlation information, and radial and tangential filtering and two-dimensional fusion are performed on the pre-processed data to obtain target filtered data. Then, adaptive threshold calculation is performed on the target filtered data, and four-level threshold segmentation processing is set and weighted enhancement is performed to generate target detection results. Finally, based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model, the target detection results are subjected to scan line correlation analysis, target consistency evaluation and radial and tangential fusion verification to generate the final target detection and verification results.

[0007] The advantages and benefits of the aforementioned technical solution include at least: The core innovations of the system lie in its adaptive multi-threshold layered processing architecture, dual-dimensional filtering mechanism, spatiotemporal cross-validation mechanism, and parameter adjustment strategy for distance perception. This technology improves the signal-to-noise ratio (SNR) of radar data processing and target detection reliability through multi-dimensional processing. It is suitable for processing millimeter-wave radar data with 8-bit AD sampling and has significant application value in complex scenarios such as marine environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] The accompanying drawings illustrate exemplary implementations of the present application of embodiments of the present invention and, together with the description, are used to explain the principles of the present application. These drawings are included to provide a further understanding of the present application, and the drawings are included in and constitute a part of this specification.

[0009] Figure 1 A flowchart of a radar detection method based on multi-threshold filtering and spatiotemporal cross-validation provided by an embodiment of the present application is shown; Figure 2 A schematic structural diagram of a radar detection device based on multi-threshold filtering and spatiotemporal cross-validation provided in one embodiment of the present application is shown. DETAILED DESCRIPTION

[0010] The following describes embodiments of the present application in more detail with reference to the accompanying drawings. Although certain embodiments of the present application are shown in the accompanying drawings, it should be understood that the present application can be implemented in various forms and should not be construed as limited to the embodiments described herein. Instead, these embodiments are provided to provide a more thorough and complete understanding of the present application. It should be understood that the drawings and embodiments of the present application are for illustrative purposes only and are not intended to limit the scope of protection of the present application.

[0011] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0012] The names of the messages or information exchanged between multiple devices in the embodiments of the present application are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0013] Reference Figure 1 , an embodiment of the present invention provides a flowchart of a radar detection method based on multi-threshold filtering and spatiotemporal cross-validation, including: S101, obtaining millimeter-wave radar raw data, noise statistical characteristics, and scan line spatiotemporal correlation information.

[0014] In one implementation, a millimeter-wave radar deployed in a certain ocean monitoring scenario detects the sea surface using 3600 scan lines per cycle. Each scan line contains 1024 sampling points, and the data format is an 8-bit unsigned integer (amplitude range 0-255). The raw data has a background noise mean of approximately 20 and a noise standard deviation of approximately 8. The effective target signal amplitude ranges from 50 to 200, with a close-range blind spot (the first 30 sampling points) and a long-range attenuation zone (the last 100 sampling points). This raw data is acquired in real time by the radar data acquisition module and serves as the basic input for subsequent processing.

[0015] Obtain noise statistical characteristics and perform noise analysis on the raw data. The 1024 sampling points of each scan line are divided into 32 segments (32 points per segment). For each segment: extract the amplitude values ​​of the 32 points within the segment, sort them, and calculate the standard deviation of the top 50% of low-amplitude points (16 points) to serve as the noise standard deviation σ(r) of the segment. If the segment is located in an invalid area (such as a close-range blind spot), the standard deviation of the adjacent valid segment is used as a substitute value. Through the above method, the noise standard deviation characteristics of the entire scan line (such as σ(400)≈8 for the 400th point in the mid-range area) and the noise amplitude distribution characteristics (such as the low-amplitude points are concentrated between 0 and 30) are generated for adaptive filtering parameter calculation.

[0016] Obtain the temporal and spatial correlation information of the scan lines. In radar data, the temporal and spatial correlation of adjacent scan lines is obtained by: for each scan line i, select its adjacent scan lines i-1 and i+1, and calculate the signal amplitude values ​​of the same sampling point j in the three scan lines. 、 、 For example, in the close distance region (j=100), the spatial domain correlation window size is set to =5, the time domain correlation window size is =3, the calculated correlation coefficient of the three scan lines at this point is 0.75, the variance is 0.1, and the normalized consistency index (correlation coefficient) is 0.82, reflecting that the spatiotemporal consistency of the signal in this area is high.

[0017] S102: Preprocess the millimeter-wave radar raw data to generate preprocessed data.

[0018] In one implementation, millimeter-wave radar raw data is processed for invalid region marking to generate an invalid region marking result. In a certain ocean monitoring scenario, millimeter-wave radar raw data contains short-range blind spots (the first 30 sampling points) and long-range attenuation zones (the last 100 sampling points). These areas are marked as invalid due to unreliable signals or excessive noise. By setting a spatial range (e.g., j < 30 or j > 923), the corresponding sampling points in the raw data are marked as invalid, generating an invalid region marking result (for example, by setting the corresponding positions in the marking matrix to "0" or a special identifier).

[0019] Millimeter-wave radar raw data is filtered for outliers to generate the filtering results. The 3σ principle is used to filter out outliers whose amplitudes significantly deviate from the normal range (e.g., noise points with amplitudes > 200 or abnormally low amplitudes < 10). For example, using a background noise mean of μ = 20 and a standard deviation of σ = 8, a threshold range of [μ - 3σ, μ + 3σ] = [−4, 44] is set. Samples outside this range (e.g., amplitudes of 250 or 5) are identified as outliers and replaced with the mean of the adjacent valid points (e.g., median filtering) to generate the filtering results. For example, the outlier point is corrected to the mean of the five adjacent points.

[0020] Background estimation and removal are performed on the raw millimeter-wave radar data to generate background estimation and removal results. Background noise is estimated using a sliding average method: each scan line is divided into 32 segments (32 points per segment). The mean of the low-amplitude points within each segment (the top 50% after sorting) is calculated as the background value (for example, the background mean for a segment is 18). The background value for the corresponding segment is then subtracted from the raw data, retaining the target signal (for example, an effective signal amplitude of 50-200 becomes 32-182 after background subtraction). This generates the background estimation and removal results.

[0021] Preprocessed data is generated through comprehensive processing based on the invalid region marking results, outlier filtering results, and background estimation and removal results. Based on the original data characteristics ("30 sampling points before the near-range blind zone and 100 sampling points after the long-range attenuation zone"), an invalid region (j < 30 or j ≥ 1024 - 100 = 924) is delineated by spatial range. For invalid sampling points, a conservative zeroing strategy is adopted or the value after background removal is retained. Zeroing directly sets the amplitude of sampling points in the invalid region to 0 to completely eliminate unreliable data (for example, sampling points with j = 10-20 are set to 0). Background values ​​are retained. If background noise characteristics need to be retained (for example, for subsequent analysis of ambient noise), the invalid region is filled with background estimation results (such as a segment mean of 18), meaning the invalid point amplitude equals the background mean. For a scan line with j = 10 (in the near-range blind zone), the original amplitude is 25 and the background mean is 18. After zeroing, the amplitude of this point becomes 0, while after retaining the background value, the amplitude of this point becomes 18.

[0022] Based on the statistical characteristics of noise (e.g., background noise mean μ = 20, standard deviation σ = 8), the 3σ criterion was used to define the normal range as [μ - 3σ, μ + 3σ] = [-4, 44] (because amplitude ≥ 0, the actual range was [0, 44]). Outliers (e.g., amplitude = 200 or 5) were corrected using a 5-point median filter or the mean of the adjacent valid points. For example, if the amplitude at point j = 500 was 200 (an abnormally high value), the median of 150 of the adjacent 5 points (j = 498-502) was used instead. For background removal, for each valid segment (e.g., j = 30-923), the mean of the low-amplitude points (the top 50%) within the segment was calculated as the background value (e.g., the background mean for a segment was 18). The background value was then subtracted from the original amplitude to obtain the background-removed amplitude. For example, if the original amplitude at point j = 400 was 60 and the background mean was 18, the post-removal amplitude would be 60 - 18 = 42. The corrected outliers (e.g., points j = 500 were corrected from 200 to 150) were superimposed with the background removal results (e.g., 150-18=132) to ensure that both outliers and background noise were eliminated from the valid points.

[0023] The preprocessed data results are as follows: Spatial range: only valid regions j = 30-923 are retained, for a total of 923 - 30 + 1 = 894 sampling points. Amplitude adjustments include the following: Invalid regions are strategically set to zero or padded with background values ​​(e.g., 0 or 18). Valid regions: Amplitude = original amplitude - background mean, ranging from the minimum valid signal (50 - 18 = 32) to the maximum valid signal (200 - 18 = 182), i.e., [0, 182] (including the value 0 after the invalid regions are zeroed). By removing outliers (e.g., noise points with amplitudes > 44) and interference from invalid regions, the background noise is reduced from a mean of 20 to near zero, significantly improving the signal-to-noise ratio (e.g., original SNR = 5dB, SNR ≈ 8dB after preprocessing).

[0024] S103 , processing the noise statistical characteristics and the spatiotemporal correlation information of the scan lines to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds.

[0025] In one implementation, feature extraction is performed on the noise statistical characteristics to generate noise standard deviation features and noise amplitude distribution features. In a certain ocean monitoring scenario, each scan line of a millimeter-wave radar contains 1024 sampling points, which are divided into 32 segments (32 points per segment). Taking the segment where the 400th point in the mid-range area is located as an example: Extract the amplitude values ​​of the 32 points in the segment, sort them, and select the top 16 low-amplitude points (e.g., within the amplitude range of 5-30). Calculate their standard deviation, σ(r), = 8 (consistent with the document's "noise standard deviation is approximately 8"). If a segment is in an invalid area (e.g., a close-range blind spot with j < 30), use the standard deviation, σ = 8, of the adjacent valid segment (e.g., the segment with j = 30-61) as a replacement value.

[0026] Through statistics of the entire scan line, low-amplitude points (≤30) account for about 60%, mainly distributed in the invalid area and background area, and the amplitude values ​​are concentrated in the range of 0-20 (such as the background noise mean of 20), forming a normal distribution feature centered on the mean.

[0027] The feature extraction process of the spatiotemporal correlation information of the scan line is performed to generate the scan line correlation coefficient feature and the distance area feature. For the scan line i=1000, j=400, the adjacent scan lines i-1=999, i+1=1001 are selected, and the correlation coefficients of the three-point amplitude values ​​S_1000(400)=80, S_999(400)=78, and S_1001(400)=82 are calculated: The mean μ=(80+78+82) / 3=80; the variance σ²=[(80-80)²+(78-80)²+(82-80)²] / 3≈2.67; the correlation coefficient corr=1 (the three points are completely positively correlated), and the consistency index after normalization is 1.0, indicating that the signal at this point has high spatiotemporal consistency.

[0028] The distance regions are divided according to the position of sampling point j, as follows: Close distance: j=30-150 (e.g. j=100), corresponding to =3 (radial half window width), =7 (tangential half window width); Medium distance: j=151-700 (e.g. j=400), corresponding to =5, =5; Long distance: j=701-923 (such as j=800), corresponding to =7, =3.

[0029] Based on the noise standard deviation characteristics, noise amplitude distribution characteristics, combined with the scan line correlation coefficient characteristics, and distance area characteristics, a comprehensive analysis is performed to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds. Among them, the adaptive filtering parameters are used to characterize the filter window size and weight coefficient of different distance areas, and the spatiotemporal cross-validation threshold is used to characterize the judgment standard of signal credibility. In the medium distance area (such as j=400), radial filtering uses a half-window width =5, weight coefficients are [0.1, 0.2, 0.4, 0.2, 0.1], by taking weighted average of the current sampling point and the 4 adjacent points (a total of 11 points), the center pixel has the highest weight and decreases towards both sides, effectively suppressing mid-range noise and retaining signal details; tangential filtering also uses half-window width =5, the weight coefficient is the same as the radial direction, and the weighted average of the same sampling point position of 5 adjacent scan lines is performed to enhance the coordination between adjacent scan lines and reduce spatial domain interference.

[0030] In the long-distance region (e.g. j = 800), the radial filter half-window width increases to =7, the weight coefficient is adjusted to a distribution with a higher center weight (such as [0.05, 0.1, 0.15, 0.2, 0.2, 0.15, 0.1]), and the long-distance attenuation noise is suppressed by expanding the time domain smoothing range while avoiding excessive target blur; the tangential filter half-window width is reduced to =3, and the weight coefficient is set to [0.2, 0.3, 0.5], focusing on the details of a single scan line and reducing cross-line interference. In terms of spatiotemporal cross-validation, taking the close distance point (j=100) as an example, the time domain correlation window is set =3 (covering 7 sampling points from j-3 to j+3) and spatial correlation window =5 (covering a total of 11 scan lines from i-5 to i+5), calculate the correlation coefficient of the three scan lines (i-5, i, i+5) at j=100. When the correlation coefficient is ≥0.65, the signal is considered credible and retained. If it is below this threshold, it is considered as noise and removed. By combining the distance area characteristics with the correlation analysis in this way, the credibility evaluation and interference removal of signals at different distances can be achieved.

[0031] S104: Process the pre-processed data to generate target filtered data.

[0032] In one embodiment, radial filtering is performed on the pre-processed data to generate radial filtering results, wherein the radial filtering process adaptively adjusts the half-window width according to the distance region where the sampling point is located, and adopts the corresponding weight coefficient array for weighted averaging, and the boundary processing adopts the boundary mirror method to ensure that the result range is between 0-255, and the distance region includes the near distance j<150, the middle distance 150≤j<700, and the far distance j≥700. The near distance half-window width is 3, the middle distance is 5, and the far distance is 7. The radial filtering process is as follows: a certain scan line is i=500, and the sampling point j=100 (belongs to the near distance region, j<150). The distance region is divided into near distance (j<150), so the radial half-window width =3, the corresponding window range is j-3 to j+3 (i.e. 97-103). =5, the weights are [0.1, 0.2, 0.4, 0.2, 0.1]", close distance ( =3) weights are [0.2, 0.3, 0.5] (center pixels have higher weights to meet the need for close-range focusing details).

[0033] ,in, is the value of the j-th sampling point of the i-th scan line after radial filtering, It is the original sampling value, and the data range is 0-255. is the weight coefficient. The amplitudes of the preprocessed data at j=97-103 are [25, 30, 35, 40, 38, 32, 28] respectively. 0.2×25+0.3×30+0.5×35=5+9+17.5=31.5. If j=1 (exceeds the left boundary), the mirroring method is used, and the value of j=2 is used as the mirror filling point (for example, if j=1, the corresponding mirror point is j=2). The calculated result of 31.5 is rounded to 32 to ensure that it is within the range of 0-255.

[0034] The pre-processed data is tangentially filtered to generate the tangential filtering results. The tangential filtering window size varies with distance, which is opposite to the radial one. The half-window width is 7 for close distance, 5 for medium distance, and 3 for long distance. The boundary processing uses the nearest scan line data to ensure that the result range is between 0-255. For the same sampling point j=100, the scan line i=500 (close distance area, j<150). The distance area is divided into close distance (j<150), and the tangential half-window width is opposite to the radial one, so =7, the corresponding scan line range is i-7 to i+7 (i.e. 493-507).

[0035] Similar to the radial direction, symmetric weights are used: [0.05, 0.08, 0.12, 0.2, 0.25, 0.15, 0.15] (the center scan line has the highest weight). The amplitudes of adjacent scan lines at j = 100 are [22, 25, 28, 35, 40, 37, 33, 30, 27, 24, 20, 18, 16] (there are 15 scan lines in total, and the corresponding weights are taken from the middle 13).

[0036] ,in, is the value of the j-th sampling point of the i-th scan line after tangential filtering, is the half-window width of the tangential filter, which varies with distance (the opposite of radial filtering). is the weight coefficient.

[0037] =0.05×22+0.08×25+0.12×28+0.2×35+0.25×40+0.15×37+0.15×33=32.45. If i=1 (exceeds the upper boundary), the data of the most recent valid scan line i=2 is used for padding. 32.45 is rounded to 32 to ensure it is within the range of 0-255.

[0038] Perform dual-dimensional filtering fusion processing on the radial filtering results and the tangential filtering results to generate the target filtering data, where the fusion formula is: , is the value of the j-th sampling point of the i-th scan line after radial filtering, is the value of the jth sampling point of the i-th scan line after tangential filtering, the fusion coefficient α is 0.7, and the final result is composed of the weighted fusion of the radial and tangential filtering results. Substituting the above parameters into the fusion formula, we can get: , α=0.7, substitute the radial result 32 and the tangential result 32: 0.7×32+0.3×32=32. The fusion result combines the filtering effects of time domain (radial) and space domain (tangential). In the close-range area, a larger tangential window ( =7) to enhance the coordination of adjacent scan lines, while using radial small windows ( = 3) to preserve local details, ultimately outputting the target filtered data as 32. Radial filtering adaptively adjusts the window size based on distance (a small window preserves details at close range, a large window suppresses noise at long range), while tangential filtering adjusts the window size inversely to balance spatial correlations. Weighted fusion (α = 0.7) of the two achieves complementary temporal and spatial information. The resulting target filtered data suppresses noise while preserving signal characteristics.

[0039] S105: Process the target filtering data and the adaptive filtering parameters to generate a target detection result.

[0040] In one embodiment, the target filter data is subjected to adaptive threshold calculation processing to generate thresholds at various distances, wherein the threshold calculation is based on the local signal-to-noise ratio, and the formula is: , Take 38, is 2.5, 512 points, noise standard deviation Calculated from the low-amplitude points in the segment. Point j = 400 in the mid-range region (belongs to the valid region, the segment index is segment 13, each segment has 32 points, 13 × 32 = 416 points, covering points j = 400-431).

[0041] Extract the 32 amplitude values ​​of the 13th segment, sort them and take the top 50% (top 16) low amplitude points, whose values ​​are [20, 22, 25, 28, 30, 32, 33, 35, 36, 38, 40, 42, 43, 45, 48, 50].

[0042] Calculate the standard deviation: , where the mean 35.

[0043] Distance decay factor: , =512 is the midpoint of the scan line.

[0044] Adaptive threshold calculation: .

[0045] Based on the threshold value at each distance, four threshold levels are set, and the target filter data is segmented by multiple thresholds, and the signal is divided into five levels and assigned corresponding weights. Based on \(T(400)=49\), four threshold levels are set: (T1(400)=0.60 49 29 (weak signal threshold); (T2(400)=0.85 49 42 (medium signal threshold); (T3(400)=1.00 49 49 (strong signal threshold); (T4(400)=1.20 49 59 (super strong signal threshold).

[0046] An example of signal level classification is as follows: If the amplitude of a sampling point (S(i,400)=35: due to T1(29) 35 < T2(42), classified as L1 level, with a weight of 0.50.

[0047] If S(i, 400) = 45: Due to T2(42) 45 < T3(49), classified as L2 level, with a weight of 0.75.

[0048] If S(i, 400) = 55: Due to T3(49) 55 < T4(59), classified as L3 level, with a weight of 0.90.

[0049] If S(i, 400) = 65: Due to T4(59), classified as L4 level, with a weight of 1.00.

[0050] Perform weighted enhancement processing on each level of signals, calculate the weighted value , generate the target detection result. Among them, the target detection result is used to characterize the target distribution of different signal intensities. Take the sampling point (S(i, 400) = 45 (L2 level, weight 0.75)) as an example: Calculate the weighted value, (S'(i, 400) = 45 0.75 = 33.75 34. After processing, the amplitude of this point is 34, corresponding to a medium signal (L2 level), marked with a specific color (such as yellow) in the target distribution map, the intensity level is "medium", and its spatial position (i = 500 lines, j = 400 points) is recorded.

[0051] For the 1024 sampling points on each scan line, calculate their threshold levels and weighted values one by one. The finally generated target detection result is a two-dimensional matrix, and each element contains: spatial coordinates (i, j), intensity level (L0 - L4), weighted amplitude (0 - 255), signal type label (weak / medium / strong / ultra-strong).

[0052] The adaptive threshold calculation is dynamically adjusted based on local noise characteristics and distance attenuation. The four-level threshold realizes the fine-grained classification of signal intensities. The weighted enhancement processing improves the visibility of weak signals while retaining strong signals. The finally generated target detection result can not only suppress background noise (such as setting L0 level to 0), but also highlight targets of different intensities through hierarchical enhancement (such as amplifying the amplitudes of L1 - L4 levels according to weights), improving the signal-to-noise ratio and the reliability of target detection.

[0053] S106, based on the adaptive multi-threshold filtering and spatio-temporal cross-validation model, process the target detection result, spatio-temporal cross-validation threshold, and scan line spatio-temporal correlation information to generate the final target detection and verification result.

[0054] In one embodiment, a scan line correlation analysis is performed on the target detection results based on an adaptive multi-threshold filtering and spatiotemporal cross-validation model to generate a correlation coefficient, mean, and variance; In the close-range area (j=100), scan line i=500, the target detection result shows an amplitude of 34 (L2 level, medium signal). The amplitudes of the three adjacent scan lines (i-1=499, i=500, i+1=501) at j=100 are: S_499(100)=32 (L2 level); S_500(100)=34 (L2 level); and S_501(100)=30 (L1 level).

[0055] Mean calculation: .

[0056] Variance calculation: .

[0057] Using the Pearson correlation coefficient formula, since the mean of the three points is 32, calculate the deviation of each point from the mean: , (because the numerator and denominator are the same, the normalized correlation coefficient is 1.0).

[0058] The target consistency evaluation of the correlation coefficient is performed based on the time-space cross-validation threshold, and the signal credibility is judged to see whether it meets the threshold requirements, and the evaluated signal is generated. The time-space cross-validation threshold is set to If the correlation coefficient is 0.65, the signal is considered highly credible and retained as a valid signal. If the correlation coefficient at a point is 0.5<0.65, it is considered noise and its amplitude is set to 0 or marked as unreliable. After evaluation, the signal retains the amplitude of 34 at point j = 100 and is marked as "high" (correlation coefficient 1.0).

[0059] The evaluated signals are verified by radial and tangential fusion combined with the spatiotemporal correlation information of the scan lines to generate the final target detection and verification results, which are used to characterize the spatial distribution, intensity level and credibility quantification indicators of the target.

[0060] Spatiotemporal correlation information: temporal correlation window of close distance points (j=100) =3 (covering j-3 to j+3), airspace-related window =5 (covering i-5 to i+5). The fusion verification process is as follows: Radial verification: Check the amplitude continuity between point j = 100 and adjacent sampling points (j = 97-103). If the amplitudes of adjacent points are all between 29-42 (T1-T2 threshold), they meet the L1-L2 signal characteristics and are considered radially consistent. Tangential verification: Check the amplitude trend between line i = 500 and the five adjacent scan lines (i-5 to i+5) at j = 100. If most scan lines have an amplitude between 30-35 at this point, the tangential consistency is considered good. Combined correlation analysis (corr = 1.0) and radial and tangential consistency verification confirm that this point is a valid target. The final detection results are: Spatial distribution: i = 500 lines, j = 100 points; Intensity level: L2 (medium signal); Confidence quantification: Correlation coefficient 1.0 (highest confidence).

[0061] In the long-distance region (j=800), the scan line i=2000, the amplitude of the point in the target detection result is 60 (L3 level, strong signal), and the amplitudes of the three adjacent scan lines are 58, 60, and 62, with a correlation coefficient of 1.0, a mean of 60, and a variance of approximately 2.67. Spatiotemporal window parameters: long-distance point time domain window =5\), airspace window =3, verifying a larger temporal neighborhood and a smaller spatial neighborhood. Fusion verification is as follows: radial verification reveals that the amplitudes at points j = 800 and j = 795-805 are both ≥ 49 (T3 threshold). Tangential verification shows strong signals at this point across all three adjacent scan lines. The final confidence level is 1.0, confirming it as a true target.

[0062] The following example shows how to remove noise points. For example, a point with an amplitude of 25 (L0 level, noise) and a correlation coefficient of 0.3 < 0.65 is removed directly, and its amplitude is set to 0. The point is marked as "noise" in the target detection results and is not included in the target distribution statistics.

[0063] In the above example, the spatiotemporal cross-validation model first quantifies signal consistency through correlation analysis of adjacent scan lines (for example, a correlation coefficient of 1.0 indicates perfect correlation). This model then compares the results with a preset threshold (0.65) to screen for reliable signals. Finally, radial and tangential fusion validation is performed using spatiotemporal window parameters (for example, σ_angle = 5 for close range and σ_radius = 5 for long range). The resulting detection results not only include the target's spatial location and intensity level, but also quantify signal reliability through confidence metrics such as the correlation coefficient. This process effectively eliminates false alarms (such as noise points with a correlation coefficient < 0.65) and enhances confidence in the detection of real targets, meeting the document's design goal of "reducing false alarms and improving weak target detection."

[0064] In another implementation, taking the target detection results for a mid-range region (j=400) and scan line i=1000 as an example, after multi-threshold processing, the amplitude of this point is 45 (L2 level, medium signal). First, the amplitudes of three adjacent scan lines (i-1=999, i=1000, i+1=1001) at j=400 are extracted as 43, 45, and 47, respectively. The mean is calculated to be 45, and the variance is [(43-45)²+(45-45)²+(47-45)²] / 3≈2.67. The correlation coefficient is 1.0 (perfect positive correlation between the three points) obtained using the Pearson formula. Next, based on the spatiotemporal cross-validation threshold γ_threshold=0.65, since the correlation coefficient is 1.0≥0.65, the signal is determined to be highly credible, and the amplitude of 45 is retained and marked as valid. Subsequently, fusion verification was performed using spatiotemporal correlation information. In the radial direction, the amplitudes at point j = 400 and five adjacent sampling points (j = 398-402) were all between 42 and 49 (T2-T3 thresholds), consistent with L2-L3 signal continuity. In the tangential direction, the amplitude trends of five adjacent scan lines (i = 998-1002) at j = 400, centered around i = 1000, revealed moderate signal strengths of 40 to 50, demonstrating good tangential consistency. The resulting detection and verification results indicated that the target was located at line i = 1000 and point j = 400, with an intensity level of L2. The confidence quantification indicator was a correlation coefficient of 1.0 (highest confidence). Its spatial distribution and intensity level were confirmed using multi-threshold filtering, and its confidence was further verified using spatiotemporal correlation analysis, effectively eliminating noise interference.

[0065] The present application proposes a radar detection method and system based on multi-threshold filtering and spatiotemporal cross-validation, which is used to solve the problems of low signal-to-noise ratio and difficult target detection when processing millimeter-wave radar data in marine environments. The method first obtains the millimeter-wave radar raw data, noise statistical characteristics and scan line spatiotemporal correlation information, and then pre-processes the raw data, and obtains the pre-processed data by marking invalid areas, filtering outliers, and estimating and eliminating the background. Then, based on the noise statistical characteristics and spatiotemporal correlation information, adaptive filtering parameters and spatiotemporal cross-validation thresholds are generated, and radial and tangential filtering and two-dimensional fusion are performed on the pre-processed data to obtain target filtered data. Then, adaptive threshold calculation is performed on the target filtered data, and four-level threshold segmentation processing is set and weighted enhancement is performed to generate target detection results. Finally, based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model, the target detection results are subjected to scan line correlation analysis, target consistency evaluation and radial and tangential fusion verification to generate the final target detection and verification results.

[0066] The system's core innovations lie in its adaptive multi-threshold layered processing architecture, dual-dimensional filtering mechanism, spatiotemporal cross-validation mechanism, and parameter adjustment strategy for distance perception. This multi-dimensional processing improves the signal-to-noise ratio (SNR) of radar data processing and the reliability of target detection. Suitable for processing millimeter-wave radar data with 8-bit AD sampling, it has significant application value in complex scenarios such as marine environments.

[0067] In one embodiment, Figure 2 As shown, the present application also provides a radar detection device based on multi-threshold filtering and spatiotemporal cross-validation, comprising: An acquisition module 201 is used to acquire millimeter-wave radar raw data, noise statistical characteristics, and scan line spatiotemporal correlation information; The processing module 202 is used to preprocess the millimeter-wave radar raw data to generate preprocessed data, wherein the preprocessed data consists of invalid area marking results, outlier filtering results, and background estimation and elimination results; process the noise statistical characteristics and scan line spatiotemporal correlation information to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds; process the preprocessed data to generate target filtering data; process the target filtering data and the adaptive filtering parameters to generate target detection results, wherein the target detection results are used to characterize the target distribution of different signal strengths; and process the target detection results, spatiotemporal cross-validation thresholds, and scan line spatiotemporal correlation information based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model to generate the final target detection and verification results.

[0068] The computer-readable storage medium provided in the above-mentioned embodiment of the present application and the radar detection method based on multi-threshold filtering and spatiotemporal cross-validation provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the method adopted, run or implemented by the application program stored therein.

[0069] Each embodiment of this application is described in a related manner. References to the common and similar parts between the various embodiments are sufficient. Each embodiment focuses on the differences from the other embodiments. In particular, the embodiments of the radar detection method, electronic device, electronic device, and readable storage medium for evaluating multi-threshold filtering and spatiotemporal cross-validation are generally similar to the aforementioned embodiments of the radar detection method based on multi-threshold filtering and spatiotemporal cross-validation, so the description is relatively simple. For relevant parts, references to the partial description of the aforementioned embodiments of the radar detection method based on multi-threshold filtering and spatiotemporal cross-validation are sufficient.

[0070] Those skilled in the art should understand that the above embodiments are only for clearly illustrating the present application and are not intended to limit the scope of the present application.

Claims

1. A radar detection method based on multi-threshold filtering and spatiotemporal cross-validation, characterized by: Obtain millimeter-wave radar raw data, noise statistical characteristics, and scan line spatiotemporal correlation information; Perform data preprocessing on the millimeter-wave radar raw data to generate preprocessed data, where the preprocessed data consists of invalid area marking results, outlier filtering results, and background estimation and elimination results; The noise statistical characteristics and the spatiotemporal correlation information of the scan lines are processed to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds; Process the preprocessed data to generate target filtered data; Processing the target filtering data and the adaptive filtering parameters to generate a target detection result, wherein the target detection result is used to characterize the target distribution at different signal strengths; Based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model, the target detection results, spatiotemporal cross-validation thresholds and scan line spatiotemporal correlation information are processed to generate the final target detection and verification results.

2. The method according to claim 1, wherein: Perform data preprocessing on the millimeter-wave radar raw data to generate preprocessed data, including: Perform invalid area marking processing on the millimeter wave radar raw data to generate invalid area marking results; Perform outlier filtering on the millimeter-wave radar raw data to generate outlier filtering results; Perform background estimation and elimination processing on the millimeter-wave radar raw data to generate background estimation and elimination results; Comprehensive processing is performed based on the invalid area marking results, outlier filtering results, and background estimation and elimination results to generate preprocessed data.

3. The method according to claim 1, wherein: The noise statistical characteristics and the spatiotemporal correlation information of the scan lines are processed to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds, including: Perform feature extraction on the noise statistical characteristics to generate noise standard deviation features and noise amplitude distribution features; Perform feature extraction on the spatiotemporal correlation information of the scan line to generate scan line correlation coefficient features and distance area features; Based on the noise standard deviation characteristics and noise amplitude distribution characteristics, combined with the scan line correlation coefficient characteristics and distance area characteristics, a comprehensive analysis and processing is performed to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds. Among them, the adaptive filtering parameters are used to characterize the filter window size and weight coefficient of different distance areas, and the spatiotemporal cross-validation threshold is used to characterize the judgment standard of signal credibility.

4. The method according to claim 1, wherein: Process the preprocessed data to generate target filtered data, including: Perform radial filtering on the preprocessed data to generate radial filtering results. The radial filtering process adaptively adjusts the half-window width according to the distance zone where the sampling point is located, and uses the corresponding weight coefficient array for weighted averaging. The boundary processing uses the boundary mirror method to ensure that the result range is between 0 and 255. The distance zone includes the near distance j < 150, the medium distance 150 ≤ j < 700, and the long distance j ≥ 700. The half-window width for the near distance is 3, the medium distance is 5, and the long distance is 7. Perform tangential filtering on the pre-processed data to generate tangential filtering results. The tangential filtering window size varies with distance. In contrast to the radial window size, the half-window width is 7 for close distance, 5 for medium distance, and 3 for long distance. The boundary processing uses the nearest scan line data to ensure that the result range is between 0 and 255. Perform dual-dimensional filtering fusion processing on the radial filtering results and the tangential filtering results to generate the target filtering data, where the fusion formula is: , is the value of the j-th sampling point of the i-th scan line after radial filtering, is the value of the j-th sampling point of the i-th scan line after tangential filtering, the fusion coefficient α is 0.7, and the final result is composed of the weighted fusion of the radial and tangential filtering results.

5. The method according to claim 1, wherein: The target filtering data and adaptive filtering parameters are processed to generate target detection results, including: The target filter data is processed by adaptive threshold calculation to generate the threshold at each distance. The threshold calculation is based on the local signal-to-noise ratio, and the formula is: , Take 38, is 2.5, 512 points, noise standard deviation Calculated by the low amplitude point of the segment; Four threshold levels are set based on the thresholds at each distance, and the target filtered data is segmented by multiple thresholds to divide the signal into five levels and assign corresponding weights; Perform weighted enhancement processing on each level of signal and calculate the weighted value , generating target detection results, where the target detection results are used to characterize the target distribution with different signal strengths.

6. The method according to claim 5, characterized in that: The target detection results, spatiotemporal cross-validation thresholds, and scan line spatiotemporal correlation information are processed based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model to generate the final target detection and verification results, including: Based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model, the scan line correlation analysis of the target detection results is performed to generate the correlation coefficient, mean and variance; Based on the spatiotemporal cross-validation threshold, the correlation coefficient is evaluated for target consistency, and the signal credibility is judged to be in compliance with the threshold requirement, and the evaluated signal is generated; The evaluated signals are verified by radial and tangential fusion combined with the spatiotemporal correlation information of the scan lines to generate the final target detection and verification results, which are used to characterize the spatial distribution, intensity level and credibility quantification indicators of the target.

7. A radar detection device based on multi-threshold filtering and spatiotemporal cross-validation, characterized in that: The device comprises: The acquisition module is used to obtain millimeter-wave radar raw data, noise statistical characteristics, and scan line spatiotemporal correlation information; The processing module is used to preprocess the millimeter-wave radar raw data to generate preprocessed data, wherein the preprocessed data consists of invalid area marking results, outlier filtering results, and background estimation and elimination results; process the noise statistical characteristics and scan line spatiotemporal correlation information to generate adaptive filtering parameters and spatiotemporal cross-validation thresholds; process the preprocessed data to generate target filtering data; process the target filtering data and adaptive filtering parameters to generate target detection results, wherein the target detection results are used to characterize the target distribution of different signal strengths; based on the adaptive multi-threshold filtering and spatiotemporal cross-validation model, the target detection results, spatiotemporal cross-validation thresholds and scan line spatiotemporal correlation information are processed to generate final target detection and verification results.

8. An electronic device, characterized in that: include: a first processor; and a memory for storing executable instructions of the first processor; The first processor is configured to execute the radar detection method based on multi-threshold filtering and spatiotemporal cross-validation according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the second processor, the radar detection method based on multi-threshold filtering and spatiotemporal cross-validation according to any one of claims 1 to 6 is implemented.

Images